I am trying to perform ordinary Kriging using gstat package on a hourly particulate matter(PM) concentration dataset. my dataset contains PM concentration for 1 hour of 250 sites.After performing ordinary kriging, I did cross validation. but I got the correlation coefficient between observed values and predicted values is low (around 0.5).
After trying every possible condition and spending many days, I understand that probably my data does not only depend on location. That means probably location is not only predictor for my dataset.
My goal is to well predict my dataset by any type of Kriging interpolation. Here, some people suggested me to consider other factor as a predictor. But I don't know actually how to do that! Is that possible to do by using kriging interpolation?
I can't found any documentation on this type of problem. I am using gstat package in R to perform kriging. Could anyone please help me out from this kind of deadlock situation?
[I am attaching my dataset (urban030101) and R code. Please take look. What I miss?]
I agree, co-kriging would be most useful. If you have access to ArcGIS you can do co-kriging directly within the program. Otherwise, consider gstat package in R. I've attached technical notes. Best of luck! Ingrid
Is your goal to predict or to predict by any type of Kriging interpolation? There are many ways to improve prediction using kriging with external drift or co-kriging ets. I think you need first to analyze your data for consistensy - there may be the answer.
If other factor can be measured or estimated in any location, you can try kriging with non-constant mean, using krige.conv function of the geoR package. Very helpful examlpe in the book Model-based Geostatistics (2007) Diggle, Ribeiro, on the 151 p. God luck.
Cokriging is a good solution if you individuate just one more predictor correlated with your target variable. I think it should be better to adopt a hybrid method like regression kriging, which allows you to consider a number of different auxiliary variables. For this purpose, in R you can use the gsif package; at the following link you can find all you need to work with it. I hope this will be useful!
https://cran.r-project.org/web/packages/GSIF/index.html
Dear Vera Petrovna Samsonova , my goal is to predict by any type of kriging. Could you please tell me bit more about consistency analysis? Why and what kind of aspect I should check ? Thanks.
Location is not a predictor in any form of kriging (simple, ordinary, universal, cokriging, kriging with external drift). Instead it is the spatial pattern of the data locations and the spatial correlation quantified by the variogram. Even in a Trend Surface interpolator is it not the location(s) that are used but rather the coordinates of the data locations. The Ordinary kriging equations are based on the assumptions that (1) the expected value of the random function is a constant (and does not depend on the location) and that the variogram is a only function of the separation vector (for a pair of locations). Unfortunately it is not possible to statistically test either of these assumptions but there are indicators as to whether these assumptions are reasonable. You need to look at a good reference book, e.g. "Geostatistics" by Chiles and Delfiner (J Wiley) or the older "mining Geostatistics" by Journel and Huijbregts (Academic Press). The tutorials for gstat are very good but they assume you already know the basics of geostatistics.
Kriging with external drift means that you are assuming that the expected value of the random function is dependent on another spatially distributed variable,e.g if you are looking at air temp close to the ground then it is likely affected by the surface characteristics (plowed ground, paved with concrete, paved with asphalt, grassland, etc). In the case of precipitation, it is obvious that it is affected by elevation but it has been treated both with external drift and with cokriging, there is no absolute answer as to which is better
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication