Global Analysis Of Spatial And Temporal Variations Of Fire Patterns And The Mechanisms

Quantifying the spatial and temporal patterns of fire and the mechanisms driving its variations at large scales are important to reduce the costs of human and economy, and guide the policy making and resource allocation. This study analyzed the spatial and temporal patterns of fire and the mechanisms driving its variations on the global scale, via the Moderate Resolution Imaging Spectroradiometer (MODIS), natural (climate, vegetation, lightning, and topographic roughness) and anthropogenic factors (population density, economy, and agricultural land use, etc.).The main results and conclusions are as follows:1. The spatial patterns of global fire were quantified and the mechanisms driving its global variations were analyzed, combining fire intensity and burnt area together. Two fire metrics were calculated to stand for global burnt area and fire intensity:BA and WAFI. A map of four categories of global fire patterns was then delineated:High burnt area and high fire intensity zone (HH), which was mainly distributed in the central South America, Northern Australia. This region was characterized by low human activity, long span of dry period and flat surface. The high burnt area and low fire intensity zone (HL), mainly concentrated in the African savannas. The low burnt area and high fire intensity zone (LH), mainly distributed in the remote boreal forests in Siberia and Canada. Finally, the low burnt area and low fire intensify zone (LL), scattered in almost every continent, and featured with high population density. Few studies have considered burnt area and fire intensity together, but similar burnt area often comes with startling different fire intensity, thus different impacts. This study quantified fire spatial patterns on the global scale and revealed preliminarily the interactions between burnt area and fire intensity.2. The temporal patterns of global fire were quantified, and the relationships between the temporal patterns and extreme climate events were analyzed. Also, the relationships between the temporal patterns and climate change were investigated in Canada as a case study. The time series of global burnt area in each biome in the last15years were analyzed, and the variation coefficient (Ⅳ) was calculated to represent the interannual variations. The results show that, the magnitude, length of fire season, and the interannual variation in different biomes were different, and related with the conditions of climate and vegetation. Among them, the burnt area in the tropical rainforests was the largest, and the fire season was the longest. Anthropogenic activities influencing the interannual variations of fire, which tended to smooth fire activities. In addition, the relationships between fire and the extreme climate events were investigated via the cross correlation analysis (CCF). The burnt area in the rainforests were large coincided with ENSO events, in contrast, there was a negative and lagged relationships (about10months) between peaks of burnt area and ENSO event in dry shrublands. The burnt areas in the tropical savannas were the largest, and were consistent among years. There were significant relationships between fire and temperature in Canada.3. Globally, the threshold relationships between fire and the influencing factors were investigated via the random forest and regression tree models. The recent theoretical analysis and small scale experiments pointed out the non-continuous relationships between fire and the influencing forces. Thresholds effects, rather than direct continuous relationships, exist, which challenge the related fire and earth system modeling. There are few studies quantified the thresholds between fire and the influencing forces. This study calculated a fire metric, Mean Annual Fire Density (MAFD) to quantify the global fire pattern. Next, the random forest and regression tree methods, which are based on threshold splitting were introduced to explore the non-continuous relationships. Also, the varying primary controlling conditions among different regions were quantified. The random forest model explained the observed global fire density patterns well (variance explained were78.33%). Previous studies suggested that fire activity would increase with temperature even when the precipitation increasing as well, but our result demonstrated that it is not applied for the regions with temperature higher than19.3℃. Finally, three intervals of tree density were identified with significantly different fire density:lower than9%,9%-53%, and higher than53%. Only within the intermediate intervals of tree density, can fire density exceed7counts per100km2per year.