Molecular signature evolution of coal-derived dissolved organic matter under geothermal conditions: FT-ICR MS and machine learning.

The accurate identification of water sources and tracing of inrush pathways in deep coal mines remain challenging, as elevated geothermal temperatures can alter the molecular fingerprints of coal-derived dissolved organic matter (Coal-DOM), a potential organic tracer. To address this, we investigated the evolution of Coal-DOM from coals of different ranks (long flame coal, lean coal, anthracite) under simulated geothermal conditions (25 and 50°C). By integrating ultrahigh-resolution mass spectrometry (FT-ICR MS) with an interpretable machine learning framework (XGBoost-SHAP and reactomics), we decoded the rank-specific molecular transformation pathways. Results showed that warming diversified the DOM pool from low-rank coals via fragmentation and oxidation, while it selectively enriched condensed aromatic and sulfur-containing structures in high-rank anthracite DOM, forming a stable and distinct fingerprint. Key molecular descriptors (O/C, NOSC, AImod, sulfur content) were identified as robust predictors of thermal reactivity. Overall, this integrated framework enables molecular-scale prediction of DOM reactivity in coal-bearing aquifers under geothermal perturbation. In addition, it yields quantifiable organic fingerprints that complement conventional indicators for mine-water source identification and water-inrush tracing. These capabilities can support environmental risk assessment and guide management of deep mine water systems.