Machine learning in stream and river water temperature modeling: a review and metrics for evaluation

Abstract

As climate change continues to affect stream and river (henceforth stream) systems worldwide, stream water temperature (SWT) is an increasingly important indicator of distribution patterns and mortality rates among fish, amphibians, and macroinvertebrates. Technological advances tracing back to the mid-20th century have improved our ability to measure SWT at varying spatial and temporal resolutions for the fundamental goal of better understanding stream function and ensuring ecosystem health. Despite significant advances, there continue to be numerous stream reaches, stream segments, and entire catchments that are difficult to access for a myriad of reasons, including but not limited to physical limitations. Moreover, there are noted access issues, financial constraints, and temporal and spatial inconsistencies or failures with in situ instrumentation. Over the last few decades and in response to these limitations, statistical methods and physically based computer models have been steadily employed to examine SWT dynamics and controls. Most recently, the use of artificial intelligence, specifically machine learning (ML) algorithms, has garnered significant attention and utility in hydrologic sciences, specifically as a novel tool to learn undiscovered patterns from complex data and try to fill data streams and knowledge gaps. Our review found that in the recent 5 years (2020–2024), more studies using ML for SWT were published than in the previous 20 years (2000–2019), totaling 57. The aim of this work is threefold: first, to provide a concise review of the use of ML algorithms in SWT modeling and prediction; second, to review ML performance evaluation metrics as they pertain to SWT modeling and prediction to find the commonly used metrics and suggest guidelines for easier comparison of ML performance across SWT studies; and, third, to examine how ML use in SWT modeling has enhanced our understanding of spatial and temporal patterns of SWT and examine where progress is still needed.