Identifying trustworthiness challenges in deep learning models for continental-scale water quality prediction

Water quality is foundational to environmental sustainability, ecosystem resilience, and public health. Deep learning offers transformative potential for large-scale water quality prediction and scientific insights generation. However, their widespread adoption in high-stakes operational decision-making, such as pollution mitigation and equitable resource allocation, is prevented by unresolved trustworthiness challenges, including performance disparity, robustness, uncertainty, interpretability, generalizability, and reproducibility. In this work, we present a multi-dimensional, quantitative evaluation of trustworthiness benchmarking three state-of-the-art deep learning architectures: recurrent (LSTM), operator-learning (DeepONet), and transformer-based (Informer), trained on 37 years of data from 482 US basins to predict 20 water quality variables. Our investigation reveals systematic performance disparities tied to process complexity, data availability, and basin heterogeneity. Management-critical variables remain the least predictable and most uncertain. Robustness tests reveal pronounced sensitivity to outliers and corrupted targets; notably, the architecture with the strongest baseline performance (LSTM) proves most vulnerable under data corruption. Attribution analyses align for simple variables but diverge for nutrients, underscoring the need for multi-method interpretability. Spatial generalization to ungauged basins remains poor across all models. This work serves as a timely call to action for advancing trustworthy data-driven methods for water resources management and provides a pathway to offering critical insights for researchers, decision-makers, and practitioners seeking to leverage artificial intelligence (AI) responsibly in environmental management. Broader context: Water quality is a critical environmental and public health concern, with increasing pollution threatening ecosystems, drinking water supplies, and biodiversity. This study addresses a pressing challenge in environmental science: building trustworthy AI models that can reliably predict water quality at large scales. Current deep learning approaches focus on prediction accuracy, often lacking analysis of performance disparity, robustness, and interpretability, which are key elements needed for real-world decision-making in areas such as pollution control and resource management. By conducting a multi-dimensional quantitative evaluation of trustworthiness in continental-scale water quality prediction models, this work highlights key limitations and proposes actionable recommendations to address them.The research sets the stage for future advancements in AI-driven environmental monitoring, emphasizing the need for interdisciplinary collaboration among data scientists, hydrologists, ecologists, and policymakers. The findings offer insights into developing more reliable and responsible AI tools, enabling proactive water management under growing climate and socioeconomic pressures. In the long term, this work could support smarter data-driven policies and technologies that enhance water security and sustainability, critical steps toward safeguarding one of Earth’s most vital resources.