Interpolation approaches underlie most attempts to create surface maps based on point samples or observations. To the best of my knowledge, Kriging associated with cross-validation is at least one of the most powerful methods. However it is too slow and we have to set a lot of initial values in order to get the best results. I would be thankful if anyone could introduce me a better and faster method (if it exists).
You should check out the Empirical Bayesian Kriging which is build to automate the parameter estimation part
i guess krigging is very reasonable for environmental data interpolation. If you change your grid spacing the issue of it being slow will be solved.
The computation time is not my only concern. For example, when I opted Gaussian model in ordinary kriging, the model was fitted more agreeably with the experimental semiovarigram near the origin than exponential model. It means it demonstrated very subtle variations of the observations in dense areas; by contrast, it had very poor extrapolation. Besides, there are some initial values should be defined for the model, which would not be appropriate for all cases.
I've recently been informed that a new interpolation method surpasses the kriging. I cannot recall its name (maybe was EOF). That's why I started asking this question.
BTW, thank you.
OK but still krigging is more appropriate for environmental data handling. there are positive and negative sides of all these methods. But krigging is tried and tested over years I am not aware of EOF, will look at it.
Kriging is effective only if you have lots of data in order to properly define the underlying variogram.
If a lag distance is short the krigging equations may occur not stable. One will obtain chopped jigsaw instead of smooth map. Try then the IDW (inverse distance weighted) method. It is also simpler in terms of equations, so may be a bit faster than krigging.
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing