Monitoring forest fund lands using remote sensing combined with CollectEarth point data
Abstract
Uzbekistan’s mountainous forests, particularly within the Gissar Range, provide vital ecosystem services such as soil erosion control and biodiversity conservation. However, these semi-arid ecosystems are increasingly pressured by anthropogenic activities, necessitating efficient monitoring tools. This study develops a robust methodology for forest area evaluation in the Dekhkanabad forestry organization using a multi-source remote sensing approach. The methodology integrates Sentinel-2 multispectral imagery with high-resolution Kompsat-3 data and topographic variables derived from an ALOS PALSAR Digital Elevation Model (DEM). To address the spectral heterogeneity of the mountainous terrain, an Object-Based Image Analysis (OBIA) framework was employed. Ground truth data were established using the FAO’s Collect Earth tool, through which 1,980 plots were classified according to IPCC and FAO Forest Resources Assessment guidelines. A supervised classification model was implemented using a 70/30 training-to-validation split. The results yielded an overall accuracy of 76 % and a Kappa coefficient of 0.66. While Pasture and Cropland classes showed high reliability, the Forest class (0.198 error) experienced spectral confusion with pastures due to the “open-canopy” nature of local juniper forests, where the understory influences the spectral signature. Settlements presented the highest classification challenge (0.731 error) due to spectral mixing with rural vegetation. Despite these challenges, the OBIA approach significantly reduced “salt-and-pepper” noise and improved boundary definition compared to pixel-based methods. This study provides a cost-effective, statistically reliable baseline for the Dekhkanabad State Forest Fund, offering a scalable workflow for sustainable forest management and conservation planning in Central Asia’s semi-arid regions.