Predicting native fish diversity in river Bharathapuzha: Non-linear responses to environmental drivers revealed by Generalised Additive Models

Predicting how biodiversity responds to environmental changes is crucial for effective conservation; however, the functional relationships underpinning diversity metrics often remain obscure. In this study, we developed a robust predictive model for Shannon diversity in fish in the Bharathapuzha River, Kerala, India, a system experiencing significant anthropogenic pressure. Utilising a comprehensive dataset (N=108), we applied Generalised Additive Models (GAMs) to predict Shannon diversity in native fish, a metric chosen for its sensitivity to community evenness in a system impacted by dominant invasive species. Our final model demonstrated an Adjusted R² of 0.171, a deviance explained by 20.7%, and was validated using 10-fold cross-validation (CV-R² = 0.138). Focusing on native species diversity, the model revealed a complex bimodal response to environmental drivers. Native fish diversity exhibited a significant nonlinear relationship with dissolved oxygen and nitrate, reflecting a fundamental longitudinal phase shift from sensitive upstream specialists to tolerant downstream generalists. These findings provide critical management insights; while habitat restoration offers direct, proportional gains in diversity, water quality management requires targeting specific ecological optima to maximise biodiversity benefits. This study underscores the utility of GAMs for developing mechanistically informative predictive models that are essential for guiding effective global river restoration and conservation. Keywords: Fish community, Bioassessment, GAM, Predictive modelling, Water quality