Successive Estimation Model of Residential Flood Damage Based on Staged Acquisition of Field Information: Integration of Pre-Event Hazard Maps, Early Inundation Observations, and Roughly Identified Inundation Areas

抄録

Immediately after a flood, actionable information emerges sequentially over several days, progressing from initial detection to early on-site observations and ultimately to broadly delineated inundation extents provided by responsible agencies. This study presents a three-phase model that updates estimates in step with this progression by integrating (i) pre-event inundation assumption areas (data of flood hazard area) prepared before disasters, (ii) on-site point observations of inundation depths or edges, and (iii) a roughly identified inundation area grasped on-site or from above by disaster response agencies such as municipalities, prefectures, and Geospatial Information Authority of Japan, obtained and shared over several days. Assuming a locally horizontal water surface, we infer water-surface elevations from point depths and propagate them to house-level depths, classifying above/below-floor with a 0.5 m threshold. With multiple points, we report H_max/H_min and H_b-max/H_b-min to express uncertainty. Applied to three districts in Nagano Prefecture affected by Typhoon Hagibis (2019), above-floor estimates in Phases 2–3 achieved ∼10% error in one district. Errors tended to increase in flat, high-density urban areas but were mitigated by sub-domain partitioning along levees and arterial roads. The framework successively updates estimates as datasets are obtained and shared by relevant agencies, supporting early, depth-indexed tasks such as resource allocation and damage recognition.