Flood Event Model predicts where emergency resources should be located during extreme weather conditions. A flood event is defined as a single requirement, based on location, for emergency resources, during a river flood, a flash flood or coastal flooding. The resources can therefore be anything from human rescue service resources to sandbags and warning signage.
It is a statistical model, built from static geographic data, variable weather condition data and historic emergency incident data. Currently, the area of Hampshire in the south of England has been modelled and the Flood Event Model is capable of predicting where these flood events will occur with approximately 85% accuracy. The resultant data can be projected onto a map or output as locations which integrate into resource management systems.
Flood Event Model has been designed from the outset to save lives as well as money. Initial development and testing has focussed on events caused by cars driving through fords in the road. 32% of flood-related deaths are by drowning in a vehicle and the cost of sending a rescue appliance (typically a fire engine) can reach £20,000.
There is an immediate intention to scale out the capability of Flood Event Model to the rest of the UK and in time, to all parts of the planet. There is also no reason why the tool can’t scale up to include other environmental emergencies such as Wildfires, Tornadoes and Earthquakes.
At the moment Flood Event Model is a statistical model running on a supercomputer. In the coming weeks it will be developed as a consumer application to allow people to see the risk levels in their own area.
Flood Event Model was developed by KnowNow Information in collaboration with the Science and Technology Facilities Council (STFC) and IBM. It currently uses only data that is open and freely available to build a series of correlations that lead to a predictive model.
Data was available and usable from 1999 to 2013. The data from the period 1999 - 2007 was analysed for relevant correlations which could then be used to build a hypothetical model that would then be tested against the period 2008 - 2013. The data sets used were taken from the Ordnance Survey, the environment agency, the met office and various sources stored on the Hampshire Hub including Hampshire County Council, Hampshire Police and Hampshire Fire and Rescue Services.
The model runs on a supercomputer at the Hartree Centre, part of the STFC’s Daresbury Laboratories. The computer is an IBM iDataplex cluster and the model uses 39 cores. The software that was used to create the model includes IBM’s SPSS Modeler and BigSheets.
Current development is focussed on creating a customer and public facing application that will allow people to enquire the model for their location. This is being created in IBM Bluemix and currently takes weather predictions from forecast.io.
There are plans to include some paid data sets which it is hoped will increase the accuracy of the model to around 90%
The original intention of Flood Event Model is to save lives. Delaying response to a water rescue from 5 minutes to more than 10 minutes multiplies risk of fatality by more than 4 times. We are trying to use a predictive model so that the resources can be in place or nearby before the event is likely to occur.
By accurately predicting the location of flood events we hope to demonstrate that you can dramatically cut the cost of dealing with the impact of flooding. With a callout to deal with an incident costing up to £20,000 it would only take a few successful predictions to ensure that the development of Flood Event Model pays for itself.
The fundamental principal demonstrated by the Flood Event Model is that you can take historical data for the environment and the emergency resources that have been deployed in the past and create a predictive model. This has not required any geographic or geological expertise up to this point. It is therefore clear that you could expand upon the Flood Event use case to develop similar systems for other environmental effects such as Earthquakes, snowfall or even traffic.
There is an investment opportunity for an organisation to help to get the whole planet within scope. There has been interest from India in particular in how the model might be applicable to various regions there. We anticipate that the cost of expanding the model for the whole UK would be in the region of £1m and it would probably be a similar figure for other territories.
There are numerous industries that have expressed an interest in using FEM. This includes the industries for which it was originally designed - Fire & Rescue Services and Local Authorities. But it also includes the Resource Management companies that serve those organisations (in Hampshire where the system is piloted the Resource Management company is Amey and they are responsible for deploying the warning signs in the event of flooding causing fords in the road). Re-Insurance Companies are also highly interested as a system such as Flood Event Model will allow them to calculate the risk per area so that they can take action to mitigate those risks before having to pay out claims to insurance companies.