$1,790,271 AI and Geosciences, GEO CI - GEO Cyberinfrastrctre

About $1.79M to predict how carbon and water cycles interlock — an agentic AI with physics built into it (University of Florida)

University of Florida Florida Started Sep 2026

Carbon and water cycling jointly sustain many ecosystem services, and the fundamental challenge is predicting how these intertwined processes scale from local observations to landscape and regional dynamics. The project combines physics-based models with data-driven AI to predict carbon-water ecosystem services across space and time.

Grant overview (primary data)

  • Award amount$1,790,271
  • RecipientUniversity of Florida (Florida)
  • ProgramAI and Geosciences, GEO CI - GEO Cyberinfrastrctre
  • Period2026-09-01 〜 2029-08-31
  • FunderU.S. National Science Foundation (NSF) / NSF

Key points

  • The record states that tradeoffs and synergies in carbon-water ecosystem services vary across space and shift over time.
  • Physics-based models are combined with data-driven AI, with adaptive agents learning from both.
  • Building physical law in as a constraint suppresses predictions that are plausible yet physically impossible where observations are thin.
  • Outcomes include datasets and modeling infrastructure alongside an interactive gamification decision-support platform and stakeholder engagement.
  • The framework is applied in the Florida coastal plains. Estimated total and obligated amount are both $1,790,271, running 2026-09-01 to 2029-08-31.
  • Prediction from data alone can be plausible yet physically impossible, so physics-based models are combined in as a constraint.

1Treating the two cycles separately is not enough

Carbon and water are usually handled as independent cycles. Yet how much carbon dioxide a forest takes up depends on the water available, and how water moves depends on vegetation. When the record says tradeoffs and synergies in carbon-water ecosystem services vary across space and shift over time, it is pointing at that interlock. Because optimizing one can degrade the other, prediction has to hold both at once.

2Building physics into the model

The AI here is not a construction that learns from data alone. Physics-based models are combined with data-driven methods, and adaptive agents learn from both high-resolution data and process-based models. Where observations are thin, prediction resting on data alone can return answers that are plausible yet physically impossible.

Building physical law in as a constraint is the approach for suppressing that class of error, and it is the central question in applying AI within the geosciences.

3Designed through to being used

The listed outcomes are not only datasets and models. An interactive gamification decision-support platform sits alongside stakeholder engagement. The premise reads as: an accurate prediction that never reaches a resource-management decision achieves nothing.

Among the 120 NSF awards this site holds as of 2026-08-28 are records for digital twins of earthquakes, landslides and floods, explainable AI for flood and hurricane evacuation, and support for wildfire evacuation — each weighted toward the step that turns a prediction into a decision. In geoscience AI, how something is conveyed is treated as a problem on par with accuracy.

4Learn from data alone and impossible answers appear

The AI here is not built to learn from data alone. Physics-based models are combined with data-driven methods, and adaptive agents learn from both high-resolution data and process-based models. The reason sits at the centre of how AI is used in the earth sciences.

Prediction learned from data alonePrediction with physical law as a constraint
Strong where observations are denseErrors stay bounded where observations are sparse
Can return answers that are plausible but physically impossibleAnswers violating conservation are structurally excluded
Hard to explain why a value came outExplicable against the process-based model

Carbon and water are often treated as separate cycles, yet how much carbon dioxide a forest takes up depends on the water available to it. Because optimising one can damage the other, prediction has to handle both at once.

Why it matters

An environmental prediction produces no effect if it is never used, however much its accuracy improves. That this project counts decision support among its outcomes, alongside data and models, makes that distance part of the design. Building physical law in as a constraint, as a way to stay reliable where observation is sparse, is an approach referenced well beyond environmental work.

FAQ

Why handle carbon and water together?
Because how much carbon dioxide a forest takes up depends on available water, and water movement depends on vegetation. The record states their tradeoffs and synergies vary by place and over time.
What does physics-informed mean here?
That the design combines process-based models with learning from data rather than learning from data alone, to suppress plausible but physically impossible predictions where observations are thin.
Why gamification?
The record lists it as an interactive decision-support platform, covering the step from prediction to resource-management decisions. The mechanism itself is not described in the record.

Sources (primary)

Source: NSF Award Search (U.S. National Science Foundation, public domain). Amounts are the obligated amount. For privacy, we do not handle principal investigator names.

#NSF#Awards#AI#Geosciences#Environment
Disclaimer: This site independently summarizes and classifies information based on official data sources. Always verify the latest and accurate information with the official sources. Content on finance, health, legal, and security is information, not advice. This site is not an official website of the U.S. government.