Search

Climate change and anthropogenic activities have impacts on fish habitat suitability, demanding more accurate modeling of species abundance for effective conservation and management. In this study, we applied Machine Learning techniques to model the habitat-abundance relationship of juvenile Atlantic salmon (Salmo salar) in the Teno catchment in Finland and Norway. To capture the complexity and nonlinearity of the habitat-abundance relationship, we employed Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GB), and Support Vector Classification (SVC) and compared their performances. Among the regression models considered, those incorporating input variables such as substrate, shade, and vegetation demonstrate higher performance. Support Vector Regression yields the highest mean cross-validation score (R2 = 0.58), and Gradient Boosting produces the highest test score (R2 = 0.6) among the regression techniques. The mean cross-validation and test scores obtained for the classification models are notably higher compared to the regression models across all scenarios. A comparison between regression and classification results highlights the challenges of accurately modeling the habitat-abundance relationship. This study provides insights into the challenges and potential of machine learning techniques for juvenile Atlantic salmon habitat-abundance modeling in complex riverine habitat environments. The findings emphasize the importance of considering the limitations of machine learning models, particularly in ecological contexts, and the need for further research to address temporal variations and improve the precision of habitat-abundance modeling.

• ML models recognize complex salmon habitat-abundance relationship, using small data. • Regression models struggle with intricate relationships in habitat data. • Models with substrates, shades, and vegetation variables offer higher accuracy. • Support vector classification excels in modeling accuracy. • The study highlights ML challenges in salmon habitat-abundance modeling. Climate change and anthropogenic activities have impacts on fish habitat suitability, demanding more accurate modeling of species abundance for effective conservation and management. In this study, we applied Machine Learning techniques to model the habitat-abundance relationship of juvenile Atlantic salmon (Salmo salar) in the Teno catchment in Finland and Norway. To capture the complexity and nonlinearity of the habitat-abundance relationship, we employed Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GB), and Support Vector Classification (SVC) and compared their performances. Among the regression models considered, those incorporating input variables such as substrate, shade, and vegetation demonstrate higher performance. Support Vector Regression yields the highest mean cross-validation score (R2 = 0.58), and Gradient Boosting produces the highest test score (R2 = 0.6) among the regression techniques. The mean cross-validation and test scores obtained for the classification models are notably higher compared to the regression models across all scenarios. A comparison between regression and classification results highlights the challenges of accurately modeling the habitat-abundance relationship. This study provides insights into the challenges and potential of machine learning techniques for juvenile Atlantic salmon habitat-abundance modeling in complex riverine habitat environments. The findings emphasize the importance of considering the limitations of machine learning models, particularly in ecological contexts, and the need for further research to address temporal variations and improve the precision of habitat-abundance modeling. [ABSTRACT FROM AUTHOR]