SPATIAL VARIATION IN ENVIRONMENTAL DRIVERS ON FISHERY YIELDS IN THE AMAZON RIVER-FLOODPLAIN by Meaghan Rupprecht B.S., Colorado State University, 2018 THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN NATURAL RESOURCES AND ENVIRONMENTAL STUDIES UNIVERSITY OF NORTHERN BRITISH COLUMBIA April 2024 © Meaghan Rupprecht, 2024 Meaghan Rupprecht UNBC April 2024 Abstract Previous research has suggested that the presence of flooded forests in Amazon River floodplains is associated with higher fish biomass and species richness and is widely recognized as important to the productivity of the fishery. However, flooded forests and other floodplain habitats of the Amazon River are not distributed homogeneously in space, making comparisons between regional fisheries based on habitat relationships difficult. My research aims to: (i) assess spatial trends of fishery catch in relation to floodplain habitat type, (ii) quantify the relationship between floodplain habitat type and multispecies fish catch in the Amazon River fishery, and (iii) quantify the relationship between floodplain habitat type and the catch (probability of being caught and quantity when caught) of commercially valuable species. To address the complexity of a spatially heterogeneous fishery, landing data from monitoring agencies in different regions of the river-floodplain fishery were integrated and associated with floodplain habitat composition using the Hess Wetland Mask (Hess et. al 2015). Generalized Additive Mixed Models (GAMM) were used to describe the non-linear relationship between fish catch, fishing effort, surrounding floodplain habitat cover, and fishing location; seasonal and long-term temporal trends in the data were also considered. The results of my models suggested a gradient of fish catch with the magnitude of fish catch declining from West to East, consistent with patterns of degradation within the environment. My results also supported the positive relationship between multispecies fish catch and presence of surrounding floodplain forests, while species-specific catch varied with habitats related to their feeding habits. Whereas previous research primarily focuses on regional fishing activities of the Upper Amazon, Central Amazon, and Lower Amazon, this research provides a comprehensive assessment of available fisheries data to address basin-wide variation in fish catch. The results of this research emphasize the importance of ecosystem-level ii Meaghan Rupprecht UNBC April 2024 management and further demonstrate that floodplain forests are vital habitats for one of the most productive fisheries in the world. iii Meaghan Rupprecht UNBC April 2024 TABLE OF CONTENTS Abstract ........................................................................................................................................... ii Table of Contents ........................................................................................................................... iv Table of Figures ............................................................................................................................. vi List of Tables ................................................................................................................................. xi Acknowledgements ...................................................................................................................... xiv 1. Introduction ........................................................................................................................... 1 2. Methods .................................................................................................................................. 8 2.1. Overview of Study Extents .............................................................................................. 8 2.2. Sources of Data ............................................................................................................. 12 2.2.1. Floodplain Land Cover Data ....................................................................................... 12 2.2.2. Fisheries Data.............................................................................................................. 12 2.3. Dataset Integration and Management .......................................................................... 15 2.4. Statistical Analyses ....................................................................................................... 16 2.4.1. Variable Preparation and Selection ............................................................................. 16 2.4.2. Multispecies Catch Model Structure ........................................................................... 17 2.4.3. Species-Specific Catch Model Structures ................................................................... 18 2.4.4. Model Selection .......................................................................................................... 20 2.4.5. Model-Averaged Predictions ...................................................................................... 20 3. Results .................................................................................................................................. 21 3.1. Multispecies Catch ........................................................................................................ 21 3.1.1. Multispecies Catch Model Selection .......................................................................... 22 3.1.2. Partial Effects .............................................................................................................. 23 3.1.3. Spatial Variation in Multispecies Catch ..................................................................... 25 3.2. Species-Specific Catch .................................................................................................. 28 3.2.1. Aracu ........................................................................................................................... 28 iv Meaghan Rupprecht UNBC April 2024 3.2.2. Curimatã...................................................................................................................... 39 3.2.3. Mapará ........................................................................................................................ 49 3.2.4. Pescada ........................................................................................................................ 60 3.2.5. Tambaqui .................................................................................................................... 70 4. 5. Discussion............................................................................................................................. 80 4.1. Multispecies Catch ........................................................................................................ 80 4.2. Species-Specific Catch .................................................................................................. 85 References ............................................................................................................................ 95 APPENDIX A- BOAT ID SENSITIVITY ANALYSIS .............................................................. 99 APPENDIX B- MODELS .......................................................................................................... 100 APPENDIX C- FISH CATCH COMPOSITION ....................................................................... 101 APPENDIX D- SPECIES LOOKUP TABLE ............................................................................ 103 v Meaghan Rupprecht UNBC April 2024 Table of Figures Figure 1. Landing data was collected from 11 municipalities within this study which represent fishing activity in regions of the Upper, Middle, and Lower Amazon. Fishing locations from the data were found in 12 river sub-basins (Level 4) as defined by Venticinque et al. (2016). ......... 10 Figure 2. The municipalities represented in the fishing data are displayed in relation to their region and its surrounding habitat. The green represents floodplain forest, blue represents water, orange represents bare ground or herbaceous vegetation, pink represents aquatic macrophytes, and yellow represents shrub. Extent of these regions are defined in Section 2.2.1. ..................... 11 Figure 3. Log-transformed values of multispecies fish catch observations show that the Lower Amazon had the densest fishing activity (as indicated by the opacity of observations) and fishing trips typically brought in lower magnitudes of fish catch within that region. .............................. 22 Figure 4. Partial effects of multispecies fish catch showed an increase in catch over time. Habitat partial effects show a negative relationship between multispecies fish catch and the proportion of shrub and a positive (before leveling) relationship between multispecies catch and the proportion of forest in the sub-basin where fishing took place. ..................................................................... 24 Figure 5. Random effect plots show that Basin 8 demonstrated the highest catches of the subbasins considered. Catch of fishing trips landing in Manaus were also the highest amongst the landing municipalities where data had been collected. ................................................................. 25 Figure 6. Model-averaged predictions for the magnitude of multispecies fish catch showed higher magnitudes of multispecies fish catch in the Upper Amazon relative to the Lower Amazon. These predictions were consistent with patterns displayed in the plots of observed fish catch (Figure 3). ............................................................................................................................ 26 Figure 7. Predictive maps of multispecies fish catch in the Amazon Basin based on variations in seasonality and effort show increased magnitudes of catch with effort, and higher magnitudes of catch during dry seasons compared to wet seasons. ..................................................................... 27 Figure 8. Partial effect plots show several relationships with habitats based on the top models. Positive relationships are shown between the probability that aracu are successfully caught during a trip and aquatic macrophytes, bare/herbaceous, open water, shrub, and population density variables. A negative relationship is seen between the probability of catch and forest cover.............................................................................................................................................. 30 Figure 9. Random effect plots show minimal variation in the probability of successfully capturing aracu between the sub-basins considered. Probability of successfully capturing aracu was highest in fishing trips landing in Santarém and Manaus, respectively.Figure 10. Predicted vi Meaghan Rupprecht UNBC April 2024 probability of successfully catching aracu during a fishing trip was generally low and minimal spatial variation was observed. ..................................................................................................... 31 Figure 11. Partial effect plots are shown in log-scale. Effects showed variable relationships between the quantity of aracu catch and different habitat types; only bare ground/herbaceous habitat appeared to have a positive relationship with the quantity of catch, although aquatic macrophytes had a positive relationship until there were higher proportions of aquatic macrophytes. ................................................................................................................................. 35 Figure 12. Quantity of aracu catch was highest from fishing trips landing in Manaus, and those which conducted fishing in Basin 8. Maps of partial effects for random factors show little variation by sub-basin, and the highest catches in the Central Amazon municipality of Manaus. ....................................................................................................................................................... 36 Figure 13. If catch was successful, the quantity of aracu catch (kg) appeared higher in southern reaches of the study area, primarily in the sub-basins associated with the Purus and Madeira Rivers. A small area along the mainstem of the Amazon River showed higher aracu catch (kg) between its confluences with the Purus River and the Negro River. ............................................ 37 Figure 14. The quantity of aracu catch was generally higher in the dry season compared to the wet season. Regardless of season, predicted quantities of catch increased with increasing effort. ....................................................................................................................................................... 38 Figure 15. Partial effect plots showed variable relationships between the probability of catching curimatã and different habitat types. Slightly positive relationships were observed with forest cover and aquatic macrophytes; a more pronounced positive relationship was shown with population density. Slightly negative relationships are shown with bare ground/herbaceous cover, open water, and to a lesser extent, shrubs. .................................................................................... 41 Figure 16. Partial effects for random factors show the highest probability of catch taking place in Basin 86, close to Manaus, with the lowest catches in the adjacent Basin 97. The probability of catch was generally higher in mainstem Amazon sub-basins. The highest probability of catch by municipality occurred in Alenquer, and the lowest probability of catch occurred in Obidos. ..... 42 Figure 17. The probability of successfully catching curimatã was highest in the southwest reaches of the study area. The Central and parts of the Lower Amazon, between the confluences of the Amazon River with the Negro and Tapajos Rivers had the lowest predicted probability of catch. Probability of catch increased downstream of the Amazon River confluence with the Tapajos River. ............................................................................................................................... 43 Figure 18. Partial effect plots showed a non-linear relationship between the quantity of curimatã catch and fishing effort. Partial effects of month showed seasonality in fish catch that peaked vii Meaghan Rupprecht UNBC April 2024 around October and were lowest around June. A negative relationship is observed between curimatã catch and bare ground/herbaceous cover. ...................................................................... 45 Figure 19. Partial effects for random factors show little variation in most sub-basis but the lowest catches were observed in Basin 114 where the Purus River is located and the highest were observed in Basin 86, northeast of Manaus. By municipality, catches in Manaus were higher relative to any other municipality. ................................................................................................ 46 Figure 20. Quantity of curimatã catch varied by basin; higher catches were observed in subbasins of tributaries of the Amazon River when compared to the mainstem sub-basins of the Amazon River. .............................................................................................................................. 47 Figure 21. The quantity of curimatã catch varied based on seasonality and effort quartiles; catch was generally higher in the dry season compared to the wet season and increased with increasing effort. ............................................................................................................................................. 48 Figure 22. Partial effects with the probability of catching mapará showed a non-linear relationship with effort. Probability of catch had generally been increasing between 1990 and 2011. The probability of mapará catch varied seasonally. Partial effect plots showed positive relationships between the probability of catching mapará and all habitat types except open water, which had a negative relationship. ................................................................................................ 51 Figure 23. Partial effects for random factors showed minimal variation in the probability of catch between basins, although slightly higher in Basin 6, a mainstem sub-basin including the Lower Amazon. The highest probability of catch by municipality occurred in Obidos, and the lowest probability of catch occurred in Manaus. ..................................................................................... 52 Figure 24. The probabilities of successfully catching mapará were highest in the Lower Amazon above its confluence with the Tapajos River, near Obidos. Predicted probability of catch was minimal in the Central and Upper Amazon regions. .................................................................... 53 Figure 25. Partial effect plots showed a non-linear relationship between the quantity of mapará catch and fishing effort. Partial effects of month did not show seasonality in catch. All habitat variables and population density showed slightly positive relationships with catch; only forest cover had a negative relationship with the quantity of mapará catch. .......................................... 56 Figure 26. Partial effects for random factors show little variation in most sub-basins. By municipality, catches in Obidos were higher relative to any other municipality, and lowest in Oriximina. ..................................................................................................................................... 57 Figure 27. Predicted mapará catches were highest in the northern sub-basin of the Lower Amazon. High catch was also predicted in an area along the Japura River. The lowest predicted viii Meaghan Rupprecht UNBC April 2024 quantity of catch occurred in the southern Tapajós River basin, although catch in the Central Amazon was also less than surrounding areas. ............................................................................. 58 Figure 28. The distribution of mapará catch was predicted based on variations in season and effort. Catch of mapará was generally the same across seasons but catch increased with increasing effect. ........................................................................................................................... 59 Figure 29. Partial effects with the probability of catching pescada showed a non-linear relationship with effort which dropped substantially at higher levels of effort. The probability of catch fluctuated over time but generally increased between 1990 and 2011. Probability also varied seasonally with higher catches around December and January and lower catches between May and September, except for a secondary peak in probability in July. Partial effect plots showed positive relationships between the probability of catching pescada and all habitat types except forest, which had a slightly negative relationship. Probability of catch was also positively related to population density. ........................................................................................................ 62 Figure 30. Partial effects for random factors showed some variation in the probability of catch between basins, with higher probability in Basin 114 which is associated with the Purus River. The highest probability of catch by municipality occurred in Oriximina, and the lowest probability of catch occurred in Tefé. ........................................................................................... 63 Figure 31. The probability of catching pescada was around 10% for most of the study area, with little spatial variation observed. .................................................................................................... 64 Figure 32. The quantity of pescada catch during fishing trips had a non-linear relationship with effort. Slight seasonal variation in catch was also observed. In relation to habitat, the quantity of pescada catch had a slightly positive relationship with shrub cover and slightly negative relationship with forest cover........................................................................................................ 66 Figure 33. Partial effects for random factors show little variation in most sub-basins. By municipality, catches in Santarém and Manaus were higher relative to any other municipality, and lowest in Alenquer. ................................................................................................................ 67 Figure 34. Higher catches of pescada were predicted in the southwestern parts of the study area, further from the mainstem of the Amazon River. The lowest catches of pescada were predicted in the Lower Amazon and on the Tapajós River closer to its confluence with the Amazon River. . 68 Figure 35. Catch of pescada was generally higher in the dry season compared to the wet season; catch generally increased with increasing effort. .......................................................................... 