Urban Expansion and Green Cover Loss in Hyderabad, India: Sentinel 2 Based LULC Analysis and MLP–Markov Urban Growth Modelling

Abstract

The present study analyses land use and land cover (LULC) dynamics and urban growth in Hyderabad, India, using multi-temporal Sentinel 2 imagery and a multilayer perceptron–Markov chain (MLP–MC) modelling framework. Sentinel 2 Level 1C data for 2018, 2020, and 2022 were pre-processed and classified into four LULC classes—water, green infrastructure, built-up, and barren land—using supervised maximum likelihood classification, achieving overall accuracies of 94.8%, 94.7% and 96.3% and kappa coefficients of 0.92, 0.90, and 0.93, respectively. Change analysis within the Outer Ring Road (ORR) revealed a substantial increase in built-up area from 844.36 km² (57.81%) in 2018 to 948.95 km² (64.97%) in 2022, primarily at the expense of green infrastructure, which declined from 301.04 km² (20.61%) to 173.45 km² (11.87%); the dominant transitions were green infrastructure→built-up (8.74% of total area) and barren land→built-up (2.3%). Urban growth was modelled in the Land Change Modeler (TerrSet) by deriving transition potential maps using an MLP with distance to roads, distance to existing built-up areas, and distance to ORR as driving factors, and by coupling these with Markov chain analysis to simulate future LULC. Model validation against the observed 2022 LULC produced kappa values of 84.98% (K_standard), 87.83% (K_no), and 89.01% (K_location), and a built-up prediction accuracy of 98.6%, indicating strong agreement in both quantity and spatial allocation. The validated model was then used to project LULC scenarios for 2030, 2040, and 2050, showing built-up areas potentially expanding to 1135.46 km² (77.73%) by 2030 and exceeding 90% of the ORR region by 2050,