Land managers have been using fire behavior and simulation models to assist in several fire management tasks. These widely-used models use average attributes to make stand-level predictions without considering spatial variability of fuels within a stand. Consequently, as the existing models have limitations in adequately modeling crown fire initiation and propagation, the effects of fuel treatments can only be evaluated based on average conditions, where the effects of thinning design (e.g., cut-tree locations) on changing fire behavior are largely ignored. To overcome these limitations, we coupled an advanced physics-based fire behavior model with light detection and ranging (LiDAR) technology to capture the spatial distribution of trees within stands and model crown fire initiation and propagation in more detail. Advanced physics-based fire behavior models are computationally demanding, and it is not currently feasible to run such models for large landscapes (thousands of hectares) at which fuel treatments are often considered. Thus, to extend the capabilities of these fine scale models to larger landscapes, we developed logistic regression models based on tree data and fire behavior model output to predict crown fire initiation and propagation for given tree locations and attributes for two weather scenarios, representing average and severe conditions, for our study area. We applied these regression models and used tree-level fuel connectivity prediction as measures to evaluate the effectiveness of thinning treatments for reducing crown fire potential. We demonstrate this method using LiDAR-derived stem map and tree attributes developed for a 4.6-ha forest stand in western Montana, USA.