Stream water temperature (SWT) is a key indicator of water quality with implications for public use and health, ecosystem function, and aquatic life survival. Advances in modeling have helped improve our understanding of SWT dynamics, but challenges remain. Recently, machine learning (ML) has been used in SWT modeling, but questions remain about how the use of ML improves insight into SWT causes and effects. This work reviews ML-SWT modeling studies (and metrics) and considers future directions.

Global soil information introduces uncertainty into models that simulate soil hydrothermal changes. Using the Noah with Multiparameterization (Noah-MP) model with two different global soil datasets, we find under-represented soil properties in wet loam, causing a dry bias in soil moisture. This bias is more pronounced and drought categories are more severe in the SoilGrids dataset. We conclude that models should incorporate detailed, region-specific soil information to minimize model uncertainties.

The data-driven approach enhances soil water content measurements by improving the precision of cosmic ray neutron sensors (CRNSs). This study demonstrates a new method to account for dynamics in air pressure, atmospheric humidity, and incoming neutron intensity. Soil water content measured by CRNSs showed reduced errors compared to reference measurements. The findings highlight the need for precise adjustments to better measure soil moisture for agricultural, water, and climate monitoring.