Multi-Sensor Landsat Time-Series Classification for Long-Term Urban Expansion Mapping: A Case Study of Tashkent, Uzbekistan

Tashkent, the capital of Uzbekistan, has undergone substantial landscape transformation since the collapse of the Soviet Union in 1991, yet long-term assessments of its urban dynamics remain limited.This study analyzes the spatial patterns and rate of urban expansion in Tashkent Region from 1991 to 2026 using multi-sensor Landsat imagery (Thematic Mapper [TM], Enhanced Thematic Mapper Plus [ETM+], and Operational Land Imager [OLI]) processed in Google Earth Engine.Eight optical spectral indices (Normalized Difference Vegetation Index [NDVI], Normalized Difference Built-up Index [NDBI], Builtup Index [BUI], Built-up Area Extraction Index [BAEI], New Built-up Index [NBI], Normalized Built-up Area Index [NBAI], Modified Built-up Index [MBI], and Urban Index [UI]) were stacked and used to train and compare five machine learning classifiers: Random Forest, XGBoost, K-Nearest Neighbors, Naïve Bayes, and Support Vector Machine.Random Forest achieved the highest classification performance (area under the curve [AUC] = 0.925, based on 2026 reference labels) and was used to generate built-up maps for nine observation years; per-year classification accuracy for the historical maps was not independently validated.Built-up area increased from 229.78 km² in 1991 to 271.80 km² in 2026, representing a cumulative increase of 18.29%.The Mann-Kendall test confirmed a statistically significant increasing trend (τ = 0.667, p = 0.0165), while Sen's slope estimated an average increase of 0.387 km²/year.Expansion remained limited and fluctuating before 2017, with negative annualized inter-period expansion rates during 1995-2000 (-0.19%/year) and 2009-2013 (-0.37%/year), reflecting reductions in mapped built-up extent relative to the preceding observation years.In contrast, expansion accelerated substantially after 2017, reaching 2.27%/year during 2017-2022 and 1.28%/year during 2022-2026, with growth remaining concentrated within the low-lying valley corridor while the eastern and northeastern mountainous zone remained largely non-built-up.This study applies multi-index spectral stacking and machine learning for long-term urban monitoring in semi-arid environments.

Hali tarjima qilinmagan