FloodNET is best-described as an 'object-oriented' flood modelling system, currently being developed in the Soils and Environmental Engineering Science Laboratory at the University of LiverpoolInstitute for Risk and Uncertainty. Rather than aiming for a precise but computationally expensive hydraulic description of a flood event - for which there are many well-developed models already in existence - FloodNET aims eventually to deliver rapid, interactive and on-demand flood risk models for any location in the world. While lacking some of the physical details attainable by fully hydraulic modelling, the approach is focused on providing data essential to flood risk planning and response, with the minimum of parameterisation. We therefore aim to fulfil the modelling axiom 'simplify as much as possible; but do not oversimplify'.
HOW DOES FLOODNET WORK?
The core of FloodNET is a novel agent-based model (ABM) constructed in the Netlogo programming environment. ABM are more typically used in the social and biological sciences, but they hold many attractive features for modellers of abiotic physical systems. We use ABM to describe 'packets' of flood water, typically each of constant volume and depth, which interact with each other and with a digital terrain model using simple, physically-based rules to move between adjacent 'patches' - fixed areas of land surface each with a characteristic elevation. Conceptually, this is similar to a classical Lagrangian particle-based scheme, or a Lattice-Boltzmann approach to the discretization of fluid flow across the spatial domain.
Several different modes of operation are being developed within the FloodNET model. However, the approach underpinning most of the maps shown on the FloodNET.eu website is a high-efficiency 'Unlimited Steady-State Flood' assumption. In this mode, FloodNET estimates a 'worst-case' flood for any specified river stage (level above base-flow) which assumes that the flood hydrograph has reached a steady maximum state and that the discharge arriving from upstream is not a limiting factor in determining the amount of flow across the floodplain.
The user specifies points on the map corresponding to the river reach being modelled, and these serve as 'seeds' for the position and local elevation of the flood. The model therefore is not simply a 'DEM slice' but is responsive to the steady fall of the channel with distance downstream. For a simple simulation, the user specifies a 'flood stage' (the level of the flood above the river base flow) and runs the model. Each river 'seed' creates water 'agents' with depth equal to the flood stage and 'total head' equal to its depth plus the local elevation. Each agent then looks for adjacent unoccupied patches and creates a new agent on that patch. The new water agent has the same total head, but the water depth is calculated by subtracting the new local elevation. The process is repeated and the flood propagates across the floodplain, with a 'total head' corresponding to the flood stage at a 'hydraulically connected' point on the river (usually, but not always, the nearest straight-line position on the river reach). The flood propagation halts when the calculation 'total head - local elevation' results in a zero or negative value. You should be able to see a video, below, which illustrates this model in action. The video is in near-real time, and also serves to demonstrate the rapid execution of the model. Significantly faster operation is achieved with 'live' view updates switched off.
WHAT IS 'FLOOD RISK INTENSITY'?
In FloodNET, we define flood risk intensity as 'the percentage of floods within a range from baseflow to a specified flood stage, for which a given location is inundated'. It is obtained by running the FloodNET model iteratively over a continuous range of flood stages, and is usually combined with a stochastic varying of elevation data within the error bounds of the applied DEM, resulting in a rich dataset with mean and error bounds on estimations for each flood stage and location. Flood risk intensity is not:
likelihood of inundation for a specific flood event.
Flood risk intensity is a measure of the relative risk of flooding for different parts of a floodplain, based on a knowledge (or estimation) of the maximum probable flood stage for the adjacent river reach and a basic assumption that there is an equal probability for any of these flood stages occurring. In practice, this assumption is highly unlikely to be valid, since it is evident that more extreme events occur relatively less frequently. However, the assumption therefore yields a reasonable 'worst case' scenario for flood risk under high-flood-stage conditions. Where reliable flood frequency analysis is available for a location, this can be mapped onto the FloodNET outputs to yield a specific flood risk intensity; this is a subject of current development [23-5-16].
WHERE DO THE DATA COME FROM?
The emphasis of FloodNET is on simplification. Therefore the only data inputs required to run a flood simulation are a digital elevation model (DEM) covering the required area to be modelled. For display and interpretation purposes, the FloodNET Netlogo model allows the user to import a basemap to overlay the DEM data, but this does not affect the simulation. To run a simulation in Unlimited Steady-State Flood mode, the user must specify the location of the river channel - so this must be reasonably identifiable within the DEM or the base map. To get reasonable results, the user should also have an estimation of the range of flood heights experienced by the river reach in question. (Note that failure to properly identify a river, or specify reasonable flood stage, does not prevent the model being run - but the outputs must be interpreted on the basis of 'nonsense in = nonsense out'!)
