NDBI Based Prediction of Land Use Land Cover Change

Urbanization plays a major role in the governance policies of the civic bodies. Predicting the growth of the cities helps the government to provide the infrastructure facilities. Getting to know the number of built-up areas indirectly points to population increase or decrease. The percentage of the land occupied by built-up areas or non-built-up areas can be found out using the remotely sensed images. Land use and land cover (LULC) of these images point out how the land has been used, whether the land is a water body, vegetation, soil or built-up area. This work explores the combination of the remote sensing images and the use multilayer perceptron (MLP)-based cellular automata (CA) for predicting the changes in the LULC on a later date, using two different initial LULC maps as the input. Four basic classes, namely water, vegetation, built-up and soil, have been considered in this work. Machine learning algorithm (Random Forest Classifier) is used to obtain the LULC map of all the 3 years under consideration, namely 2013, 2016 and 2019. The LULC map of the years 2013 and 2016 acts as the initial rasters for the prediction, whereas the LULC map of the year 2019 is used for validation. The novelty of the work lies in the use of Normalized Difference Built-up index raster along with the roadway map, as the spatial variables in the MLP-based CA. The predicted results show an increase of 12.65% in the built-up areas, decrease of the soil class by 5.23%, decrease of the vegetation by 6.63% and decrease in the water bodies by 0.79% compared the base year considered (2013). The predicted LULC map using the proposed model is then validated against an independently obtained LULC map of 2019. The Jaccard similarity coefficient was found to be 0.93, which indicates the validity of the algorithm for future predictions.

Automated summary

Urbanization is crucial for governance policies, and predicting city growth and land use changes is essential. This study utilizes remote sensing images and machine learning algorithms to forecast and assess land use and cover in cities, indicating an increase in built-up areas and decreases in vegetation and water bodies.