Overview

TERRAHydro is an AI-powered water cycle simulation system being actively developed and deployed on NASA’s cloud computing environment. The system will be used to analyze real-world scenarios such as water availability in the Himalayas and near-real-time integration of live satellite data. Underpinning TERRAHydro is CREST, a software framework that allows scientists across government, industry, and academia to seamlessly blend traditional modeling approaches with modern AI, providing the technical foundation that makes TERRAHydro possible. Together, they represent a significant step forward in NASA’s ability to simulate and monitor Earth’s environment at scale.

Science Area

Accurately monitoring and predicting the terrestrial water cycle, including soil moisture, evapotranspiration, and river runoff, is critical for understanding water availability, flood risk, and the impacts of climate change. Current Earth system models struggle to fully capture the complex, interconnected nature of these processes, often relying on simplified physical assumptions that introduce significant errors. The growing volume of satellite observations remains underutilized, leaving a critical gap between the data available and our ability to turn it into actionable, real-time environmental insight.

Technology

TERRAHydro’s key technical breakthrough lies in its ability to seamlessly couple AI-driven and traditional physics-based models into a single, unified simulation system, something no open-source community platform has achieved at this scale. By integrating live satellite observations, such as NASA’s SWOT data, directly into the model for continuous real-time updating, TERRAHydro is able to learn and preserve physically realistic relationships between variables like soil moisture and evapotranspiration. Early results show it outperforms all current state-of-the-art land surface models on community benchmarks.

Advancements

• A new interoperability layer will allow models written in different AI software languages, TensorFlow, PyTorch, and JAX, to work together within a single system.
• A newly designed web-based user interface will give scientists the ability to configure, run, and visualize Earth system simulations directly through a browser without needing specialized computing expertise.
• A traditional physics-based photosynthesis model integrated into TERRAHydro will directly test and validate the team’s approach to combining classical science models with modern AI.

Principal Investigator

Dr. Craig Pelissier joined NASA in 2013 as a member of ASTG after completing receiving PhD in numerical simulations of Quantum field theories from The George Washington University. In 2018, he became the lead of ASTG and manages its diverse portfolio of projects. He also actively conducts research in machine learning for land surface modeling (TERRAHydro), and developing numerical algorithms to simulate melting precipitation and their radiative properties to support the NASA Precipitation Measurement Mission. He is an avid badminton player, enjoys playing the piano, and spending time outdoors.

Selected Publications

Pelissier, Craig; Olson, William; Kuo, Kwo-Sen; Adams, Ian, “A Physically Based, Meshless Lagrangian Approach to Simulate Melting Precipitation,” Journal of the Atmospheric Sciences, 2022.

Pelissier, Craig; Frame, Jonathan; Nearing, Grey, “Combining Parametric Land Surface Models with Machine Learning,” IGARSS, 2020.

Pelissier, Craig; Alexandru, Andrei; “Resonance Parameters of the rho-meson from asymmetrical lattices,” Physical Review, 2012.