Cryo-Bench
Benchmark suite for geospatial foundation models on cryosphere tasks — glaciers, glacial lakes, sea ice, and calving fronts. 14 GFMs evaluated across 5 datasets.
projects
Selected research projects, datasets, and open-source tools.
Prithvi-CAFE
Dual-branch architecture combining the Prithvi ViT with CNN features via a Convolutional Attention Module — for flood mapping on Sen1Floods11 and FloodPlanet.
NISAR Polarimetric Pipelines
Production pipelines for NISAR GCOV/GSLC: Freeman–Durden & Cloude–Pottier H/α decomposition, flooded-vegetation mapping in the Florida Everglades and Atrato/Amazon basins.
Glacial Lake Bench (GLB)
Comparative benchmark across Prithvi, TransNorm, UNet, UViT, DeepLabV3+, Panopticon, and DOFA on 1,907 common glacial-lake test tiles with rigorous Wilcoxon statistics.
GEE Multi-Sensor Mosaics
Large-scale Google Earth Engine pipelines for 10-band mosaic exports (S1 + S2 + DEM/slope) at 10 m UTM across South Andes, Central Asia, Svalbard, Antarctica, and Alaska.
GLACIA
Multimodal LLM framework that uses instance-aware positional reasoning to segment glacial lakes — bridging vision-language models with Earth-observation segmentation.
DL4EO (PyPI package)
Open-source Python package for deep-learning workflows on Earth-observation data — data loading, augmentations, segmentation/regression baselines.
AI for Sustainability (Lab 06)
Course materials for UW–Madison: full Jupyter notebooks for GFM-based flood mapping using Sen1Floods11, IBM TerraTorch, and TerraMind-small fine-tuning on Google Colab T4.