Authors: Claudia Corona – Colorado School of Mines
Title: Evaluation of Machine Learning Applications and Assessment Metrics in Stream Water Temperatures Models
Presentation Type: Lightning Talk
Abstract: Stream water temperature (SWT) is an increasingly important indicator of ecosystem health and function. Advances in science and technology have improved our ability to measure SWT, but there remain numerous stream reaches and entire catchments that are difficult to observe. Moreover, access issues, financial constraints, and temporal and spatial inconsistencies or failures with in-situ instrumentation warrant alternative solutions.
In response to these limitations, statistical methods and physically-based computer models have been increasingly employed to examine SWT dynamics and controls. Recently, the use of machine learning (ML) algorithms has garnered interest in hydrology, specifically to identify unknown patterns from complex data and attempt to fill data streams and knowledge gaps. A review of publications over the last 25 years, focusing on SWT modeling, identified a similar number (~27) of publications utilizing ML in the last 5 years (2020-2024) as in the previous 19 years (2000-2019).
The objective of this talk is to showcase our review of ML performance evaluation metrics as it pertains to SWT modeling, where we identify RMSE, R2, and NSE as the most commonly used metrics and suggest guidelines for comparisons. We also outline the need for a more consistent standard of comparison in SWT modeling, which we call “Temporal, Unseen, Ungaged Region Tests” (TUURTs), where temporal and spatially focused testing can strengthen model robustness and transferability.