AI-Assisted Livestock and Ger Detection — Mongolia's National Statistics Office Satellite Census Pilot

Mongolia's National Statistics Office (NSO) has long relied on labor-intensive door-to-door enumeration for its agricultural census, a costly exercise in a country where roughly a third of the population lives a nomadic or semi-nomadic herding lifestyle across vast, sparsely connected terrain. To explore a cheaper, faster alternative, NSO partnered with the Partnership in Statistics for Development in the 21st Century (PARIS21) to pilot machine-learning-based detection of livestock and dwellings from satellite and drone imagery.

In Arkhangai province, NSO trained machine-learning models using the Spectral Angle Mapper (SAM) method for image classification, drawing on "ground truth" data generated during the 2022 agriculture census to validate model outputs against known conditions on the ground. The models were tasked with two distinct detection problems from remote-sensing imagery: identifying large livestock (cattle, horses, camels) and identifying traditional ger dwellings, whose distinctive circular shape makes them a useful proxy for locating herder households.

The pilot reported an 81.63% accuracy rate for identifying large livestock and a 98.11% accuracy rate for identifying gers. NSO staff have said the approach could eventually reduce reliance on costly high-resolution imagery and specialized drone operations, though the agency notes that training staff in image processing and AI remains necessary to manage the technology sustainably in-house. NSO is now working with the Ministries of Food, Agriculture and Light Industry to develop an open-source system intended to scale remote-sensing-based enumeration nationwide.

As of mid-2026 this remains a single-province pilot rather than a national rollout, and the 81.6% livestock-detection accuracy, while promising, implies a meaningful error rate that would need further improvement before the method could fully substitute for ground enumeration. Separately, unaffiliated academic researchers have applied a similar ger-detection computer-vision approach to poverty mapping in the capital Ulaanbaatar, lending independent support to the underlying technique's viability, though that research is a distinct study rather than part of the NSO pilot itself.

Read the full analysis: https://www.paris21.org/our-impact/remote-close-home-how-local-data-initiatives-are-making-livestock-monitoring-more