Data, Code & Research Software

This page provides selected interactive tools, code, models, and data products developed through my research.

Interactive Tools

🌐 GSM3 Viewer — Global Soil Moisture Explorer

• Interactive visualization of daily global soil moisture predictions at 9 km resolution.
• Supports spatial exploration, temporal animation, and data access.
• Based on the GSM3 global multitask soil moisture dataset.

Code & Models

🌍 Geospatial Foundation Embeddings for Hydrologic Prediction

• Evaluates geospatial foundation model embeddings as transferable catchment descriptors for rainfall–runoff prediction.
• Uses representations from AlphaEarth and StefaLand across U.S. and global catchments.
• Demonstrates the value of learned embeddings for spatial transfer and prediction in ungauged basins.

🔄 Transformer Models for Hydrologic Prediction

• Benchmarks Transformer architectures against LSTM for large-sample rainfall–runoff prediction.
• Shows that architectural modifications substantially improve Transformer performance relative to the vanilla architecture.
• Provides a benchmark for exploring non-recurrent and pretrained architectures in hydrologic prediction.

🛰️ Multiscale Deep Learning for Soil Moisture

• Multiscale deep learning framework that jointly learns from satellite and in situ soil moisture observations.
• Produces daily soil moisture predictions at 9 km resolution across the conterminous United States.
• Provides a general framework for learning from observations available at different spatial scales.

Data Sets

🌐 Global Soil Moisture from a Multitask Model (GSM3)

GSM3 is a global daily soil moisture dataset generated using a multitask deep learning model trained jointly with satellite and in situ observations.