气候变化背景下江西省林火空间预测

林火是森林生态系统中重要的干扰因子之一,深刻地影响森林景观结构和功能。在全球气候化背景下,揭示气候变化对林火空间分布格局的影响,可为林火管理和防火资源分配提供科学指导。因此,基于江西省2001—2015年MODIS火影像数据(MCD14ML)和年均气温、年均降水量、植被、地形、人口密度、距道路距离、距居民点距离7个因子数据,利用增强回归树模型:(1)分析林火发生影响因子的相对重要性及其边际效应;(2)将GFDL-CM3和GISS-E2-R气候变化模式中的年均气温和年均降水量作为未来的气象数据,在3个温室气体排放量情景(RCP2.6、RCP4.5、RCP8.5)下,对2050年(2041—2060的平均值)和2070年(2061—2080的平均值)江西省林火分布进行预测,生成林火发生概率图。并采用受试者工作特征(ROC曲线)和混淆矩阵评估模型预测的精度。研究结果表明:(1)年均气温和海拔与江西省林火发生的相关性较强,年均降水量、居民点距离、人口密度、道路距离与林火发生的相关性较弱,但是与林火发生密切相关的如降水、风速等也应重点关注;(2)训练数据(70%)和验证数据(30%)的AUC值(ROC曲线下面积值)均为0.736,混淆矩阵对火点预测的正确率为67.8%,表明模型能够较好地预测研究区林火的发生;(3)在RCP8.5排放情景中林火发生的增幅最明显,其增幅较大的区域由赣南向赣北移动;(4)未来2050年和2070年林火发生与当前气候(2001—2015年)下相比,赣州市、鹰潭市的增幅较为明显,其他区域不明显。江西省各林业管理部门要加强林火高发区及潜在发生区的森林监测和管理,加大防火宣传力度,提升民众的森林防火意识。

Prediction research of the forest fire in Jiangxi province in the background of climate change

Forest fire is one of the most important disturbances in forest ecosystems, which significantly alters forest landscape structure and function worldwide. The spatial patterns of forest fire occurrence are closely associated with climate change. Researches have shown that fire frequency and burned area could substantially increase with prolonged growing seasons in warming climate scenarios. Revealing influences of climate change on the spatial distribution patterns of forest fires can provide scientific guidance for formulating feasible forest and fire management strategies. Therefore, based on the MODIS fire image (MCD14ML) data and 7 climatic (annual average temperature and precipitation), vegetation (forest type), topography (elevation) and human activities (population density and distances to the nearest roads and settlements) data from 2001 to 2005 in Jiangxi province, this study used the Boosted Regression Tree (BRT) model to:(1) quantify relationships, i.e., relative importance, marginal and incorporative effects, between fire occurrence and the explanatory variables; (2) project and generate fire occurrence maps under current (2001-2015) and two GCMs'' (GFDL-CM3 and GISS-E2-R) future climate RCPs scenarios (RCP2.6, RCP4.5 and RCP8.5) in 2050 and 2070. We evaluated the performance of the BRT model using the area under curve (AUC) of a receiver operating characteristic curve (ROC). An alternative method to evaluate models was achieved by comparing the observed with the predicted fire occurrence with confusion matrixes method. The results showed that:(1) the annual temperature and altitude strongly correlated with the occurrence of forest fire in Jiangxi province, and the annual precipitation, distance to the residential areas, population density, and distance to roads had weaker correlations with the occurrence of forest fires; (2) the AUC values both of the training data (70%) and verification data (30%) were 0.736. The accuracy of the confusion matrix in predicting fire occurrence was 67.8%. The mod evaluation results indicated that the BRT model fitted well and could be used to predict forest fire occurrence in future climate scenarios in the study area; (3) the increasing in forest fire occurrence was the highest under the future climate scenario of RCP8.5; (4) fires would increase significantly in the cities of Ganzhou and Yingtan under future climate scenarios in both 2050 and 2070 compared with current climate scenarios (2001-2005). Forest and fire managers in Jiangxi province should strengthen their monitoring and management of forests in areas with high occurrence of forest fires based on the results from this study. And then they should strengthen public''s awareness of fire prevention especially in southern Jiangxi province.