(Theme session) Academic exchange session in English

[1301-08-01] Hybrid AI–Geostatistical Modeling for 3D Ash Prediction and Zoning in Sparse Drillhole Environments: A Case Study Integrating 3D Block Model and Machine Learning

○Angesom Gebretsadik Abraha^{1[Student presentation: Doctoral course],2}, Natsuo Okada ¹, Ryo Itano¹, Yoko Ohtomo¹, George Mathews³, Youhei Kawamura¹ (1. Hokkaido University, 2. Aksum University, 3. Nazarbayev University)

司会：パク　イルファン（北海道大学）

Keywords:Hybrid modeling, Coal quality, Machine Learning, Geostatistics, Smart mining

Accurate ash content prediction and spatial zoning are essential for coal quality control, mine planning, and environmental management. However, in areas with sparse drilling data, conventional geostatistical approaches such as kriging often fail to capture non-linear geological variability, leading to inaccurate predictions. This study presents a hybrid modeling framework that integrates geostatistics with artificial intelligence (AI) to improve the accuracy of ash content prediction and delineation of ash-rich zones in a 3D geological context. Using data from a coal deposit characterized by limited borehole coverage, we first performed kriging-based estimation of ash content on a 3D block model to capture spatial continuity. Then, machine learning algorithms including XGBoost, Random Forest, and LightGBM were trained on residuals to capture non-linear patterns missed by geostatistics. These models were optimized using hyperparameter tuning strategies such as GridSearchCV and Optuna. The outputs were combined into a hybrid model that leverages both spatial and contextual features. Performance metrics (R², RMSE, MAE) demonstrate that the hybrid model significantly outperforms both standalone kriging and AI models. The hybrid approach significantly improves predictive performance compared to standalone methods, as evaluated using R², RMSE, and MAE. Ash zoning maps generated from the hybrid model provide clearer identification of ash zones, supporting more effective selective mining strategies. This integrated methodology demonstrates strong potential for sustainable and cost-effective resource management in data-limited mining environments.