Subprojects

integrates data-driven learning with physics-based models, addressing generalization issues in deep learning (DL) by enforcing physical laws.

identifies cause-effect relationships, mitigating ML limitations in distribution shifts and interpretability, but methodological challenges remain.

enhances interpretability by uncovering symbolic, mathematical relationships in data, incorporating physical constraints for better generalization.

maps high-dimensional data to a lower-dimensional space, improving efficiency and interpretability; adaptation to spatial time-series is needed.

assesses model confidence, crucial for early-warning systems. It ensures that models recognize when predictions require further fine-tuning.

develops a hybrid physics-AI model that combines neural networks with differentiable hydrologic equations for runoff generation and flow routing.

develops hybrid AI-physics models that encode competing physical constraints and quantify predictive uncertainties across spatial and temporal scales.

generates an ensemble of AI-powered phenological models, comprising new model formulations through equation discovery, new causal model representations, and physically-consistent hybrid model setups coupled in Earth system models.

couples process-based and DL models to resolve spatial controls on vegetation-process parameterizations for ecosystem fluxes.