Although the vast majority of contemporary wildfires in the Upper Midwest of the United States have a human origin, there has been no comprehensive analysis of the roles played by abiotic, biotic, and human factors in determining the spatial patterns of their origins across the region. The Upper Midwest, a 2.8 × 105 km2 area in the northern, largely forested parts of the states of Minnesota, Wisconsin, and Michigan, contains regions of varied land cover, soil type, human settlement densities, and land management strategies that may influence differences in the observed spatial distribution of wildfires. Using a wide array of satellite‐ and ground‐based data for this region, we investigated the relationship between wildfire activity and environmental and social factors for >18 000 reported fires of all sizes between 1985 and 1995. We worked at two spatial scales to address the following questions: (1) Which abiotic, biotic, and human variables best explained decade‐scale regional fire activity during the study period? (2) Did the set of factors related to large fires differ from the set influencing all fires? (3) Did varying the spatial scale of analysis dramatically change the influence of predictive variables? (4) Did the set of factors influencing the number of fires in an area differ from the set of factors influencing the probability of the occurrence of even a single fire?
These data suggest that there is no simple “Lake States fire regime” for the Upper Midwest. Instead, interpretation of modern fire patterns depends on both the fire size considered and the measurement of fire activity. Spatial distributions of wildfires using two size thresholds and viewed at two spatial scales are clearly related to a combination of abiotic, biotic, and human factors: no single factor or factor type dominates. However, the significant factors for each question were readily interpretable and consistent with other analyses of natural and human influences on fire patterns in the region. Factors seen as significant at one scale were frequently also significant at the other, indicating the robustness of the analysis across the two spatial resolutions. The methods for conducting this spatially explicit analysis of modern fire patterns (generalized linear regression at multiple scales using long‐term wildfire data and a suite of environmental and social variables) should be widely applicable to other areas. Results of this study can serve as the basis for daily, seasonal, or interannual studies as well as the foundation for simulation models of future wildfire distribution.