42-12 Flow-Connectivity Improves Stream Temperature Predictions in a Karst Watershed: A Spatial Stream Network Approach

Abstract:

Stream temperature drives chemical and biological processes, so accurate predictive models are essential for stream-network management. However, karst systems are challenging to model because cool-spring groundwater discharge affects downstream surface water via flow connectivity. Previous regression and Euclidean modeling techniques do not explicitly represent this directional thermal transport. This study applies a moving-average, flow-connected stream network model to Spring Creek, a mixed-land-use karst watershed in central PA, and compares its performance with Euclidean and nonspatial regression approaches under high thermal variability. Initial models use long-term temperature observations and available site and watershed-level predictors, but ongoing model development will evaluate additional hydrologic, groundwater, and landscape controls. Among the models evaluated to date, the tail-up model had the strongest fit and predictive performance, reducing cross-validation prediction error by ~25% and ~20% relative to the nonspatial and Euclidean models, respectively. These results highlight the value of incorporating flow-connected thermal transport when predicting thermal conditions in karst stream networks and provide a foundation for identifying thermally important reaches for targeted monitoring and management.

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.