I’m working in R using the gstat package and I’m using the universal kriging to interpolate temperatures and precipitations in Europe. While for temperatures the issue is quite rare, I often have final raster surfaces with some high-concentration values (peaks) where some meteostations are located.
Does anybody know how to solve it? Is this connected to the spatial distribution of points or to variogram fitting? How could I work to generate more smoothed surfaces?
Thanks
As rainfall data etc. will not change such quickly, I suggest just running an additional low pass filter on the result of kriging
Dear Maurizio,
The resulted downscaled surface is more smoothed if the variogram model the gstat::krige() function uses has a larger nugget. You should
1) use your own variogram model created with vgm();
2) fit only the range of the variogram but not the sill/nugget with fit.variogram(object, model, fit.sills = FALSE, fit.ranges = TRUE); or
3) manually increase the nugget parameter of a fitted variogram model.
If nugget is 0, kriging is an exact interpolator. I.e., downscaled values will be equal to the observed values in the original, coarse resolution grid, which may result in peaks.
HTH,
Ákos
You can also try to use a different variogram function when it makes sense and increases your prediction accuracy (e. g. Gauss instead of quadratic).
I presume you talk about a bulls eye effect. Try to use covariates to interpolate your variables: co-kriging with elevation as co-variable, repression-kriging with elevation, latitude, longitude as predictors, local regression like GWR with elevation as predictor.
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity