Geothermal energy potentiality modeling using GIS-based machine learning algorithm concept in Southwestern Nigeria: insights from geophysical and remote sensing data sets

Abstract

Abstract The paucity of energy orchestrated by the demerits of non-renewable energy sources has posed significant challenges to global demand for energy. Salvaging this difficulty, geothermal renewable energy resource is considered as an alternative. This study investigated the effectiveness of a GIS-based machine learning algorithm to analyze remote sensing and geophysical datasets to address this task. The acquired remote sensing dataset was processed to derive surface-induced geothermal conditioning factors (GCFs): land use land cover, normalized difference vegetation index, lineament density, land surface temperature, and slope percent. Spectral analysis was carried out on aeromagnetic data to derive sub-surface induced GCFs, including Curie point depth, heat flow, and geothermal gradient. Geospatial analysis module, the thematic maps for the GCFs were produced in a GIS environment. With MATLAB program coding, the optimized weights for the produced GCFs thematic maps were determined using machine learning algorithms by applying adaptive neuro-fuzzy inference system (ANFIS) models incorporated with metaheuristic optimization mechanisms. The output of various metaheuristic optimization algorithms, such as Genetic Algorithm (GA), Invasive Weed Optimization (IWO), and Particle Swarm Optimization (PSO), was processed in the GIS platform to create maps of the geothermal potential predictive index (GPPI) for the study area. Receiver Operating Characteristics (ROC) curve, Root Mean Square Error (RMSE), and multifaceted geology were used to validate the produced GPPI model maps. The multifaceted geology approach validation results revealed that a high probability of geothermal manifestation predominates the Ifewara shear zone. The results of the ROC-AUC on the optimized ANFIS models, namely: IWO-ANFIS, GA-ANFIS, and PSO-ANFIS, are 77.2%, 81.1%, and 77.5%, respectively, compared to 73.5% from the ANFIS model. The observed RMSE validation results also determined the prediction values of 0.041803, 0.10281, and 0.10734 for IWO-ANFIS, GA-ANFIS, and PSO-ANFIS compared to 0.021803 for the conventional ANFIS model. The GA-ANFIS model performed better than all investigated machined learning algorithm models. The study shows that the GPPI model map created by GA-ANFIS could be employed for accurate decision-making in geothermal resource exploitation in the investigated area and other regions with comparable geologic terrain.