Introducing One Concern’s live compound flood forecast pipeline in Japan

Dr. Yi Liu, Data Scientist at One Concern

Making disasters less disastrous is One Concern’s core mission, and one specific natural disaster we focus on is flooding. Currently, an effort is well underway at One Concern to deploy a national live compound flood forecast pipeline in Japan, as an expansion of last year’s pilot flood pipeline in Kumamoto City, Japan, a successful proof of concept product. The ability to forecast compound floods in a real-time manner with sufficient lead time is critical for emergency managers to make well-informed decisions about evacuation and resource allocation before a flood event. To that end, our Japan flood pipeline is designed to ingest weather forecast data (including rainfall, temperature, wind and pressure etc.) whenever available automatically and make live forecasts of hourly compound flood maps with an integrated consideration of storm surge, streamflow and urban runoff with a forecast period of up to 72 hours (Figure 1). A flood alerting mechanism that triggers the inundation prediction is implemented in the pipeline. The forecasted flood depths are then used to calculate impact statistics such as overflow points, buildings inundated and population affected.

The Japan flood pipeline is undergoing extensive calibration and validation against major flood events across Japan to ensure the quality of its forecasts. One selected validation event is Typhoon Hagibis that made landfall in the Greater Tokyo Area on October 12, 2019 (Figure 2), and caused widespread destruction across its path in Japan. According to press reports, Hagibis caused dozens of deaths, hundreds of injuries, flood damage to at least 10,000 homes, and power outages in many more. River embankments collapsed in at least 66 places on 47 different rivers and streams. Hakone, one of the areas hit hard by Hagibis, received 922.5 millimeters (36.3 inches) of rainfall in just one day. Here we show some representative validation results of our coastal model and inundation model for Typhoon Haigibs:

  • Coastal model: our coastal model based on an advanced large-scale hydrodynamic model SCHISM (Semi-implicit Cross-scale Hydroscience Integrated System Model) for the entire Japan Coast, validated using tide data at major tide gauges from the Japan Meteorological Agency (JMA) across Japan with an average root mean square error of 0.15 m, and tested during Typhoon Haishen in 2020, is again validated during Typhoon Hagibis with difference of the maximum water level within 0.1 m when compared with the JMA Tokyo tide gauge, and that of peak surge timing within an hour, using the GFS 0.25-degree forecast as input.
  • Inundation model: a large-scale SCHISM inundation model co-developed with Virginia Institute of Marine Science was used. In Nagano, our inundation model was able to capture the inundation pattern of the riverine overflow flooding of Chikuma River when compared with observation (Figure 3) with an overall hit rate of 0.8 when using observed streamflow and reanalysis rainfall as weather input. Details and more results are included in future publications.

Flood defense: appropriately representing flood defense in the inundation model is essential to accurately capture the inundation pattern and river overflow. Japan is a country with extensive flood defense structures such as riverine and coastal levees, yet no comprehensive data of all the levee location and height exists. Multiple methods including machine learning models were explored to develop a scalable way for identifying levees and estimating their heights from other datasets such as high-resolution DEM.

We’re advancing science and technology to build global resilience, and make disasters less disastrous