Have you considered using semivariogram methods in your geostatistical approach?

(see http://resources.arcgis.com/de/help/main/10.1/index.html#//00310000003600000)

Hello Vitalii Mokin,

one geostatistical software package is called GEOPACK:

http://www2.epa.gov/water-research/geostatistical-software-package-geopack

I have experiences with it during my doctorate and it is of acceptable quality and hence can be recommended.However it is running under DOS. But i think it should be possible to get it running in a virtual machine with the free software VM ware player. when you had installed a DOS before.

Then i have seen, that variogram analysis also seems to be possible with R. Also Kriging interpolation, Spatial BLUP (best linear unbiased prediction) is possible with R. I have seen, that the packages geoR and gstat and a package Kriging are available with R. Usually routines, written for R shold have a good quality, so also this should give you an opportunity.

kind regards,

Ludwig

You can try to create a new variable - distance from the beginning of the river and then use it for geostatistical analysis in new 2d space - distance_from_the_river_beginning*distance_to_the_river_stream, instead of the lon*lat space.

In this case, you can use traditional geostatistics and take into account spatial autocorrelation along the river stream..

I dont know, how well you can use variogram in this case, because the autocorrelation presents in one direction - along the stream, maybe by using anisotropical variogram,

but you can try to use gls or something else, where you can add a correlation structure into the model...It depends on the task

Best regards.

Some time ago I came across the R package SSN, which should be optimized for this sort of analysis. However, I never tested it, so I do not actually know how it works.

Here is the link to the description: http://www.jstatsoft.org/v56/i03/paper

I am interested in this sort of research, therefore may I ask you to keep me posted on your results?

Thanks, Fabio

Dear colleagues,

I'm already in 2011 published an article on this issue (unfortunately there are only Ukrainian version).

I wanted to know how to solve this problem more.

Article Improvement of the method of ordinary kriging of geostatisti...

Several previous commenters have noted the problems that arise in this kind of an application,e.g. distance along a river is not the same as "straight line distance" hence complicating the idea of spatial correlation. In addition the river is presumably flowing and hence there is not only spatial correlation but also temporal correlation and likely a spatial trend (non-stationarity) . To deal with tributaries you may need to split the river up into sub regions and analyze each separately. The flow rates may vary significantly and another reason why you may want to partition the river into subregions. Since it is likely that no two rivers are alike, other examples can be helpful in suggesting methods of analysis but ultimately the analysis will have to be tailored to the specific river.

For some purposes you might want to look at sediments in the bottom of the river which might not be as strongly effected by the river flows

One way of dealing with the directional flow and the non straight line distances is to think of a (non linear) map of the center line of the river onto a two dimensional plane. Ideally this map should be in analytic form and invertible