An inquiry about Interpolation Model Validation.
I ran the interpolation of IDW, Ordinary kriging, and EBK. But the R(sq.) values for the all these models (including the semivariograms for OK) rarely exceed 0.1 (even sometimes 0.007 is the highest).
Is model with R(sq.) value of 0.007 good for publication? I think this value indicates too poor prediction, but none of these models is showing a decent R(sq.) value.
On the other hand, the RMSSE is really close to 1, and mean error is around 0.009.
What should I do know?
What can be the possible reason? Am I missing something? Or I should follow more complex models? Is the spatial distribution being controlled more by extrinsic factors (e.g. human interference)?
[120 samples were collected randomly from this study area].
I would gladly appreciate any suggestion.
Just a suggestion, overlay your data with topography (contours or elevation) and soil type, and perhaps land cover type as forest, grassland, and sometimes even vegetation types might produce some differences, such as pine vs deciduous forest, or areas of study that were severely burned, or severely eroded in past land uses, etc. The ground based history and conditions and spatial differences may help explain some of the differences found.
Show your original sample points. Interpolation is based on an assumption that there is some sort of continuity between the values. Any interpolated surface is suspect at the edges. Basically the same question asked in October, see my respnse https://www.researchgate.net/post/Which_one_is_better_between_IDW_and_Kriging_for_Interpolation the same response and especially read "An Introduction to Various Spatial Analysis Techniques" Chapter An Introduction to Various Spatial Analysis Techniques
As @William-Hansen-4 suggested overlaying a different data set may help you determine that, like soilgrids.
What did you compare your result with? That will determine why you have a low r square
William F. Hansen Excellent suggestions, the Soilgrids folks have very extensive documentation of their methods of interpolation. My 'guess' just looking at the plots is it is very under-sampled and the algorithm is dominated by edge effects and having insufficient distribution of neighboring points. The ArcGIS documentation also has a fairly good explanation of sufficient data characteristics. An actual physical description of the 'sampling' and what other values were taken would also offer some clues. Around here you can hit a 100% carbon layer ( forest fires ) and 100% silica ( volcanic ash ), LoL :-)
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts