ML methods for census data: lessons learned from mapping global livestocks

Machine Learning methods for census data: lessons learned from mapping global livestock dynamics at 1 km resolution 2000–2022

1 hr webinar with 2 presenters giving insights into how to use cutting edge ML methods for areal agression: to downscale census (polygon-based) data; OpenGeoHub is building Open global consistent time-series data on livestock dynamics (cattle, goats, sheep and horses) at 1-km annual for 2000-2022+ with uncertainty (prediction intervals). We used correlation between census data (+50k census polygons) and large stack of environmental covariates to downscale sub-national livestock inventories. The modeling pipeline is open source and available on Github.

Process of mapping per-pixel counts of livestock has shown to be complex, often sensitive to large data gaps and inconsistencies. Livestock distribution is not only determined by environmental factors, but also determined by different livestock management cultures, making it difficult to produce global models. Reserve your spot and get to ask questions. Created for the purposes of the Open-Earth-Monitor Cyberinfrastructure and Land Carbon Lab (Global Pasture Watch) projects.