Modeling the Spatiotemporal Distribution of Livestock Grazing Density in Kazakhstan Using Machine Learning

Abstract

Background and Aim. Kazakhstan possesses one of the largest pasture resources in Central Asia, yet effective management of these lands is hindered by the lack of spatially detailed livestock distribution data. The growth in the number of small ruminants and horses, along with increased pressure on pasture ecosystems, necessitates new monitoring approaches. The objective of this study is to model the spatiotemporal distribution of pasture livestock density in Kazakhstan from 2000 to 2019 at a high spatial resolution (1 km2) using the Random Forest algorithm.

Materials and Methods. District-level livestock statistics forsheep, goats, and horses were used along with 13 socio ecological variables. The Random Forest algorithm was applied to generate annual livestock density maps. Spatial variables include climatic, vegetative, demographic, and  infrastructural factors. Model validation was performed using R2, RMSE, and MAE metrics with various training and testing sample configurations.

Results. Accurate livestock density maps were developed, and “hotspots” of pasture pressure were identified in the southern and southeastern regions of Kazakhstan. The model with a 10 km buffer and 90:10 data split showed the highest accuracy. Significant increasing trends in livestock density were revealed using Mann-Kendall trend test and Sen’s slope.

Conclusion. The resulting spatiotemporal data can be applied in pasture management, agricultural policy, ecological monitoring, and veterinary planning. The developed methodology can be adapted for other regions with similar conditions.