Modelling framework

The iTURF framework is composed of three main elements. The details and research underlying each will be described here in detail as the project develops. A brief description of each component is provided below.

  • Social media harvesting

Data are collected in real time from Twitter and stored within a spatial database structure. Relevant hashtags such as #toonflood are continuously monitored, as are keywords such as 'flood' within Tyne and Wear. Each tweet is automatically analysed for words which infer a spatial location or a severity of flood, or if an appropriately accurate geolocation is attached to the tweet this is also used. Street names, postcodes, landmarks (e.g. shopping centres, transport hubs) are all recognised, and a measure of relative confidence in the location is calculated.

  • Hydrodynamic modelling

Simulation for floods of differing intensities are run sumultaneously on a high-performance computing system at Newcastle University. This system takes advantage of graphics processing units (GPUs) to perform the same calculation millions of times across a grid, thus providing results several times faster than more traditional central processing unit (CPU) based software. The hydrodynamic behaviour of water is simulated using the non-linear shallow water equations, which are solved using a finite-volume approach capable of accurately representing rapidly varying and transient flow conditions. The Open Computing Language (OpenCL) ensures compatibility with a huge range of modern computer processors. 

Parameters are perturbed in each simulation to provide differing rates of rainfall intensity and losses to the sewer and drainage systems. The best match to the data obtained from social media is then identified, and a higher resolution, more refined flood map can be generated.

  • Web publishing and feedback cycle

The resulting flood map is made available through a web interface for the public and authorities to use. Strategic transport routes and areas likely to flood if rain continues can also be identified from the results. Members of the public can help improve future results by identifying errors in the model results, or in the interpretation of tweets. Such corrections if approved by the project team are then fed in to the system for processing any future flood events, and the cycle repeats.

Should you require any further information regarding these components, please get in touch.