1. INTRODUCTION Mineral exploration has always been a high-risk and resource-intensive activity. Considerable time, money, and energy are invested long before a discovery has any chance of becoming a producing mine (Groves and Santosh, 2015). In such settings, the way we handle uncertainty is often as important as the geological model itself. Exploration success depends on how well we recognise, quantify, and incorporate different types of uncertainty into our datasets and decision-making frameworks. One of the main challenges stems from the sparse and uneven nature of sampling. Geological terrains are complex, access is often restricted, and budgets rarely allow dense coverage. As a result, the datasets used for decision-making carry several layers of uncertainty—from laboratory analytical noise and mapping errors to inconsistencies between software platforms, modelling algorithms, and even subjective geological interpretations. These uncertainties move through every stage of the workflow: interpolation, anomaly delineation, ranking, and ultimately the investment decisions that determine whether a prospect advances or not. Spatial uncertainty, in particular, remains a dominant issue during early exploration (Chile`s and Delfiner, 2012). Any attempt to extrapolate limited samples into unsampled areas introduces estimation uncertainty. As shown by Sadeghi et al., 2023, this uncertainty can be separated into stochastic components, driven by natural geological variability, and systemic components, which arise from data acquisition and modelling choices. Distinguishing between these helps us build a more realistic view of mineralization patterns rather than relying on deterministic maps that imply confidence we do not actually have. Uncertainty also plays a critical role in selecting analytical and interpretive methods (Caers, 2011). Different clustering or classification algorithms can produce very different outcomes, even on the same dataset. As demonstrated in Sadeghi (2025), evaluating the robustness of clustering performance is fundamental for choosing the most reliable method instead of assuming one technique will always perform better. The same principle applies when comparing kriging variants, machine learning models, multifractal approaches, or hybrid AI–geostatistical workflows (Sadeghi, 2024). What matters is understanding when and why a particular method is appropriate, and how sensitive it is to the structure and quality of the data. These considerations highlight a broader point: modern exploration cannot rely solely on deterministic maps or method-driven workflows. We need adaptive, uncertainty-aware approaches that reflect the real state of our knowledge, especially in frontier terrains where information is scarce. By embedding uncertainty into spatial modelling, method selection, and decision-making, exploration teams can allocate resources more effectively and defend their choices more transparently. This perspective naturally leads to the development of new AI- and data-driven frameworks that integrate heterogeneous datasets and quantify uncertainty in a systematic way. The Return– Risk concept in this research builds on this need by combining geoscientific understanding with probabilistic modelling and explainable AI to identify targets that maximize geochemical “Return”
RkJQdWJsaXNoZXIy MTM0Mzk2