π Ph.D. student in Climate Resilience & Hydroclimate Extremes at the University of Florida
π Geospatial & Environmental Data Scientist | Remote Sensing | AI for Disaster Management
π‘ Passionate about bridging science and real-world decision-making through intelligent tools
I work at the intersection of climate science, remote sensing, and machine learning, with a strong emphasis on:
- π Flood mapping and forecasting using satellite data (SAR, optical) and ground sensors (e.g., traffic cameras, rain gauges)
- π°οΈ Multi-sensor data fusion for risk-informed infrastructure planning
- π§ Geospatial foundation models & real-time decision support tools
- π» Custom web GIS dashboards (ArcGIS Experience Builder, APIs, automation)
My current project involves integrating AI with remote sensing and on-the-ground data to predict and monitor floods in near real-time, helping communities and agencies improve resilience.
| Geospatial | Data Science | Dev Tools | Visualization |
|---|---|---|---|
| GEE, ArcGIS Pro, QGIS | Python, JS, SQL, R | Git, GitHub, Jupyter, Conda | Dashboards, StoryMaps, Notebooks |
| SNAP, HEC-RAS, Hazus | TensorFlow, Scikit-learn | Docker, VS Code | Matplotlib, Plotly, Leaflet |
-
π§ Cross-Modality Transfer Learning with Geospatial Foundation Models
Investigated the transferability of pre-trained optical features from the IBM-NASA Prithvi foundation model to a Synthetic Aperture Radar (SAR) flood mapping task.
The study fine-tuned the Prithvi modelβs encoder on a small, custom-built SAR dataset and compared its performance against a baseline UNet model trained from scratch.
Results showed that Prithvi achieved significantly higher segmentation accuracy (IoU ~2.2 vs. 0.6 for UNet), demonstrating the efficiency and potential of cross-modality transfer learning for geospatial applications in data-scarce regions.
π Read Report (PDF) -
π End-to-End Flood Dataset Generation Pipeline
Created a fully automated, modular pipeline for preparing flood datasets from SAR imagery using Google Earth Engine, thresholding methods, and permanent water masks.
Supports machine learning workflows for flood mapping model training and evaluation.
π View GitHub Repo -
πΊοΈ NLCD Land Cover Analysis Tool
Built a cloud-based Google Earth Engine app that calculates land cover class percentages (wetlands, forest, developed areas, etc.) across any U.S. region at the state, county, or tract level.
This tool transforms a task that once took hours in ArcGIS Pro into seconds through scalable remote computation.
π View GEE Code
π₯οΈ My personal portfolio: HatamiMatt.github.io
If you're working on climate resilience, geospatial AI, or disaster risk reduction, I'm open to collaborations, internships, and consulting opportunities.
π« Email: [email protected]
π LinkedIn: linkedin.com/in/HatamiMatt
βTurning data into actionable insights, one pixel at a time.β