How can we asses the reliability of an interpolated surface using geo-statistical analyst option of Arc GIS? For, example, if I take 2/3 dataset for interpolating a suface using different methods (e.g IDW, Kriging, spline etc.) I would expect similar results when I change the dataset in a good fit. Is there any method to validate interpolated results in ArcGIS or any other spatial analysis software?
See my article Article GIS-based Methods for Estimating Missing Poverty Rates & Pro...
to assess if the methodology used for validation of interpolation may help your efforts
Srini Vasan paper is a good read and shows how to validate different interpolation methods. You may have to do your own investigation because different data types will respond differently. If your input data has complete coverage, then randomly remove one or more input data points then do your different interpolations. Next check which type of interpolation had the best fit to the actual removed data points. The more random points removed and the more interpolation comparisons the better you can validate which interpolation method works best.
the reliability of an interpolation method depends on the representativeness of the study area and above all the sampling mesh must be enriched by a considerable amount of data (with the aim of minimizing the white spaces between the points)
First of all you need to ask whether the interpolation method is based on any statistical assumptions and if so are those assumptions satisfied. IDW is just a heuristic method (no statistics), splines are a special case of radial basis functions which in turn is equivalent to universal kriging. Secondly what are you going to use the results for?
In the case of some form of kriging you need to estimate/fit a variogram and there is no best automatic method. With splines and radial basis functions you can't use a moving search neighborhood, with kriging you have to make choices about the moving neighborhood.
You need to think carefully about the phenomenon that generated the data and what you know about it. You need to do exploratory analysis of the data set.
"Representativeness" is a meaningless term with respect to spatial data.
There is no best nor complete method to evaluate different interpolation methods for a given data set. If you look at the literature you will find a number of examples where the authors have compared different methods but no over all method.
Reuben Reyes Can we use an iterative process to randomly remove data and check the model performance. ArcGIS's geostatistical analyst is already having a cross validation option where it is going through an iterative process by removing one point and estimating the results based on interpolation of other points. Accordingly, it is calculating errors based on that prediction.
Samiul
Most any geostatistical package includes cross validation but the big question is how to use the different possible statistics obtained from cross validation
See 1991, Myers,D.E., On Variogram Estimation in Proceedings of the First Inter. Conf. Stat. Comp., Cesme, Turkey, 30 Mar.-2 April 1987, Vol II, American Sciences Press, 261-281
and
1990, A.W. Warrick, R. Zhang, M.M. Moody and D.E. Myers, Kriging Versus Alternative Interpolators: Errors and Sensitivity to Model Inputs - in Field-Scale Water and Solute Flux in Soils (Monte Verita),(eds)Roth, Fl�hler, and Jury � Birkhauser Verlag Basel
Cross validation is not explicitly designed " for reliability assessment of spatial interpolation techniques? "
It is mis-leading to think that you can "assess the reliability of spatial interpolation techniques
Note that splines is a special case of Radial Basis functions and in turn Radial Basis functions is equivalent to Universal kriging.
- Badges
- Science topic