Mongolia's National Statistics Office piloted satellite-and-drone machine-learning models in Arkhangai province, identifying large livestock with 81.6% accuracy and traditional ger dwellings with 98.1% accuracy, a step toward replacing costly manual agricultural censuses.
Mongolia's National Statistics Office (NSO), facing costly door-to-door agricultural census enumeration across a nomadic/semi-nomadic herding population, partnered with PARIS21 to pilot machine-learning detection of livestock and dwellings from satellite and drone imagery in Arkhangai province, using the Spectral Angle Mapper (SAM) method validated against ground-truth data from the 2022 agriculture census.
Results
The pilot reported an 81.63% accuracy rate for identifying large livestock (cattle, horses, camels) and a 98.11% accuracy rate for identifying traditional ger dwellings from remote-sensing imagery.
Conclusions
As of mid-2026 this remains a single-province pilot rather than a national rollout; the 81.6% livestock-detection accuracy implies a meaningful error rate that would need further improvement before the method could fully substitute for ground enumeration, though NSO is now working with the Ministries of Food, Agriculture and Light Industry on an open-source system intended to scale nationally, and independent academic research has separately applied a similar ger-detection technique to poverty mapping in Ulaanbaatar.
Implementation
Indicative cost
Low (< €50k) — No cost figures published; the pilot reused ground-truth data already collected for the 2022 agriculture census, implying incremental rather than large new cost.
Time to results
Short (< 1 year) — Pilot conducted 2022-2024 in Arkhangai province; nationwide open-source scale-up is a stated future goal, not yet implemented.
Staffing & skills
National Statistics Office of Mongolia (NSO), with PARIS21, NSO staff note that training in image processing and AI remains necessary to manage the technology sustainably in-house
Conditions for success
Availability of ground-truth data from the 2022 agriculture census enabled model validation
Partnership with an international technical body (PARIS21) supported methodology
Coordination with the Ministries of Food, Agriculture and Light Industry is underway to build an open-source system for national scale-up
Common failure modes
Pilot covers a single province (Arkhangai); NSO itself describes nationwide rollout as a future goal
81.6% livestock-detection accuracy implies a meaningful error rate not yet resolved
Reliance on costly high-resolution imagery and specialised drone operations remains a constraint NSO is trying to reduce
Where it fits
Governance type
national statistics agency with international donor partner
Scale
sub-national pilot (single province)
Income level
lower-middle income
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.
FAO and Timor-Leste's government built an XGBoost-corrected seasonal drought index that triggered the country's first Anticipatory Action protocol in October …
