Machine learning for building extraction and integration of particle swarm optimization with sleuth for urban growth pattern visualization for liveable cities

Abstract

Rapidly increasing population and migration from rural areas to nearby urban agglomerations develop tremendous pressure on system of the existing cities without compromising socioeconomic and cultural linkages. Policy interventions, both at global and local scale, have created newer avenues for the researchers to explore real-time solutions for problems world-wide. For instance, the outcome of 2015 United Nations agenda for the achievement of the Sustainable Development Goals (SDGs) by the year 2030 primarily focuses on urbanization issues and probabilistic modelling of future scenarios to obtain a robust alternative for resource utilization and further for maximizing sustainability through land use pattern analysis. This is the clear indication toward the very important role of “ever dormant” urban planning, especially in the case of a rapidly developing country such as India. Remote sensing and geo informatics along with Machine learning can provide extremely relevant information about the pattern change in cities and as input to visualize the future growth pockets. In this context, potential of cellular automata (CA) in urban modelling has been explored by various researchers across the globe. In the recent past, models have been drawing majority of the attention along with geographic CA processes about urban growth and urban sprawl studies. Most recent approaches include optimization of transition rules based on machine learning techniques and evolutionary algorithms that follow nature-inspired mechanism such as Genetic Algorithm, Ant colony optimization, Particle Swarm Optimization (PSO), simulated annealing, Grey Wolf optimizer etc. Irrespective of any modelling technique, model calibration remains one of the challenging and most crucial steps towards obtaining realistic results. This research communication tries to demonstrate a novel idea of integrating PSO with SLEUTH post calibration of the spatial-temporal footprint of urban growth from the year 1990 to 2017 for Kolkata, a historical megacity of Eastern India. Results were evaluated and validated using statistical fit measuresreveals PSO-SLEUTH performed substantially better compared to traditional Brute Force calibration method (BFM). Another significant development was in terms of computation time of optimized values from days (BFM) to hours (PSO). The study identifies Kolkata region to be sensitive to spread and road gravity coefficients during calibration procedure. Results indicate growth along the transport corridors with multiple agents fuelling the growth. Further, with the aid of high spatial resolution data, buildings were extracted to understand the growth parameters incorporating neural networks. Using the results, renewable energy aspects were explored to harness and provide a suitable local solution for energy issues in energy gobbling cities. Pattern of landscape change, development of better process of modeling and extraction of building from machine learning techniques for planning smart cities with self-sustaining energy is presented in this research work.