AXA-IRSTEA Joint Research Initiative "A Countrywide Probabilistic Flood Forecasting System"

Keeping Watch on Flooding, Countrywide

Major floods are dreadful natural hazards that spare no region of the world. The year 2014 began with high-impact flooding in the UK and in the Balkans; between 1987 and 1997, almost 230,000 people in Asia lost their lives to floods that created $136 billion in losses. The enormous burden of flood damage, along with environmental changes that seem to be increasing the frequency and intensity of flooding, makes the need for forecasting and warning systems increasingly vital. In a joint initiative with AXA Global P&C, Dr. Vazken Andréassian is working to create a country-scale river flood forecasting system integrating multiple types of hydrological data and capable of providing flooding maps. 

The team will begin by developing these methods for France, where Dr. Andréassian will attempt to tackle the difficult question of how different floods along a river affect one another: how downstream floods are reduced by upstream flooding, and how upstream flood defences such as dikes can aggravate downstream floods. To efficiently handle the huge volume of data of a countrywide model, Dr. Andréassian will assess to what extent it can be handled automatically. He will develop automatic data quality filters, inspired by the methods of expert human forecasters. These should help detect and correct problems with the data more easily and quickly, saving time for analysis. In any location, many waterways will not benefit from real-time data collection; in France 70,000 kilometers of streams are not monitored. Dr. Andréassian will, however, need to integrate these sites into his forecasting model. Working with AXA, he will develop methods to use flood damage estimates for these regions to replace the lacking data on river flows.

In flood forecasting, two types of models exist that provide different information. For example, one can simulate the predicted area of inundation, while the other translates rainfall into increased river flow. They can be used simultaneously, but, at the scale of an entire country, the computational costs of this become too great. Dr. Andréassian will also identify the most reasonable approaches and scales on which to couple these models. This will allow his system to offer the best compromise, capable of operating in real time to forecast and map flooding on a large scale. 

The end goal will be a system that constantly has its eye on watercourses countrywide, making it possible to provide early warning by distributing flood information broadly, to town officials and insurers, like AXA. Every party involved could be better prepared for a coming flood, reducing property damage and loss of life. Dr. Andréassian’s forecasting system will also be designed with the intention of adapting it to other countries’ context. Similar tools could be deployed around the world to prepare for and mitigate the great and increasing risk of watercourses rushing over their banks.