Global Spatial Pattern of Wildfire Dominant Precursors

Predicting wildfires is a complex task because the causations are typically diverse and the time window when the multiple wildfire precursor variables prevail is inconsistent with each other. In this study, we applied a causation discovery method, Peter & Clark Momentary Conditional Independence (PCMCI), to figure out the dominant wildfire precursors globally and their time lags to wildfires. We used data from 2003 to 2018 and divided the 12 selected wildfire precursors into three groups, that is, atmospheric group, hydrologic group, and vegetation group. Results show that the atmospheric group dominates the other two groups globally, especially in the north high-latitude and tropic regions. Because the environment there is usually cold and wet, wildfires are limited by unsuitable fire weather and short fire season, and the changes in atmospheric conditions can promote the ignition and spread of wildfires. The hydrologic group is usually dominant in dry or wet regions where wildfires are limited by fuels or high ecosystem water content. Hydrologic conditions there influence the growth of vegetation and therefore, the accumulation of fuels. In wet ecosystems, the short time lags there indicate that wildfires are more likely to occur and spread when the ecosystem water content is lower. The vegetation group, which is usually not considered in wildfire causation studies, shows dominance in the north temperate zone and around the Equator of Africa where the land cover types are grassland, shrubland, and cropland. Most importantly, the time lags of the vegetation precursors for wildfires are much larger than the other groups, we identified time lags up to more than one year. This indicates that when predicting wildfires, vegetation information which is related to atmospheric and hydrologic conditions can be seen as a separate group because of the complex nonlinear relationships among them, and these kinds of relationships usually cannot be detected and expressed by simple models. Moreover, it is important to take time lags into account, otherwise, the prediction of wildfires will be limited by its accuracy.