Predicting the Storm Surge for Super Typhoon Haishen
How we used our emerging large-scale coastal flood model to forecast Super Typhoon Haishen’s impact on Japan
Authors: Zhuo Liu, Ph.D., Yi Liu, Ph.D., Shabaz Patel
On August 29, 2020, the Japan Typhoon Warning Services began tracking a very disorganized tropical disturbance in the Pacific Ocean. Over the course of the next few days, the disturbance quickly intensified into a tropical depression, and then to a tropical storm, when forecasters gave the developing typhoon its name: Haishen. Super Typhoon Haishen, which became the strongest typhoon on record in 2020, was initially projected to make landfall in Western Japan — bringing concerns of a catastrophic disaster for a large region of the country.
One Concern’s Emergency Support team, led by our Chief Emergency Management Officer and former FEMA Administrator Craig Fugate, ensures that we can quickly adapt to these kinds of developing events. With team members from all corners of the organization — from data science to customer success to senior executives — this group convenes quickly and regularly in the event of an emerging disaster, providing critical information and platform support to the customer, and capturing key data for validation and enhanced product development.
Thankfully, the impacts from the typhoon were not catastrophic. This event, however, provided an outstanding test of our real-time operational capabilities as a company, and an opportunity for our Data Science Team to obtain validation data for our developing large-scale coastal flood model. Running this model for predicted impact on Kyushu island, our prediction fell within 0.2 meters of the actual gauge observations from the Japan Meteorological Agency (JMA).
Case Study: Storm Surge Analysis of Super Typhoon Haishen
One Concern’s hyperlocal flood model, currently deployed in our product in Kumamoto City, operates on a threshold basis: flooding predictions are only generated if the forecast data reaches a certain threshold, indicating a potential flooding event. For Typhoon Haishen, none of the forecast data indicated a potential major flood, so no prediction was generated for our product in Kumamoto City.
However, to test the capabilities of our preliminary large-scale coastal flooding model (currently in development) against the reality of the incoming storm, the One Concern Data Science Team performed several analyses of the data received from weather forecast services. As Typhoon Haishen was developing, the team applied this model — an advanced large-scale hydrodynamic model, SCHISM (Semi-implicit Cross-scale Hydroscience Integrated System Model) — with typhoon forecasts as inputs, to predict the storm tide along the coastlines of Japan.
It is important to note that One Concern is not a weather forecasting service. Our team utilizes data and information provided by external meteorological services to predict the potential storm tide (storm surge + astronomical tide) given inputs from those external typhoon forecasts.
Initial Predictions and Uncertainty
Typhoon forecast tracks often change frequently and dramatically during an event, giving uncertain predictions that underpin the necessity of continuous operational monitoring. During this event, early forecasts indicated a track towards central Japan, then shifted to indicate a direct hit to Kyushu Island, and later forecasts showed the storm steering away from Kyushu, but remaining just off the southern coast.
The predicted track history of Super Typhoon Haishen is shown below, with the final track — or eye locations — displayed as yellow dots.
Given the high uncertainty in the input typhoon forecasts, the One Concern team decided to monitor the storm and run 5-day storm tide predictions each day, beginning September 2nd, to have the most up-to-date results available should they be needed. As noted, the predicted results for each five day period changed daily, as typhoon forecasts were updated.
One Concern Model Outputs
Each of our SCHISM predictions were based on the National Oceanic and Atmospheric Administration (NOAA) GFS forecast initialized at 12:00 UTC each day. Below, we present all 5 predictions of total water level to demonstrate the dynamics of storm surge prediction during a typhoon event (shown in the GIF below). The accuracy and confidence in the predictions are typically expected to improve as updated hurricane forecasts are generated and integrated.
At the early stage of Haishen, GFS forecasts showed the typhoon would make a potential direct hit to the Kyushu Island of Japan. So, in predictions 1 and 2, higher water level (~5 m above Mean Sea Level or MSL) occurred in the Northern Ariake Sea, while the max water level along the Kumamoto City coastline was predicted lower, at ~3 m above MSL. This made sense as the typhoon track aligned with the geometry of the Ariake Sea, pushing water up North, resulting in higher water level in the North.
As Super Typhoon Haishen’s forecast changed over the next 48–72 hours, the track was predicted to steer away (a bit westward) from making a direct landfall in Kyushu. Thus, in our subsequent predictions (Prediction #3, #4, #5), the peak water level (relative to mean sea level) along the Kumamoto City coastline reduced to ~2.2 m above MSL.
Our prediction matched well with Japan Meteorological Agency’s (JMA) preliminary observation (9/3–9/8) at Oura (a region near the border of Saga Prefecture and Nagasaki Prefecture, facing the Ariake Sea coast). Our model produced a prediction with an error of less than 0.2 m: our predicted peak water level value was 2.6 m MSL, compared to the observed value of 2.5 m MSL.
Conclusion
Our mission is to make disasters less disastrous. Doing so requires a continuous commitment to deploying the most effective, efficient, and realistic techniques to model disaster events, and evolving our technology alongside our understanding of these phenomena. At One Concern, we’re conducting research and developing our capabilities in coastal storms — an increasingly important field of flood modeling for Japan — in order to generate inundation extents during typhoons and other events. SCHISM, the model detailed in this case study, is one of the models used in our pipeline to capture storm tides, and this analysis serves as part of the company’s ongoing research in this area.
In this case study, we demonstrated how our preliminary large-scale coastal flooding model can predict the potential coastal storm tide, based on the specific typhoon forecasts for Super Typhoon Haishen. In the above results, we were able to provide the potential impacted areas in SW Japan, the peak water levels along Kumamoto prefecture coastline at least 48 hours in advance prior to the event, and update our predictions with newer typhoon forecasts as they became available. Our validation process revealed that our results matched well with preliminary JMA tidal gauge observations, with an error value of less than 0.2 meters.
As our models are inherently based on uncertain typhoon forecast predictions, we are working hard on both model improvements as well as engineering improvements to run models efficiently as soon as new predictions are available, and to provide uncertainty analyses in the future. As our technological capabilities grow, so too does our team and our ability to collaborate with partners around the world — helping us scale, evolve, and ultimately live out our mission to build resilience to disasters, everywhere.
Note to reader: The model described above is in development by One Concern, as part of the company’s ongoing research into the challenging and increasingly important field of coastal flood modeling for Japan. All results are preliminary and for discussion purposes only.
Interested in learning more about One Concern’s research? Head over to our website or our Medium page to check out some of our recent projects!
