In this study, we used satellite-derived soil moisture maps from SMAP-HydroBlocks to explore whether patterns of wet and dry areas can help explain how nitrate is transported from land to streams. We trained a deep learning model using data from nine watersheds across the United States, combining information on streamflow, soil moisture patterns, topography, and land use, including agriculture and urban areas.

The model predicted daily nitrate levels with high accuracy and showed that spatial patterns in soil moisture play a key role in explaining nitrate movement. Using explainable AI, we found that these patterns were especially important during wet periods, and that near-stream zones were the most informative areas in the catchments. Our results take a step forward in identifying where within a watershed soil moisture dynamics are most strongly related to nitrate transport, offering new tools to support water quality management.