Decadal analysis and simulation of land use and land cover changes in Taiwan using machine learning and markov chain models

This paper is a long-term, large-scale spatio-temporal study and considers demographic factors as no previous study has done. We have examined the decadal Land Use Land Cover (LULC) changes 1990–2020 and computed the LULC simulation for 2030 and 2050 using Landsat series satellite images under four different scenarios, 2-variables, 3-variables, 5-varialbes, & 6-variables, comprising of six independent variables i.e., aspect, elevation, hillshade, distance to river, distance to road, and slope. The maximum likelihood classifier, supervised classification method, has been used to do the LULC classification for the years 1990, 2000, 2010, and 2020. While the Support Vector Machine led Markov-chain method is used for LULC simulation for the years 2030 & 2050 under different scenarios and for 2020 the LULC was also simulated for validation of the methods. From the accuracy assessment it is found that the simulated LULC for 2020 has recorded overall accuracy of 0.87 & kappa index 0.74. The results show that the built-up area increased to 140% in 2020 from 1990 and in the simulation result under 6-variables scenario conditions it would increase up to 223% from 1990. The combination of Taiwan’s negative population growth rate and the simultaneous increase in impervious surfaces creates a significant topic for discussion.