The efficiency of a coal bed as a reservoir of methane and CO2 sink primarily depends on its fracture permeability. Conventional techniques of permeability estimation from drilled-core plugs in the laboratory and analytical computations are often different from in-situ measurements by multiple orders. Hence, pre-drill estimation of coal bed permeability remains one of the major challenges for the operators. Researchers usually try to solve this issue by building DFN models based on the direct use of µCT images of millimeters scale coal chips. However, such small-scale DFN models also overestimate permeability by many folds. As a solution, we propose the deployment of calibrated Discrete Fracture Network (DFN) modeling technique based on mesoscale coal images. Deterministic and stochastic DFN models are simulated by replicating thirteen coal specimens collected from the Gondwana coal seams in the Raniganj coalfield of India. Different model outputs such as directional permeabilities, permeability anisotropy, connectivity, etc. are used to distinguish and calibrate these models. The derived average permeabilities from these models closely match with available field measurements. Thus, the proposed technique helps in minimizing the huge gap between the laboratory and well-test measurements. In the process, we derive the threshold sample/model size that would sufficiently address the average cleat permeability at the well-drainage scale. Additionally, sensitivity analysis of key fracture attributes is carried out to quantify their effect on coal permeability.