69 Figure 36. Partial effects with the probability of catching tambaqui showed a non-linear relationship with effort which initially increased before leveling off. The probability of catch declined between 1990 and 2011. The probability also varied seasonally with higher catches ix Meaghan Rupprecht UNBC April 2024 around April and lower catches between December and January. Partial effect plots showed negative relationships between the probability of catching tambaqui and all habitat types included in the top models except forest, which had a positive relationship. ............................... 72 Figure 37. Partial effects for random factors showed some variation in the probability of catch between basins, with higher probability in Basin 86 which near Manaus. The highest probability of catch by municipality occurred in similarly in Parintins, Oriximina, and Alenquer; the lowest probability of catch occurred in Obidos........................................................................................ 73 Figure 38. The highest probability of catch for tambaqui was located in the western sub-basins of the Upper Amazon. The probability of catch declined in eastern reaches of the mainstem Amazon River and associated tributaries such as the Madeira and Tapajós Rivers. Other areas of higher probability included a sub-basin outside of Manaus and downstream of the confluence between the Amazon and Tapajós Rivers. .................................................................................... 74 Figure 39. The quantity of tambaqui catch during fishing trips had a non-linear relationship with effort. Seasonal variation in catch was also observed in the quantity of catch. In relation to habitat, the quantity of tambaqui catch had a slightly positive relationship with bare ground/herbaceous cover and forest cover, but a slightly negative relationship with open water cover.............................................................................................................................................. 76 Figure 40. Partial effects for random factors show little variation in most sub-basins. By municipality, catches Manaus were substantially higher relative to any other municipality, and lowest in Monte Alegre. ................................................................................................................ 77 Figure 41. Higher catches of tambaqui were predicted in the western parts of the study area and declined to the east along a gradient. The lowest catches of tambaqui were predicted in the Lower Amazon and on the Tapajós River. ................................................................................... 78 Figure 42. Catch of tambaqui was generally higher in the dry season compared to the wet season; catch generally increased with increasing effort. .......................................................................... 79 x Meaghan Rupprecht UNBC April 2024 List of Tables Table 1. Definition of habitat types in accordance with Hess et al. (2015) descriptions and classifications (e.e. L. Hess, University of California, Santa Barbara, pers. comms.). ................ 12 Table 2. The top 15 species caught by weight during fishing trips throughout the Amazon Basin cumulatively represent approximately 91% of the total fish caught within the dataset. Species were identified to be migratory or not migratory, and their associated trophic guilds were noted (Goulding et al. 2019; Amazon Waters Alliance). ....................................................................... 14 Table 3. Species groups selected for further analysis. The scientific names of all species in the dataset which belong to that group, their feeding strategy, and the number of records available from the dataset are described....................................................................................................... 15 Table 4. Correlation Matrix of sub-basin variables. Landscape variables from sub-basins of interest were compared using a correlation matrix to determine the relationship between other variables in the dataset. Significance of correlation is denoted. ................................................... 16 Table 5. AICc summary statistics for models retained in the 95% confidence set for the best model of multispecies catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .............. 23 Table 6. AICc summary statistics for models retained in the 95% confidence set for the best model for the probability of aracu catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................................................................................................................................... 28 Table 7. AICc summary statistics for models retained in the 95% confidence set for the best model of aracu catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................... 33 xi Meaghan Rupprecht UNBC April 2024 Table 8. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of curimatã catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................................................................................................................................... 39 Table 9. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of mapará catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................................................................................................................................... 49 Table 10. AICc summary statistics for models retained in the 95% confidence set for the best model of mapará catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................... 54 Table 11. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of pescada catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................................................................................................................................... 60 Table 12. AICc summary statistics for models retained in the 95% confidence set for the best model of pescada catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate xii Meaghan Rupprecht UNBC April 2024 set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................... 65 Table 13. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of tambaqui catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................................................................................................................................... 70 Table 14. AICc summary statistics for models retained in the 95% confidence set for the best model of tambaqui catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). .......................... 75 Table 16. Variance of the Boat ID random factor and residuals are shown according to different methods used to create unique Boat IDs when fitting a global model for fish catch. The Boat ID factor captured the most variance and minimized variance in the residuals best when a unique combination of boat name, length, and landing municipality were applied to create the Boat ID. ....................................................................................................................................................... 99 Table 19. Fifteen models were fit to fish catch data using a combination of variables to assess which model best fit out-of sample data. Fixed effects included s(e) as the smooth for effort, s(m) s the smooth for month, s(y) as the smooth for year, s(h) as the smooth for the landscape variable assessed (either habitat proportion or population density), and s(X,Y) as the spatial smooth for the approximate coordinates of fishing location. All models were fit with the same random effects for unique boat ID, unique basin ID, and the landing municipality of catch. The landscape variable of interest is described for each model as well. ............................................................ 100 Table 20. Fish catch by species indicated the presence of 41 species groups within the data. The highest proportion of fish catch between 1991 and 2011 belonged to mapará, which accounted for a quarter of all fish catch by species. .................................................................................... 101 Table 21. The lookup table for species used Species as the grouping variable. The names of species recorded in all datasets (i.e., ProVarzea, IARA, Manaus (Vandick), Manaus (Bayley), and Mamiraua) were then classified according to the species group they belonged to. These classifications were used to standardize records of fish catch for further analysis. ................... 103 xiii Meaghan Rupprecht UNBC April 2024 Acknowledgements I would like to thank my advisor, Dr. Eduardo Martins, and co-advisor, Dr. Leandro Castello, for the opportunity to take on this project. Your valuable support and guidance during this study has helped me learn so much about the processes of research and scientific writing. To my committee member, Dr. Oscar Venter, thank you for your valuable input and time during this process. I would also like to thank many of the researchers working with the SABERES Project, including the fellow student on this project, Gabriel Borba, who met with me throughout my research and gave me direction and advice on many of the nuances present within our data. This research was enabled in part by support provided by Simon Fraser University’s Cedar system (cedar.alliancecan.ca) and the Digital Research Alliance of Canada (alliancecan.ca). I am also grateful to the GIS department at UNBC, Dr. Anthony Jjumba and Matt McClean, who met with me many times as I was getting the GIS portion of my research underway. Your advice and counseling set me up to accomplish many of the tasks during the duration of my research. To the members of the Freshwater Fish Ecology Lab at UNBC, you have all been an incredible support system during this experience. I am so grateful to have had fellow students to learn from and troubleshoot with. Finally, to my friends and family, who never let me forget that I could do this. I appreciate you always. xiv Meaghan Rupprecht 1. UNBC April 2024 Introduction The Amazon River is the world’s largest river by volume and contributes approximately 17% of the world’s freshwater discharge into the oceans (Dai & Trenberth 2002). The Amazon and its tributaries comprise large expanses of wetlands throughout the basin with conservative estimates of approximately 8.4x105 km2, or 14% of the Amazon Basin, being classified as wetland habitat (Hess et al. 2015). Within these wetlands exists a dynamic river-floodplain ecosystem influenced by seasonal flooding (i.e., flood pulses), which is one of the main ecological processes that drive fish production and migratory behaviors (Junk et al. 2007; Ruffino 2014). This seasonal flooding connects the mainstem of rivers to surrounding vegetation and periodically isolated lakes (Junk et al. 1989; Hess et al. 2015). Such periods of inundation integrate the aquatic and terrestrial interfaces, allowing fishes to migrate laterally out of rivers and lakes into the newly available floodplain environments, gaining access to important feeding opportunities, nursery and refuge habitats, and breeding grounds (Goulding 1980). The abundance of fish supported by floodplains is exploited by extensive commercial and subsistence fisheries, which are a major source of protein and employment to the people of the Amazon River Basin (Isaac & Almeida 2011; Ruffino 2014). The people which rely upon the fishery for their livelihoods are one of the primary sources of fisheries data within the Amazon River, whereby fishers are interviewed upon landing catches at major ports (IARA/IBAMA Project; Natural Resources of the Várzea Management Project ProVárzea/IBAMA). The floodplain forests provide particularly important sources of allochthonous inputs such as fruit, seeds, and leaves of trees, which are major components in the seasonal diets of fish species living within the river-floodplain ecosystem (Forsberg et al. 1993; Oliveira et al. 2006). Previous research has suggested that inputs of plant material from flooded forests and aquatic macrophytes are highly important carbon sources for fish of the Amazon River within floodplain 1 Meaghan Rupprecht UNBC April 2024 lakes of the Central Amazon (Oliveira et al. 2006). Furthermore, dietary preferences of some omnivores and frugivores suggest that fruits and seeds from the flooded forest – some of the more nutritious food items available in those habitats – are important to seasonal diets during periods of high water in floodplain lakes (Oliveira et al. 2006; Forsberg et al. 1993). Even piscivorous fish, which predominately consume fish throughout the year, display carbon signatures of plant material based on preferred prey species associations with the vegetation (Oliveira et al. 2006). In studies considering larger floodplain areas, research indicates that most of the carbon assimilated by commercially important species comes from sources such as phytoplankton, tree fruit and seeds, aquatic macrophytes, and periphyton (Forsberg et al. 1993; Benedito-Cecilio et al. 2002). The results of these studies point to floodplain forest cover being integral to Amazon fishes and the commercial and subsistence fisheries which rely upon them. Indeed, several studies have shown that floodplain habitats are vital to fish catch. Most notably, floodplain forests have been consistently identified as one of the floodplain habitats supporting these multispecies fisheries (Arantes et al. 2019; Castello et al. 2018; Goulding et al. 2019; Barros et al. 2020). Assessments of fish-habitat relationships based on such data have found that forest cover in Lower Amazon lake systems is positively associated with multispecies catch per unit effort (CPUE); not only was there a positive relationship between the two, but forest cover had the greatest influence on multispecies CPUE amongst all variables considered, likely due to increased biomass being available for harvest in areas surrounded by forest cover (Castello et al. 2018). Another study utilizing such data in the Lower Amazon predicted that loss of forested areas surrounding lakes would result in declines of fish catch; this effect depended on the size of the lake system as well, where smaller lake systems with less forest cover would 2 Meaghan Rupprecht UNBC April 2024 experience a greater decline in fish catch if the valuable habitat was removed (Barros et al. 2020). Prior studies have also suggested that species-specific catch responds to floodplain habitats, although responses varied based on functional traits or diets of the species. An assessment of commercial fish catch in relation to surrounding habitat types indicated that CPUE from nine of the 10 commercial species in the study were also positively related to forest cover in the Lower Amazon, suggesting that the relationship between fish and floodplain forests is not an isolated phenomenon amongst species (Castello et al. 2018). Other studies have also found that species which relied upon resources in forests (e.g., herbivorous, detritivorous, and invertivorous fishes) exhibited strong associations with forest cover (Arantes et al. 2017). However, certain groups with preference for prey items which are abundant in open water, such as planktivores and piscivores, had negative associations with forest cover (Arantes et al. 2018). Considering one of the primary drivers of freshwater degradation in the Amazon is landuse change and deforestation, a key question is how the fishery is vulnerable to land-use changes (Castello & Macedo 2016; Castello et al. 2013). While some commercially targeted species may be adaptable and resilient to land use change, others with traits strongly associated with certain habitats may not be able to adapt as well to a changing aquatic environment (Arantes et al. 2019). Generalist species appear to be negatively related to forest cover and instead prefer areas of less forest and more herbaceous vegetation cover; these species would likely not be as vulnerable to land-use changes compared to feeding strategists which rely on terrestrial inputs (such as detritivores, herbivores, invertivores, and omnivores), and have a positive relationship with the flooded forest (Arantes et al. 2017). The consequences of this could be a loss of biodiversity within the fishery and the loss of favored commercial species, wherein the loss of 3 Meaghan Rupprecht UNBC April 2024 habitats which support specialist species such as frugivores results in replacement by generalist species which are more adaptable to the degraded environment (Arantes et al. 2017). Assessing the relationship between fish catch and surrounding habitat compositions could therefore identify species at risk of further decline and improve the understanding of how species composition within the multispecies fishery may shift in response to land-use changes. Despite the role of land-use change in freshwater degradation, efforts to quantify responses of the fishery to the loss of floodplain forests has historically been limited by the lack of widespread and contiguous land-use data. More recently, satellite imagery and historical imagery analyses have both identified higher proportions of floodplain forest in western reaches of the Amazon River compared to eastern reaches (i.e., the Lower Amazon) (Hess et al. 2015; Renó & Novo 2019). While some of these differences are natural, they also are attributed to human activity such as jute and cattle farming which required the clearing of native vegetation (Renó et al. 2011). Between the 1970’s and 2010’s the Upper, Middle, and Lower Amazon landscapes have lost 1.7%, 29%, and 70% of floodplain forest, respectively (Renó & Novo 2019). The resulting patterns of forest depletion show western-most regions of the basin had less overall deforestation and thus now possess better habitat integrity, while eastern landscapes such as the Lower Amazon have been highly impacted by anthropization and land-use change (Renó & Novo, 2019). This gradient of forest depletion throughout floodplains of the Amazon River has not been incorporated into assessments conducted at smaller regional scales, but it may be informative to the consequences of deforestation on fish catch at a large scale (Renó & Novo 2019). Currently, many of the studies on the relationship between floodplain habitat and fish catch have been focused on regional geographical scales within the Lower Amazon (Arantes et 4 Meaghan Rupprecht UNBC April 2024 al. 2017; Arantes et al. 2019; Barros et al. 2020; Castello et al. 2018; Lobón-Cerviá et al. 2015). Studies within the Lower Amazon floodplain suggest that floodplain forests are an important habitat within a degraded landscape. However, within a larger spatial framework that encompasses a myriad of habitat compositions and stages of degradation, uncertainty exists surrounding the relationships between habitat and fish catch. One such study which has attempted to address the spatial limitation of previous research expanded its analyses to include the floodplain lakes of the Central Amazon; results revealed that shrubs seemed to be the primary habitat type describing trends in fish catch at a larger spatial scale (Pereira et al. 2023). Yet despite expanding analyses into additional regions, this study was limited to fewer years of data compared to those which have been conducted at smaller scales; the temporal limitations of the data thus limit our understanding of historic trends and relationships between fish catch and habitats at larger spatial scales (Pereira et al. 2023). Between the temporal and spatial limitations of available data, existing studies have yet to compare historic fish catch in pristine landscapes compared to highly degraded landscapes, and little is known about the large-scale distribution of multispecies or species-specific catch over time. Consequently, considerations of cumulative impacts of land-use change and relative comparisons between regional fisheries in the Amazon are left neglected, although such information is key to any attempts at ecosystem-level management and policy. Additionally, uncertainty exists due to how fisheries data has been assessed. Previous research has been conducted primarily using linear modeling approaches, which is restricted in its ability to account for non-linear relationships within data and variables, an important aspect of assessing the effects of effort within multispecies fisheries (Lorenzen et al. 2006). Previous research on spatial variation has also been conducted on different spatial scales (i.e., lake 5 Meaghan Rupprecht UNBC April 2024 systems, buffer areas surrounding ports, etc.) and typically without the consideration of human population density’s effects on fish catch. The various spatial units and scales used in previous studies have not incorporated the coordinates of fishing location in the analysis, which addresses issues such as spatial autocorrelation within the records of fish catch, nor have these results been used to generate spatially informed predictions of fish catch. Additionally, previous studies have primarily relied upon artificially generated buffer zones, municipal boundaries, or small-scale lake systems that do not necessarily reflect the habitat composition of the areas where fishing took place. However, the use of habitat data for the area or river basin of fishing is necessary to improve our understanding of habitat-fishery yield relationships, as they better represent the hydrological components of the ecosystem and are more ecologically related to fisheries than methods such as municipal boundaries or artificial buffers. The purpose of this research was to assess the relationship between floodplain habitat or population density and fish catch throughout the Amazon Basin within the context of the existing landscape gradient. The primary objectives of this research were to quantify the relationships between floodplain habitats and multispecies catch as well as the relationships between floodplain habitats and species-specific catch of five commercially important species groups. These species groups were selected based on their contributions to multispecies catch, their associated feeding strategies, and the availability of data. A longitudinal analysis was conducted to assess spatial variations in respective fish catch within the context of existing landscape gradients by analyzing fish landing records from 11 municipalities distributed along an extent of approximately 2,000 km of the Amazon River. Using a space-for-time substitution, large-scale floodplain habitat data were used in lieu of before-and-after land-use data to assess the 6 Meaghan Rupprecht UNBC April 2024 differences in fish catch amongst landscapes at varying stages of deforestation and habitat composition as described by Renó & Novo (2019) (Lovell et al. 2023). I addressed three hypotheses: first, that higher multispecies fish catch would be observed in regions where flooded forest remains abundant relative to regions which have experienced extensive deforestation, such as the Lower Amazon. Second, that the quantity and probability of species-specific fish catch would vary depending on habitat types associated with the food resources of respective feeding strategies. Finally, that the spatial distribution of species-specific fish catch would vary based on floodplain habitat composition throughout the Amazon Basin. The results will inform conservation efforts within the Amazon River fishery and identify species that are more vulnerable to land-use changes within the Amazon River floodplain. 