In geostatistical models that use for example kriging or IDW algorithms, are there particular rules that indicate what relationship should be between the number of sampled sites and the size of the cells in the grid used for the predictive model?
@Paolo, This may helps:
https://www.researchgate.net/post/How_do_I_choose_an_appropriate_cell_size_for_interpolation_from_point_data
I think you need to look at the literature on "block kriging"
A key question is what do you want to do with the interpolated cell values?
Note that IDW is not a geostatistical method/tool. It is just heuristic, there are no statistical assumptions
Are you assuming that all the data locations must be inside the "cell"? That is not true for block kriging.
Dear Donald, Many thanks for your suggestions. I will go through the literature on block kriging I didn't know sufficiently before. At first glance, it seems fitting very well to my question.
To explain a little better the context, we have ca. 30 sampling sites in a 20x20 km area, where we measured the trace elements accumulation in lichens. Then we ran some interpolations (right! IDW is just a heuristic method, not a geostatistical model, thank you!).
To answer your question, we want to determine whether the modelled accumulation in (non-sampled) grid cells is below or over a given threshold. I wonder how the cell dimension is affecting the final results and if there is any method to calculate the uncertainty for each cell formally.
As we have not true replicates at the site level (meaning that we have, unfortunately, a single sample for each point), we have not the opportunity to directly interpolate a descriptor of sampling uncertainty. I suppose it could have been a further option.
Many thanks, Donald and everybody, for any further suggestion.
Kind regards, Paolo
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Climate Science
- Climate Change