I know the issue of resampling has been raised before but I’ve noticed something which concerned me a bit in my latest use of the SDR model.
I am aware that input datasets are resampled to the resolution of the DEM, which in this case is 30 m. For the R factor, I have been using the global R Factor map from ESDAC which is recommended in the user guide and forums. As many of you will know gives rainfall erosivity in MJ mm ha-1 h-1 yr-1 at 1 km resolution.
My worries emerged when examining the “aligned_erosivity” raster intermediary output. I see the cell size for this raster was changed correctly to match the DEM, however the values on each pixel are the same as the original raster (aside from some interpolation). I could be getting this wrong, but in my mind should the per pixel value not go down when this dataset is resampled to 30 m resolution? For example, if the initial value is 1000 MJ mm/ha/year, should it not be divided by 11.1 (i.e. 100^2 / 30^2) in order to convert the per hectare value to a per pixel value (at 30 m resolution) of around 90 MJ mm/ha/year?
I am not sure whether my thinking is correct here but my worries emerged when I was comparing the USLE raster output from the model with other USLE studies. The InVEST output seemed like it was an order of magnitude or more too high. For example over Lesotho, the average USLE value from the model is 16.9 t/pixel/year (30 m), which translates to around 188 t/ha/year! This is confirmed by doing Zonal stats of my USLE layer for the whole country, which gives a figure of around 600 Million t of soil loss across Lesotho (which has an area of a little over 3 million ha). Lesotho is a mountainous country but this still seems really high. It’s almost as if the per pixel value is in fact the per ha value, and the total erosion is thus 11 times too high. My settings for the C and P factors seemed reasonable (I have added a file with a land cover legend and these values for reference). My K factor raster shouldn’t be the issue either, as the maximum value of this is around 0.05.
Given the error seems so big, I’m wondering if it could be something to do with a resampling issue like the one I mention with the R factor. If that is indeed the case, then it seems to me the same logic would apply to the per pixel values in the K factor raster, since they are also calculated as a per ha value. I could be really getting the wrong end of the stick with all of this though.
Looking forward to your thoughts.
Biophysical.csv (952 Bytes)