7 Meaghan Rupprecht UNBC 2. Methods 2.1. Overview of Study Extents The data used in this study were compiled from fisheries monitoring programs April 2024 throughout the Amazon River floodplain to represent the diverse spatial scale of commercial fisheries. Three major fishing regions are represented within the data: the Upper Amazon, Middle Amazon, and Lower Amazon (Figure 1). The Upper Amazon was defined by areas of the Solimões River (as the Amazon River is known in Brazil before joining the Negro River), above the confluence with the Negro River. The Middle Amazon was defined as the region just below the confluence of the Solimões and Negro, primarily the region surrounding the city of Manaus. The Lower Amazon was defined consistent with Renó et al. (2011), which refers to about 600 km of the Amazon River between the state boundaries of Pará and Amazonas in Brazil, to the confluence of the Amazon River with the Xingu River. Fish landing records were obtained from a total of 11 landing municipalities throughout these regions. Fishing records from these municipalities were associated with sub-basins of the Amazon River drainage basins where fishing took place based on approximate coordinates of fishing location using the SNAPP River Basin Framework, developed by Venticinque et al. (2016). This framework prioritizes natural spatial units within the Amazon aquatic ecosystem and provides several levels of river basin classifications dependent on the size of the basin and its tributaries (Venticinque et al. 2016). Unlike other basin classification systems such as the Pfafstetter basin coding system, which is often used by government management agencies, the SNAPP River Basin Framework recognized distinct spatial units within the mainstem Amazon River, which was an integral part of assessing the Amazon River fishery and floodplain habitat types (Goulding et al. 2019; Venticinque et al. 2016). 8 Meaghan Rupprecht UNBC April 2024 For the purposes of my research, Basin Level 4 was selected as the appropriate level. This layer distinguishes the mainstem of the Amazon River into sub-basins and recognizes them as distinct aquatic and ecological units, while also maintaining river basin sizes that incorporate multiple fishing locations (Venticinque et al. 2016). These sub-basins were between 10,000 km2 and 100,000 km2 and their associated floodplains, including sub-basins of the mainstem Amazon River (Basin Level 4; Venticinque et al. 2016). Of the 198 Level 4 River Basins in the Legal Amazon, multispecies fish catch records from this study were observed within 12 basins which collectively represent approximately 250,000 km2 of the Amazon Basin. Records of fish catch by species were limited to as few as five sub-basins or observed in as many as 11 sub-basins. The extent of floodplain habitat within the study was defined by the Amazon Wetland mask which was developed by Hess et al. (2015). Sub-basins from the previously described framework were used to determine the proportions of floodplain habitat present near fishing locations. 9 UNBC April 2024 10 Figure 1. Landing data was collected from 11 municipalities within this study which represent fishing activity in regions of the Upper, Middle, and Lower Amazon. Fishing locations from the data were found in 12 river sub-basins (Level 4) as defined by Venticinque et al. (2016). Meaghan Rupprecht UNBC April 2024 11 Figure 2. The municipalities represented in the fishing data are displayed in relation to their region and its surrounding habitat. The green represents floodplain forest, blue represents water, orange represents bare ground or herbaceous vegetation, pink represents aquatic macrophytes, and yellow represents shrub. Extent of these regions are defined in Section 2.2.1. Meaghan Rupprecht Meaghan Rupprecht UNBC April 2024 2.2. Sources of Data 2.2.1. Floodplain Land Cover Data Wetland inundation and floodplain habitat for the full extent of the Amazon Basin were obtained from NASA’s Large-Scale Biosphere-Atmosphere Experiment (LBA-ECO), which compared habitat classes between imagery from a low-water period in 1995, before the flood pulse, and a high-water stage in 1996, during the flood pulse (Hess et al. 2015). Habitat types were grouped based on similar cover during high-water stages, except shrubs which were maintained even if the high-water class was water (Table 1). This approach was adapted from previous research conducted by Castello et al. (2018). Table 1. Definition of habitat types in accordance with Hess et al. (2015) descriptions and classifications (L. Hess, University of California, Santa Barbara, pers. comms.). Habitat Class Aquatic Macrophytes Forest Shrub Bare/Herbaceous Open Water Description Flooded herbaceous vegetation; includes woodlands of low tree density where aquatic macrophytes are likely present in high-water stage Vegetation with height >5m; canopy coverage >70% Vegetation with height < 5m (>70% cover) Areas with sparse canopy coverage or bare land (4-20% cover) Floodplain lakes or mainstem rivers consisting of open water 2.2.2. Fisheries Data Inland fisheries catch data collected from 11 landing municipalities throughout the Brazilian Amazon Basin were used in this study. Landing data from the Upper Amazon were represented by records of fishing landings in Tefé and Alvarães, State of Amazonas, between 1991 and 2022 and were provided by the Mamirauá Sustainable Development Institute. Central Amazon landing data were obtained from the Amazon Fisheries Management and Sustainability Project; these data represented records of fish landings from Manaus in Amazonas State collected from 1994 to 2004. Lower Amazon landing data collected from 1993 and 2011 were provided by the Administration of the Middle Amazon Fishery Resources (IARA Project) and 12 Meaghan Rupprecht UNBC April 2024 Brazilian Institute for Environment and Natural Resources (IBAMA), representing eight municipalities between Parintins and Almeirim, State of Pará. In each of these municipalities, fish landing records were compiled based on interviews of fishers at landing locations. Variables of interest from fisheries data included year, total quantity of catch (kg), port of landing, approximate coordinates of fishing activity, effort (fishers/day), and the type of gear used (Isaac et al. 2008). Approximate coordinates of fishing activity were identified by the monitoring programs using information provided by fishers during the interview process regarding waterbodies where fishing was conducted during their trip; coordinates were typically the center point of the waterbody which was identified or a point in the waterbody nearest to their described fishing location. Terminology used to describe gear types and species of catch were standardized according to classes of fishing gear outlined by the Food and Agriculture Organization of the United Nations (FAO) (FAO 2020). Approximately 81% of the records were caught using gillnets; because of this, and to eliminate sources of variation within the data, only data corresponding to gillnet catch were maintained for analysis. Records of catch were grouped into 44 species groups using common names, scientific names, and previous classification techniques (ProVárzea; Goulding et al. 2019; Appendix D). These species groups were aligned with market names that are used to refer to species which frequently occur at commercial markets (Goulding et al. 2019). The landing data indicated that 41 species of fish were caught in gillnets, 15 of which represented approximately 91 % of the total gillnet catch (Table 2). Based on the availability of records within the data, one species from each trophic guild was selected for further analysis from the top species; these were used to assess the potential relationship between trophic guild (i.e., food preferences) and habitat (Table 3). Catch quantities and proportions for all observed species can be found in Appendix C. 13 Meaghan Rupprecht UNBC April 2024 Table 2. The top 15 species caught by weight during fishing trips throughout the Amazon Basin cumulatively represent approximately 91% of the total fish caught within the dataset. Species were identified to be migratory or not migratory, and their associated trophic guilds were noted (Goulding et al. 2019; Amazon Waters Alliance). Species Sum of Proportion Cumulative Sum Migratory Trophic Guild Group Catch (kg) of Catch of Proportions Mapará Jaraqui Curimatã 6,146,881.7 2,888,303.7 2,013,430.4 0.25 0.118 0.082 0.25 0.3675 0.4494 Yes Yes Yes Tambaqui 1,999,679.3 0.081 0.5307 Yes Pacu 1,404,173.7 0.057 0.5878 Yes Dourada 1,221,408 0.050 0.6375 Yes Pescada 1,210,013.6 0.049 0.6867 Unknown Aruana Surubim Moela Aracu Tucunaré Sardinha Pirapitinga 968,982.9 856,712.6 750,607 673,113.6 671,582.2 617,252.6 468,678.1 0.039 0.035 0.031 0.027 0.027 0.025 0.019 0.7261 0.7609 0.7914 0.8188 0.8461 0.8712 0.8903 No Yes Yes Yes No Yes Yes Matrinchã 465,041.4 0.019 0.9092 Yes 14 Planktivore Detritivore Detritivore Frugivore/ zooplanktivore Omnivore/ frugivore Piscivore Piscivore/ crustivore Omnivore Piscivore Omnivore Omnivore Piscivore Omnivore Frugivore Omnivore/ frugivore Meaghan Rupprecht UNBC April 2024 Table 3. Species groups selected for further analysis. The scientific names of all species in the dataset which belong to that group, their feeding strategy, and the number of records available from the dataset are described. Species Group Scientific Name(s) Feeding Strategy Number of Records Mapará Hypophthalmus marginatus; H. edentates Planktivore 4,497 Curimatã Prochilodus nigricans Detritivore 4,881 Tambaqui Colossoma macropomum Frugivore/ zooplanktivore 4,748 Pescada Plagioscion spp.; Pachypops spp. Piscivore/crustivore 6,497 Aracu Anostomoides laticeps; Leporinus friderici; Leporinus trifasciatus; Schizodon fasciatum; Schizodon vittatum; Rhytiodus argenteofuscus; Rhytiodus microlepis; Leporinus affinis; Leporinus fasciatus Omnivore 3,088 2.3. Dataset Integration and Management Available fisheries datasets were standardized and characterized to identify the records which were available for analysis based on the variables of interest. Records which were missing key attributes (such as those related to fishing effort, locational information, or habitat information) were identified and removed if supplemental data to derive them were not available. Vocabulary used to identify attributes was standardized across datasets for consistency. Datasets which included geospatial information were standardized for variations such as projection, resolution, and units of area. Coordinates of fishing trips were standardized for projection differences. The data were integrated in a relational database designed in PostgreSQL (v. 14.2) with PostGIS (v. 3.1) as a plug-in to enable the storage and manipulation of data with spatial attributes. The database enabled a flexible approach to data extraction based on shared 15 Meaghan Rupprecht UNBC April 2024 characteristics between the datasets, ensured data and referential integrity, and reduced the redundancy present in the data. Approaches to database building using disparate datasets followed the methods outlined by Soranno et al. (2015). 2.4. Statistical Analyses 2.4.1. Variable Preparation and Selection The area of each floodplain habitat type within a basin (Level 4) was extracted from the dual-season land cover raster for wetlands in the Amazon Basin (Hess et al. 2015). Population data for Basin Level 4 was provided by Lopes 2021 and used to calculate population density by basin. The proportions of surrounding habitat and population density were extracted from the river basin shapefile (Venticinque et al. 2016) and associated with the corresponding fishing trip using the geographic coordinates provided in the dataset. Preliminary analyses indicated a high correlation among habitat variables and between some habitat variables and population density (Table 4). In order to assess how habitats or population density relate to fish catch, the proportion of habitats and population density variables were considered in separate models to avoid the effects of collinearity. Table 4. Correlation Matrix of sub-basin variables. Landscape variables from sub-basins of interest were compared using a correlation matrix to determine the relationship between other variables in the dataset. Significance of correlation is denoted. Variables 1 2 3 4 5 6 1 1. Aquatic Macrophyte 2. Bare soil/ 0.706* 1 Herbaceous -0.131 -0.593* 1 3. Forest -0.199 0.328 -0.934*** 1 4. Open Water -0.027 0.394 -0.806** 0.715** 1 5. Shrub 0.558 0.727** -0.297 0.064 0.276 1 6. Population Density * p-value < 0.05 ** p-value < 0.01 *** p-value < 0.001 16 Meaghan Rupprecht UNBC April 2024 Fishing boats were identified with a unique boat ID; ambiguity in how unique boats could be identified within the data was addressed in Appendix A, where a sensitivity analysis was conducted to determine the best approach. Results of the sensitivity analysis suggested that unique combinations of boat names, lengths, and their associated municipality of landing best captured variation in the data and were thus used to generate unique boat IDs for the model. 2.4.2. Multispecies Catch Model Structure The effects of landscape variables such as proportion of specific habitat, fishing location, and population density of river basin on multispecies fish catch were estimated with Generalized Additive Mixed Models (GAMM). A Gamma response with log link function was used as fish catch (kg) is a continuous and strictly non-negative random variable. The GAMM used to analyze total fish catch in the Amazon Basin followed the format: ‫ܥ‬௜,௝,௞,௟ = ߙ + ߛ௝ + ߛ௞ + ߛ௟ + ݂(݁௜ ) + ݂(݉௜ ) + ݂(‫ݕ‬௜ ) + ݂൫ℎ௜,௞ ൯ + ݂(ܺ௜ , ܻ௜ ) + ߝ௜,௝,௞,௟ γ௝ ~ N (0, σଶஓೕ ) γ௞ ~ N (0, σଶஓೖ ) γ௟ ~ N (0, σଶஓ೗ ) ߝ௜,௝,௞,௟ ~ N (0, σଶக ) Where ‫ܥ‬௜,௝,௞,௟ is the multispecies fish catch (kg) for fishing trip ݅ conducted by boat ݆ in basin ݇ whose catch was landed at municipality ݈; γ௝ , γ௞ , and γ௟ are, respectively, the random deviation from the intercept by fishing boat, river basin, and landing municipality. The functions ݂(·) denote smoothing functions which were used to fit localized splines to characterize nonlinear relationships (Pederson et al. 2019). Smoothing functions were fit for fishing effort (݁௜ ), month when fishing occurred (݉௜ ), the year of the fishing trip (‫ݕ‬௜ ), either the proportion of a habitat variable or the population density within the river basin where fishing took place (ℎ௜,௞ ), 17 Meaghan Rupprecht UNBC April 2024 and a spatial smoother in the form of an interaction between latitude and longitude of the approximate fishing location (ܺ௜ , ܻ௜ ). The term, ε୧,௝,௞,௟ is the residual. The spatial smooth incorporated observation-level locations which helped account for correlations in spatial data in accordance with approaches in Zuur et al. (2009). The smoothing functions for month (݉௜ ) and year (‫ݕ‬௜ ) were included for capturing both seasonal variation throughout the year and variation throughout the 21 years represented in the final dataset (1991-2011). Random effects (γ௝ , γ௞ , γ௟ ) and ߝ௜,௝,௞,௟ were assumed to be normally distributed with mean zero and variances, ߪఊଶೕ , ߪఊଶೖ , ߪఊଶ೗ , and σଶக , respectively. 2.4.3. Species-Specific Catch Model Structures Fish catch by species group was treated as a zero-inflated dataset; not all species of interest were successfully caught during a fishing trip, resulting in zeroes within the speciesspecific data. As such, these data were assessed using a Gamma Hurdle Model with two models for each species group of interest. The first was a binomial (Bernoulli) GAMM with a logit link function, used to assess the probability that a species group was successfully caught during a fishing trip; the second was a Gamma GAMM with a log link function, used on the non-zero (i.e., successful) catch data to assess the relationship between catch quantity (kg) and the parameters of interest. All available records (20,979 records) with complete information for fish catch by species group were used for the binomial GAMMs. The number of total available records for species-specific catch (20,979) differed from the number of records for multispecies fish catch (23,027) as some fishing records only provided total catch quantities and not species-specific catch quantities. The Gamma GAMMs used only records from trips with successful catch for the given species. Available records by species group ranged from 18 Meaghan Rupprecht UNBC April 2024 The Binomial and Gamma response models followed the format: ‫ܥ‬௜,௝,௞,௟,௠ = ߙ + ߛ௞ + ߛ௟ + ߛ௠ + ݂൫݁௜,௝ ൯ + ݂ ൫݉௜,௝ ൯ + ݂൫‫ݕ‬௜,௝ ൯ + ݂ ൫ℎ௜,௝,௟ ൯ + ݂(ܺ௜,௝ , ܻ௜,௝ ) + ߝ௜,௝,௞,௟,௠ γ௞ ~ N (0, σଶஓೖ ) γ௟ ~ N (0, σଶஓ೗ ) γ௠ ~ N (0, σଶஓ೘ ) ߝ௜,௝,௞,௟,௠ ~ N (0, σଶக ) where ‫ܥ‬௜,௝,௞,௟,௠ is the fish catch (kg) of species group ݆ for fishing trip ݅ conducted by boat ݇ in basin ݈ whose catch was landed at municipality ݉; γ௞ , γ௟ , and γ௠ are, respectively, the random deviation from the intercept by fishing boat, river basin, and landing municipality. The functions ݂ (·) denote smoothing functions which were used to fit localized splined to characterize nonlinear relationships (Pederson et al. 2019). Smoothing functions were fit for fishing effort (݁௜,௝ ), month when fishing occurred (݉௜,௝ ), the year of the fishing trip (‫ݕ‬௜,௝ ), either the proportion of a habitat variable or the population density within the river basin where fishing took place (ℎ௜,௝ ), and a spatial smoother in the form of an interaction between latitude and longitude of the approximate fishing coordinates (ܺ௜,௝ , ܻ௜,௝ ). The term ε୧,௝,௞,௟,௠ is the residual. The spatial smooth incorporated observation-level locations which helped account for correlations in spatial data in accordance with approaches in Zuur et al. (2009). The functions for month (݉௜,௝ ) and year (‫ݕ‬௜,௝ ) were included for capturing both seasonal variation throughout the year and variation throughout the 21 years represented in the final dataset (1991-2011). Random effects (γ௞ , γ௟ , γ௠ ) and ߝ௜,௝,௞,௟,௠ were assumed to be normally distributed with mean zero and variance σଶஓೖ , σଶஓ೗ , σଶஓ೘ and σଶக , respectively. 19 Meaghan Rupprecht UNBC April 2024 2.4.4. Model Selection For each model structure, a total of 15 models representing all possible combinations of habitats and smoother functions were fitted to the data (Appendix B). Fishing effort was maintained as a variable in every model that was fitted to the data. The fit of the global model was assessed using residual plots to determine if remaining patterns were present within the residuals (Zuur et al. 2010). Model selection was conducted using the bias-corrected Akaike Information Criterion (AICc), which is an “information-theoretic” approach used to identify the model(s) in a candidate set which provided best out-of-sample predictive accuracy (Burham and Anderson 2002; McElreath 2020). Models with the lowest AICc value indicated the most parsimonious model with regards to out-of-sample predictive accuracy, while AIC weights were calculated and used to represent the probability of the model being the most parsimonious in the candidate set (Burnham et al. 2010). In the case where no candidate model was noticeably better (AICc weight>0.95), AICc weights were used to create a 95% confidence for the best models (Johnson & Omland 2004). 