Data-driven copper prospectivity mapping across Iran using an interpretable maximum entropy modeling

• Paleogene-Neogene magmatism is the key control for copper mineralization in Iran. • Intrusion proximity and Cu anomalies are the key ore-controlling factors. • Deposit-scale and type-specific modeling enhances reliability. • High-potential zones (14% area) contain 86% of known copper occurrences. Iran is situated in the central part of the Tethyan metallogenic domain—one of the three major metallogenic belts globally—where frequent Paleogene-Neogene magmatic activities have resulted in abundant copper resources, dominated by porphyry-type deposits along with hydrothermal and skarn types. To achieve quantitative prediction of copper resource potential, this study constructed a multi-source geospatial database covering the entire territory of Iran. Based on this database, a mineral prospecting geological model incorporating eight ore-controlling factors was established, followed by an integrated predictive model utilizing 14 predictor variables, 674 training samples, and 209 test samples. Three categories of training sample configurations (seven schemes in total) were proposed. Variable contribution analysis and response curve analysis were employed to embed geological knowledge into the Maximum Entropy (MaxEnt) modeling process, significantly enhancing the interpretability of the data-driven model. The results identify proximity to intrusions (Intr_P) and copper geochemical concentration (Cu) as the key ore-controlling factors. The high mineralization probability intervals are concentrated within 50 km of intrusive rocks and at Cu concentrations between 40 × 10 −6 and 130 × 10 −6 . All seven MaxEnt models constructed achieved training AUC values greater than 0.8, indicating good generalization capability. The model trained with all deposit types and scales combined yielded the lowest training AUC (0.885), while models based on specific deposit types and scales demonstrated superior predictive performance. The mineral potential was classified and the prediction efficiency was evaluated using the C-A fractal model and P-A plot method applied to the deposit-combined MaxEnt model results. The high-, medium-, and low-potential zones account for 11 %, 24 %, and 65 % of Iran’s territory, respectively. Notably, only 14 % of the total area captures 86 % of known deposits/occurrences, demonstrating a highly efficient predictive capability. Field validation further confirmed the effectiveness and practical utility of the approach for copper exploration in Iran.