Potential with kernel

configuration-examples Potential with Kernel

The potential function performs a neighborhood operation. It computes an output grid where the resulting value for each cell is the sum of the values of all the input cells within a specified neighborhood around that cell.

This specified neighborhood is defined by the kernel (also known as distance matrix). In ArcGIS the operation is known as focalsum (part of focal statistics).

Within the GeoDMS potential calculations are implemented using a convolution by Fast Fourier Transformation (FFT) algorithm. This makes potential calculations much faster than focalsum calculations, especially with large kernels.

in use for

Neighborhood operations are for instance used to:

  • model the attractivity of a land use type in a suitability map based on the distance to for example an airport or a shopping area. a distance decay function can be used in the kernel to model the diminishing attractivity.
  • spread out densities of target groups, for instance inhabitants above 65 years old. They are often concentrated on a few locations (old age or nursing homes). The potential operation can be used to get a clear view of the density of an area and help to prevent in tracing back data to individual persons (often not allowed under AVG/GDPR regulations).

example

The example originates from a performance contest of the Spatial Analysis department of the Vrije Universiteit Amsterdam. In the contest a 1 km world wide grid and a 50 km kernel were used.

container contest
{   
   unit<fpoint>  WorldBaseUnit;
   unit<float32> potential;

   container Kernel
   {
      unit<uint32> Dist2Range;
      unit<spoint> pot50km := range(spoint, point(-50s, -50s), point(51s, 51s))
      {
         attribute<Dist2Range> distMatr  := dist2(point(0s, 0s, .), Dist2Range);
         attribute<Potentiaal> AbsWeight := distMatr <= 2500 ? 1s : 0s;
      }
   }
   container SourceData
   {
      unit<wpoint> wetlands
      :   Descr           = "Arc/Info binary grid"
      ,   StorageName     = "%sourcedatadir%/SpaceToGo/wetlands1"
      ,   StorageType     = "gdal.grid"
      ,   DialogData      = "WorldBaseUnit"
      ,   StorageReadOnly = "True"
      {
          unit<wpoint>     World1kmGrid := TiledUnit(point(1024w, 1024w,.));
          attribute<uint8> ReadData  (World1kmGrid);
          attribute<bool>  IsWetLand (World1kmGrid) := ReadData == 1b;
      }
  }
  container Result
  {
    attribute<float32> Potential50km (SourceData/wetlands/World1kmGrid) := 
       potential(
            float32(SourceData/wetlands/IsWetLand)
           ,float32(kernel/pot50km/AbsWeight)
       )
     ,  StorageName = "%LocalDataProjDir%/Potential50km_float32.tif";
   }
}</B>

explanation

In the example two grid domains are in use

  1. A geographic grid with values from which the attribute isWetland is derived, read from an arc-info-binary-grid. The files for this grid are stored in a %SourceDataDir%/SpaceToGo folder.
  2. A non geographic grid domain (the values are not related to a location on the earth surface), a so-called kernel, in the example named: pot50km.

The actual potential calculations is perfomed in the bold line. Both arguments are converted to float32 value types, as this is required for the potential function.

The calculated results are stored in a geotiff file called: potential50km_tiff.

performance

Using the focalsum function in ArcGIS, this example from the contest took more than 24 hours to calculate. In the GeoDMS, on a similar machine, it took a few minutes to calculate the same results with the potential function.