2.4.5. Model-Averaged Predictions To account for model selection uncertainty, multimodel inference was used to generate model-averaged spatial predictions in the Amazon Basin (Burnham et al. 2010). Model-averaged predictions were generated for the quantity of multispecies fish catch, the probability that each species was successfully caught, and the quantity of catch for each species when successfully caught; these predictions were subsequently plotted on a predictive surface for visualization. Predictions were also made for each effort quartile and season (i.e., dry or wet season) to assess how seasonal trends and effort contributed to variation in total fish catch along the spatial 20 Meaghan Rupprecht UNBC April 2024 gradient. Model-averaged partial effects were calculated and used to demonstrate how fish catch responded to each variable. Statistical analyses on the available data were conducted in R (v. 4.4.1). Models were fit using the gamm4 package (Wood & Scheipl 2020) and selected using the package MuMin (Barton 2023). Spatial analyses and visualization of predictions were conducted using the sf, terra, exactextractr, and tmap packages (Baston 2022; Hijmans 2023; Pebesma & Biyand 2023; Pebesma 2018; Tennekes 2018). Results of modeling were visualized using GIS applications in R with the above packages to visually demonstrate spatial variation in fish catch. 3. Results 3.1. Multispecies Catch After filtering data for complete records, the final dataset comprised of 23,027 records for total fish catch collected between 1991 and 2011. Observed multispecies fish catch (kg) demonstrated lower magnitudes in the Lower Amazon compared to many other parts of the study area; this area of low catches was also where records of fishing trips were the densest (Figure 3). Most of the recorded fishing trips took place along the mainstem of the Amazon River, but major tributaries such as the Purus and Madeira River were also used for fishing trips in the Central and Upper Amazon regions (Figure 3). Unlike other tributaries, fishing trips along the Tapajos River were not recorded beyond the area closest to its confluence with the Amazon River, despite its proximity to the heavily fished areas of the lower Amazon River (Figure 3). 21 Meaghan Rupprecht UNBC April 2024 Figure 3. Log-transformed values of multispecies fish catch observations show that the Lower Amazon had the densest fishing activity (as indicated by the opacity of observations) and fishing trips typically brought in lower magnitudes of fish catch within that region. 3.1.1. Multispecies Catch Model Selection My results showed that multispecies catch had a positive relationship with forest cover, and it was negatively related to shrub cover within the basin where fishing took place; the approximate fishing location was an important factor in assessing multispecies fish catch. Of the 15 models for multispecies catch considered, three models were retained in the 95% confidence set for the best model of catch (Table 5). All top models for multispecies fish catch included coordinates of fishing location; however, the second ranked model included the proportion of shrub and coordinates of fishing location, and the final model in the 95% confidence set included the proportion of forest and coordinates of fishing location (Table 5). All other models had negligible or no support from the data. 22 Meaghan Rupprecht UNBC April 2024 3.1.2. Partial Effects Model-averaged partial effects of habitat variables indicated a positive relationship with forest cover and a negative relationship with the proportion of shrub coverage present within the sub-basin of fishing (Figure 4). An assessment of correlation between habitat variables indicated that shrub and forest habitats were strongly negatively correlated to each other (r = -0.806). Consequently, partial effects of both habitat types included in the top models support that multispecies fish catch increases in the presence of forest cover. Model-averaged partial effect plots indicated that multispecies catch had a non-linear relationship with fishing effort, where catch increased with effort before leveling out (Figure 4). Partial effects also showed that multispecies catch has generally been increasing over time; seasonal variations by month were also observed, wherein the drier months such as October demonstrated higher catches than wetter months such as June (Figure 4). A secondary increase in catch was observed around March (Figure 4). Fishing trips which took place in Basin 8, the subbasin between Manaus and Tefé, were generally those with the highest multispecies fish catch compared to fishing trips in other sub-basins (Figure 5). Fishing trips landing at Manaus also exhibited the highest catch (kg) of any municipality (Figure 5). Table 5. AICc summary statistics for models retained in the 95% confidence set for the best model of multispecies catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Model AICc Δi wi K Cumulative Weight XY shrub + XY forest + XY 319650.380 319652.509 319655.985 0.000 2.129 5.605 0.697 0.240 0.042 23 4 5 5 0.697 0.938 0.980 Meaghan Rupprecht UNBC April 2024 Figure 4. Partial effects of multispecies fish catch showed an increase in catch over time. Habitat partial effects show a negative relationship between multispecies fish catch and the proportion of shrub and a positive (before leveling) relationship between multispecies catch and the proportion of forest in the subbasin where fishing took place. 24 Meaghan Rupprecht UNBC April 2024 Figure 5. Random effect plots show that Basin 8 demonstrated the highest catches of the sub-basins considered. Catch of fishing trips landing in Manaus were also the highest amongst the landing municipalities where data had been collected. 3.1.3. Spatial Variation in Multispecies Catch Spatial predictions showed fish catch was greater in magnitude in the Upper Amazon where forests are abundant. Model-averaged predictions generated from the top models show higher magnitudes of catch near the confluence of the Japura River and Solimões River (i.e., the upper Amazon River; Figure 6). Higher magnitudes of catch were also observed in the southern reaches of basins associated with the Purus and Madeira Rivers, with the highest magnitudes of 25 Meaghan Rupprecht UNBC April 2024 catch predicted in areas which are furthest from the mainstem Amazon River (Figure 6). Lower catches were predicted in the lower Negro River and on the Solimões River just above its confluence with the Negro River; these areas are both near Manaus, which was the landing municipality where fishing trips brought in the highest magnitudes of catch according to random effect estimates (Figure 5). The lowest predicted catches occurred in the lower Amazon, near Santarém and the confluence of the Tapajos River and the Amazon River (Figure 6). Figure 6. Model-averaged predictions for the magnitude of multispecies fish catch showed higher magnitudes of multispecies fish catch in the Upper Amazon relative to the Lower Amazon. These predictions were consistent with patterns displayed in the plots of observed fish catch (Figure 3). Predictions of multispecies fish catch based on seasonality and effort quartiles indicated that catch was generally higher in the dry season compared to the wet season (Figure 7). These predictions also showed the highest fish catch throughout all basins was observed during the dry season when effort was at the 75th quartile, whereas the lowest fish catch was observed during the wet season at the 25th quartile (Figure 7). In general, fish catch increased with effort regardless of the season (Figure 7). Spatial variation in multispecies fish catch was minimal at 26 Meaghan Rupprecht UNBC April 2024 lower levels of effort, and became more pronounced with increases in effort, regardless of season (Figure 7). Figure 7. Predictive maps of multispecies fish catch in the Amazon Basin based on variations in seasonality and effort show increased magnitudes of catch with effort, and higher magnitudes of catch during dry seasons compared to wet seasons. 27 Meaghan Rupprecht 3.2. UNBC April 2024 Species-Specific Catch Results by species demonstrated unique relationships with habitats for both the probability of catch and quantity of catch, as well as unique spatial distributions. Modelling results indicated that more than one habitat and/or population density contributed to top model sets for the probability and quantity of catch for all species except for curimatã. In this case, the quantity of curimatã catch was best modelled with a single model and habitat type (bare ground/herbaceous cover). 3.2.1. Aracu 3.2.1.1. Probability of Catch Of the 15 models for species-specific catch considered, eight models were retained in the 95% confidence set for the best model of probability of catch (Table 6). The model which included bare ground/ herbaceous cover contributed the most to the top model set by weight, followed closely by aquatic macrophyte cover; other habitats and population density contributed to top model weights to a lesser extent (Table 6). Forest and shrub proportions were included twice in the top model set, both with and without fishing locations (Table 6). Table 6. AICc summary statistics for models retained in the 95% confidence set for the best model for the probability of aracu catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Aracu-Binomial Model AICc Δi wi K Cumulative Weight Bare Ground/ Herb 13266.076 0.000 0.268 4 0.2682 Aq. Macrophyte 13266.353 0.278 0.233 4 0.5017 Shrub 13267.304 1.229 0.145 4 0.6468 Open Water 13268.111 2.035 0.097 4 0.7438 Shrub + XY 13268.734 2.658 0.071 5 0.8148 Forest 13268.736 2.660 0.071 4 0.8858 Forest + XY 13269.282 3.207 0.054 5 0.9398 Pop. Density 13270.885 4.809 0.024 4 0.9640 28 Meaghan Rupprecht UNBC April 2024 Model-averaged partial effect plots of the probability aracu was successfully caught indicated a non-linear relationship with effort, with initial declines in probability before a substantial increase at higher efforts followed quickly by a steep decline (Figure 8). The probability aracu was successfully caught fluctuated over time but generally increased between 1990 and 2011 (Figure 8). The probability of aracu being successfully caught varied seasonally, with highest probabilities between May and June, and lowest in November (Figure 8). Partial effects of habitats in the top models indicated slightly positive relationships between the probability of aracu catch and several habitat types including aquatic macrophytes, bare ground or herbaceous cover, shrubs, and open water, as well as with population density (Figure 8). A slightly negative relationship was observed between the probability of catching aracu during a fishing trip and the proportion of forest in the sub-basin of fishing (Figure 8). Overall, no single habitat type or population density was indicated as having a greater effect on fish catch than any other. According to random effects, the probability of successfully catching aracu was highest in Santarém and Manaus, respectively (Figure 9). Both landing municipalities in the Upper Amazon, Tefé and Alvarães, showed lower probabilities of aracu catch relative to most other municipalities except Obidos, which had the lowest probability of aracu catch amongst municipalities (Figure 9). The probability that aracu was successfully caught varied minimally amongst basins (Figure 9). Consequently, model-averaged spatial predictions showed that the probability of catching aracu was low throughout the study area and did not display substantial spatial variation (~1%; Figure 10). 29 Meaghan Rupprecht UNBC April 2024 Figure 8. Partial effect plots show several relationships with habitats based on the top models. Positive relationships are shown between the probability that aracu are successfully caught during a trip and aquatic macrophytes, bare/herbaceous, open water, shrub, and population density variables. A negative relationship is seen between the probability of catch and forest cover. 30 Meaghan Rupprecht UNBC April 2024 Figure 9. Random effect plots show minimal variation in the probability of successfully capturing aracu between the sub-basins considered. Probability of successfully capturing aracu was highest in fishing trips landing in Santarém and Manaus, respectively. 31 Meaghan Rupprecht UNBC April 2024 Figure 10. Predicted probability of successfully catching aracu during a fishing trip was generally low and minimal spatial variation was observed. 3.2.1.2. Quantity of Catch Aracu represented approximately 3% of the total gillnet catch within the final dataset and successful catches were recorded within nine of the 12 sub-basins considered in this study (Table 2; Figure 12). Of the 15 models considered, five models were retained in the 95% confidence set for the best model of catch (Table 7). The top two models had similar AICc weights, and both included approximate fishing locations, but varied with regards to the habitat variable, indicating uncertainty in which habitat variable best predicted catch of aracu (Table 7). All possible habitat variables were present in the top model set, and all models included the approximate fishing location; population density was not present in any of the top models (Table 7). 32 Meaghan Rupprecht UNBC April 2024 Table 7. AICc summary statistics for models retained in the 95% confidence set for the best model of aracu catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Aracu-Catch AICc Δi wi K Cumulative Weight Model Aq. Macrophyte + XY Bare Ground/Herb + XY Open Water + XY Forest + XY Shrub + XY 36401.789 36401.804 36402.150 36402.940 36404.571 0.000 0.016 0.361 1.152 2.782 0.262 0.260 0.219 0.147 0.065 5 5 5 5 5 0.262 0.522 0.741 0.889 0.954 Model-averaged partial effect plots showed a non-linear relationship between the quantity of aracu caught and effort (Figure 11). The quantity of aracu caught during fishing trips had generally increased over time, however, displayed a secondary pattern of cyclical increases and decreases in catch every few years (Figure 11). Partial effects showed a small amount of seasonal variation in catch, with slightly higher quantities of aracu catch around the month of October, and slightly lower quantities of aracu catch between February and July (Figure 11). Relationships between aracu catch and habitat types varied; aracu catch displayed a positive relationship with bare ground/herbaceous cover and aquatic macrophytes up to a point before declining (Figure 11). The quantity of aracu catch had a slightly negative relationship with open water and a more pronounced negative relationship with shrub coverage; the quantity of aracu catch did not show a substantial trend with the proportion of forest cover (Figure 11). Random effects indicated that fishing trips which took place in Basin 8 were generally those with the highest aracu fish catch compared to fishing trips in other sub-basins; this subbasin is one of the mainstem sub-basins along the Amazon River, upstream of Manaus (Figure 12). Only seven of the possible 11 municipalities had landing records associated with aracu 33 Meaghan Rupprecht UNBC April 2024 catch; fishing trips landing at Manaus exhibited the highest catch (kg) of any landing municipality (Figure 12). Model-averaged predictions of aracu catch suggested the highest catches of aracu occurred along the southern periphery of the study area, particularly along the Purus and Madeira Rivers (Figure 13). A small area associated with higher predicted catches was identified along the mainstem of the Amazon River as well, between the confluences of the Amazon River with the Purus and Negro Rivers (Figure 13). The lowest catches for aracu were predicted in the Lower Amazon River near Santarém (Figure 13). Predictions of aracu catch based on seasonality and effort quartiles indicated that the higher catches occurred during the dry season compared to the wet season (Figure 14). Catch generally increased with increasing effort, regardless of season (Figure 14). The lowest predicted catches for aracu occurred during the wet season when effort was at the 25% quartile (Figure 14). The highest predicted catches for aracu occurred during the dry season when effort was at the 75% quartile (Figure 14). Similar spatial distributions as described above were observed in variations of effort and seasonality; these patterns were more pronounced in the dry season with increasing effort (Figure 14). 34 Meaghan Rupprecht UNBC April 2024 Figure 11. Partial effect plots are shown in log-scale. Effects showed variable relationships between the quantity of aracu catch and different habitat types; only bare ground/herbaceous habitat appeared to have a positive relationship with the quantity of catch, although aquatic macrophytes had a positive relationship until there were higher proportions of aquatic macrophytes. 35 Meaghan Rupprecht UNBC April 2024 Figure 12. Quantity of aracu catch was highest from fishing trips landing in Manaus, and those which conducted fishing in Basin 8. Maps of partial effects for random factors show little variation by sub-basin, and the highest catches in the Central Amazon municipality of Manaus. 36 Meaghan Rupprecht UNBC April 2024 Figure 13. If catch was successful, the quantity of aracu catch (kg) appeared higher in southern reaches of the study area, primarily in the sub-basins associated with the Purus and Madeira Rivers. A small area along the mainstem of the Amazon River showed higher aracu catch (kg) between its confluences with the Purus River and the Negro River. 37 Meaghan Rupprecht UNBC April 2024 Figure 14. The quantity of aracu catch was generally higher in the dry season compared to the wet season. Regardless of season, predicted quantities of catch increased with increasing effort. 38 Meaghan Rupprecht UNBC April 2024 3.2.2. Curimatã 3.2.2.1. Probability of Catch Of the 15 models considered, seven models were included in the 95% confidence set for the best model of probability of catch (Table 8). The top model included approximate fishing location and contributed the most to the top model set by weight (Table 8). Models with lesser weights included all possible habitat types and population density; all models in the top model set included the approximate fishing locations (Table 8). Table 8. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of curimatã catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Curimatã-Binomial AICc Δi wi K Cumulative Weight Model XY Pop. Density + XY Open Water + XY Bare Ground/ Herb + XY Forest + XY Aq. Macrophyte + XY Shrub + XY 18397.756 18399.879 18401.479 18401.504 18401.578 0.000 2.123 3.723 3.748 3.823 0.481 0.167 0.075 0.074 0.071 18401.696 18401.756 3.940 4.000 0.067 0.065 4 5 5 5 5 5 5 0.481 0.648 0.723 0.797 0.868 0.935 1.000 Model-averaged partial effects showed a non-linear relationship between the probability of curimatã catch and effort which initially declines with increasing effort before increasing to a fairly constant level (Figure 15). The probability that curimatã were successfully captured varied over time, but generally decreased between 1990 and 2011 (Figure 15). The probability of catching curimatã showed some seasonal variation, with a peak in probability observed between October and November (Figure 15). Negative relationships between the probability of curimatã catch and bare ground/herbaceous cover, open water, and shrub were observed, although the relationship with shrubs was subtle (Figure 15). A slightly positive relationship was observed 39 Meaghan Rupprecht UNBC April 2024 with forest cover and aquatic macrophytes, and a more pronounced positive relationship was observed with population density (Figure 15). Random effects indicated the probability of catch varied between sub-basins, with the highest probability in Basin 86 and lowest probability in the adjacent Basin 97, both located outside of Manaus (Figure 16). Basins along the mainstem of the Amazon River generally had higher probability of catch than tributary basins (Figure 16). Probability of catch also varied by municipality according to random effects; highest probability of catch was observed in Alenquer while the lowest probability of catch was observed in Obidos (Figure 16). Model-averaged predictions revealed the lowest probability for catching curimatã occurred in the Central and Lower Amazon, between the confluences of the Amazon River with the Negro and Tapajós Rivers as well as along tributaries associated with these areas, including the Madeira and Tapajós Rivers (Figure 17). The probability of successfully catching curimatã was highest in the southwest reaches of the study area (Figure 17). The probability of successfully catching curimatã varied from ~5% on the lower mainstem Amazon River to about 45% in areas with the highest probability (Figure 17). 40 Meaghan Rupprecht UNBC April 2024 Figure 15. Partial effect plots showed variable relationships between the probability of catching curimatã and different habitat types. Slightly positive relationships were observed with forest cover and aquatic macrophytes; a more pronounced positive relationship was shown with population density. Slightly negative relationships are shown with bare ground/herbaceous cover, open water, and to a lesser extent, shrubs. 41 Meaghan Rupprecht UNBC April 2024 Figure 16. Partial effects for random factors show the highest probability of catch taking place in Basin 86, close to Manaus, with the lowest catches in the adjacent Basin 97. The probability of catch was generally higher in mainstem Amazon sub-basins. The highest probability of catch by municipality occurred in Alenquer, and the lowest probability of catch occurred in Obidos. 42 Meaghan Rupprecht UNBC April 2024 Figure 17. The probability of successfully catching curimatã was highest in the southwest reaches of the study area. The Central and parts of the Lower Amazon, between the confluences of the Amazon River with the Negro and Tapajos Rivers had the lowest predicted probability of catch. Probability of catch increased downstream of the Amazon River confluence with the Tapajos River. 3.2.2.2. Quantity of Catch Curimatã accounted for approximately 8% of the total gillnet catch recorded within the dataset and successful catches were recorded within 10 of the 12 sub-basins considered in this study (Table 2, Figure 19). Of the 15 models fit to the dataset, a single model which included the proportion of bare ground and herbaceous cover within a sub-basin represented more than 95% of the AICc weight (weight = 0.993). All other models had negligible support. Model-averaged partial effect plots showed a non-linear relationship between the quantity of curimatã catch and effort, where catch increased steeply at initial increases in effort before starting to stabilize (Figure 18). Curimatã catch varied generally varied over time, but recently showed an increase between 2007 and 2011 (Figure 18). The quantity of curimatã catch during a fishing trip showed slight seasonal variation as well, where catch was highest around October 43 Meaghan Rupprecht UNBC April 2024 and lowest around June (Figure 18). The quantity of curimatã catch also showed a slightly negative relationship with bare ground/herbaceous habitat (Figure 18). Random effects showed some variation in catches between sub-basins, with highest catches associated with Basin 86 outside of Manaus and lowest catches associated with Basin 114 associated with the Purus River tributary (Figure 19). Additionally, the quantity of curimatã catch was substantially higher in fishing trips landing in Manaus compared to any other landing municipality (Figure 19). Nine of the possible 11 municipalities had landing records associated with curimatã catch (Figure 19). Predictions of curimatã catch showed variation by sub-basin, with the highest catch predicted in a sub-basin outside of Manaus; however, there was not a large difference between the highest and lowest predicted catches (Figure 20). Higher catches were generally predicted in the tributary sub-basins compared to sub-basins associated with the mainstem of the Amazon River (Figure 20). The lowest predicted catches were associated with the mainstem sub-basin that includes the Lower Amazon (Figure 20). Predictions of curimatã catch based on seasonality and effort quartiles indicated that the higher catches occurred during the dry season compared to the wet season (Figure 21). Lowest catches were predicted during the wet season when effort was at the 25% quartile, and spatial variation was minimal under these conditions (Figure 21). The highest catches were predicted during the dry season when effort was at the 75% quartile (Figure 21). Spatial variation was minimal amongst predictions for effort and seasonality (Figure 21). 44 Meaghan Rupprecht UNBC April 2024 Figure 18. Partial effect plots showed a non-linear relationship between the quantity of curimatã catch and fishing effort. Partial effects of month showed seasonality in fish catch that peaked around October and were lowest around June. A negative relationship is observed between curimatã catch and bare ground/herbaceous cover. 45 Meaghan Rupprecht UNBC April 2024 Figure 19. Partial effects for random factors show little variation in most sub-basis but the lowest catches were observed in Basin 114 where the Purus River is located and the highest were observed in Basin 86, northeast of Manaus. By municipality, catches in Manaus were higher relative to any other municipality. 46 Meaghan Rupprecht UNBC April 2024 Figure 20. Quantity of curimatã catch varied by basin; higher catches were observed in sub-basins of tributaries of the Amazon River when compared to the mainstem sub-basins of the Amazon River. 47 Meaghan Rupprecht UNBC April 2024 Figure 21. The quantity of curimatã catch varied based on seasonality and effort quartiles; catch was generally higher in the dry season compared to the wet season and increased with increasing effort. 48 Meaghan Rupprecht UNBC April 2024 3.2.3. Mapará 3.2.3.1. Probability of Catch Of the 15 models considered, six models were included in the 95% confidence set for the best model of probability of catch (Table 9). The model which contributed the most to the top model set by weight included the proportions of open water in a sub-basin and the approximate fishing coordinates (Table 9). Population density and all habitat types except forest were included in the 95% confidence set for the best model; approximate location of fishing was present in every model included in the 95% model set (Table 9). Table 9. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of mapará catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Mapará-Binomial AICc Δi wi K Cumulative Weight Model Open Water + XY XY Pop. Density + XY Bare Ground/Herb + XY Aq. Macrophyte + XY Shrub + XY 12840.928 12842.473 12842.749 12843.390 12843.484 12845.604 0.000 1.545 1.821 2.462 2.556 4.676 0.381 0.176 0.153 0.111 0.106 0.037 5 4 5 5 5 5 0.381 0.557 0.710 0.822 0.928 0.965 Model-averaged partial effect plots indicated a non-linear relationship between the probability of catching mapará and fishing effort (Figure 22). Probability of catching mapará generally increased over time (Figure 22). Seasonal variation in the probability of catching mapará was observed, with the highest probability occurring around March and lowest probability occurring between October and November (Figure 22). Partial effects indicated positive relationships between the probability of mapará catch and aquatic macrophytes, bare ground or herbaceous cover, shrub cover, and population density (Figure 22). Open water was 49 Meaghan Rupprecht UNBC April 2024 the only habitat with a negative relationship with the probability of catching mapará during a fishing trip (Figure 22). Random effects indicated the probability of catch varied minimally by basin but were slightly higher in Basin 6 relative to other basins (Figure 23). Basin 6 was a mainstem sub-basin of the Amazon River which incorporated the Lower Amazon region and municipalities (Figure 23). Probability of catching mapará varied by landing municipality as well, with the highest probability of catch in Obidos and lowest probability of catch in Manaus (Figure 23). Model-averaged predictions suggest the probabilities of successfully catching mapará were highest in the Lower Amazon above its confluence with the Tapajós River, near Obidos (~25%; Figure 24). The predicted probability of catch declined substantially to <5% in most other regions outside of the Lower Amazon, including the Central and Upper Amazon regions (Figure 24). 50 Meaghan Rupprecht UNBC April 2024 Figure 22. Partial effects with the probability of catching mapará showed a non-linear relationship with effort. Probability of catch had generally been increasing between 1990 and 2011. The probability of mapará catch varied seasonally. Partial effect plots showed positive relationships between the probability of catching mapará and all habitat types except open water, which had a negative relationship. 51 Meaghan Rupprecht UNBC April 2024 Figure 23. Partial effects for random factors showed minimal variation in the probability of catch between basins, although slightly higher in Basin 6, a mainstem sub-basin including the Lower Amazon. The highest probability of catch by municipality occurred in Obidos, and the lowest probability of catch occurred in Manaus. 52 Meaghan Rupprecht UNBC April 2024 Figure 24. The probabilities of successfully catching mapará were highest in the Lower Amazon above its confluence with the Tapajos River, near Obidos. Predicted probability of catch was minimal in the Central and Upper Amazon regions. 3.2.3.2. Quantity of Catch Mapará were the most abundant species in the dataset and catch from mapará contributed to ~25% of the total fish catch with gillnets; however, successful catches were recorded in the fewest sub-basins of any of the species with only five of the 12 sub-basins being represented (Table 2; Figure 26). Of the 15 models considered, eight models were retained in the 95% confidence set for the best model of catch (Table 10). The model including the proportion of aquatic macrophytes and the approximate fishing coordinates contributed the most to the top model set by weight; the second model by weight included the proportion of open water and approximate fishing coordinates (Table 10). All other habitat proportions and population density of the basins contributed to a lesser extent, as did open water and aquatic macrophytes when they were modelled without approximate fishing locations (Table 10). 53 Meaghan Rupprecht UNBC April 2024 Table 10. AICc summary statistics for models retained in the 95% confidence set for the best model of mapará catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Mapará-Catch AICc Δi wi K Cumulative Weight Model Aq. Macrophyte + XY Open Water + XY Bare Ground/Herb Aq. Macrophyte Shrub Forest Pop. Density Open Water 67682.080 67683.222 67685.517 67686.034 67686.038 67686.631 67687.722 67688.487 0.000 1.142 3.437 3.954 3.958 4.551 5.642 6.407 0.431 0.243 0.077 0.060 0.060 0.044 0.026 0.017 5 5 4 4 4 4 4 4 0.431 0.674 0.751 0.811 0.870 0.914 0.940 0.957 Model-averaged partial effect plots indicated a non-linear relationship between the quantity of mapará catch and fishing effort (Figure 25). The quantity of mapará catch did not seem to vary substantially over time but shows a recent decline according to partial effects (Figure 25). The quantity of mapará catch also did not seem to show seasonal variation by month (Figure 25). Slightly positive relationships were observed between all habitat types and population density except for forest cover, which was negatively related to the quantity of mapará catch (Figure 25). Random effects showed minimal variation in catches between sub-basins (Figure 26). Seven of the possible 11 municipalities had landing records associated with mapará catch (Figure 26). The quantity of mapará catch was also substantially higher in fishing trips landing in Obidos compared to any other landing municipality, and lowest in Oriximina (Figure 26). Neither Upper Amazon municipalities, Alvarães or Tefé, had fish landing records with recorded mapará catches (Figure 26). 54 Meaghan Rupprecht UNBC April 2024 Model-averaged predictions showed higher mapará catches were observed on the northern sub-basins of the Lower Amazon, as well as an area of higher catch along the Japurá River (Figure 27). The lowest predicted quantity of catch occurred in the southern Tapajós River basin, although catch in the Central Amazon was also less than in surrounding areas (Figure 27). Predictions by effort and season indicated no seasonal variation between dry and wet season, consistent with partial effects lacking indications of seasonality, but catch did increase with increasing effort (Figure 28). The lowest catches were predicted when effort was at the 25% quartile, while highest catches were predicted when effort was at the 75% quartile (Figure 28). 55 Meaghan Rupprecht UNBC April 2024 Figure 25. Partial effect plots showed a non-linear relationship between the quantity of mapará catch and fishing effort. Partial effects of month did not show seasonality in catch. All habitat variables and population density showed slightly positive relationships with catch; only forest cover had a negative relationship with the quantity of mapará catch. 56 Meaghan Rupprecht UNBC April 2024 Figure 26. Partial effects for random factors show little variation in most sub-basins. By municipality, catches in Obidos were higher relative to any other municipality, and lowest in Oriximina. 57 Meaghan Rupprecht UNBC April 2024 Figure 27. Predicted mapará catches were highest in the northern sub-basin of the Lower Amazon. High catch was also predicted in an area along the Japura River. The lowest predicted quantity of catch occurred in the southern Tapajós River basin, although catch in the Central Amazon was also less than surrounding areas. 58 Meaghan Rupprecht UNBC April 2024 Figure 28. The distribution of mapará catch was predicted based on variations in season and effort. Catch of mapará was generally the same across seasons but catch increased with increasing effect. 59 Meaghan Rupprecht UNBC April 2024 3.2.4. Pescada 3.2.4.1. Probability of Catch Of the 15 models considered, seven were included in the 95% confidence set of models for the best model of probability of catch (Table 11). The top model considered the proportion of forest in the sub-basin where fishing took place, while the second ranked model considered the proportion of shrub in the sub-basin where fishing took place (Table 11). All habitat types and population density were included in the top model set, with the proportion of forest cover being included in two models, one of which also included the fishing location as well (Table 11). Table 11. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of pescada catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Pescada-Binomial AICc Δi wi K Cumulative Weight Model Forest Shrub Pop. Density Open Water Aq. Macrophyte Bare Ground/ Herb Forest + XY 21777.473 21777.568 21777.764 21777.839 21777.905 21778.257 21780.672 0.000 0.095 0.291 0.367 0.433 0.785 3.200 0.180 0.171 0.155 0.150 0.145 0.121 0.036 4 4 4 4 4 4 5 0.180 0.351 0.506 0.656 0.801 0.922 0.958 Model-averaged partial effect plots indicated a non-linear relationship between the probability of catching pescada and fishing effort in which the probability of catch dropped substantially at higher levels of effort (Figure 29). Partial effects also showed that the probability of catch fluctuated over time, although it generally increased between 1990 and 2011 (Figure 29). The probability of catching pescada varied seasonally, with higher catches around December and January and lower catches between May and September, except for a secondary peak in 60 Meaghan Rupprecht UNBC April 2024 probability in July (Figure 29). Partial effect plots showed positive relationships between the probability of catching pescada and all habitat types except forest, which had a slightly negative relationship; additionally, a positive relationship was observed with population density (Figure 29). Random effects indicated the probability of catch varied minimally by basin but were slightly higher in Basin 114 relative to other basins (Figure 30). Basin 114 was the sub-basin associated with the Purus River, a major tributary of the Amazon River (Figure 30). Probability of catching pescada varied by landing municipality as well, with the highest probability of catch in Oriximina and lowest probability of catch in Tefé (Figure 30). The probability of catch was generally higher in Lower Amazon municipalities relative to the Central and Upper Amazon municipalities of Manaus, Alvarães, and Tefé (Figure 30). Model-averaged predictions showed a catch probability of approximately 10% for most of the study area, with little spatial variation observed for most of the basin but slightly higher predicted probability of catch (13%) in the Lower Amazon near the confluence of the Amazon River and the Tapajós River (Figure 31). 61 Meaghan Rupprecht UNBC April 2024 Figure 29. Partial effects with the probability of catching pescada showed a non-linear relationship with effort which dropped substantially at higher levels of effort. The probability of catch fluctuated over time but generally increased between 1990 and 2011. Probability also varied seasonally with higher catches around December and January and lower catches between May and September, except for a secondary peak in probability in July. Partial effect plots showed positive relationships between the probability of catching pescada and all habitat types except forest, which had a slightly negative relationship. Probability of catch was also positively related to population density. 62 Meaghan Rupprecht UNBC April 2024 Figure 30. Partial effects for random factors showed some variation in the probability of catch between basins, with higher probability in Basin 114 which is associated with the Purus River. The highest probability of catch by municipality occurred in Oriximina, and the lowest probability of catch occurred in Tefé. 63 Meaghan Rupprecht UNBC April 2024 Figure 31. The probability of catching pescada was around 10% for most of the study area, with little spatial variation observed. 3.2.4.2. Quantity of Catch Pescada contributed ~5% of the total gillnet catch in the dataset and were recorded within nine of the 12 sub-basins considered in this study (Table 2; Figure 33). Of the 15 models considered, three models were retained in the 95% confidence set for the best model of catch (Table 12). The model including the approximate fishing coordinates contributed the most to the top model set by weight (Table 12). The proportion of forest and proportion of shrub in the subbasin where fishing took place were the only habitat variables included in the top model set (Table 12). 64 Meaghan Rupprecht UNBC April 2024 Table 12. AICc summary statistics for models retained in the 95% confidence set for the best model of pescada catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Pescada-Catch AICc Δi wi K Cumulative Weight Model XY Forest + XY Shrub + XY 75065.824 75068.042 75069.104 0.000 2.218 3.280 0.653 0.215 0.127 4 5 5 0.653 0.868 0.995 Model-averaged partial effects indicated a non-linear relationship between the quantity of pescada catch and fishing effort (Figure 32). The quantity of catch fluctuated over time but did not show substantial trends between 1990 and 2011 (Figure 32). Slight seasonal variation was observed, with higher catches of pescada in November and lower catches between May and June (Figure 32). A slightly positive relationship was observed between quantity of catch and shrub cover, while a negative relationship was observed between quantity of catch and forest cover (Figure 32). Random effects showed minimal variation in catches between sub-basins (Figure 33). Nine of the possible 11 municipalities had landing records associated with pescada catch (Figure 33). The quantity of pescada catch was substantially higher in fishing trips landing in Santarém and Manaus, respectively, relative to other landing municipalities (Figure 33). The lowest catches by municipality occurred in Alenquer (Figure 33). Model-averaged predictions of pescada catch suggested higher catches in basins near the southwestern reaches of the study area, distant from the mainstem of the Amazon River (Figure 34). The lowest catches were predicted in the Lower Amazon and on the Tapajós River closer to its confluence with the Amazon River (Figure 34). 65 Meaghan Rupprecht UNBC April 2024 Predictions by effort and season indicated higher catch during the dry season compared to the wet season, and increased catch with increasing effort (Figure 35). Spatial patterns of catch across effort and seasonality were consistent with those described previously. The lowest catches were predicted during the wet season when effort was at the 25% quartile, while highest catches were predicted during the dry season when effort was at the 75% quartile (Figure 35). Figure 32. The quantity of pescada catch during fishing trips had a non-linear relationship with effort. Slight seasonal variation in catch was also observed. In relation to habitat, the quantity of pescada catch had a slightly positive relationship with shrub cover and slightly negative relationship with forest cover. 66 Meaghan Rupprecht UNBC April 2024 Figure 33. Partial effects for random factors show little variation in most sub-basins. By municipality, catches in Santarém and Manaus were higher relative to any other municipality, and lowest in Alenquer. 67 Meaghan Rupprecht UNBC April 2024 Figure 34. Higher catches of pescada were predicted in the southwestern parts of the study area, further from the mainstem of the Amazon River. The lowest catches of pescada were predicted in the Lower Amazon and on the Tapajós River closer to its confluence with the Amazon River. 68 Meaghan Rupprecht UNBC April 2024 Figure 35. Catch of pescada was generally higher in the dry season compared to the wet season; catch generally increased with increasing effort. 