AI-Enhanced Science

AI has been transformative to RC-SFA science, accelerating our ability to understand complex hydrobiogeochemical processes across scales.   

Examples of AI Applications from the RC-SFA 

  • We leverage AI to enhance multiple aspects of our watershed science in the following ways:  
    • Using industry-developed geospatial foundation models to create computationally efficient surrogates for watershed flow and transport processes 
    • Integrating AI into process models to decrease bias and reveal unknown processes 
    • Taking advantage of explainable AI to reveal influential processes governing watershed responses to perturbation 
    • Employing object-detection models to quantify streambed characteristics from smartphone and drone imagery to enable AI-learned scale closure functions for multiscale modeling 
    • Training AI algorithms to monitor stream intermittency via automated imaging 
    • Using AI-based image fusion techniques to detect surface water in narrow streams. 
  • Our AI-ready data publishing practices ensure that our datasets are optimized for AI applications, facilitating model development and enabling rapid ModEx (Model–Experiment) iteration.  

Examples of AI-Advantage Impacts from the RC-SFA 

Graphic showing the approach for ICOM ModEx
Figure from Malhotra et al. 2025
  1. AI grain size mapping is 100,000 times faster than traditional methods and deepens understanding (Chen et al., 2026. High-throughput AI video surveys enable reproducible multiscale sediment size mapping, with implications for hydrobiogeochemical parameterization. Water Resources Research, Submitted). 
  2. AI-guided CONUS-wide ModEx increases AI model performance by 10-fold (Malhotra et al., 2025. Harnessing artificial intelligence to automate environmental predictions. In review; bioRxiv, https://doi.org/10.1101/2025.11.06.684583). 
  3. Embedding AI in physics model reduces bias >60% and reveals biogeochemical controls (Zheng et al., 2026. Continental-scale controls on hyporheic respiration revealed by knowledge-guided machine learning. In review). 
  4. Fine-tuning time series foundation models can double the value of field-collected data (Niroula et al., 2026. Increasing the value of collected data with Time Series Foundation Models. In review). 

The Future of AI-Enhanced Science 

Moving forward, we will continue to integrate AI with process-based models in hybrid modeling frameworks that maintain and provide mechanistic understanding while opening science opportunities that could not be pursued without AI. This integration allows us to robustly test hypotheses about how wildfire, variable inundation, and biophysical setting influence watershed hydrobiogeochemistry. The knowledge that is unlocked via use of AI ultimately improves our capacity to predict environmental responses (e.g., altered water availability) to perturbations across the hillslope-to-stream continuum.