Uncertainty quantification and causal reasoning remain key challenges in deep learning for geospatial optimization. Traditional models for Mineral Prospectivity Mapping (MPM) often lack interpretability, struggle with out-of-distribution (OOD) generalization, and yield unreliable uncertainty estimates, leading to inefficient resource allocation and false positives. To address these limitations, this study introduces Conformalized Causal Learning (CCL), a novel deep learning framework that integrates causal feature selection, conformal prediction, and attention-based neural architectures to enhance robustness and trustworthiness. Unlike conventional models that prioritize predictive accuracy alone, CCL incorporates calibrated uncertainty estimation to support confidence-aware decision-making. Experimental evaluations across four diverse geological datasets show that CCL reduces predictive uncertainty by up to 90% while maintaining competitive AUROC scores. The framework consistently outperforms baseline methods in both robustness and interpretability. CCL offers a generalizable solution for high-stakes geospatial applications, including mineral exploration, environmental monitoring, climate modeling, and remote sensing—advancing the development of trustworthy, uncertainty-aware AI.