69 Meaghan Rupprecht UNBC April 2024 3.2.5. Tambaqui 3.2.5.1. Probability of Catch Of the 15 models considered, five models were retained in the 95% confidence set for the best model of probability of catch (Table 13). The model which contributed the most to the top model set by weight included the proportion of forest cover and approximate fishing locations; however, other habitats included in the 95% confidence set included bare ground and herbaceous, open water, and shrub habitats (Table 13). All models in the top model set included the approximate fishing locations, and population density was not included in any top model (Table 13). Table 13. AICc summary statistics for models retained in the 95% confidence set for the best model of the probability of tambaqui catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Tambaqui-Binomial AICc Δi wi K Cumulative Weight Model Forest + XY XY Bare Ground/Herb + XY Open Water + XY Shrub + XY 17277.634 17280.908 17281.646 17281.775 17282.708 0.000 3.275 4.012 4.142 5.074 0.627 0.122 0.084 0.079 0.050 5 4 5 5 5 0.627 0.749 0.833 0.912 0.962 Model-averaged partial effect plots indicated a non-linear relationship between the probability of catching tambaqui and fishing effort in which the probability of catch initially increased with higher levels of effort before levelling out and declining (Figure 36). Partial effects also showed that the probability of catch decreased substantially over time between 1990 and 2011 (Figure 36). The probability of catching tambaqui varied seasonally, with higher probability in April and lower probability between December and January (Figure 36). Partial effect plots showed negative relationships between the probability of catching tambaqui and bare 70 Meaghan Rupprecht UNBC April 2024 ground and herbaceous cover, open water, and shrub cover (Figure 36). However, the probability of tambaqui catch had a positive relationship with the proportion of forest cover (Figure 36). Random effects indicated the probability of catch varied minimally by basin but were slightly higher in Basin 86 relative to other basins (Figure 37). Basin 86 was the sub-basin located just outside of Manaus near the confluence of the Amazon River and Negro River (Figure 37). The probability of catching tambaqui varied by landing municipality as well, with the highest probability of catch in the Lower Amazon municipalities of Parintins, Oriximina, and Alenquer, and lowest probability of catch in Obidos (Figure 37). Model-averaged predictions showed distinct spatial patterns in the probability of catching tambaqui during a fishing trip (Figure 38). The highest probability of catch for tambaqui was predicted in the western sub-basins of the Upper Amazon which are associated with the Juruá River (Figure 38). Other areas of higher probability included a sub-basin outside of Manaus and downstream of the confluence between the Amazon and Tapajós Rivers (Figure 38). Predicted probability of catch declined in eastern reaches of the mainstem Amazon River and associated tributaries such as the Madeira and Tapajós Rivers (Figure 38). The lowest predicted probability in the study area was along the Amazon River, between its confluences with the Negro River and Tapajós River (Figure 38). Predicted probability of catch varied from less than 15% in the Lower and Central Amazon to more than 75% in the tributary sub-basins of the Upper Amazon (Figure 38). 71 Meaghan Rupprecht UNBC April 2024 Figure 36. Partial effects with the probability of catching tambaqui showed a non-linear relationship with effort which initially increased before leveling off. The probability of catch declined between 1990 and 2011. The probability also varied seasonally with higher catches around April and lower catches between December and January. Partial effect plots showed negative relationships between the probability of catching tambaqui and all habitat types included in the top models except forest, which had a positive relationship. 72 Meaghan Rupprecht UNBC April 2024 Figure 37. Partial effects for random factors showed some variation in the probability of catch between basins, with higher probability in Basin 86 which near Manaus. The highest probability of catch by municipality occurred in similarly in Parintins, Oriximina, and Alenquer; the lowest probability of catch occurred in Obidos. 73 Meaghan Rupprecht UNBC April 2024 Figure 38. The highest probability of catch for tambaqui was located in the western sub-basins of the Upper Amazon. The probability of catch declined in eastern reaches of the mainstem Amazon River and associated tributaries such as the Madeira and Tapajós Rivers. Other areas of higher probability included a sub-basin outside of Manaus and downstream of the confluence between the Amazon and Tapajós Rivers. 3.2.5.2. Quantity of Catch Tambaqui contributed to approximately 8% of the total gillnet catch in the dataset and were recorded within 11 of the 12 sub-basins included in this study (Table 2; Figure 40). Of the 15 models considered, five models were retained in the 95% confidence set for the best model of catch (Table 14). The top two models contributed similarly to the top model set by weights. The top model considered the proportion of bare ground and herbaceous cover and the approximate fishing locations, while the second ranked model considered just the approximate fishing locations (Table 14). The only habitat variables included in the top model set were bare ground and herbaceous cover, open water cover, and forest cover (Table 14). 74 Meaghan Rupprecht UNBC April 2024 Table 14. AICc summary statistics for models retained in the 95% confidence set for the best model of tambaqui catch. AICc results are displayed for the models which contribute to at least 95% of the cumulative weight amongst all candidate models. Delta AICc(Δi) is the difference between the model with the lowest AICc and a given model within the candidate set. AICc weight (wi) indicates the probability of the model being the most parsimonious in the candidate set. Additionally, K represents the number of terms included in the given model. All models contained the effects of effort, month and year (omitted from column Model). Tambaqui-Catch AICc Δi wi K Cumulative Weight Model Bare Ground/Herb + XY XY Open Water + XY Open Water Forest + XY 47808.814 47808.878 47810.898 47811.105 47812.793 0.000 0.064 2.083 2.291 3.978 0.343 0.332 0.121 0.109 0.047 5 4 5 4 5 0.343 0.675 0.796 0.905 0.952 Model-averaged partial effects indicated a non-linear relationship between the quantity of tambaqui catch and fishing effort (Figure 39). The quantity of catch generally declined between 1990 and 2011, although there was a slight increase in catch quantities around 1998 before declining again in the following years (Figure 39). Seasonal variation was observed, with higher catches of tambaqui in between December and January and lower catches around July (Figure 39). A slightly negative relationship was observed between quantity of catch and open water cover, while positive relationships were observed with both the proportion of bare ground or herbaceous cover and forest cover (Figure 39). Random effects showed minimal variation in catches between sub-basins (Figure 40). All 11 municipalities had landing records associated with tambaqui catch (Figure 40). The quantity of tambaqui catch was substantially higher in fishing trips landing in Manaus relative to other landing municipalities (Figure 40). The lowest catches by municipality occurred in Monte Alegre (Figure 40). Model-averaged predictions of tambaqui catch indicated higher catches in the western Amazon Basin and declined longitudinally from West to East (Figure 41). The lowest catches of tambaqui were predicted in the Lower Amazon and on the Tapajós River (Figure 41). 75 Meaghan Rupprecht UNBC April 2024 Predictions by effort and season indicated higher catch during the dry season compared to the wet season, and increased catch with increasing effort (Figure 42). Spatial patterns of catch across effort and seasonality were consistent with those described previously. The lowest catches were predicted during the wet season when effort was at the 25% quartile, while highest catches were predicted during the dry season when effort was at the 75% quartile (Figure 42). Figure 39. The quantity of tambaqui catch during fishing trips had a non-linear relationship with effort. Seasonal variation in catch was also observed in the quantity of catch. In relation to habitat, the quantity of tambaqui catch had a slightly positive relationship with bare ground/herbaceous cover and forest cover, but a slightly negative relationship with open water cover. 76 Meaghan Rupprecht UNBC April 2024 Figure 40. Partial effects for random factors show little variation in most sub-basins. By municipality, catches Manaus were substantially higher relative to any other municipality, and lowest in Monte Alegre. 77 Meaghan Rupprecht UNBC April 2024 Figure 41. Higher catches of tambaqui were predicted in the western parts of the study area and declined to the east along a gradient. The lowest catches of tambaqui were predicted in the Lower Amazon and on the Tapajós River. 78 Meaghan Rupprecht UNBC April 2024 Figure 42. Catch of tambaqui was generally higher in the dry season compared to the wet season; catch generally increased with increasing effort. 79 Meaghan Rupprecht UNBC April 2024 4. Discussion 4.1. Multispecies Catch One of the most important findings of this research was that multispecies fish catch varies greatly throughout the Amazon Basin, and this variation is largely related to the presence of floodplain forests in the river basins where fish are caught. These findings were consistent with one of the primary hypotheses assessed during this research, which was that multispecies fish catch would vary spatially based on floodplain habitat composition. The spatial variation in multispecies fish catch was driven by relationships with fishing locations, and either the proportion of shrubs or proportion of forest in the sub-basins where fishing took place according to the top models. These statistical models indicated a positive effect of floodplain forests and a negative effect of shrub habitats on multispecies fish catch. These two habitat types were further strongly negatively and oppositely related to each other. Consequently, since shrub habitat was negatively related to floodplain forest and negatively related to multispecies fish catch, it further suggests that this model also supported that the opposite is true for floodplain forests, such that multispecies fish catch would be positively related to floodplain forests. Predictions of catch also showed patterns that were fairly consistent with habitat compositions of sub-basins. Particularly, areas with the highest magnitudes of predicted catch were primarily in the Upper Amazon reaches, where relatively pristine habitats with intact forest are more abundant. The magnitude of catch declined along the mainstem of the Amazon River towards the Lower Amazon, which is a highly degraded landscape characterized by historic deforestation and high proportions of shrub habitats. These results, using a space-for-time approach similar to that described in Lovell et al. (2023), demonstrate that multispecies fish catch varies between landscapes at varying stages of degradation; specifically, multispecies fish catch declined in areas affected by deforestation and which had less floodplain forests present in the areas where fishing was conducted. 80 Meaghan Rupprecht UNBC April 2024 It is also important to consider that ecological phenomena, such as interactions between fish catch and habitat, take place at varying degrees within different scales according to the concepts of multiscale patterns and processes in ecology (Levin 1992). This is acutely evident within the Amazon fishery, which relies heavily upon migratory fish that interact with floodplain habitats throughout a range of geographic scales whether they are local, mid-distance, or longdistance migratory species (Goulding et al. 2019). The relationships between fishes and their surrounding habitats can vary in importance and magnitude at the different scales in which they are assessed and depending on the scales of activity in which their environments are being used, such as for long-distance migrations or more local spawning activity. As such, it is important to incorporate cross-scale studies which would combine fine-scale assessments or experimentation with broad-scale modeling of patterns to generate a more holistic understanding of the ecological mechanisms driving the Amazon fishery (Levin 1992). In this case, existing research has primarily focused on the implications of floodplain habitat type within the Lower Amazon (i.e., Castello et al. 2018, Arantes et al. 2019, Barros et al. 2020, etc.), which is a degraded landscape that has experienced substantial deforestation of the floodplain (Renó & Novo 2019; Renó et al. 2011). Research at larger scales beyond the Lower Amazon is typically limited by availability of data and the heterogeneity of data collected from various sources at various scales. However, these challenges were addressed using a vast dataset which combined and standardized such variation, allowing my research to be conducted on a scale larger than previous studies. This broader, large-scale study provided spatial patterns of multispecies fish catch and showed similar catch-habitat relationships, which can be explained by the underlying ecological driver of floodplain forest cover. When considered in conjunction 81 Meaghan Rupprecht UNBC April 2024 with studies conducted at smaller scales, these results confirm that floodplain forests are critical habitats for multispecies fish catch across geographic scales. However, unlike findings from previous research at smaller scales, it appeared that rather than the direct assessment of floodplain forests, the model using the proportion of shrub coverage best captured this relationship. A possible explanation for shrubs being the habitat type which best captured the fish catch-habitat relationships is the Modifiable Areal Unit Problem (MAUP). The Modifiable Areal Unit Problem (MAUP) states differences in spatial aggregation techniques or areal units used for analysis, which is inherently present in most spatial analyses, can cause differences in results (Wong 2011). In this case, previous research was approached using lake systems units or buffer areas surrounding features such as municipalities, which were smaller geographical scales. However, my approach used larger spatial units which were selected based on the extent of data, thus incorporating multiple regional fisheries, and requiring larger units for computational efficiency during analyses (Basin Level 4 sub-basins between 10,000 and 100,000 km2; Venticinque et al. 2016). This resulted in a coarser scale of analysis compared to regional assessments with finer resolution spatial units and the aggregation of habitat data. With spatial aggregation, the scale effect of MAUP can result in higher correlations between variables (Wong 2011). This effect was observed in the correlations between shrub and forest cover within my own data and could explain why the model including the proportion of shrubs best captured the relationship between fish catch and habitat. While the proportion of shrub cover was the habitat type with the most support in the top model set, the underlying ecological conclusions from these models remain consistent with research that has been conducted at smaller geographic scales, which is that multispecies fish catch is positively related to floodplain forests. 82 Meaghan Rupprecht UNBC April 2024 My results also demonstrated non-linear relationships between multispecies fish catch and effort, which is a particularly important relationship that has been described in previous research (Lorenzen et al. 2006). However, previous research on fish catch in the Amazon has mostly been conducted using linear modeling techniques, which lack the ability to account for non-linear relationships within data. Using non-linear modeling techniques, my research was able to accurately characterize the effects of fishing effort within a multispecies fishery, and how other variables being assessed within the model also contribute to understanding the underlying relationships with multispecies fish catch as well. Understanding the effects of fishing effort on the fishery is particularly vital to understanding how fishing-driven exploitation is occurring within the Amazon River, and accurately characterizing these relationships with the appropriate modeling tools was an important step taken by this research. Like many other studies on this topic, one of the major limitations of this research is that there is no direct comparison of fish catch before and after land cover changes occurred. Therefore, no direct causal relationship can be implicated between floodplain land cover and fish biomass. However, the space-for-time approach used the existing gradient of forest depletion to address a major question about the largescale effects of deforestation on multispecies fish catch which has previously been neglected due to the lack of available data beyond local or regional scales. Ecosystem-level management and cumulative effect assessments have become increasingly emphasized in recent years, particularly within the terrestrial environments of the Amazon Basin and more recently within the realm of freshwater conservation (Goulding et al. 2019). With ecosystem-level management, however, comes the question of adequate or appropriate spatial scale which can best inform management of aquatic ecosystems (Delacámara 83 Meaghan Rupprecht UNBC April 2024 et al. 2020). For example, coarser scale assessments such as this may be more informative to national efforts to prioritize management and protection of the greater fishery, whereas regional efforts such as co-management with fishers may be best served by patterns and processes within the local landscapes in which they most frequently operate. Thus, understanding the implications of scale on these patterns and processes can be used to improve policy coordination at different levels or provide a more holistic account of the ecosystem at hand depending upon the approach (Delacámara et al. 2020). Careful consideration of results at both scales will ultimately contribute to an improved understanding of the dynamic ecology behind Amazon floodplainfisheries. However, despite variations in spatial scales and resolution, my research indicates there is a consistent relationship observed between habitat and fish catch which supports the presence of an important ecological phenomenon between the floodplain forests and multispecies fish yields of the Amazon River fishery. In the end, this approach quantified the differences in multispecies fish catch from the relatively pristine landscapes of the western Amazon River to the highly deforested floodplains of the eastern Amazon. I found that the magnitude of multispecies fish catch declines within areas that have been subjected to deforestation and increases with the availability of floodplain forest in the landscape. These results can be used to further support that floodplain deforestation has negative consequences to the multispecies fish catch that is available to the commercial fishing industry across a larger landscape gradient. Ultimately, the results of this research provide irrefutable evidence that floodplain forests are a vital resource to multispecies fisheries in the Amazon River Basin, and this resource has widespread impacts on the fishery which transcend the spatial scales at which they are traditionally assessed. 