人工智慧在都市淹水預測之應用與展望

汪立本 ； Susana Ochoa Rodriguez ； Yuting Chen ； Christian Onof

《土木水利》 48卷2期 (2021/04) Pp. 56-63

https://doi.org/10.6653/MoCICHE.202104_48(2).0008

摘要

本文分享歐盟計畫FloodCitiSense於過去幾年，應用人工智慧（Artificial Intelligence, AI）相關技術以輔助都市淹水預測、預警（urban flood forecasting and warning）模型之發展。計畫執行期間，探索多種類型數據驅動（data-driven）淹水預測模式，從單純預測某場暴雨是否為致災暴雨事件，到近一步結合都市淹水水理模式以預測淹水的空間分布。本文將詳細說明其中二種模型。第一種為純粹資料驅動模型，僅使用歷史降雨資料及實際收集到的淹水報告，藉由學習過去致災暴雨事件之統計特徵，預測該暴雨是否可能造成淹水，再近一步預測可能的淹水空間分布。第二種為氣候類比（analogue）模型，該模型主要由歷史致災暴雨事件相關之大氣參數及雷達降雨影像驅動，藉由分析、分類暴雨事件大氣參數之相似性，以及雷達降雨影像之特徵，建立出致災暴雨類比模型，並結合都市淹水水理模式預先產出之空間淹水模擬地圖（flood maps），即時估算都市淹水機率之空間分佈圖。此計畫證明應用人工智慧於都市淹水預測之可行性，然而也發現僅依靠歷史淹水報告，在時間及空間上，無法提供完整及連續之淹水分布，造成模式訓練之困難；此外，隨著都市基礎建設發展，歷史淹水報告亦無法正確地反映現況。此缺點可以藉由傳統都市淹水水理模型之模擬結果改善。

並列摘要

In this article, we would like to share the main findings of an EU-funded research project, named FloodCitiSense, in developing data-driven urban flood forecasting and warning system. A range of data-driven approaches were explored throughout the project period; two of them were found to be feasible for the future operation and are further explained here. The first one is a purely data-driven model. It aims at predicting urban flooding relying merely on historical rainfall data, flood registry records and some hydraulic features of a given city. The result suggests that this model can well predict if a given storm may lead to flooding, but it could not further predict the spatial distribution of a flood. The second model is an analog model. This model characterises the features of the underlying atmospheric variables and rainfall patterns of historical flood-inducing events. By checking similar features of the current weather condition and rainfall data, this model can determine if there are similar flooding events in the history, and based upon them, flood and flood probability maps can be produced. The result suggests the analog model can predict spatial distribution of a flood with a more than 70% accuracy, which shows a great potential to be used operationally.

參考文獻

Pitt, M.(2008).Lessons learned from the 2007 floods.,未出版.

Wang, L.-P.,Simoes, N.,Rico-Ramirez, M.,Ochoa-Rodriguez, S.,Leitao, J.,Maksimovic, C.(2012).Radar-based pluvial flood forecasting over urban areas: Redbridge case study.Weather Radar and Hydrology.351,532-637.

Evers, M.,Jonoski, A.,Maksimovič, Č.,Lange, L.,Ochoa-Rodríguez, S.,Teklesadik, A.,Cortes-Arevalo, J.,Almoradie, A.,Simões, N. E.,Wang, L.-P.,Makropoulos, C.(2012).Collaborative modelling for active involvement of stakeholders in urban flood risk management.Nat. Hazards Earth Syst. Sci..12,2821-2842.

Ten Veldhuis, J. A. E.,Ochoa Rodriguez, S.,Bruni, G.,Gires, A.,vanAssel, J.,Wang, L.-P.,Reinoso Rodinel, R.,Kroll, S.,Schertzer, D.,Onof, C.,Willems, P.(2014).Weather radar for urban hydrological applications: lessons learnt and research needs identified from 4 pilot catchments in North-West Europe.International Weather Radar and Hydrology Symposium.(International Weather Radar and Hydrology Symposium).:

Verbeiren, B.,Dagnachew Seyoum, S.,Lubbad, I.,Xin, T.,ten Veldhuis, J. A. E.,Onof, C.,Wang, L.-P.,Ochoa-Rodríguez, S.,Veeckman, C.,Boonen, M.,See, L.,Nalpas, D.,O’Brien, B.,Johnston, J.,Willems, P.(2018).FloodCitiSense: Early Warning Service For Urban Pluvial Floods For And By Citizens and City Authorities.11th International Conference on Urban Drainage Modelling.(11th International Conference on Urban Drainage Modelling).: