Digital insights improve flood resilience in Bangkok

Stormwater floods are tricky to predict

Intense, highly localised tropical rainstorms can occur with little warning, overwhelming Bangkok’s drainage network. Storms may last only an hour or two but can bring huge disruption and massive economic losses – not to mention the impact on public health and safety.

While Thailand’s capital also suffers river flooding from the Chao Phraya, this is easily forecast as the flood peak can be tracked as it travels downriver. By contrast, stormwater flooding from intense rainfall is challenging to predict.

Unlike coastal or river flooding, which starts from a known source and propagates out from that location, stormwater flooding can occur nearly anywhere.

Sharla McGavock

Principal flood risk specialist

“Unlike coastal or river flooding, which starts from a known source and propagates out from that location, stormwater flooding can occur nearly anywhere – particularly for localised tropical storms,” says Sharla McGavock, our principal hydrologist.

“We cannot simply use trigger levels based on level or flow gauge data because this doesn’t capture the spatial variability of stormwater flooding, and the lead time is not long enough to prepare for a flood event.”

Stormwater can quickly flood Bangkok’s streets

Urbanisation is compounding the problem as green spaces and floodplains are lost to development, and expansion in hard surfacing exacerbates runoff. Drainage network capacity is being lost too, due to encroachment on canals, with rising population levels increasing the volumes of wastewater and solid waste. 

Climate change will also increase the risks of stormwater flooding in Bangkok, as extreme rainfall events become more frequent and severe, while rising sea levels interfere with discharge from the drainage network.

An early warning pilot

To respond proactively to flooding, Bangkok Metropolitan Administration (BMA) needs rapid, accurate information on where and when it will occur – but the city had no early warning system for stormwater flooding.

So, a pilot project to develop a decision support system (DSS) for flood management was initiated under the Global Future Cities Programme, funded by the UK Foreign, Commonwealth & Development Office (FCDO).

Intense, highly localised tropical rainstorms can occur with little warning in Bangkok, bringing disruption and loss

Mott MacDonald, as FCDO lead delivery partner for the programme in southeast Asia, designed and delivered what would become the world’s first flood warning system for fast-developing convective rainfall.

We did this by optimising Bangkok’s rain gauge network and rainfall radar assets, then applying machine learning to dramatically speed up predictive modelling. Finally, we integrated all the real-time rainfall data and flooding model outputs in Moata Smart Water, on our award-winning digital platform, Moata.

The resulting flood management decision support system, which can be accessed online from anywhere, uses a map-based interface to show past, present and predicted rainfall data and generate flooding alerts.

Key features – decision support system:

• Secure, private web portal
• Dashboard and interface tailored to client’s needs
• In-built flood alerts based on rainfall and water level triggers
• Data can be interrogated and analysed
• All observed data and forecasting information clearly visualised
• Real-time rainfall and flood forecasting maps and animations

Optimising local data sources

The city had already put in place various systems for measuring rainfall, but the data was not being used as well as it could – and that had to change.

Bangkok has a flood operation control centre that analyses data from multiple sources about conditions on the ground.​ These sources include real-time sensors measuring canal and river levels and flows, as well as water levels on critical roads. The city also has a rain gauge network providing data on average rainfall and rain accumulation and intensity.

Unfortunately, these various datasets were all stored in different locations and different formats, making it difficult to access and analyse information rapidly. We knew our digital tools could make a big difference here.

We also saw that we could improve the accuracy of rainfall estimates by making better use of Bangkok’s rainfall radar monitoring system. The city had invested heavily in C-band radar technology, but the outputs were inaccurate and only available as images, requiring manual interpretation. This meant radar data could only be used for post-event analysis.  

Vertically profiling radar calibrates the permanent C-band radars operated by the city

Calibrating estimates for rainfall

Working in partnership with Weather Radar New Zealand, we installed a vertically profiling radar to provide continuous calibration of the rainfall estimates generated by the C-band radar, which meant the system could produce data usable for prediction too.

The recalibrated radar data provides better statistical results than the rain gauge network so can identify faulty gauges. Those recording erroneously high or low rainfall are now automatically detected and their data withheld.

Because the radar measures rainfall away from the gauge locations, it also gives a more detailed picture of the spatial distribution of rain and so can show more clearly its pattern and direction of travel.

We created a workflow that uses the radar derived data to automatically generate highly accurate rainfall maps in 500m cells across the city. And we further developed these to produce ‘nowcasts’, live two-hour forecasts of how a storm will develop at a local scale.

Rainfall radar map showing estimated rainfall across central Bangkok

These targeted interventions have delivered a world class quantitative precipitation estimation system. Flood management staff can easily see, understand and interrogate the outputs from the radar system, gaining an immediate understanding of rainfall events and associated flash flooding.

