Abstract
The article develops a cloud-based workflow for preliminary cadastral monitoring of agricultural lands in Almaty Region using Google Earth Engine and the Random Forest algorithm. The study focuses on the remote-sensing diagnosis of actual land-surface conditions rather than on potential land suitability assessment. Landsat 8/9 Surface Reflectance images were processed for 2015, 2020 and 2025, and Sentinel-2 imagery was used as an auxiliary visual reference for sample interpretation. Seasonal composites were generated, cloud-contaminated pixels were masked, and NDMI and BSI indicators were calculated to describe moisture-related conditions and exposed or sparsely covered surfaces. The spectral bands and derived indicators were then combined in a Random Forest classifier to distinguish four monitoring classes: stable vegetation cover, open or sparsely covered soil, water bodies, and built-up or non-agricultural areas. Validation was performed using an independent sample, and class areas were calculated from the classified raster. The resulting map is proposed as a diagnostic layer for identifying territories that may require additional cadastral or field verification. The approach does not replace official cadastral records, but it supports land-use control through reproducible satellite-based evidence.