Pappu Paul

Climate models exhibit substantial differences in their sensitivity to increased greenhouse gas concentrations, arising in part from uncertainties in atmospheric and surface feedback processes. My research aims to better understand and reduce these uncertainties by investigating how surface warming patterns interact with atmospheric circulation, clouds, and top-of-atmosphere radiation. I use climate model experiments to trace these coupled feedbacks and determine how the spatial pattern of surface warming influences the Earth's radiative response.

A major component of my recent work focuses on parameter optimization and uncertainty quantification in climate models. I have developed Bayesian approaches to constrain uncertain model parameters and characterize parameter correlations and equifinality, with the goal of distinguishing robust physical constraints from compensating parameter combinations. This work provides a framework for understanding how parametric uncertainty contributes to uncertainty in simulated climate feedbacks.

I am also developing machine-learning approaches to improve Earth system model physics. In particular, I use high-fidelity microphysical simulations to develop neural-network corrections for simplified cloud microphysics, while retaining the underlying physical structure of the climate model. My current research explores how differentiable, GPU/TPU-enabled climate models can combine mechanistic physics with machine learning and gradient-based parameter calibration. Together, these approaches aim to develop climate models that are more physically realistic, computationally efficient, and better characterized in terms of uncertainty, ultimately improving our understanding of the processes that determine the magnitude of global warming.