Abstract
Vegetation mapping is a key task in remote sensing for environmental protection. Modern remote sensing technologies offer numerous advantages, including significant time savings, the ability to cover large areas, and effective long-term monitoring. These methods can greatly accelerate tasks compared to traditional field-based approaches. This article presents a method for land cover classification using time series analysis of Planet Scope satellite imagery from 2019 to 2023. The aim of the study is to examine how different input data affect classification outcomes and to analyze vegetation dynamics over the selected period. The research was conducted in the northeastern part of East Kazakhstan using the Random Forest (RF) algorithm to determine optimal parameters and track changes in vegetation types. The results include land cover classifications based on five different feature combinations, along with an overall accuracy assessment. The findings indicate that the best results in land cover mapping are achieved by combining spectral bands, topographic indices, and tree crown height layers. Vegetation was classified into three categories: trees, shrubs, and grass cover, with water bodies and bare soil areas also identified. The analysis revealed that from 2019 to 2023, the total area of trees and water bodies decreased, while grass cover, shrublands, and bare areas expanded. These findings can aid in further ecosystem analysis and highlight significant changes in vegetation structure and ecosystem processes.