DEWSat

A four-member NSL team was selected as a finalist in NASA’s Space to Soil challenge for DEWSat, a Drought Early Warning Satellite. DEWSat carries a red and near-infrared multispectral imager and an onboard edge-AI compute stack. It splits its imagery into tiles, computes vegetation indices on an FPGA, and runs a quantized Bi-LSTM (compressed ~20–50×) to give each tile a drought-risk score. Only the high-risk tiles and their metadata get downlinked, over S-band, with telemetry and commands over UHF. Inference runs only over priority regions during sunlit periods to save power.

Doing the analysis in orbit means faster responses without heavy ground infrastructure, and the same architecture extends to wildfire, flood, and urban-heat monitoring, especially as a constellation.

Our competition entry went to AIAA as an extended abstract. This semester the team is finding a PI to work with, developing and quantizing the model, laying out the satellite in CAD as a 6U design (down from 12U), and writing a full paper due December 1 that we’ll also submit to AFRL’s University Nanosatellite Program.