Landuse/landcover monitoring and spatiotemporal modelling using multilayer perceptron and ‘multilayer perceptron’-Markov Chain ensemble models: A case study of Dausa City, Rajasthan

Abstract The present work is an attempt to the LULC classification, monitoring, and spatiotemporal prediction using Artificial Neural Network - Multi-Layer Perceptron (MLP) and MLP-Markov Chain (MC) models. Dausa city and its surroundings of Rajasthan, India has been selected for this study for several reasons including arid climatic setting being a sensitive precursor to the climate change scenarios and the huge population pressure experienced by the area. The MLP based supervised classification for two periods 2001 and 2018 have been analyzed using Landsat 7 Thermal Mapper (TM) and Landsat 8 OLI satellite images. The images were classified into six LULC categories viz. Built-up (Settlements), Cultivated Lands (Agricultural/Cropland), Water Body, Uncultivated/Fallow Lands, Barren Lands, and Forest/Vegetation cover. The accuracy assessment for both classified images was performed using confusion matrix led Kappa Coefficient (K) technique. Reasonable accuracies, K=0.82 (2001) & K = 0.91 (2018), have been achieved for datasets selected for both periods of time. The MLP-MC model based spatiotemporal LULC prediction for the year 2045, using the trends in the classified LULC results for the period 2001-2018, prophecies that the ‘built-up land’ would increase to reach 76.10 km2 (67.60% increase) in 2045 with the reference year 2001 whereas the increase in this class of LULC would only be 39.34% during the period 2018-2045. The ‘cultivated land’ (2001-2045: -83.86%; 2018-2045: -65.20%), ‘barren land’, (2001-2045: -54.70%; 2018-2045: -4.86%), ‘water body’ (2001-2045: -96.43%; 2018-2045: -84.42%), and ‘forest/vegetation’ (2001-2045: -81.94%; 2018-2045: -20.59%), categories would experience continuous areal decline over this period, though some at faster pace and other at comparatively lower rate. The projected unprecedented exponential increase in ‘follow land/uncultivated land’ (2001-2045: +372.45%; 2018-2045: +6.39%) presents worrisome future picture of this ecologically sensitive and fragile region. The results of this study indicate and warrant intensive management and policy, and local level participation of communities to help maintain the deteriorating ecological balance in this ecologically sensitive arid ecosystem with fragile agricultural and natural vegetation traits.