84 Meaghan Rupprecht 4.2. UNBC April 2024 Species-Specific Catch My results indicated that the relationship between fish catch and the surrounding floodplain habitat varied substantially by species groups, resulting in large spatial variability in catch by species groups across the Amazon Basin. The relationships identified between speciesspecific catch and floodplain habitat were generally consistent with the habitats where food resources were likely to be more abundant for the respective feeding habits of the study groups. These findings were consistent with the hypothesis that the probability and quantity of speciesspecific fish catch would vary based on the respective feeding habits, as well as the hypothesis that spatial variation was present in species-specific fish catch. There were several observations which were consistent across all species groups, regardless of the associated feeding habit. All groups showed non-linear responses to fishing effort for both the probability and the quantity of species-specific catch. This is likely an artifact of these species being caught within a multi-species fishery, where fishing effort and aggregated fish catch typically displays non-linear relationships (Lorenzen et al. 2006). While catch was assessed on a species-specific level, fishing trips are conducted within the multispecies fishery where gear types are often not species selective. Particularly, this research used records of fish catch conducted with gillnets, which were the primary gear type used within the Amazon River fishery and accounted for 80% of commercial catches in this study. This relationship was present within both the probability and quantity of catch for each species group assessed. In general, the probability and quantity of catch for a particular species group were related to more than one habitat type and/or population density. However, the quantity of curimatã catch was best assessed using just the proportion of bare ground or herbaceous habitat at the sub-basin level, without considering the approximate fishing locations. This was unique as most other species groups included at least one model with the approximate fishing locations, 85 Meaghan Rupprecht UNBC April 2024 indicating that fishing locations capture important information in modeling species-specific fish catch. Additionally, all species groups except for mapará displayed seasonal variation in the quantity of catch, with higher quantities of catch in the dry season when fish are concentrated and more easily targeted by fishing activity, and lower quantities of catch during wet seasons when fish can disperse and seek out refuge in floodplain habitats made available by rising water levels of the flood pulse. This suggests that mapará are consistently exploited year-round, which may explain why mapará were the most commercially caught fish within the dataset and contributed to ¼ of commercial catch between 1991 and 2011. The lack of seasonal variation in catch quantity suggests that mapará have little reprieve from fishing pressure during flood pulses as many other species do. Interestingly, the probability of successfully capturing mapará did display seasonal variation, which was consistent with the flood pulse, but opposite to that observed for other species groups. In the case of mapará, the probability of catch was higher during periods of high water and lower during periods of low water, when open water habitats were more limited. Two of the five species groups displayed similar relationships for probability and quantity of catch: aracu and pescada, the omnivorous and piscivorous species groups in this study, respectively. Omnivorous species generally rely upon a variety of food items such as plant material, detritus, arthropods/invertebrates, and smaller fish, while piscivorous species rely upon fish at varying life stages for most of their diet (Arantes et al. 2019; Amazon Waters Alliance). Each of the respective feeding groups had fairly low and constant probabilities of being caught throughout the river sub-basins considered in the study and displayed little spatial variation. Both species groups also demonstrated higher quantities of catch in the sub-basins of the southern 86 Meaghan Rupprecht UNBC April 2024 tributaries of the Amazon River, except for a single area with higher catch for aracu along the mainstem of the Amazon River between the confluences with the Purus and Negro Rivers. However, previous research has indicated opposite habitat relationships between fish attributed to each feeding habitat, wherein omnivores were positively associated with forest cover but piscivores were negatively associated with forest cover in the Lower Amazon (Arantes et al. 2017). My results suggest that within the larger landscape, both species groups were slightly less likely to be caught in areas with more forest cover. Regarding the quantity of catch, pescada were negatively related to floodplain forests, as seen in previous research, but aracu displayed little change related to floodplain forests, and were more strongly associated with other habitat types. Particularly, aracu catch had positive associations with aquatic macrophytes and bare ground or herbaceous habitat, while displaying negative associations with open water and shrub habitats. Despite having differing relationships with the surrounding environments and different feeding habits, both species groups showed similar spatial distributions, which was unexpected given the relationships observed. The detritivorous species group in this study, curimatã, showed some unique habitat relationships between the probability and quantity of catch compared to the other species. Previous research has shown that detritivorous species, particularly curimatã, are strongly positively associated with forest cover and their diets rely heavily upon organic material (Arantes et al. 2017; Arantes et al. 2019; Oliveira et al. 2006). Results of this study further support these findings. While the probability of catching curimatã was best modelled using the approximate fishing locations, all habitat types and population density contributed to the top model set, with only population density having substantially more support compared to floodplain habitats. The resulting predictions also indicated that the probability of catch was lowest in the Central and 87 Meaghan Rupprecht UNBC April 2024 Lower Amazon mainstem, while it was higher in peripheral basins to the West and East ends of the study area. These observations suggested that curimatã are unlikely to be caught in the two regions where human population in the Amazon Basin is typically highest, and where fishing trips are often concentrated. However, curimatã were also the only species in which a single habitat best modelled fish catch: bare ground and herbaceous vegetation. The quantity of curimatã catch was negatively related to the proportion of bare ground and herbaceous habitat at a sub-basin level; similar to the distributions for probability of catch, the quantity of curimatã caught were lowest in the mainstem sub-basins associated with the Central and Lower Amazon regions. The similarities between probability and quantity of catch despite differing modelling results are likely due to the positive correlation between bare ground and herbaceous habitats with areas of higher population density (Table 4). Given that the removal of vegetation is often a consequence of expanding urbanization and agriculture (Renó & Novo 2019), these results suggest that expanding populations and subsequent absences of floodplain vegetation (such as shrubs and forests) in its wake will have detrimental impacts to curimatã, and likely to other species which rely upon organic inputs from vegetation. The final two species groups, mapará and tambaqui, demonstrated distributions which closely aligned with observations of a forest depletion gradient throughout the Amazon Basin, although in opposite manners. The forest depletion gradient suggests that floodplain forests become increasingly lost and altered from west to east along the Amazon River, a phenomenon which has been taking place over decades in response to increasing anthropogenic activity (Renó & Novo 2019). Indeed, this pattern of forest loss is displayed within the habitat data used for this study, wherein forested habitats were substantially higher in western sub-basins and declined to the east (Hess et al. 2015). However, previous research shows that mapará and tambaqui tend to 88 Meaghan Rupprecht UNBC April 2024 have positive habitat associations in opposition of each other, one with open water and the other with floodplain forests, respectively. Yet, the proportions of these habitats within the landscape are nearly inverse of each other (Table 4). This presented a unique opportunity to compare the two species groups against the deforestation gradient which has been described and assess how each species might be interacting with the environment on a larger scale given their disparate relationships to the surrounding habitats. Mapará were the planktivorous species group considered in this research, and previous research in the Lower Amazon has shown positive associations between open water and mapará, likely due to the availability of phytoplankton within such habitats (Arantes et al. 2017). However, my results suggested that the probability of catching mapará declines with open water habitat, which may be indicating that capture efficiency is lower when the species is not in more isolated bodies of water during periods of low water, despite favoring these locations for feeding activity. Yet the quantity of mapará catch was positively related to open water and negatively related to forest cover, which was consistent with previous findings (Arantes et al. 2017). Despite being the most commercially caught species group throughout the fishery, successful capture of mapará was recorded in the fewest river basins out of all the species assessed (5 of 12 basins). Records of successful catch were almost exclusively within mainstem Amazon River basins, except for the basin of the Tapajós River, both of which were associated with higher proportions of open water relative to other sub-basins. These observations could also be related to the disproportionate fishing activity which takes place in the Lower Amazon compared to other regions, where eight of the eleven landing municipalities were located, and higher human population densities are generally observed. My results further suggested that the Lower Amazon was the region with the highest quantities and probability of mapará catch, 89 Meaghan Rupprecht UNBC April 2024 particularly in the immediate areas surrounding the eight municipalities; landscapes in this region are more degraded and open water is more abundant relative to other regions of the Amazon Basin (Renó & Novo 2019; Hess et al. 2015). The large contribution of mapará to the total fish yield of the Amazon River may be indicating a strong preference towards this fish and consequently a heavy reliance on this fish for commercial catches, specifically from fishing fleets in the Lower Amazon. The most fishing activity was located within the same area that mapará were the most likely to be caught during a fishing trip, and it appears that mapará are an abundant species group which are frequently caught during fishing trips in the Lower Amazon, but are also caught at relatively consistent quantities throughout the year, making them a reliable source of fish catch relative to the other species groups which exhibited more seasonal variability. However, the lack of seasonal reprieve in fish catch and reliance upon mapará for a high proportion of commercial fish catch in the Lower Amazon also brings up cause for concern. Increasing demands for fish from the surrounding populations may disproportionately fall on mapará compared to other species which make up smaller proportions of the fishery, for which fishing effort appears to be more evenly distributed upon. Since mapará appear to be reliably and consistently captured year-round, there is a great potential for overexploitation of this species group in the following years if the stocks are not able to take advantage of seasonal reprieves during flood pulses as the catch of other groups seemed to demonstrate. Tambaqui were the frugivorous species group considered in this study, although tambaqui also rely upon zooplankton as the water levels decline and fruits and seeds of the flooded forests become more limited (Oliveira et al. 2006). Indeed, previous research has shown positive associations between tambaqui and floodplain forests (Arantes et al. 2017). My results 90 Meaghan Rupprecht UNBC April 2024 confirmed these relationships beyond the floodplain lakes of the Central and Lower Amazon, where previous studies had been focused, and showed similar relationships throughout the Amazon Basin. Particularly, I found that tambaqui were more likely to be caught in higher quantities during a fishing trip when forest cover was more abundant in the river basin where fishing took place. Out of all of the species groups considered, tambaqui had the highest possible probability of catch within the study area, with some basins exhibiting over 75% predicted probability of tambaqui being successfully caught during a fishing trip. These sub-basins were also those with the highest proportions of floodplain forest present, particularly in the sub-basins associated with the highest proportions of forest cover such as near the Juruá River and Upper Solimões River. These regions were also amongst those identified in Renó & Novo (2019) which demonstrated the most intact forests and little fragmentation over time. Quantity of tambaqui catch was also skewed to western areas with higher proportions of floodplain forest; along the mainstem of the Amazon River, catch was highest in the Upper Amazon, but when including major tributaries, predictions indicated that catch was highest in the Negro River further from the mainstem of the Amazon River. One of the limitations of this study which may affect the accuracy of this prediction is the lack of data incorporated to address the different types of rivers within the Amazon Basin which have different physical and chemical characteristics. For example, the floodplains of the Amazon River are based on a whitewater river which is rich in nutrients, whereas the floodplains of the Negro River are associated with a blackwater river which is typically acidic and low in nutrients (Venticinque et al. 2016). As such, while predictions based on floodplain habitat composition may indicate that catch of tambaqui were higher in the sub-basins of the Negro River, 91 Meaghan Rupprecht UNBC April 2024 incorporating chemical and physical attributes of rivers would improve upon these predictions to better reflect fish catch distributions in the context of the existing aquatic environments as well as the periodically available terrestrial environments. Tambaqui (C. macropomum) were once the leading species in the Central Amazon fishery and remain a highly valuable commercial species to this day (Prestes et al. 2022). However, the quantities of catch throughout the basin were relatively low compared to other species, consistent with the small proportion of total catch attributed to tambaqui (~ 8.1%). Tambaqui have been heavily exploited throughout the fishery, and drastic declines in yield have been attributed to the loss of these flooded forests in combination with gross overfishing (Ruffino 2014; Prestes et al. 2022). The results of my research further suggest that declines in the probability of catch and quantity of catch are not only associated with the loss of floodplain forests, but both have also been declining over time. The impacts of floodplain forest loss can be further associated with the patterns of forest loss and fragmentation which has been described previously by Renó & Novo (2019), which were consistent with the predicted distributions of probability and quantity of tambaqui catch. The results present a dichotomy between mapará and tambaqui, planktivore and frugivore, wherein one thrives amongst the open waters and deforested landscapes of the Lower Amazon while the other relies heavily upon the intact forests of the Upper Amazon which have yet to experience impacts on functions within the floodplain ecosystem (Renó & Novo 2019). Ultimately, my results indicate that the catch of commercially important species groups varied throughout the Amazon River Basin, and that habitat associations and distributions appeared to be unique to species with different feeding habits. Floodplain forests are integral to several commercially important species, while their absence has also been an opportunity for other 92 Meaghan Rupprecht UNBC April 2024 species, such as mapará, to rise in prominence within the fishery. Furthermore, these results suggest that some species are more vulnerable to the loss of floodplain forests than others, and certain species may in turn benefit from the removal of forests and other vegetation due to their preferred habitats and feeding habits. These relationships are vital to understanding the consequences that widespread deforestation may have on the composition of fish catch throughout the Amazon Basin, and how commercial catches may change or respond to land use change and degradation. Given the spatial variation in the probability and quantity of catch for commercially important species group, management within these contexts would be important to address regions where species may need additional protection from land use change or overexploitation. For example, using this information to identify areas where floodplain forests should be protected to support the remaining tambaqui fishery. Or to identify the disproportionate amount of fishing pressure on mapará in the Lower Amazon, perhaps suggesting that the species should be considered for additional regulation to ensure sustainable fishing practices to support such a large proportion of the Amazon fishery. Ecosystem-level management has become increasingly more important as the cumulative impacts within ecosystems, particularly in the Amazon Basin and its fishery. Yet management of these species and the basin-wide fishery are not conducted at an ecosystem-level scale and are rather focused on the smaller regional extents where research is typically conducted (Goulding et al. 2019; Castello et al. 2013). Management at larger spatial scales is particularly important considering many commercially important fish species are migratory in nature and can travel moderate and long distances into different river sub-basins or migrate laterally into the surrounding floodplains (Goulding et al. 2019). Yet the management of these species does not reflect the dynamisms of the fishery throughout the landscape, and often large-scale impacts are 93 Meaghan Rupprecht UNBC April 2024 not considered in management efforts (Castello et al. 2013). The lack of available information and research conducted at a large landscape level leaves the Amazon fishery, and many of the species which contribute to it, particularly vulnerable to landscape-level changes throughout its expanse. As such, improving upon the extent and availability of research on the Amazon fishery should be continue to be prioritized as research in this field progresses; furthermore, understanding the spatial distribution of fish catch and its relationship to the surrounding environments is imperative to ensuring a sustained Amazon River-floodplain fishery. 94 Meaghan Rupprecht 5. UNBC April 2024 References Amazon Waters Alliance. Aguas Amazonicas. [accessed 2023 Aug 1]. https://en.aguasamazonicas.org/. Arantes CC, Winemiller KO, Asher A, Castello L, Hess LL, Petrere M, Freitas CEC. 2019. Floodplain land cover affects biomass distribution of fish functional diversity in the Amazon River. Scientific Reports. 9(1). doi:https://doi.org/10.1038/s41598-019-52243-0. Arantes CC, Winemiller KO, Petrere M, Castello L, Hess LL, Freitas CEC. 2017. Relationships between forest cover and fish diversity in the Amazon River floodplain. Arlinghaus R, editor. Journal of Applied Ecology. 55(1):386–395. doi:https://doi.org/10.1111/13652664.12967.Barros et al. 2020 Barros D de F, Petrere M, Lecours V, Butturi-Gomes D, Castello L, Isaac VJ. 2020. Effects of deforestation and other environmental variables on floodplain fish catch in the Amazon. Fisheries Research. 230:105643. doi:https://doi.org/10.1016/j.fishres.2020.105643. Bartoń K. 2023. _MuMIn: Multi-Model Inference_. R package version 1.47.5, . Daniel Baston. 2022. _exactextractr: Fast Extraction from Raster Datasets using Polygons_. R package version 0.9.1, . Batista V Da S, Petrere Júnior M. 2003. Characterization of the commercial fish production landed at Manaus, Amazonas State, Brazil. Acta Amazonica. 33(1):53–66. doi:https://doi.org/10.1590/1809-4392200331066. Benedito‐Cecilio E, Araujo‐Lima CARM, Forsberg BR, Bittencourt MM, Martinelli LC. 2000. Carbon sources of Amazonian fisheries. Fisheries Management and Ecology. 7(4):305– 314. doi:https://doi.org/10.1046/j.1365-2400.2000.007004305.x. Burnham KP, Anderson DR. 2002. Model selection and multimodel inference: a practical information-theoretic approach. New York: Springer. Burnham KP, Anderson DR, Huyvaert KP. 2011. AIC model selection and multimodel inference in behavioral ecology: some background, observations, and comparisons. Behavioral Ecology and Sociobiology. 65(1):23–35. doi:https://doi.org/10.1007/s00265-010-1029-6. Castello L, Macedo MN. 2016. Large-scale degradation of Amazonian freshwater ecosystems. Global Change Biology. 22(3):990–1007. doi:https://doi.org/10.1111/gcb.13173. Castello L, McGrath DG, Hess LL, Coe MT, Lefebvre PA, Petry P, Macedo MN, Renó VF, Arantes CC. 2013. The vulnerability of Amazon freshwater ecosystems. Conservation Letters. 6(4):217–229. doi:https://doi.org/10.1111/conl.12008. Castello L, Hess LL, Thapa R, McGrath DG, Arantes CC, Renó VF, Isaac VJ. 2018. Fishery yields vary with land cover on the Amazon River floodplain. Fish and Fisheries. 19(3):431–440. doi:https://doi.org/10.1111/faf.12261. Dai A, Trenberth KE. 2002. Estimates of Freshwater Discharge from Continents: Latitudinal and Seasonal Variations. Journal of Hydrometeorology. 3(6):660–687. doi:https://doi.org/10.1175/1525-7541(2002)003%3C0660:eofdfc%3E2.0.co;2. [accessed 2021 Jan 21]. 95 Meaghan Rupprecht UNBC April 2024 Delacámara, G, O’Higgins TG, Lago M, Langhans S. 2020. Ecosystem-Based Management: Moving from Concept to Practice. In: O’Higgins, T., Lago, M., DeWitt, T. (eds) Ecosystem-Based Management, Ecosystem Services and Aquatic Biodiversity. Springer, Cham. https://doi.org/10.1007/978-3-030-45843-0_3. FAO Fisheries & Aquaculture. wwwfaoorg. [accessed 2023 Mar 25]. https://www.fao.org/fishery/en/geartype. Forsberg BR, Araujo-Lima CARM, Martinelli LA, Victoria RL, Bonassi JA. 1993. Autotrophic Carbon Sources for Fish of the Central Amazon. Ecology. 