Creating a robust hydraulic model

The 17km² pilot area selected for the project was in Wang Thonglang, a district in the east of Bangkok which, like much of the city, is low lying and heavily urbanised. It is bounded by external canals, with its internal canals gated and discharged by pumping.

We developed a conventional hydraulic flood model of the pilot area using TUFLOW simulation software. Major elements of the drainage network, including road drainage pipes, pump stations and canals, are explicitly represented. The model outputs spatial flood predictions, such as the extent, depth and duration of flooding, in the form of flood maps and animations.

This mathematical model was calibrated against previous flood events using historical canal gauge and radar data, and further validated by historical flood records, images and videos posted to social media by members of the public and consultation with BMA. The calibration and validation demonstrated that the model can confidently predict the onset of flooding and the areas at risk.

Despite recent advances in computational power, run times for this type of model are too long to provide predictions within the timeframes needed for effective flood warning and preparation. So our next step was to apply machine learning to produce a surrogate model able to skip the steps in which equations must be solved and emulate the hydraulic model at a fraction of its speed.

AI for accelerated modelling

Unlike the hydraulic model, the surrogate model works fast enough to predict flooding in real time. It is so quick that it can perform predictions for 41 rain events in 30 seconds – the hydraulic flood model would take 30 hours to do the same task.

To achieve this, the original model was run repeatedly to provide a dataset for training the surrogate through machine learning. ​The two models agree on the locations, depths and extent of flooding, and the surrogate has an average error relative to the hydraulic model of just 20mm. 

We have speeded up the time it takes to convert rainfall forecasts to flood maps from hours to just seconds.

Sharla McGavock

Principal flood risk specialist

The radar-derived live forecasting ‘nowcasts’ can be applied to the machine learning model to produce – in mere seconds – real-time stormwater flood maps that show the predicted locations, depth and extent of flooding over the next two hours.

“We have speeded up the time it takes to convert rainfall forecasts to flood maps from hours to just seconds,” says Sharla. “The system processes up to 20bn data points each day to provide real-time flood insights for Bangkok’s operators, with continuous rainfall radar calibration.”

Frequent updating is critical to the system. Observations are refreshed every five minutes and forecasts every 10 minutes – which is essential, given that a storm event rarely lasts more than two hours. ​By contrast, standard weather forecasts usually update only every three to six hours.

“This application of machine learning to enable real-time stormwater flood forecasting is ground-breaking,” adds Sharla. “We’re starting to see increasing use of machine learning for individual sensors. But applying it for spatial flood prediction maps is really at the cutting edge of flood forecasting.

Cities can get ahead of the flood and take a proactive response to stormwater flood events and target resources efficiently and effectively.

Sharla McGavock

Principal flood risk specialist

“We get a complete view of how a flood event is likely to unfold over the next two hours, which provides greater accuracy and a longer lead time for response than a gauge-based forecasting system. This means that cities can get ahead of the flood and take a proactive response to stormwater flood events and target resources efficiently and effectively.”

Key features – machine learning model

• Flood predictions in one second
• Relative error of 20mm
• 20bn data points processed each day
• Forecasts output every 10 minutes

Screengrab showing radar accumulation data (estimated rainfall) and level and flow gauge locations

Integrating data for early flood alerts

The final step in creating the flood management decision support system was to integrate, in one system, all the data from the gauges and sensors, radar-based rainfall estimates and modelling outputs.

For this we used Smart Water, an app hosted on our own digital solutions platform, Moata. It provides a graphic interface that presents all the forecasting data geospatially for ease of use. And because it is web-based, the DSS is available on demand.

Forecast flood depths and extents are automatically generated and displayed in map form on Smart Water. Animations illustrate how rainfall is expected to move across the city and how flood depths and extents are expected to develop over the following two hours.

The DSS generates alerts to give early warning of heavy rainfall or flooding; these can be set up based on actual or predicted rainfall or flood depth at any location. Faulty gauges or sensors in need of maintenance or recalibration will also trigger a notification.

Single source of truth

“The DSS provides a single source of truth for flood management staff, reducing the risk of miscommunication or confusion during an extreme weather event,” says Mott MacDonald’s Nasrine Tomasi, technical director for Smart Water. “They are now able to see accurate estimates of rainfall quantities and rainfall nowcasts without the need for judgment or interpretation.

“The Moata Smart Water platform makes the data available with minimal delay and in a visual format that is easily accessible. It is independent of available technology on site – if you have a laptop or a tablet and an internet connection, you are good to go.”

The system also displays the annual exceedance probability of observed and forecast rainfall, helping decision makers quickly get a feel for the magnitude of an event.

Rainfall and flooding forecasts can be interrogated for any location across the city, and users can extract catchment averages. For extra insight they can also ask the system to plot graphs, for instance showing cumulative rainfall for a specific area.

