Bio

I am a Principal Research Scientist in the Microsoft AI for Good Research Lab where I co-lead the Geospatial ML research group and focus on tackling large scale applied problems at the intersection of remote sensing and machine learning/computer vision. Generally, I’m interested in research topics that facilitate using remotely sensed data more effectively in conservation, sustainability, and damage response application. For example: self-supervised methods for training deep learning models with large amounts of unlabeled satellite imagery, human-in-the-loop methods for creating and validating modeled layers, and domain adaptation methods for developing models that can generalize over space and time. Similarly, I am also interested in creating open-source tools that facilitate using remotely sensed data in machine learning pipelines – I am a creator/maintainer of the torchgeo library and “satellite imagery labeling tool”.

I graduated from the Georgia Institute of Technology with a PhD in 2020 under the supervision of Bistra Dilkina with a dissertation titled, “Large scale machine learning for geospatial problems in computational sustainability”.

I am also an adjunct faculty member at Arizona State University, where I taught CSE 475: Foundations of Machine Learning with Hannah Kerner.

Projects

Global Renewables Watch

A temporal dataset of global solar and wind energy installations derived from satellite imagery, with Planet and The Nature Conservancy.

Visualizer / Paper / GitHub

TEMPO: Building Density and Height

Global temporal building density and height estimation from satellite imagery.

Visualizer / Paper / GitHub

Field Boundary Delineation

Fields of The World: benchmark datasets, models, and global 10m maps for agricultural field boundary segmentation.

Visualizer / Website / GitHub / Papers: Global Map · PRUE · FTW · FTP

Building Damage Assessment

Rapid post-disaster mapping of building damage from satellite and aerial imagery, used by responders in the hours after an event.

Visualizer / Paper / GitHub / HASTE

Demos

Interactive demos of some of our recent work — click through to try them:

Throughput Bench

How fast can a deep learning model map the planet? Benchmark results for geospatial models across GPUs, precisions, and batch sizes.

14   1 contributor

DeltaBit

Label, train, and predict per-pixel change detection in the browser over compressed AlphaEarth embedding-difference tiles.

2   2 contributors

Sentinel-2 Paint

Recreates any photo as a mosaic of real Sentinel-2 satellite imagery patches.

6   1 contributor

Software

TorchGeo

PyTorch datasets, samplers, transforms, and pre-trained models for geospatial data. I am a co-creator and maintainer.

4.1k   128 contributors

torchgeo-bench

Lightweight benchmarking of frozen geospatial foundation models on the GeoBench suites, with KNN and linear-probe metrics.

24   5 contributors

Satellite Imagery Labeling Tool

Lightweight web interface for creating and sharing vector annotations over satellite and aerial imagery scenes.

294   7 contributors

Dynamic World in PyTorch

PyTorch port of Google's Dynamic World 10m land cover model — a bit-exact match to the official TensorFlow weights.

14   1 contributor

s2-superres

Multi-temporal Sentinel-2 super-resolution by optimization.

13   1 contributor

MapLibre GL Components

Single-file MapLibre GL JS plugins: swipe map comparisons and Cloud Optimized GeoTIFF rendering, no build step or tile server.

0   1 contributor

vsrecent

Tiny Windows launcher for VS Code's "Open Recent" projects.

0   1 contributor

Talks

• Keynote at the EarthVision Workshop at CVPR 2026 — From Local to Global Maps from Satellite Imagery: ML Techniques and Applications (June 2026, Denver, CO)
• Invited talk at the NOAA Northeast Fisheries Science Center AI 101 Symposium — Geospatial ML for Sea Lions, Whales, and Buildings (May 2026)
• Keynote at the CV4EO Workshop at WACV 2026 — Applied GeoML: From Local to Global (March 2026, Tucson, AZ)
• Invited talk at the ITU AI for Good webinar — Mapping Connectivity for Saving Lives: The Early Warning Connectivity Map (EWCM) (January 2026, online)