Predicting land use land cover dynamics using machine learning and satellite imagery: A case study

Abstract

Land use and land cover (LULC) change is an important factor when solving environmental issues, such as water resources, agricultural productivity, and soil preservation. This study looked at the spatial and temporal trends of LULC over a 33-year period, in an area of the Sebou basin in Morocco, while forecasting future patterns, and investigating predictive performance from several machine learning classification models. A supervised classification was applied using satellite imagery from 1993, 2003, 2013, and 2023. The three different classification methods used the gradient boosting machine (GBM), support vector machine (SVM), and random forest (RF) to classify the land use and land cover into four major LULC categories: vegetation/forest, built area, bare land, and water. In predicting the land use and land cover class for 2033, a Cellular Automata – Artificial Neural Network (CA-ANN) was applied in QGIS using the MOLUSCE plugin. The GBM model was more accurate than the others with Kappa coefficients of 95.9 in 1993, 93.3 in 2003, 94.4 in 2013, and 99.8 in 2023. The overall Kappa index for the validation of the 2023 classification was 77%. Results indicate that bare land and built-up areas are on the rise as a consequence of anthropogenic activities, at the expense of vegetation and forest cover. Such changes present a source of concern for long-term sustainability, and resource availability. The study accuracy depends on the quality and resolution of the satellite data used, while assumptions in the simulation models can also introduce uncertainty in future projection. The findings of this analysis provide critical information that will support and help implement and informed policy and planning in the area of sustainable land management and protection. The research incorporated various methodological approaches in one study, thus combining LULC analyses into one study, and performing comparative analyses utilizing machine learning methods, as well as predictive modeling studies in an applied context to carry out monitoring and forecasting land use dynamics in susceptible areas.