Exploring continental-scale relationships between stream temperature signatures and watershed characteristics

Water temperature is vitally important for the health of rivers and streams, influencing the quality and integrity of habitats and, in turn, ecosystems and the services they provide. While climate has been regarded traditionally as the primary organizing driver of stream temperature regimes, other factors such as hydrology, watershed structure, and human impacts modify water temperature signatures. To better understand how and why stream temperatures vary both spatially and temporally, we sought to disentangle the underlying drivers of thermal signatures, including maximum temperature, monthly range, and thermal sensitivity, using a random forest machine learning approach. Random forest is an analytical technique that leverages an ensemble of classification and regression trees to model target values using a group of predictor variables. We applied a set of random forest models to identify the primary drivers of various stream temperature signatures at 412 sites spanning the conterminous United States. For our analysis, we selected 26 relevant predictor variables from the GAGES-II database of watershed characteristics (Falcone, 2011) and grouped these variables into four key categories: climate, human impacts in terms of dams and land use, hydrology, and watershed structure. The results of our random forest models revealed seasonal and spatial variability in the relative importance of watershed predictor variables to stream temperature signatures. At continental scales, the presence of dams had a strong organizing influence on thermal signatures such as maximum water temperature and thermal sensitivity, though the importance of dams differed between streams in the eastern and western US. Beyond climate, characteristics related to local hydrology (including streamflow and groundwater contributions) were particularly important predictors across all stream temperature signatures, emphasizing that hydrology may mediate the role of climate in organizing thermal regimes. Our work demonstrates that continental US water temperatures are not solely organized by climate response and are influenced by hydrological processes and landscape modification by humans.