标题：A comprehensive framework for assessing the spatial drivers of flood disasters using an Optimal Parameter-based Geographical Detector–machine learning coupled model
期刊：Geoscience Frontiers（中科院一区）
时间：2024
作者：Yang, LY
洪水灾害对全球人类生命财产安全构成严重威胁。从宏观尺度探究洪水灾害的空间驱动机制对防灾减灾具有重要意义。本研究提出一个融合驱动因子优化与模型可解释性、兼顾空间异质性的综合分析框架，通过集成最优参数地理探测器（OPGD）、递归特征消除（RFE）和轻量化梯度提升机（LGBM）模型，构建OPGD-RFE-LGBM耦合模型来识别关键驱动因子并模拟洪灾空间分布格局，同时采用SHAP可解释性算法定量解析洪灾空间分异的驱动机制。以中国西南典型山原区——云南省为案例区，基于整理的7332条历史洪灾事件数据库，初选降水、地表环境和人类活动等三大类22个潜在驱动因子展开实证研究。结果表明：云南省洪灾空间异质性显著，其中地貌分区可解释历史洪灾空间变异的66.1%；相较于单一LGBM模型，耦合模型在关键驱动因子识别与影响量化方面具有明显优势，即使在因子数据精简情况下仍能保持较优的模拟性能（RMSE平均降低6%，R²平均提升1%）；因子解释性分析显示，不同子区域的关键驱动因子组合存在差异，但降水强度指数（SDII）、强降水日数（R10MM）和五日最大降水量（RX5day）等降水类因子始终是控制洪灾空间分异的主导因素。本研究为具有显著异质性的大尺度区域洪灾空间驱动机制研究提供了定量分析框架，可为灾害管理部门制定宏观防灾策略提供科学依据。
Fig. 1. Location and topography of the study area.
Fig. 2. Spatial zoning scheme in Yunnan Province. (a) Eco, the definition of ecology zoning codes following Fu et al. (2001). (b) Geo, the definition of geomorphic zoning codes following Wang et al. (2020). (c) Climate, the definition of climate zoning codes following Zheng et al. (2010).
Fig. 3. Schematic illustration of the comprehensive analysis framework for the driving factors of flood disasters.
Fig. 4. Quantitative representation of the spatial distribution of flood disasters in Yunnan Province from 1986 to 2020.
Fig. 5. Ranking of the q values for the flood disaster driving factors at the provincial scale based on the OPGD factor detector.
Fig. 6. Interaction effects of the flood disaster drivers at the provincial scale based on the OPGD interactive detector.
Fig. 7. Ranking of q values for flood disaster driving factors at the subregional scale based on the OPGD factor detector. Panels (a), (b), (c) and (d) refer to SUB1, SUB2, SUB3, and SUB4, respectively.
Fig. 8. Identification of the essential driving factor sets in the four subregions of Yunnan Province based on the OPGD–RFE–LGBM coupled model. The pink points indicate the minimum RMSE. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 9. Manners and degrees of influence of the essential driving factors on the distribution of flood disasters. (a), (b), (c) and (d) refer to SUB1, SUB2, SUB3, and SUB4, respectively. Each point represents a data sample, its color showing the magnitude of feature value. The point’s horizontal placement reflects the SHAP value of the driving factor, signifying its impact intensity. Together, the x-axis position and point color depict the link between factors and flood disasters.
Fig. 10. Venn diagram showing overlaps in essential flood disaster driving factors among the four subregions of Yunnan Province.
喜欢也行，不喜欢也行；如果觉得有用处的话，还请点点右下角的赞/在看，记得关注我哟！！！
