Comparative and exploratory study of ATR and diffuse reflectance mid-infrared spectroscopy for coal property analysis

To evaluate mid-infrared sampling geometries for rapid coal analysis, attenuated total reflectance (ATR) and diffuse reflectance FTIR (DRF) were systematically compared, and multimodal fusion was explored. A total of 200 coal samples were analyzed for six key quality indices: ash, calorific value, volatile matter, fixed carbon, moisture, and sulfur. During data preprocessing, extended multiplicative scatter correction (EMSC) was applied to improve spectral stability, followed by correlation-based wavelength selection and cross-validated optimization of latent variables to construct partial least squares (PLS) regression models for each property. Notably, this study establishes a unified and reproducible benchmarking framework to disentangle sampling-geometry effects (surface-sensitive ATR and bulk-sensitive DRF) under strictly identical preprocessing, variable-selection, and cross-validation rules, and interprets the observed performance differences via chemically meaningful spectral contribution. In addition, we systematically benchmark three fusion levels (low/mid/high) within the same framework to clarify when multimodal integration is beneficial and when it is not. DRF achieved the most accurate ash prediction, whereas ATR performed better for volatile matter and moisture; calorific value and fixed carbon were comparable. Sulfur prediction remained challenging for both modalities. Low- and mid-level fusion showed no consistent synergistic gain, while high-level fusion improved prediction for five properties. Overall, the study provides actionable guidance for selecting FTIR modality and fusion strategy for practical coal quality assessment.