Recipient Organization
UNIVERSITY OF FLORIDA
G022 MCCARTY HALL
GAINESVILLE,FL 32611
Performing Department
(N/A)
Non Technical Summary
Extreme weather events, such as hurricanes, can severely disrupt agricultural systems, impacting food production and livelihoods. Florida, for example, experienced extensive agricultural damage during Hurricane Milton, with losses totaling an estimated $2.5 billion. Traditional methods for assessing such damage are often slow, complex, and expensive, limiting their effectiveness in time-sensitive disaster response efforts. This project aims to address these challenges by developing an AI conversational platform that enables farmers, policymakers, and other non-experts to assess crop damage and monitor recovery using satellite data and artificial intelligence. Through this intuitive platform, users can ask natural-language questions, such as identifying flooded areas or comparing crop health before and after a storm, and receive accurate and timely insights. The system enhances satellite imagery to a resolution of one meter, providing highly detailed information for decision-making. By making advanced technology accessible and actionable, this platform will help users reduce costs, improve resilience to extreme weather events, and support sustainable agricultural practices in affected communities.
Animal Health Component
100%
Research Effort Categories
Basic
(N/A)
Applied
100%
Developmental
(N/A)
Goals / Objectives
The overarching objective of this project is to develop a conversational Artificial Iintelligence (AI)system that makes Earth Observation (EO)data easily accessible to non-expert users through a dedicated chat interface for geospatial data. Our system will integrate Large Language Models with specialized AI models for the automated processing, and analyzing of satellite data to perform tasks such as assessing flood extent across the field and determining crop damage and recovery compared to past seasons with enhanced accuracyObjective 1: Develop a chat-based geospatial platform that provides non-expert users with fast, seamless access to EO data and novel AI algorithms for high quality EO-driven insights to enhance the resilience of agricultural production systems during and after disastersObjective 2: Develop AI-driven services in support of agricultural system's resilience
Project Methods
The project will build upon existing tools, prototypes, and algorithms, leveraging synergies to enhance agricultural resilience during and after disasters. By incorporating strong ties with local growers and extension agents, the methodology ensures practical relevance and successful integration of technology.Objective 1: Develop a Chat-based Geospatial Platform for Non-expert UsersThe primary goal is to create a chat-based platform that provides non-expert users with seamless access to Earth Observation (EO) data and Artificial Intelligence (AI)-driven insights. This platform will assist in enhancing the resilience of agricultural production systems, particularly in disaster-prone areas.To achieve this, we will utilize a custom digital infrastructure for large-scale land monitoring, which automates data retrieval, processing, and analysis using satellite imagery. The data infrastructure will be compatible with major EO data archives (Sentinel-2), ensuring efficient integration of data streams for timely insights. This infrastructure will provide the foundation for developing AI-powered analytics, offeringInitially, the focus is on developing a platform that interprets user queries--often expressed in simple, non-technical terms--into actionable research questions that can be answered using remote sensing technology. By fine-tuning existing large language models (LLMs), the system will be able to process general queries related to crop damage, flooding, and recovery, and convert them into geospatially interpretable questions. To ensure the platform is intuitive, we aim to minimize the need for repeated refinement of user inputs. We will implement a feedback-driven retraining approach, enabling the system to adapt based on user input.Subsequently, the focus is on improving the spatial resolution of satellite imagery for agricultural monitoring. The project will build on previous work by applying AI-based super-resolution techniques to Sentinel-2 imagery (generative adversarial neural networks), increasing the spatial resolution below the 10m. This enhanced imagery will be used to generate detailed maps of agricultural areas, enabling precise assessment of crop damage and recovery, especially in regions impacted by disasters. The process will involve neural networks trained on a time series of Sentinel-2 images to separate static features from transient ones, such as cloud cover. A synthetic ground truth strategy will be used to assess the quality of the upscaled imagery by comparing it with both original satellite data and high-resolution field measurements. The enhanced imagery will be incorporated into the platform's data processing pipeline, allowing users to generate vegetation indices and other indicators of crop health for damage assessments. The system will provide farmers with near real-time insights on the state of their crops, supporting targeted disaster recovery efforts.Objective 2: Develop AI-Driven Services to Support Agricultural ResilienceThis objective focuses on creating AI-driven services that help agricultural stakeholders assess and respond to disaster-related challenges. Two key services will be developed: flood extent mapping and crop damage assessment.Flooding poses a significant threat to agricultural systems. To address this challenge, the project will enhance existing data infrastructure to enable the generation of flood susceptibility maps using AI. By incorporating geospatial factors such as land use, soil types, and topography, the system will identify areas at risk of flooding, providing valuable insights for disaster management. Using AI-based models, the system will classify areas into different levels of flood susceptibility. An explainable AI approach will be used to highlight the factors contributing to flood risk, enabling users to interpret and act on the results. The system will allow farmers to ask specific questions, such as "What percentage of my fields are flooded?" and receive tailored responses based on real-time data.Simultaneously, traditional crop damage assessments are limited to post-event snapshots, failing to account for the ongoing recovery process. The project will develop a continuous monitoring system that tracks crop health over time using satellite-derived vegetation indices (e.g., NDVI). This system will enable farmers to assess both immediate damage and longer-term recovery trends, comparing current crop conditions with past seasons. The platform will allow users to ask questions like, "What percentage of my crop is damaged?" or "How does this season compare to last year?" The system will provide interactive maps and graphs to visualize crop health trends and recovery rates, guiding farmers in making informed decisions about replanting or other recovery actions.