Satellites are now pricing rain, rice, and wetlands
Three preprints used satellite data and machine learning to size rainwater tanks across 41,592 Ugandan facilities, map Bangladeshi rice within 2.34% of official statistics, and price 30 years of wetland loss in Ethiopia. Measurement is getting cheap.
- An ML model trained on simulations predicted the required rainwater tank size within 25% for 97.0% of facilities in spatial cross-validation. Self-reported m:…5351
- 77.1% of schools and 62.2% of health facilities in Uganda could reach water supply targets with a tank of 200 m³ or less. Self-reported m:…5351
- Satellite-derived Boro rice area in Manikganj, Bangladesh, differed from official government statistics by 2.34%. Self-reported m:…3199
- Wetland ecosystem service values in Ethiopia's West Arsi Zone fell from USD 1.17 million to USD 0.05 million between 1995 and 2025; total landscape value fell 3.9%. Self-reported m:…2666
The measurement layer
Three research groups put new work on EarthArXiv this week: one priced rooftop rainwater in Uganda, one mapped rice in Bangladesh, and one put a dollar figure on wetland loss in Ethiopia. Different continents, different questions, the same instrument — satellite imagery plus machine learning. Things once measured with clipboards and guesswork are now being measured from orbit, and the numbers are getting precise enough to plan budgets against. m:…5351 m:…3199 m:…2666
All three are preprints, so none of this is peer-reviewed yet. But the methods are no longer exotic. The teams worked from public satellite archives and standard machine-learning tooling, with two of the three building inside Google Earth Engine. That is the point. This kind of measurement is getting cheap, and for anyone putting capital into climate adaptation, food security, or development infrastructure, it changes what is knowable before the money moves. m:…5351 m:…3199 m:…2666
Rainwater, priced by the building
In a preprint study of rooftop rainwater harvesting in Uganda, a machine-learning model trained on simulations predicted the required tank size within 25% for 97.0% of facilities in spatial cross-validation. m:…5351
77.1% of schools and 62.2% of health facilities in Uganda could reach water supply targets with a tank of 200 m³ or less, according to the study — a concrete budget number for anyone funding decentralized water infrastructure. m:…5351
Regional rainfall patterns matter more than annual totals: facilities receive about 40% less water yield than Eastern facilities at the same annual rainfall due to seasonal concentration, and reliability of rainwater harvesting showed a significant trend at 15.1% of facilities from 1981 to 2025. m:…5351
Rice, audited from space
A study mapping Boro rice cultivation in Manikganj, Bangladesh, found that Sentinel-1 VH polarization backscatter reductions during transplantation, in late February to early March, provide a reliable signature for flooded rice fields, and its Random Forest model achieved overall classification accuracies between 88.1% and 94.9%. m:…3199
The number that matters for food statistics: satellite-derived rice area estimates showed a 2.34% difference from official government statistics. That suggests satellites could serve as an independent check on food production figures, refreshed far more often than ground surveys. m:…3199
Wetlands, on the books
In Ethiopia's West Arsi Zone, wetland ecosystem service values declined from USD 1.17 million to USD 0.05 million between 1995 and 2025, while total landscape ecosystem service value decreased by 3.9% over the same period. m:…2666
The researchers identify agricultural expansion and urban development as the primary drivers of land use and land cover change in the zone — the trade-off that policymakers usually argue about in the abstract now has a number. m:…2666
What it means for executives
Together these three preprints describe a measurement layer arriving for development economics. Rainwater infrastructure can be screened and sized per building from orbit. Crop statistics can be independently checked. Ecosystem loss can be priced in dollars. If you fund, build, or insure climate-adaptation or development infrastructure, the data bottleneck is shifting from collecting measurements to acting on them — and the edge moves to whoever acts on the measurement first. m:…5351 m:…3199 m:…2666
The caveats deserve equal weight. These are preprints, not peer-reviewed results. Satellite measurement prices what can be seen; it does not replace ground truth, and policy built on these numbers should treat them as screening tools until validated. m:…5351 m:…3199 m:…2666
What is still unresolved
- None of the three papers is peer-reviewed; all findings rest on the authors' own reported results.
- The Uganda team says its screening estimates require on-site validation to confirm local feasibility.
- The Bangladesh and Ethiopia teams did not provide public code or datasets with their papers.
- The Uganda authors note seasonal forecasting skill is limited outside September–November.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.