Metaheuristic algorithms due to flexibility can be applied to a wide range of complex engineering optimization problems. The effectiveness, efficiency, and adaptability of such algorithms can significantly be enhanced through the modified variants. In this paper a novel modified bat algorithm (MoBA) using the concept of expectation value is proposed and evaluated using different benchmark functions, and then compared and ranked among other previously improved variants. Subsequently, the proposed MoBA was hybridized with a pretrained multitask adaptive deep learning model to generate 3D spatial subsurface mapping of geothermal temperatures in Catalonia, Spain. The success, effectiveness and superiority of the presented MoBA in compare with previously modified firefly algorithm was confirmed using different accuracy performance criteria by at least 1.71% improvement.