Stormwater flood maps show the predicted depth and extent of flooding in the pilot area

Harness the power of data with Moata

Our digital platform, Moata, hosts a range of solutions that use the power of data to solve today’s most pressing infrastructure challenges around the globe. It is open, secure, scalable and adaptable, delivering predictive power in a geospatial context, through advanced analytics and machine learning.

Moata Smart Water is a suite of cloud-based analytics and visualisation products providing real-time and predictive insights into water asset performance to enable proactive maintenance and better decision making.

It enables feeds from sensors, social data, weather data and modelling software and more to be combined with an accurate digital representation of systems and their components in their geospatial context.

The data derived in any Moata Smart Water digital twin can be easily accessed through Moata’s comprehensive API (application programming interface) and directly integrated into any enterprise system.

Future expansion options

The decision support system has been designed in a modular way to allow connection of additional data feeds and analytics, so it can be expanded for further functionality.

Future developments could include integrating data streams such as meteorological office weather forecasts and river flood forecasts, socio-economic and environmental data, or SCADA (supervisory control and data acquisition) data for better operational control.

By integrating all the flood monitoring and forecasting data in a single platform, we have opened up opportunities for development of the DSS. Essentially, it will keep on growing, and keep on improving.

Sharla McGavock

Principal flood risk specialist

The system could be linked to other digital twins, such as a transport digital twin to bring in live traffic data to enable better traffic management during flood events and keep the city moving.

It could also be scaled up and connected to public communications channels to provide flood warnings directly to communities across Bangkok.

Sharla says: “By integrating all the flood monitoring and forecasting data in a single platform, we have opened up opportunities for development of the DSS. Essentially, it will keep on growing, and keep on improving.”

Delivering better social and environmental outcomes

The decision support system not only increases urban flood resilience but can also help improve social equity and deliver environmental benefits.

Flooding widens social inequalities: cheaper or informal housing is often built on flood prone land, meaning the poorest people are hit more frequently. Older and disabled people, meanwhile, may find it harder to evacuate in an emergency. And those responsible for childcare, disproportionately women, will be most affected by school closures as a result of flooding.

Richer data analysis can ensure the needs of vulnerable people are considered in decision making, and that specific vulnerabilities of a city’s population are prioritised when developing flood mitigation and response measures. Looking at flood risks through this wider social lens can inform more inclusive adaptation strategies, aligning them better with desired social outcomes such as promoting gender equality.

In addition, inclusion of environmental data will facilitate a more holistic approach to water management, making it easier to identify opportunities for nature-based solutions.

Aligned with UN goals

The outcomes of the Bangkok project are closely aligned with the UN’s Sustainable Development Goals, primarily SDG 11: Make cities and human settlements inclusive, safe, resilient and sustainable, and SDG 13: Take urgent action to combat climate change and its impacts.

The SDG Project Assessment Tool, developed by UN-Habitat for the Global Future Cities Programme, helps cities to develop more inclusive, sustainable and effective urban projects and maximise their outcomes in terms of all 17 SDGs.

In a joint evaluation exercise between Mott MacDonald, BMA and UN-Habitat, the tool was used to evaluate the DSS and found it scored well against all key sustainability principles, reflecting the wider social, economic and environmental benefits of the project.

Throughout the programme we organised a series of webinars giving BMA access to global flood protection experts

A global blueprint for flood forecasting

The DSS project in Bangkok has accomplished its objective of proving that a real-time stormwater flood forecasting system can work. “We have shown how digital developments in machine learning and data integration can make better use of sensor data to improve flood management,” says Sharla.

The real-time forecasting delivered on this project really pushes the envelope of stormwater flood management.

Sharla McGavock

Principal flood risk specialist

“Furthermore, this was successfully demonstrated in a tropical climate where intense rainfall presents significant prediction challenges. The real-time forecasting delivered on this project really pushes the envelope of stormwater flood management. 

“The DSS serves as a proof of concept for cities across the world. This technology can be used globally to address an existing challenge that has until this point been deemed impossible. Previously it has been accepted across the industry that these events propagate too quickly to develop a warning system to allow proactive responses.

“The DSS provides a blueprint for scaling up flood warning systems for cities around the world facing increasing flood resilience challenges.”

The framework we developed will help Bangkok mitigate the impacts of climate change and improve the quality of life for its residents

The framework we developed in Bangkok alongside our partners RAB Consultants provides clarity on the best way to develop a flood response plan. Flood managers can make quick, proactive decisions based on the best available information of how an event is likely to unfold – rather than just where the flood occurs first. An effective warning system enables them to prepare in advance and target emergency response activities effectively.​

The system also helps cities plan for the longer term because it can identify flood mitigation opportunities and optimise existing assets for flood management, as well as being used to prioritise infrastructure investment and promote more sustainable development.

The ultimate outcome is a more resilient city that is better able to mitigate the impacts of climate change and improve the quality of life for its residents.