Multi attribute refined identification of flood-affected bodies based on multi-source data fusion

Abstract

Lands and populations are the most direct and core disaster-bearing bodies in floods. Accurate and comprehensive identification their attributes is critical for differentiated flood prevention and mitigation strategies. However, two key challenges persist in current practices. First, the accuracy of urban land function (ULF) identification based on machine learning is constrained by data and grid scale, yet research on their impacts remains insufficient. Second, location-based service (LBS) data has sampling bias and nighttime distortion in characterizing dynamic population distribution, and its spatial resolution is insufficient for high-precision flood simulations. For ULF identification, abundant comparison schemes are generated through data traversal and multi-scale fusion, and an ensemble learning model is constructed to select the optimal ULF identification scheme. This avoids the accuracy uncertainty caused by subjective selection, and the results provide reliable data support for economic loss assessment and subsequent population spatial interpolation. For dynamic population distribution, a human-land relationship matching method based on spatiotemporal behavioral laws is proposed to reduce the impact of data bias. Meanwhile, spatial downscaling is achieved through regional division, land type and area weight calculation, generating dynamic population distribution maps with high spatiotemporal resolution. The results support the analysis of population mobility’s impact on flood risk. Hydraulic simulation is coupled with GIS (geographic information system) analysis to construct a grid-based diagnostic framework for multi-attributes of disaster-bearing bodies, including land function, population size, water depth, and spatial location. Case studies show that this framework provides reliable support for the accurate and comprehensive identification of flood disaster-bearing bodies.