Indonesia's flood, hazard, and land-change data exists, in satellite archives, disaster-agency portals, and consultants' reports. Almost none of it is readable by the millions of people making the country's most consequential land decisions. Earthmetrics turns that archive into a risk story anyone can read, starting with one regency: Sleman, Yogyakarta.
A working MVP on open data, built and verified end to end:
Indonesia records thousands of flood events every year, and land conversion at the urban fringe, rice paddies becoming housing with unresolved drainage, keeps manufacturing new exposure. The information asymmetry sits at the household level: hazard maps are published for planners, not for the family deciding where to live next month. Closing that gap changes where people build and buy, which is one of the few demand-side levers on flood exposure that scales without regulation.
The concrete wedge: families relocating to Yogyakarta for its schools, moving fast, deciding on land whose flood record, zoning status, and land-use history they cannot see. One family, one readable risk report, repeated across a country where most of the biggest decisions land on land.
Annual 10 m satellite embedding fields replace hand-tuned index thresholds, so one model reads land conversion (paddies to housing, wetland loss, drainage change) consistently across all of Indonesia. This powers the "what happened to this land" timeline behind every report at national scale.
Gemini drafts each land report grounded strictly in the computed layers: what changed and when, which hazard zones apply, what remains unverified. The model writes for a family, not a GIS analyst, and every report carries its sources and limitations.
Yearly composites, DEM-derived hydrology, and change layers computed once, nationally, and refreshed as new imagery arrives. GEE's commercial licensing is budgeted as a runway line item before any paid phase, not assumed away.
Two production decisions in the current MVP illustrate how we treat data credibility: