FastFlood Docs

Compute

Local compute

The FastFlood method provides such an advance in compute time, that we can run the model on the users hardware (client-side). The current version of the FastFlood model carries out all computations on the central processing unit of the users device, wether that is a phone, tablet, laptop, computer or server. We facilitate this by running the model through WebAssembly technology, reaching near speeds of the C++ programming language compiled to native byte code. If you are reaching the limits of our web platform, please contact us for options in stand-alone executables.

Compute requirements

Expect the following memory requirements and compute time for various grid sizes:

  1. 1000x1000px 48 mb memory, 1 second compute
  2. 2500x2500px 300 mb memory, 5 second compute
  3. 10000x10000px 4.8 gb memory, 100 second compute
  4. 25000x25000px 30 gb memory, 1000 second compute

If you think this is a lot of compute, try running a traditional full flood simulation, for example using the www.lisemmodel.com model, and find it is about 1500x slower! Note that the model runs on your own device. There is no server doing calculations for you, which allows for this tool to be free. Having a stronger cpu and more memory can help with compute time.

The compute time for the model depends on a couple factors

  1. Domain size (pixel count)
  2. Settings

For higher domain sizes, more compute is required. In addition, to this linear increase with pixel count, the flow accumulation and hydro-correction schemes must carry out more iterations as well. The settings play a crucial role as well. The non-linear correction makes heavy use of power-law estimation on the CPU, which slows down the model by a factor 2 approximately.

If you are have trouble with simulation time, try to lower the resolution of your simulation. It is typically pointless to simulate large regions at anything more than 10 meter resolution. Hybrid approaches (feeding a small detailed simulation with output from a regional simulation) are much more suited then.

Where is the data

Because the FastFlood model runs all calculations on the users side, no data needs to be up and downloaded. While it may seem like the browser is taking up and downloads to the fastflood.org website, in reality the geotiff files provide by the users are parse and stored in user-machine memory. Thus, there is no reason to worry about data rights and confidentiality, as the FastFlood servers never see any of your data, processing steps, simulations details or anything else. All that is known is the usage of the global datasets that are available for automatic download.