Quantifying urban growth in 10 post-Soviet cities using Landsat data and machine learning

Abstract

ABSTRACTUrbanization is one of the most irreversible anthropogenic forms of land use. Unplanned and rapid urban growth can result in environmental degradation, sprawl, and unsustainable production and consumption practices. The unique challenges facing the post-Soviet countries throughout the transition period highlight the critical need for a quantitative assessment of urban dynamics. Total of 32 Level-1 precision terrain corrected (L1T) Landsat scenes with 30 m resolution and auxiliary population and economic data were utilized to quantify the urban expansion dynamics in 10 cities across nine post-Soviet republics. Land cover was classified by using Support Vector Machine (SVM) learning algorithm with overall accuracies ranging from 87% to 97% for 29 classification maps over three time steps. The initial time step was the year 1989 ± 2, the middle time step was the year 2000 ± 2, and the final time step was the year 2015 ± 2. The results demonstrated several spatial and temporal urban expansion patterns across the post-Soviet region. The urban land area in several cities increased significantly over the study period. The average annual urban expansion rate was 1.6 ± 0.7 % per year for 10 cities over the study period and the average area of land converted to new urban environment was 227 ± 224 km2 with a corresponding average per cent increase of 54.5 ± 26.7%. Furthermore, the results demonstrated significant decrease in overall population densities across the 10 cities with an average decrease of −26.9 ± 14.8% over the study period. The urban expansion rates considerably outpaced the urban population growth rates in all 10 cities during the last quarter of a century, indicating more expansive urban growth patterns.