Divergent burned-area trends in Southern and Northern-Hemisphere Africa over the first half of the 21st century

About 50% of global biomass-burning carbon emissions result from wildfires, with approximately 70% of the world’s wildfire burned area originating in Africa. This study applied machine learning models based on random forest (RF), support vector machine (SVM), and k-nearest neighbors (KNN) algorithms to simulate future burned areas using wildfire drivers and historical burned-area data from MODIS fire products for 2003–2020. The RF model outperformed SVM and KNN in burned-area prediction. The RF model was then used to project burned areas over the period 2021–2050 in Africa under a low (SSP1-2.6) and a high (SSP5-8.5) emissions scenario. During the reference period, burned areas generally declined in Northern-Hemisphere Africa (NH: −2.6 Mha yr⁻¹) and Southern-Hemisphere Africa (SH: −1.6 Mha yr⁻¹). The projections over the period 2021–2050 indicate a systematic increase in burned area in SH, particularly in savannas (0.3 Mha yr⁻¹ under SSP1-2.6; 0.5 Mha yr⁻¹ under SSP5-8.5); while in NH, the burned area is projected to decline (−0.2 Mha yr⁻¹ under SSP1-2.6; −0.5 Mha yr⁻¹ under SSP5-8.5). Linking wildfire dynamics with climate is critical for developing effective fire management strategies. 全球约50%的生物质燃烧碳排放源于野火, 而全球约70%的野火过火面积发生在非洲.本研究应用随机森林(RF),支持向量机(SVM)和k近邻(KNN)算法的机器学习模型, 利用2003年至2020年中分辨率成像光谱仪(MODIS)野火燃烧产品的历史过火面积和野火驱动要素数据, 模拟了非洲未来的过火面积.结果表明, 随机森林模型在过火面积预测方面优于支持向量机和k近邻算法模型.随后, 本研究利用随机森林模型, 预估了低排放(SSP1-2.6)和高排放(SSP5-8.5)两种气候情景下, 2021年至2050年非洲的过火面积.在2003至2020的参考时期内, 非洲北半球地区(−2.6 百万公顷/年)和南半球地区(−1.6 百万公顷/年)的过火面积总体呈下降趋势.2021年至2050年的过火面积预测表明, 南半球的过火面积将发生转折, 呈现增加趋势, 尤其是在热带稀树草原地区(SSP1-2.6情景下为0.3百万公顷/年;SSP5-8.5情景下为0.5百万公顷/年);而北半球的过火面积预计将会继续减少, 但是减少趋势变缓(SSP1-2.6情景下为-0.2百万公顷/年;SSP5-8.5情景下为-0.5百万公顷/年).该研究表明, 将野火动态与气候条件联系起来, 对于制定有效的火灾管理策略至关重要.