Pore structure and microstructural insights on higher heating value prediction for Indonesian low-rank coals: A multi-technique case study with cross-validated statistical assessment

Indonesian low-rank coals (LRCs) are a strategic energy resource, but their high moisture, oxygen-rich organic matrix, and heterogeneous pore-mineral architecture complicate higher heating value (HHV) prediction by classical correlations. This study addresses not which equation predicts HHV most accurately, but why classical correlations deviate for tropical LRCs. Microstructural characterization of an Aceh LRC (A1) was combined with cross-validated evaluation of a curated Indonesian LRC dataset. XRD reveals a dominantly disordered carbonaceous matrix with subordinate quartz and kaolinite; SEM-EDX shows organic-rich and mineral-rich domains with localized Fe-S co-occurrence; N₂ adsorption indicates a high BET surface area (418 m²/g) with mesopore-dominated porosity. These observations suggest HHV is controlled by coupled moisture-porosity-oxygen-mineral interactions, not elemental composition alone. Five classical correlations, three MLR variants, and three machine-learning algorithms were evaluated under leave-one-out cross-validation. Gradient Boosting achieved the lowest error (R² = 0.82, RMSE = 1.68 MJ/kg, MAPE = 6.0%); Modified Dulong was the most reliable closed-form correlation (MBE = −1.31 MJ/kg). Other classical correlations under-predicted systematically (MBE = −2.4 to −4.2 MJ/kg), reflecting structural bias from oxygenated functional groups, moisture-retaining mesoporosity, and phase-specific mineral effects compressed into elemental or ash terms. The work provides a mechanistic framework for improving HHV models for Indonesian LRCs.