Project reports and highlights

One month after the end of an allocation period and required for provision of an additional allocation, project PIs are requested to submit via email to support@empireai.edu in PDF format

• a 1-page project report written for a non-expert audience using the template below, and
• a 1-slide project highlight using this template.

These materials are an important part of our reporting to the state and will be made public; thus, please consider what information you include (e.g., email addresses and other contact information are optional and at your discretion).

[FYI: When we switch to ColdFront for allocation management we will likely also move report submission there too.]

Please ensure your publications and presentations acknowledge use of Empire AI resources as outlined here.

Required elements for project report

1. Project Title
2. Project PI Name, Department, Institution
3. Summary of project activities
4. Summary of major findings
5. Outcomes (e.g., patents, publications, presentations, students graduated, proposals submitted or awarded, etc.
   [For publications, please include DOI or PMID]
6. Acknowledgments (incl. award numbers and agencies for any external support for this research)

Example report (generated by Google Gemini and lightly edited)

Fine-Tuning a Foundational AI Model for Enhanced Ocean-Atmosphere Energy Transfer Analysis from Satellite Imagery

Dr. Tanya Mythical, Department of Oceanography, Famous University Oceanographic Institution

Summary of Project Activities: This hypothetical research project focused on advancing our ability to quantify ocean-atmosphere energy transfer, a critical component of Earth's climate system, using readily available satellite imagery. The core of the project involved fine-tuning a pre-existing, state-of-the-art foundational AI model, initially trained on a vast and diverse dataset of general images, to specifically interpret and extract relevant features from satellite-derived oceanic and atmospheric data.

The project commenced with the meticulous curation of a specialized dataset. This dataset comprised co-located satellite imagery (e.g., infrared sea surface temperature, visible cloud patterns, microwave wind speed and direction) and corresponding in-situ measurements of latent and sensible heat fluxes obtained from buoys and research vessels. This multi-modal dataset was crucial for training the model to recognize the subtle visual cues in satellite data that correlate with energy transfer processes. The fine-tuning process adopted a transfer learning approach, leveraging the model's pre-existing knowledge of image features while adapting it to the specific nuances of oceanographic and atmospheric patterns. Data augmentation techniques were also applied to enhance the robustness and generalization capabilities of the fine-tuned model, accounting for variations in cloud cover, sun glint, and sensor noise. Subsequently, the fine-tuned model was applied to large archives of historical and near real-time satellite imagery. The model was designed to output gridded estimates of latent and sensible heat fluxes, alongside uncertainty metrics. Qualitative and quantitative validation against independent in-situ measurements and established reanalysis products was a continuous activity throughout the project, ensuring the accuracy and reliability of the model's outputs.

Summary of Major Findings: The project yielded significant advancements in the field of ocean-atmosphere interaction. The fine-tuned AI model demonstrated a remarkable ability to extract subtle features from satellite imagery that are indicative of energy transfer processes, surpassing the performance of traditional algorithms. Specifically, the model showed:

1. Improved Latent Heat Flux Estimation: The model significantly reduced the root-mean-square error (RMSE) in latent heat flux estimations by 25% compared to existing satellite-based methods.

2. Enhanced Sensible Heat Flux Detection: For the first time, the model showed promising capabilities in directly inferring sensible heat fluxes from satellite imagery with reasonable accuracy.

3. Identification of Fine-Scale Energy Transfer Events: The model was able to identify and delineate regions of intense energy transfer, such as those associated with strong wind events or frontal passages, at higher spatial resolutions than previously achievable from satellite data alone.

4. Robustness to Data Gaps and Noise: The model demonstrated greater resilience to missing data (e.g., due to cloud cover) and sensor noise, providing more continuous and reliable energy flux estimates.

Outcomes:

• Publications:

  • Sharma, A., et al. (2025). "Fine-Tuned AI for Enhanced Ocean-Atmosphere Heat Flux Estimation from Satellite Imagery." Journal of Geophysical Research: Oceans. (Submitted)

  • Smith, J., & Sharma, A. (2024). "Leveraging Foundational Models for Remote Sensing Applications in Oceanography." Remote Sensing of Environment (In Review).

• Presentations:

  • American Geophysical Union (AGU) Fall Meeting 2024: "Advancing Air-Sea Interaction Studies with Deep Learning and Satellite Data." (Oral Presentation)

  • Ocean Sciences Meeting 2025: "High-Resolution Energy Fluxes from Space: A Novel AI Approach." (Poster Presentation)

• Students Graduated:

  • Dr. Emily Chen (Ph.D. in Oceanography, December 2024) - Thesis: "Deep Learning for Quantifying Air-Sea Energy Exchange."

• Proposals Submitted or Awarded:

  • Submitted: National Science Foundation (NSF) Grant Proposal: "Towards Operational Ocean-Atmosphere Energy Flux Products from AI-Enhanced Satellite Data." (Under Review)

  • Awarded: NASA Earth Science Technology Office (ESTO) Grant: "Validation of AI-Derived Energy Fluxes using Airborne Campaigns." (Awarded, October 2024)

Acknowledgments:

• This work was supported in part by NASA grant# Oceans11.