Large language models reveal big disparities in current wildfire research

Contemporary fire-human-climate nexus has led to a surge in publication numbers across diverse research disciplines beyond the capability of experts from a single discipline. Here, we employed a generalized large language model to capture the dynamics of wildfire research published between 1980 and 2022. More than 60,000 peer-reviewed papers were scanned and analyzed. Through integrating geographical metadata extracted by the artificial intelligence and satellite wildfire datasets, we found large disparities in geographic patterns and research themes. The hottest spot of wildfire research is western United States, accounting for 15% of publications but only 0.5% of global burnt area, while the world’s most widely burnt region, like Siberia and Africa are largely underrepresented by contemporary publications. Similar discrepancies are found between the fuel of wildfire and its ignition and climatic drivers, between socioeconomic development and wildfire mitigation, raising concerns on sustainable wildfire managements and calling for further artificial intelligence-aided transdisciplinary collaborations.

Zhengyang Lin, Anping Chen, Xuhui Wang, Zhihua Liu, Shilong Piao