Abstract: To elucidate the multi-scale correlation mechanism of coal spontaneous combustion and address the bottleneck of inaccurate prediction due to the lack of a consistent physical basis for key transport parameters in macroscopic models, this study is grounded in mesoscale science theory. It focuses on the critical mesoscale structure—particle agglomeration (mesoscale Ⅱ), which bridges microscopic reactions and macroscopic phenomena. The research aims to quantitatively characterize the synergistic evolution of flow and temperature fields within coal particle agglomerations during low-temperature oxidation, and further to establish quantitative functional relationships between pore structure and macroscopic parameters. This approach reveals the intrinsic link between cross-scale parameter transfer and the intensification of macroscopic phenomena. Gas coal and coking coal were selected to prepare loose coal samples with particle sizes of 0−3, 3−7, and 7−10 mm. X-ray computed tomography (CT) scanning and three-dimensional reconstruction techniques were employed to extract the real pore structure. The representative elementary volume (REV) of particle agglomerations was determined to be 300 voxels (approximately 1.671 3×10⁹ μm³) through stability analysis of macroscopic parameters. Two-dimensional porous media models with different porosities (φ=0.5, 0.7, 0.9) were generated using the four-neighborhood method, and a mesoscale numerical model incorporating coal seepage, heat transfer, and a heat source term was constructed based on the lattice Boltzmann method (LBM) to systematically simulate the dynamic evolution of flow and temperature fields under different conditions. The results show that oxidation increases porosity, and when φ increases from 0.5 to 0.9, the surface area of 0−3, 3−7, and 7−10 mm particle agglomerations decrease by 86%, 83%, and 79%, respectively, indicating dramatic structural changes. The flow and temperature fields are jointly controlled by the pore structure, exhibiting two distinct heat transfer intensification paths. Under conditions of small particle size and low φ, flow channels are tortuous, with a peak velocity of 0.300−0.350 m/s, which induces a “fingering effect” and intensifies local heat accumulation, resulting in a cross-sectional average temperature up to 3.8 times the initial value. Under conditions of large particle size and high φ, enhanced flow channel connectivity intensifies convective heat transfer and heat transport, with the peak velocity decreasing to 0.035−0.040 m/s and the heat flux increasing by a maximum of approximately 75%, leading to faster heat dissipation. Regression analysis of the simulation data established exponential quantitative functions relating porosity φ to permeability K, inertial resistance coefficient β, and effective convective heat transfer coefficient h, with coefficients of determination (R²) exceeding 0.998 7 and root mean square errors reduced by 60%−85% compared to the classical Kozeny–Carman equation. The obtained function sets K(φ), β(φ), and h(φ) were embedded as physical closure relations into the macroscopic continuum governing equations, constructing a cross-scale parameter transfer pathway capable of quantifying mesoscale intensification effects. By comparing the structural characteristics and simulation results of gas coal and coking coal, the proposed exponential functions demonstrate strong descriptive capability and universality across different coal types. This study provides critical theoretical parameters and models for developing multi-scale prediction models of coal spontaneous combustion based on heterogeneous coal media and non-steady-state self-heating processes.