Overview

The Coastal Zone Digital Twin (CZDT) aims to create a digital replica of interconnected coastal systems across the entire eastern United States. CZDT answers “what now” (current state), “what next” (forecasts), and “what if” (hypothetical scenarios). The system integrates NASA remote sensing data into a continuously updated Essential Variable data cube. CZDT introduces an innovative agentic AI framework, transforming the system into an active decision-making partner. This allows emergency managers, resource planners, and stakeholders to use natural language to evaluate trade-offs, optimize interventions, and make informed coastal management choices.

Science Area

Coastal science seeks to understand the complex, interconnected dynamics between terrestrial watersheds and coastal environments. CZDT will help researchers answer critical questions about compounding coastal hazards, such as how upstream land surface changes, heavy precipitation, and soil moisture will interact with coastal storm surges and high tides to create severe compound flooding. This project will also be useful for understanding the impact sea-level rise will have on vulnerable ocean ecosystems. Finally, CZDT could be used to forecast how warming temperatures, marine heat waves, and polluted runoff degrade water quality might endanger public health and threaten agriculture.

Technology

Powered by a loosely coupled digital twin architecture, CZDT utilizes machine learning surrogate models (like hybrid ConvLSTM2D U-Nets) to rapidly emulate computationally expensive physics-based models (such as NASA LIS). CZDT incorporates NASA’s DEEP-VIEW model to assess water quality and expands the Essential Variable data cube with SWOT, NISAR, and OPERA datasets. The architecture also supports “bring-your-own” third-party models and socioeconomic data. Crucially, the system employs a multi-agent LLM engine using the Model Context Protocol (MCP) to automate data discovery, wrangling, model execution, and visualization, enabling rapid, conversational “what-if” scenario testing for non-technical users.

Advancements

• Agentic LLM framework for geospatial workflows introduces a cutting-edge multi-agent AI engine, transforming CZDT into an active, conversational decision support partner.
• Machine learning surrogate models enable for rapid forecasting, increasing CZDT’s utility for emergency response use cases.
• Open-platform “bring-your-own” interoperability allows for highly localized, cross-agency impact assessments.

Principal Investigator

Thomas G. Grubb is the AR/VR Product Development Lead for the NASA GSFC AR/VR Research & Development Lab, which is developing multiple AR/VR prototypes and projects in support of science and engineering, including the Roman Space Telescope and the On-orbit Servicing, Assembly, and Manufacturing (OSAM-1) mission. He was a Goddard Associate for the Earth Science Technology Office from 2008-2015. He has been a developer for many projects over his 30-year career, including the Goddard Mission Services Evolution Center (GMSEC), which was GSFC 2008 Software of the Year (SOTY) submission and the General Mission Analysis Tool (GMAT), which was GSFC 2015 SOTY submission. He was the 2016 Robert H. Goddard winner for Exceptional Achievement for Customer Support. He received his B.S. degree in Computer Science from University of Maryland Baltimore County and has a Masters Degree in Computer Science from John Hopkins University.

Selected Publications