Modeling of Land Use Land Cover Change Dynamics Using Google Earth Engine and Machine Learning Techniques in the Abaya-Chamo Sub-Basin, Ethiopia

Abstract

<title>Abstract</title> Modelling land use and land cover change analysis is employed to categorize historical and predicted spatial-temporal land cover dynamics at regional and local scales. This research examined historical and projected future land use and land cover (LULC) dynamics utilizing machine learning methods within Google Earth Engine (GEE) and a Cellular Automata-Markov model. The transition potential maps were modelled using the multi-layer perceptron (MLP) neural network in the Land Change Modeler. The study applied Remote sensing datasets collected from Landsat-7 ETM+, Landsat-8 OLI-TIRS, and Landsat-8 OLI-TIRS for the study years. Although potential driver factors, including digital elevation models, road networks, river networks, and slope maps, were considered for predicting future land use and land cover. The findings reveal that the seven categorized LULC maps achieved overall accuracy and Kappa coefficient values of more than 97% and 0.9, respectively. Owing to the spatial pattern of the historically classified mapping, water bodies and agricultural land represent the majority and minority of land, respectively. Over the past 30 years, forests and barren areas have shown a declining tendency, whereas agricultural land, built-up areas, and aquatic bodies have shown expansion trends. By the end of the projected time (2065), built-up areas, waterbodies, and agricultural land are expected to expand gradually, while forest land and barren land exhibit a pronounced and continuous decline. The predicted future LULC patterns suggest significant implications for sustainable land management, food security, and ecosystem conservation.