A novel fuzzy Harris hawks optimization-based supervised vegetation and bare soil prediction system for Javadi Hills, India

• Published in: Arabian journal of geosciences 2023, Vol.16 (8), Article 478
• Main Authors: MohanRajan, Sam Navin, Loganathan, Agilandeeswari
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Cellular automata; Classification; Classifiers; Earth and Environmental Science; Earth science; Earth Sciences; Heuristic methods; Hills; Land cover; Land surface temperature; Land use; Learning algorithms; Machine learning; Multilayer perceptrons; Neural networks; Optimization; Original Paper; Predictions; Principal components analysis; Satellite imagery; Supervised learning; Support vector machines; Surface temperature; Vegetation
• ISSN: 1866-7511; 1866-7538
• DOI: 10.1007/s12517-023-11538-3

For several decades, researchers throughout the world have been motivated and contributed to the research on land use/land cover (LU/LC) change prediction analysis. This research work builds the ensemble learning model using unsupervised metaheuristic optimization and supervised machine learning algorithms for obtaining an LU/LC classification and prediction map in the area of Javadi Hills, India. The unsupervised fuzzy Harris hawks optimization (HHO) algorithm was used for finding the unknown features from the pre-processed satellite image. The feature-extracted map was used as an input for finding the LU/LC classes by using the supervised machine learning classifiers (support vector machine, random forest, and maximum likelihood). The principal component analysis (PCA) has been used for fusing the results of the supervised machine learning classifiers. The fused results were combined with the Adjusted Vegetation and Bareness Index (AVBaI) map and the final ensemble LU/LC (E-LU/LC) map was achieved for the years 2012 and 2015 with good classification accuracy, with an average of 95.275%. The impact of the LU/LC changes through the Land Surface Temperature (LST) map has been calculated and used as the input along with the E-LU/LC map during the process of LU/LC prediction. The ensemble-based prediction (EP)-LU/LC map for the years 2018, 2021, 2024, and 2027 has been forecasted by using the Markovian-Cellular Automata with the Multilayer Perceptron Neural Network model. The result of the EP-LU/LC map provides an average prediction accuracy of 95.763%. Our research on the LU/LC prediction will assist the concerned government officials in taking essential measures for protecting the LU/LC environment.