Pluvial flash floods are fast-moving hazards and causes significant disruptions in urban areas. With the increase in heavy precipitations, the ability to proactively identify flash floods hotspots in cities is critical for flood nowcasting and predictive monitoring of risks. While rainfall runoff models and hydrologic models are useful models for flash flood prediction, these models are computationally expensive and effort intensive to be used for flood nowcasting. To address this challenge, this study presents interpretable machine learning models for predicting urban flash flood hotspots based on intertwined land and built environment features. The task of predicting flash flood hotspots is formulated as a binary classification problem, and three recent flash flood events in U.S. cities are selected for data collection and model validation. Various features related to land and built environment characteristics are constructed using diverse datasets, and the occurrences of flash floods are captured using crowdsource data from the events. Using these features and datasets, the flash flood hotspots of cities are predicted with two ensemble models based on decision trees. The results demonstrate that the models can achieve good accuracy (0.8) in identifying flooded/non-flooded locations. Especially, the models can achieve high true positive rate (0.83–0.89) and low missing rate, demonstrating the methods' practicability for accurately predicting flooded hotspots. The model interpretation results indicate that land features related to hydrological and topological features have greater impacts on flash flood risk, than built environment features. Further analysis reveals that the feature importance, model performance, and model transferability performance vary among cities and localized specifications of the models are needed for accurate prediction of flash flood for a particular city. The data-driven machine learning models presented in this study provide a useful tool for predicting flash flood hotspots based on the intertwined features of land and the built environment in cities to enable nowcasting and proactive monitoring of flash flood hotspots for emergency response and also inform integrated urban design and development towards flash flood risk reduction.

中文翻译：

---

可解释的机器学习，利用交织的土地和建筑环境特征来预测城市山洪热点

雨洪暴发是一种快速移动的灾害，会对城市地区造成严重破坏。随着强降水的增加，主动识别城市山洪热点的能力对于洪水临近预报和风险预测监测至关重要。虽然降雨径流模型和水文模型是山洪预测的有用模型，但这些模型用于洪水临近预报时计算成本昂贵且工作量大。为了应对这一挑战，本研究提出了可解释的机器学习模型，用于根据相互交织的土地和建筑环境特征来预测城市山洪热点。预测山洪热点的任务被制定为二元分类问题，并选择美国城市最近发生的三起山洪事件进行数据收集和模型验证。使用不同的数据集构建与土地和建筑环境特征相关的各种特征，并使用来自事件的众包数据捕获山洪的发生。利用这些特征和数据集，通过两个基于决策树的集成模型来预测城市的山洪热点。结果表明，该模型在识别淹没/非淹没位置方面可以达到良好的精度（0.8）。特别是，该模型可以实现较高的真阳性率（0.83-0.89）和较低的漏报率，证明了该方法在准确预测洪水热点地区的实用性。模型解释结果表明，与水文和拓扑特征相关的土地特征比建筑环境特征对山洪风险的影响更大。进一步分析表明，特征重要性、模型性能和模型可移植性性能因城市而异，需要模型的本地化规范才能准确预测特定城市的山洪。本研究中提出的数据驱动的机器学习模型提供了一种有用的工具，可以根据土地和城市建筑环境的相互交织的特征来预测山洪热点，从而能够对山洪热点进行即时预报和主动监测以进行应急响应，并为综合信息提供信息旨在减少山洪风险的城市设计和开发。