The DEM input format for FloodNET is 'headless .asc'; in other words, a standard ESRI ASCII grid file (.asc) with the header data removed. This allows for simple transfer of data to and from ArcGIS and widely-available data sources. DEMs currently in use within FloodNET:
Shuttle Radar Topography Mission (SRTM) 90m global elevation dataset - particularly suitable for large-area models, such as the Upper Tigris/Mosul Dam collapse;
ASTER global DEM (30m resolution);
a bespoke reprocessed combination of ASTER and SRTM, yielding substantial improvement in vertical and horizontal error at 30m resolution, produced at SEESLiverpool by Dr Z. Abdeldayem;
A diagrammatic workflow for the FloodNET project is given on the About page. We run standard desktop PCs operating Windows 7 Enterprise edition. Once downloaded, DEMs in .asc format are de-headed using Notepad and are then immediately ready for use with the Netlogo model. For display on this website, model output is saved pixel-by-pixel in .asc-compliant format and 're-headed' in Notepad using the original .asc header data. These are then loaded into ImageJ image processing software where a colourful lookup table is applied and the background is made transparent. Data are then saved as .png format and uploaded to the server. We use Openlayers 3 to display the data online, overlaid on Bing Maps global mapping data. Netlogo FloodNET records the inundated pixel area as a function of flood stage and this is exported to a .csv file. We use Excel to convert the data to real spatial units (based on the DEM resolution) and calculate means and standard deviations as required (this could be done within Netlogo instead). Finally this is copied directly into the html5 page code for the relevant model project, and visualised via the lovely Plot.ly. Videos are made using a standard screen recording software and converted to .mp4, then embedded using native html5.
HOW RELIABLE ARE THE FLOODNET MAPS?
Hmm... the killer question! This is a subject of continuous research work, which we plan to publish on an ongoing basis as new case studies are investigated. First off, to reiterate our mission statement: FloodNET aims to respond rapidly to flooding situations around the UK and the world, providing freely-available, scientifically-robust flood area maps. Noting everything that has been said above, we therefore cannot claim to do everything in flood modelling - for specific estimations of velocities, flow-structure interactions, supercritical flow conditions and so on, please go somewhere else (there are plenty of models available).
We would claim to be able to produce some of the most high resolution flood maps, both in terms of spatial resolution (through the in-house 'improved' DEM method by Abdeldayem) and uncertainty resolution (since FloodNET runs very quickly, it can iterate through a large number of different flood scenarios and across DEM error space). This allows us to express quantities like 'flood risk' and 'modelling uncertainty' much more rapidly and transparently than more physically-complete hydraulic models.
Finally, we believe that our innovative web-oriented workflow, which allows flood risk intensity to be interactively viewed down to street or household level, provides an unparalleled quantity of well-specified, timely information for users - which is in the end the ultimate objective of any flood modelling. Hence, our description of FloodNET as 'object-oriented flood modelling'.
You may be suspicious of too much waffle in response to a simple question. There follows some evidence in support of our estimation that FloodNET provides robust flood area models. Please feel free to comment on any of this.
Comparison between best-resolution public Environment Agency map and FloodNET - Whalley case study The image on the left is taken from the Environment Agency's UK-wide Flood Map for Planning at the highest publically-available resolution. There is no readily-available information on the models and methods used to obtain the simple flood zonation. The image on the right is the FloodNET flood risk intensity map for Whalley, scaled to the same resolution. This website provides transparent information on the method underpinning the flood map. The project page allows further zoom to building-level resolution, and provides a flood-stage relationship indicating the range of error on the FloodNET mapping.
Comparison between published flood maps for Mosul City and FloodNET maps
These four maps compare FloodNET output (explorable in detail here and here) with two publically-available internet maps of the Mosul dam-break flood in Mosul city. Top-left is output from an innovative and ambitious online flood estimation tool, floodmap.net created by Sameer Burle. Top-right is equivalent FloodNET output. Floodmap, although offering instant replotting of data, is an example of a DEM slicer - 'floods' are assumed to inundate all points below a specified elevation, regardless of location or realism. You can see that the flood area prediction is much smaller upstream and much broader downstream, which is incorrect. FloodNET uses local elevation and rules on hydraulic connectivity to plot a much more reasonable flood across the whole reach.Bottom-left is a very widely cited and adapted map by Al-Taiee and Rasheed (2009) based on the Simple Dam Break hydraulic model. This is taken from their paper and is the highest publically-available resolution. Bottom right is the equivalent FloodNET output, showing excellent areal correspondence.