Spatial wildfire ignition predictions are needed to ensure efficient and effective wildfire response but robust methods for modelling wildfire occurrence have not been fully evaluated. Here we leverage high resolution, static spatial data to predict the ignition locations of human and naturally-caused wildfires across the state of Montana (USA). We leveraged a 25-year historical wildfire dataset (1992-2017) and four high resolution spatial variables that capture fuel availability, topography, geographic location and human transport infrastructure. We combined these data to train spatial logistic regression Generalized Additive Models (GAM) designed for big datasets (BAM) for both human and natural ignitions and we tested the efficacy of incremental changes in model complexity. Results showed that the best human and natural-caused ignition models were highly capable of distinguishing locations with and without new wildfire occurrences statewide (AUC = 0.89 and 0.84 respectively). Natural-caused ignitions were strongly influenced by slope and location, while human-caused fires were best predicted by distance to roads as well as terrain and fuel availability. Finally, these spatial fire occurrence models can be combined with temporally-variant data to predict the spatial and temporal distribution of wildfires across the state with the view that these methods can be used to develop predictive models at larger scales.