I am utilizing Area-to-Area CoKriging (AtACoK) for raster downscaling in R using the atakrig package. For this purpose I have one dependent variable (let's say x) with pixel size of 400m and 2 covariates (c1 and c2) with 100m pixel size. When I perform AtACoK the resulting raster has pixel values from 0:Infinity. What might causes the problem? For comparison, when I perform AtAK, that is, using a single independent variable the downscaling works fine.
Sounds like a numerical issue. Could be many things, datatype overflow, inappropriate variogram, or some peculiarity with the method used for the CoKriging. Try to use a simple variogram such as linear, try with smaller rasters to make sure there are no memory issues. Make sure the raster data types are the same and highest range possible, say 64-bit float. If nothing works I would use a different software package.
Neetesh Sharma I guess you are right. But as I said, I did try Area-to-Area Kriging using 1 ancillary variable and the algorithm works fine. What I haven't tried yet is to set the data type to the highest range, that's a good idea. Also, I attached the variogram (1 ancillary variable) and the cross-variogram (2 covariates). To me they look okay (please correct me if I am wrong). Lastly, do you know any package in R that performs Area-to-Area CoKriging? I think gstat doesn't include this.
Dear Nikolaos,
function ataCoKriging() from R package 'atakrig' may be of your interest.
Have a nice day,
Ákos
Dear Ákos Bede-Fazekas that's the package I am using for the downscaling (I forgot to mention that, I edited my question). As Neetesh Sharma mentioned, there is a numerical issue with the CoKriging function in the package. I tried
At this point I am out of ideas so any new suggestions would be greatly appreciated. I can share my code and the data, if required.
Dear Nikolaos,
OK, sorry, it was a new info for me. In this case I would recommend to get in touch with the package maintainer (Maogui Hu, humg at lreis.ac.cn) or raise an issue in github (https://github.com/maoguihu/atakrig/issues), but as far as I see, you are a step further then me. So fingers crossed, I hope the package maintainer can help. I have no idea what to do, sorry,
Ákos
Thank you Ákos Bede-Fazekas for sending me Maogui's website. I came in touch with him yesterday and he offered to help me. I gave him my data and my R code, I told him what I have tried so far and as you said, fingers crossed. If I can't proceed with CoKriging I will rely on Area-to-Area Kriging (using 1 ancillary variable) until Dr Maogui manages to solve the problem.
- Badges
- Science topic