Space to Soil

Soil degradation drives USD 400 billion in annual global losses, yet satellite monitoring remains reactive: full scenes are processed regardless of relevance, and ground verification depends on manual inspection. SoilSentinel is a 6U CubeSat concept that shifts computation onboard, distilling NASA-IBM's Prithvi foundation model 307× into a 1.19M-parameter student that predicts vegetation and bare-soil indices on a Jetson Orin Nano. Only 50 KB alerts are downlinked, reducing data transmission by 98.7% and delivering insights within minutes. On 146 million pixels across California and Nevada, the pipeline achieves AUC-PR of 0.919, and generalizes to post-fire erosion zones without architectural changes. Applications span regenerative agriculture, crop insurance triggers, post-fire recovery, watershed and hydropower protection, and conservation verification. Market research has opened discussions with ICL Growing Solutions, IFDC, and the UNL On-Farm Research program. We propose two paths, NASA-led development and/or a commercial-public hybrid.