Current and Future Patterns of Global Wildfire Based on Deep Neural Networks

Abstract

Abstract Global climate change and extreme weather has a profound impact on wildfire, and it is of great importance to explore wildfire patterns in the context of global climate change for wildfire prevention and management. In this paper, a wildfire spatial prediction model based on convolutional neural networks (CNNs) was constructed in the reference period (1997–2014) by using wildfire driving factors and historical burned areas derived from the Global Fire Emissions Database (GFED4s). The shifting spatial patterns of global burned areas in future scenarios for the twenty‐first century was investigated by using shared socioeconomic pathways (SSPs) published by CMIP6. Projected burned areas are analyzed by using nine climate models from CMIP6 under four SSPs (SSP126, SSP245, SSP370 and SSP585) for four defined periods. The evolution of the spatial pattern of global wildfires was further described based on terrestrial ecoregions and GFED regions. The results showed that for the reference period (1997–2014), burned areas were generally distributed in tropical and subtropical regions. The projection results exhibited a systematic increasing trend under the four SSPs from a global perspective in response to climate warming. The increasing trend for the burned area in the SSP370 and SSP585 scenarios was more obvious than that for the SSP126 and SSP245 scenarios. As the severity of the emission scenarios increases, severe wildfires will gradually shift to higher latitudes in the mid‐to‐long term (2061–2080) and long term (2081–2100).

Keywords

• wildfire
• deep neural networks
• convolutional neural network
• CMIP6
• burned area