74(3):643–652. doi:https://doi.org/10.2307/1940793. Goulding M. 1980. The fishes and forest: explorations in Amazonian natural history. Berkeley: University Of California. Goulding M, Venticinque E, Ribeiro ML de B, Barthem RB, Leite RG, Forsberg B, Petry P, Lopes da Silva-Júnior U, Ferraz PS, Cañas C. 2019. Ecosystem-based management of Amazon fisheries and wetlands. Fish and Fisheries. 20(1):138–158. doi:https://doi.org/10.1111/faf.12328. Hess LL, Melack JM, Affonso AG, Barbosa C, Gastil-Buhl M, Novo EMLM. 2015. Wetlands of the Lowland Amazon Basin: Extent, Vegetative Cover, and Dual-season Inundated Area as Mapped with JERS-1 Synthetic Aperture Radar. Wetlands. 35(4):745–756. doi:https://doi.org/10.1007/s13157-015-0666-y. Hijmans R. 2023. _terra: Spatial Data Analysis_. R package version 1.7-39, . Isaac VJ, Da Silva CO, Ruffino ML. 2008. The artisanal fishery fleet of the lower Amazon. Fisheries Management and Ecology. 15(3):179–187. doi:https://doi.org/10.1111/j.13652400.2008.00599.x. Isaac VJ and Almeida MC. 2011. El consumo de pescado en la Amazonía brasileña. FAO/Copescal Documento Ocasional 13: 43. Johnson JB, Omland KS. 2004. Model selection in ecology and evolution. Trends in Ecology & Evolution. 19(2):101–108. doi:https://doi.org/10.1016/j.tree.2003.10.013. Junk W, Bayley P, Sparks R. 1989. The Flood Pulse Concept in River-Floodplain Systems. Can. Spec. Public Fish. Aquat. Sci. 106. Junk WJ, Soares MGM, Bayley PB. 2007. Freshwater fishes of the Amazon River basin: their biodiversity, fisheries, and habitats. Aquatic Ecosystem Health & Management. 10(2):153–173. doi:https://doi.org/10.1080/14634980701351023. Levin SA. 1992. The Problem of Pattern and Scale in Ecology: The Robert H. MacArthur Award Lecture. Ecology. 73(6):1943–1967. doi:https://doi.org/10.2307/1941447. Lobón-Cerviá J, Hess LL, Melack JM, Araujo-Lima CARM. 2014. The importance of forest cover for fish richness and abundance on the Amazon floodplain. Hydrobiologia. 750(1):245–255. doi:https://doi.org/10.1007/s10750-014-2040-0. 96 Meaghan Rupprecht UNBC April 2024 Lopes G. 2023. A pesca de subsistência e comercial na Amazônia brasileira [MSc Thesis]. [Instituto Nacional De Pesquisas Da Amazônia – INPA]. [accessed 2023 Aug 21]. https://repositorio.inpa.gov.br/handle/1/39140. Lorenzen K, Almeida O, Arthur R, Garaway C, Khoa SN. 2006. Aggregated yield and fishing effort in multispecies fisheries: an empirical analysis. Canadian Journal of Fisheries and Aquatic Sciences. 63(6):1334–1343. doi:https://doi.org/10.1139/f06-038. Lovell RSL, Collins S, Martin SH, Pigot AL, Phillimore AB. 2023. Space-for-time substitutions in climate change ecology and evolution. Biol Rev, 98: 2243-2270. https://doi.org/10.1111/brv.13004. McElreath R. 2020. Statistical Rethinking. CRC Press. Oliveira CB, Soares M, Martinelli LA, Moreira MZ. 2006. Carbon sources of fish in an Amazonian floodplain lake. Aquatic Sciences. 68(2):229–238. doi:https://doi.org/10.1007/s00027-006-0808-7. Pebesma E, Bivand R. 2023. Spatial Data Science: With Applications in R. Chapman and Hall/CRC. https://doi.org/10.1201/9780429459016 Pebesma E. 2018. Simple Features for R: Standardized Support for Spatial Vector Data. The R Journal 10 (1), 439-446, https://doi.org/10.32614/RJ-2018-009 Pedersen EJ, Miller DL, Simpson, Ross N. 2019. Hierarchical generalized additive models in ecology: an introduction with mgcv. PeerJ 7:e6876. Pereira DV, Arantes CC, Sousa NS, Edwar C. 2023. Relationships between fishery catch rates and land cover along a longitudinal gradient in floodplains of the Amazon River. Fisheries Research. 258:106521–106521. doi:https://doi.org/10.1016/j.fishres.2022.106521. Prestes L, Barthem R, Mello-Filho A, Anderson E, Correa SB, Couto TBD, Venticinque E, Forsberg B, Cañas C, Bentes B, et al. 2022. Proactively averting the collapse of Amazon fisheries based on three migratory flagship species. Aguirre WE, editor. PLOS ONE. 17(3):e0264490. doi:https://doi.org/10.1371/journal.pone.0264490. R Development Core Team (2008) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0. Available at http://www.R-project.org. Renó V, Novo E. 2019. Forest depletion gradient along the Amazon floodplain. Ecological Indicators. 98:409–419. doi:https://doi.org/10.1016/j.ecolind.2018.11.019. [accessed 2023 Dec 11]. Renó VF, Novo EMLM, Suemitsu C, Rennó CD, Silva TSF. 2011. Assessment of deforestation in the Lower Amazon floodplain using historical Landsat MSS/TM imagery. Remote Sensing of Environment. 115(12):3446–3456. doi:https://doi.org/10.1016/j.rse.2011.08.008. Ruffino, ML. 2014. Status and trends of fishery resources of the Amazon basin in Brazil. pp. 120. Welcomme, R.L., Valbo-Jorgensen, J. & Halls A.S. (eds). 2014. Inland fisheries evolution and management – case studies from four continents. FAO Fisheries and Aquaculture Technical Paper No. 579. Rome, FAO. 77 pp. 97 Meaghan Rupprecht UNBC April 2024 Soranno PA, Bissell EG, Cheruvelil KS, Christel ST, Collins SM, Fergus CE, Filstrup CT, Lapierre J-F, Lottig NR, Oliver SK, et al. 2015. Building a multi-scaled geospatial temporal ecology database from disparate data sources: fostering open science and data reuse. GigaScience. 4(1). doi:https://doi.org/10.1186/s13742-015-0067-4. Tennekes M. 2018. “tmap: Thematic Maps in R.” _Journal of Statistical Software_, *84*(6), 139. doi:10.18637/jss.v084.i06 . Venticinque E, Forsberg B, Barthem R, Petry P, Hess L, Mercado A, Cañas C, Montoya M, Durigan C, Goulding M. 2016. An explicit GIS-based river basin framework for aquatic ecosystem conservation in the Amazon. Earth System Science Data. 8(2):651–661. doi:https://doi.org/10.5194/essd-8-651-2016. [accessed 2021 Jul 8]. https://essd.copernicus.org/articles/8/651/2016/. Wood S, Scheipl F. 2020. _gamm4: Generalized Additive Mixed Models using 'mgcv' and 'lme4'_. R package version 0.2-6, . Wong, 2011. SAGE research methods. The SAGE handbook of spatial analysis. Zuur A, Ieno EN, Walker N, Saveliev AA, Smith GM, Springerlink (Online Service). 2009. Mixed Effects Models and Extensions in Ecology with R. New York, Ny: Springer New York. Zuur A, Ieno EN, Elphick CS. 2010. A protocol for data exploration to avoid common statistical problems. Methods in Ecology and Evolution. 1(1):3-14. 98 Meaghan Rupprecht UNBC April 2024 APPENDIX A- BOAT ID SENSITIVITY ANALYSIS Prior to running the models, a sensitivity analysis was conducted on the methods used for identifying individual boats within the data. Information associated with unique boat identification (i.e., name, length, and landing municipality) was used in combination to determine which method captured the most variability for a random factor variable of Boat ID within our model. Each method was used within the global model structure, and the standard deviation and variances of the random factors (i.e., boat ID, landing municipality, and river basin ID) and residuals were compared. Results indicated that the third method, identifying individual boats as a function of boat name, length, and landing municipality, captured more variability within the data, and reduced the variance of residuals the most. Boat ID’s generated with this method were therefore used throughout the models as a random factor. Table 15. Variance of the Boat ID random factor and residuals are shown according to different methods used to create unique Boat IDs when fitting a global model for fish catch. The Boat ID factor captured the most variance and minimized variance in the residuals best when a unique combination of boat name, length, and landing municipality were applied to create the Boat ID. Boat_ID Method 1 2 3 Variable Combinations boat_name boat_name, boat_length boat_name, boat_length, landing_mun Boat_ID Variance 0.2837 0.3248 Boat_ID Std.Dev. 0.5326 0.5699 Residual Var 0.4755 0.4620 Residual St. Dev 0.6895 0.6797 0.3597 0.5998 0.4418 0.6647 99 Meaghan Rupprecht UNBC April 2024 APPENDIX B- MODELS Table 16. Fifteen models were fit to fish catch data using a combination of variables to assess which model best fit out-of sample data. Fixed effects included s(e) as the smooth for effort, s(m) s the smooth for month, s(y) as the smooth for year, s(h) as the smooth for the landscape variable assessed (either habitat proportion or population density), and s(X,Y) as the spatial smooth for the approximate coordinates of fishing location. All models were fit with the same random effects for unique boat ID, unique basin ID, and the landing municipality of catch. The landscape variable of interest is described for each model as well. 1 2 f(e) + f(m) + f(y) + f(X,Y) f(e) + f(m) + f(y) + f(h) Aquatic Macrophyte Random Effects γbid, γbasin, γmun γbid, γbasin, γmun 3 4 5 f(e) + f(m) + f(y) + f(h) f(e) + f(m) + f(y) + f(h) f(e) + f(m) + f(y) + f(h) Bare/Herbaceous Forest Open Water γbid, γbasin, γmun γbid, γbasin, γmun γbid, γbasin, γmun 6 7 8 9 f(e) + f(m) + f(y) + f(h) f(e) + f(m) + f(y) + f(h) + f(X,Y) f(e) + f(m) + f(y) + f(h) + f(X,Y) f(e) + f(m) + f(y) + f(h) + f(X,Y) Shrub Aquatic Macrophyte Bare/Herbaceous Forest γbid, γbasin, γmun γbid, γbasin, γmun γbid, γbasin, γmun γbid, γbasin, γmun 10 11 12 f(e) + f(m) + f(y) + f(h) + f(X,Y) f(e) + f(m) + f(y) + f(h) + f(X,Y) f(e) + f(m) + f(y) + f(h) + f(X,Y) Open Water Shrub Population Density γbid, γbasin, γmun γbid, γbasin, γmun γbid, γbasin, γmun 13 14 15 f(e) + f(m) + f(y) + f(h) f(e) + f(m) f(e) + f(y) Population Density - γbid, γbasin, γmun γbid, γbasin, γmun γbid, γbasin, γmun Model Fixed Effects Landscape Variable (h) 100 Meaghan Rupprecht UNBC April 2024 APPENDIX C- FISH CATCH COMPOSITION Table 17. Fish catch by species indicated the presence of 41 species groups within the data. The highest proportion of fish catch between 1991 and 2011 belonged to mapará, which accounted for a quarter of all fish catch by species. Species Group Sum of Catch (kg) Proportion of Catch Cumulative Sum of Proportions Mapara Jaraqui Curimata Tambaqui Pacu Dourada Pescada Aruana Surubim Moela Aracu Tucunare Sardinha Pirapitinga Matrincha Acari Filhote Piramutaba Acara Salada Cujuba/cuiu-cuiu Apapa Other Branquinha Pirarara Bacu Jau/pacamum Piranha Pirarucu Piracatinga Charuto Piranambu Arraia Traira Tamoata 6146881.70 2888303.70 2013430.40 1999679.30 1404173.70 1221408.00 1210013.60 968982.90 856712.60 750607.00 673113.60 671582.20 617252.60 468678.10 465041.40 418979.68 350126.10 283205.00 216959.00 197236.00 136250.70 121445.00 103037.60 69897.20 67002.90 60129.31 45821.50 39696.80 38794.00 36267.10 16411.70 9144.00 7532.00 5159.00 3621.07 0.25 0.1175 0.0819 0.0813 0.0571 0.0497 0.0492 0.0394 0.0348 0.0305 0.0274 0.0273 0.0251 0.0191 0.0189 0.017 0.0142 0.0115 0.0088 0.008 0.0055 0.0049 0.0042 0.0028 0.0027 0.0024 0.0019 0.0016 0.0016 0.0015 0.0007 0.0004 0.0003 0.0002 0.0001 0.25 0.3675 0.4494 0.5307 0.5878 0.6375 0.6867 0.7261 0.7609 0.7914 0.8188 0.8461 0.8712 0.8903 0.9092 0.9262 0.9404 0.9519 0.9607 0.9687 0.9742 0.9791 0.9833 0.9861 0.9888 0.9912 0.9931 0.9947 0.9963 0.9978 0.9985 0.9989 0.9992 0.9994 0.9995 101 Meaghan Rupprecht Jandia Camarao Cara-de-gato Peixe-cachorro Jeju Espadarte UNBC 1147.00 999.00 941.00 96.00 59.00 1.00 April 2024 0 0 0 0 0 0 102 0.9995 0.9995 0.9995 0.9995 0.9995 0.9995 UNBC April 2024 acara-bicudo acara-boca-de-juquia acara-cascudo acara-disco acara-prata acara-roxo acara-tinga bararua Acara Acara Acara Acara Acara Acara Acara Acara Acara Acara Acara Acara - acara-acu acara-bicudo Acara Acara Acara Acara Acara Acara ProVarzea- Common Name acara acara acara acara Species Group ProVarzea-Scientific Name Acarichthys heckelli Caquetaia spectabilis Aequidens Cichlasoma amazonarum Astronotus crassipinis Satanoperca acuticeps Satanoperca jurupari Acaronia nassa Chaetobranchopsis orbicularis Symphysodon aequifasciatus Chatobranchus flavescens Heros sp Geophagus proximus Uaru amphiacanthoides - 103 acara-rosado - acara-roxo acaratinga - - - - acara-acu - - IARA cara-branco cara-preto cara-rosado - cara-roxo cara-tinga - - - - - Manaus (Vandick) cara - - - - - - - Manaus (Bayley) aca - - - - - - acara-bararua acara-roxo acara-acu acara-(outros) - Mamiraua Table 18. The lookup table for species used Species as the grouping variable. The names of species recorded in all datasets (i.e., ProVarzea, IARA, Manaus (Vandick), Manaus (Bayley), and Mamiraua) were then classified according to the species group they belonged to. These classifications were used to standardize records of fish catch for further analysis. APPENDIX D- SPECIES LOOKUP TABLE Meaghan Rupprecht apapa/sardinhao apapa-amarelo apapa-branco aracu aracu cabeca-gorda aracu cabeca-gorda aracu comum aracu comum aracu pau-de-vaqueiro aracu pau-de-vaqueiro aracu-flamengo aracu-flamengo - acari-bodo acari-pedra arari arraia arraia arraia arraia Apapa Apapa Apapa Aracu Aracu Aracu Aracu Aracu Aracu Aracu Aracu Aracu Aracu Aracu Arari Arraia Arraia Arraia Arraia - Acara Acari Acari Meaghan Rupprecht Leporinus trifasciatus Schizodon fasciatum Schizodon vittatum Rhytiodus argenteofuscus Rhytiodus microlepis Leporinus aff. affinis Leporinus fasciatus Chalceus spp. Potamotrigon constellata Potamotrygon aff. hystrix Potamotrygon motoro Potamotrygon scobina Pellona castelnaeana Pellona flavipinnis Anostomoides laticeps Leporinus friderici Liposarcus pardalis Hypostomus emarginatus Ilisha amazonica UNBC 104 - - - aracu-amarelo arraia - apapa-amarelo apapa-branco aracucabeca_gorda aracu-comum - - acari-bodo - - - - aracu-piau arraia - apapa-amarelo aracus aracu-cabecagorda aracu-comum - apapas - April 2024 ara apa - - - - - - - - - - - - - - aracu-comum - apapa-ou-sardinhao-ousarda aracu aracu-cabeca-gorda acara-tucunare acari-bodo - aruana babao/barba-chata bacu bacu/rebeca bacu-pedra rebeca/bacu bico-de-pato braco-de-moca braco-de-moca branquinha branquinha branquinha branquinha cascuda branquinha cascuda branquinha cascuda branquinha-cabeca-lisa branquinha-comum Aruana Babao Bacu Bacu Bacu Bacu Bacu Bico-de-pato Braco-de-moca Braco-de-moca Branquinha Branquinha Branquinha Branquinha Branquinha Branquinha Branquinha Branquinha Meaghan Rupprecht Osteoglossum bicirrhosum Goslinia platynema Pterodoras lentiginosus Platydoras costatus Lithodoras dorsalis Megalodoras uranoscopus Surubim lima Hemisorubim platyrhynchos Platystosmatichthys sturio Curimata inornata Steindachneria cf. bimaculata Cyphocharax abramoides Psectrogaster amazonica Caenotropus labyrinthicus Psectrogaster rutiloides Potamorhina altamazonica Potamorhina latior UNBC - - - 105 branquinhacabeca-lisa branquinhacomum - branquinhacascuda - - - - bacu-liso - - bacu-pedra aruana - - - - - - - branquinhacabeca-lisa branquinhacomum - branquinhacascuda - - branquinha - aruana April 2024 - bra aru - - - - - - - - - - - - - - - - branquinha-comum-ouchorona branquinha-cabeca-lisa - - branquinha-cascuda - - - bacu-liso bico-de-pato - - bacu-pedra aruana aviun camarao camarao ag. doce cara-de-gato charuto charuto charuto charuto charuto cubiu/charuto cujuba/ cuiu-cuiu curimata dourada espardate barbado filhote/piraiba filhote/piraiba indet jacunda jacunda jacunda jandia jaraqui-escama-fina Camarao Camarao Camarao Cara-de-gato Charuto Charuto Charuto Charuto Charuto Charuto Cujuba/ cuiu-cuiu Curimata Dourada Espadarte Filhote Filhote Filhote Indet Jacunda Jacunda Jacunda Jandia Jaraqui Meaghan Rupprecht Prochilodus nigricans Brachyplatystoma rouseauxii Pristis spp. Goslinia platynema Brachyplatystoma filamentosum Indeterminada Crenicichla sp Crenicichla aff. ornata Crenicichla reticulata Leiarius marmoratus Semaprochilodus teanurus Platynematichtys notatus Hemiodus microlepis Hemiodus ocellatus Hemiodus unimaculatus Hemiodus immaculatus Hemiodus sp. Anodus melanopogon Oxydoras niger UNBC - - - 106 jandia jaraqui-fina - - espadarte barbado filhote curimata dourada cujuba charuto aviun camarao cara-de-gato - - - jandia jaraqui-fina - piraiba indeterminado - barbado filhote - curimata dourada cubiu cuiu-cuiu charuto cara-de-gato aviun April 2024 - - pirai - fil cur dou cub cui - - - - - - - jandia jaraqui-escama-fina filhote curimata dourada cubiu cuiu-cuiu - - - camarao carataui/cara-de-gato mandi mandi mandi mandi/cachorro-depadre mandi/mandi-peruano mandube mandube mandube mapara mapara mapara jatuarana/ matrincha jatuarana/ matrincha moela/fura-calca outros outros marinhos - Jaraqui Jau/pacamum Jeju Mandi Mandi Mandi Mandi/peruano/ cp Mandi/peruano/ cp Mandube Mandube Mandube Mapara Mapara Mapara Matrincha Other Other Other Moela Matrincha jaraqui-escamagrossa jau/pacamum jeju Jaraqui Meaghan Rupprecht Pimelodina flavipinnis - - Semaprochilodus insignis Zungaro zungaro Hoplerythrinus unitaeniatus Pimelodus cf. altipinnis Pimelodus blochii Platysilurus barbatus Parauchenipterus galeatus Auchenipterus nuchalis Ageneiosus dentatus Ageneiosus aff. ucayalensis Ageneiosus brevifilis Hypophtalmus Hypophtalmus marginatus Hypophtalmus fimbriatus Brycon amazonicus UNBC - - 107 - fura-calca matrincha/jatuar ana - mapara - - - bodeco - - jatuarana matrinxa mapara - - - - - - mandi jaraquis jau jaraqui-grossa - mandube - mandi jau jeju - jaraqui-grossa April 2024 - - - - - rem jat mat - - - map - jar jau - jatuarana matrincha mapara mandube mandi - - - - - - - - jaraqui pacamum-(doce)-ou-jau jeju jaraqui-escama-grossa Pacu Pacu Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Other Pacu Pacu Pacu pacu-branco pacu-jumento pacu-manteiga/ pacu-comum pacu-manteiga/ pacu-comum pacu-marreca Meaghan Rupprecht Mylossoma duriventre Metynnis argenteus Myleus torquatus Myleus schomburgki Mylossoma aureum UNBC 108 pacu-marreca - outros puraque peixe-boi sauna pacu-jumento pacu-comum - - caratai poraque bodo bodo-de-praia saia-suja iaca tracaja nao identificado peixe-liso carau-acu pacu-jumento pacu-comum - April 2024 len lis acu bod - - - - - - - - pirabutao pirabutao-branco pirabutao-melado pirabutao-pintado bacurua orana peixe-agulha pacu-comum - pacu-marreca pacu-piranha peixe-cachorro peixe-cachorro peixe-cachorro peixe-cachorro saranha/peixe-cachorro curvina curvina pescada pescada pescada pescada pescada preta piracatinga piramutaba piranambu Pacu Pacu Pacu Pacu Pacu Peixe-cachorro Peixe-cachorro Peixe-cachorro Peixe-cachorro Peixe-cachorro Pescada Pescada Pescada Pescada Pescada Pescada Pescada Piracatinga Piramutaba Piranambu Meaghan Rupprecht Metynnis hypsauchen Catoprion mento Acestroryncus falcatus Acestrorhyncus falcirostris Rhaphiodon vulpinus Hydrolycus scomberoides Cynodon gibus Pachypops trifilis Pachypops furchraeus Plagioscion squamosissioums Plagioscion aff. surinamensis Plagioscion sp. "monte alegre" Plagioscion sp. "montei" Plagioscion auratus Calophysus macropterus Brachyplathystoma vaillantii Pinirampus UNBC - 109 piranambu piramutaba pescada-preta piracatinga - - - pescada saranha - - - pacu-olhudo peixe-cachorro - - - - - piranambu piramutaba piracatinga pescada saranha curvina - - - pacu-galo peixe-cachorro - April 2024 - - - - - - - - - - piram pir pes pac - - - - - piranambu-ou- piramutaba piracatinga/mota pescada - - - - peixe-cachorro pacu pacu-galo - - pirarucu salada sardinha comprida sardinha comum sardinha papuda peixe lenha surubim Pirarucu Pirarucu Pirarucu Salada Sardinha Sardinha Sardinha Sardinha Sardinha Surubim Surubim - pirarara piranha-preta Piranha Pirarara piranha-comprida Piranha pirapitinga piranha-branca piranha-caju Piranha Piranha - piranha-branca Piranha Piranha Piranha Piranha Pirapitinga piranha-amarela Piranha Meaghan Rupprecht pirinambu Serrasalmus spilopleura Serrasalmus aff. eigenmanni Serrasalmus calmoni Pygocentrus nattereri Serrasalmus elongatus Serrasalmus rhombeus Piaractus brachypomus Phractocephalus hemioliopterus Arapaima gigas Triportheus elongatus Triportheus albus Triportheus flavus Sorubimichthys Merodontus tigrinus UNBC 110 pirarucu salada sardinhacomprida sardinha-papuda - pirarara piranha-mafura pirapitinga piranha-preta - piranha-caju - - sardinhacumprida sardinha-comum sardinha-papuda sardinhas sardinha-chata peixe-lenha surubim pirarucu pirarara piranhas pirapitinga piranha-preta - piranha-caju - - April 2024 sur sar - - - pirar piran pirap - - - - - - sardinha sardinha-chata peixe-lenha-ou-pirauaca surubim - pirarucu pirarucu-(salmorado) pirarucu-(seco) salada sardinha-comprida pirarara piranha-xidaua pirapitinga piranha-preta - piranha-caju barba-chata surubim-lenha Surubim Tucunare Tucunare Zebra/ flamengo Ze-do-o - ze-do-o traira tucunare tucunare/ tucunare-pinima tucunare-acu zebra/flamengo tamoata Tambaqui Tambaqui Tambaqui Tambaqui Tambaqui Tambaqui Tamoata Traira Tucunare Tucunare tambaqui Tambaqui - surubim-flamengo Surubim Surubim surubim/caparari Surubim Meaghan Rupprecht Cichla monoculus Brachyplatystoma juruensis Roeboides myersi Colossoma macropomum Hoplosternum litoralle Hoplias malabaricus Cichla sp Cichla temensis Pseudoplatystoma tigrinum Brachyplatystoma juruense Sorubimichthys planiceps - UNBC 111 - tucunare-acu tucunare-tatu - tucunare-pinima traira surubimlenha/canela surubimpintado/tigre tambaquiamarelo/preto tamuata - - - - - tucunare-acu - traira tucunare - tamoata - tambaqui surubim-tigre caparari April 2024 tra tuc tam cap - - - - - - - - - - traira tucunare - - - tambaqui-(salgado) tambaqui-boco tambaqui-file tambaqui-medida tambaqui-ruelo tambaqui-sirico tamoata tambaqui caparari