For the (frequent) case of desiring no edge-filling or nodata-filling method during terrain attribute derivation, we can use a 3D convolution (already implemented with a function hidden in spatialstats.py) for fast computing of terrain attributes. This should work for almost all but fractal rugosity. Convolution can be done through SciPy or Numba (which one is fastest depends on matrix size and convolution kernel size).
Additionally, our current functions are deriving all coefficients of Zevenberg and Thorne (A, B, C, ... to I), while less than half are typically used for terrain attributes! (A, B, C and I are virtually useless, and D, E and F only required for curvature).
To-do
[ ] Add implementation for windowed attributes,
[ ] Add tests,
[ ] Homogenize current implementation to support the new functions.
Bonus?
[ ] Check if we can't derive 2D planar fit coefficients analytically for a certain grid size and turn them into kernels (we should be able to!)... To provide the means of deriving slope/aspect/curvature over any grid size! :smile:
Ongoing
Resolves #302
For the (frequent) case of desiring no edge-filling or nodata-filling method during terrain attribute derivation, we can use a 3D convolution (already implemented with a function hidden in
spatialstats.py
) for fast computing of terrain attributes. This should work for almost all but fractal rugosity. Convolution can be done through SciPy or Numba (which one is fastest depends on matrix size and convolution kernel size).Additionally, our current functions are deriving all coefficients of Zevenberg and Thorne (A, B, C, ... to I), while less than half are typically used for terrain attributes! (A, B, C and I are virtually useless, and D, E and F only required for curvature).
To-do
Bonus?