Nikos Papanikolau

Hybrid models that combine machine learning (ML) with physics-based approaches are increasingly used in Earth System Science (ESS) to balance predictive accuracy with physical generalizability. However, they face challenges such as equifinality, biases under distribution shifts, and limited adaptability. Structural Causal Models (SCMs) provide a principled framework to integrate physical knowledge with data-driven components while enabling explicit modeling of causal effects.

This project will develop causal hybrid models to study how land surface properties and land-management practices influence low-level cloud formation. By embedding hybrid ESS models into the formalism of SCMs, we aim to disentangle causal drivers from correlations, account for confounding effects, and mitigate selection biases in observational data. This approach also allows for counterfactual and intervention analysis, quantifying how changes in land use or vegetation might impact cloud dynamics.

The ultimate goal is to produce robust, versatile, and policy-relevant hybrid models that advance our understanding of land-atmosphere interactions and their role in climate regulation.