中小河流洪水预警误差溯源与阈值可信度分析——以重庆市永川区为例

Analysis of Error Attribution and Threshold Reliability of Flood Warnings for Small and Medium-Sized Rivers—A Case Study of Yongchuan District, Chongqing

摘要: 重庆市中小河流涨水风险提示产品在业务运行中存在较多空报、漏报。为厘清误差来源并定量评价雨量阈值可信度，本文以永川区为例，基于2023~2025年19期44次预警产品、92个雨量站逐日降水量和4个控制水文站水位资料，界定了误差溯源与阈值可信度的概念内涵，建立“降雨预报–产流条件–监测数据–阈值设定”四环节定量归因决策树，采用泰森加权面雨量和前期影响雨量指数逐事件定量复盘，并以命中率、空报率、临界成功指数及其95%置信区间开展预警检验。结果表明：总体命中率84.6%、空报率75.0%、临界成功指数23.9%；33次空报中降雨预报偏差占76% (复算面雨量平均低于阈值33.9 mm)，产流条件偏差占18%，数据支撑不足占6%，未发现可归因于阈值设定的失误；2次漏报均因实况降雨达到阈值而预报未报出。现有雨量阈值未见系统性偏差，误差主要来源于降雨预报不确定性，业务优化应优先提升降雨预报精度，并逐步探索考虑前期影响雨量的动态阈值。

Abstract: The flood-rise risk warning product for small and medium-sized rivers in Chongqing, China, suffers frequent false alarms and missed events in operational application. To clarify the sources of error and quantitatively evaluate the reliability of rainfall thresholds, taking Yongchuan District as a case study and using 44 warnings from 19 product issues during 2023~2025, daily precipitation from 92 rain gauges and water-level records from 4 control hydrological stations, this paper defines the concepts of error attribution and threshold reliability, and establishes a four-step quantitative attribution decision tree covering rainfall forecasting, runoff generation, monitoring data support and threshold setting. Each warning is re-analyzed event by event with Thiessen-weighted areal rainfall and the antecedent precipitation index, and evaluated by the probability of detection (POD), false alarm ratio (FAR) and critical success index (CSI) with 95% confidence intervals. The overall POD is 84.6%, the FAR 75.0% and the CSI 23.9%. Of the 33 false alarms, rainfall forecast bias accounts for 76% (recomputed areal rainfall averaging 33.9 mm below threshold), runoff-generation bias for 18% and insufficient data support for 6%; none is attributed to threshold setting. Both missed events occurred because the forecast failed to capture rainfall that reached the threshold. The existing rainfall thresholds show no systematic deficiency; errors mainly originate from rainfall forecast uncertainty, so operational improvement should prioritize improving the accuracy of rainfall forecasts and gradually exploring dynamic thresholds accounting for antecedent soil moisture.