AI-GEOSTATS: Summary to nested models?
- Thank you very much to all people who answered my question. It was very
helpful. Heres a summary of the responses.
Dear list members,
I have started using geostats only this year and therefore I appologise that
questions are too basic for this discussion list...
I am using Variowin 2.2 and GeoEAS 1.2.1 to analise my data, and an
version of Surfer to contour the distribution. However, I've encountered
1. When I use latitude and longitude in my DAT file, in what units is the
variogram x-axis (h)?
2. Surfer doesnt understand the grid file created by GeoEAS KRIGE. How
transfer my results to surfer (until now, I've used the estimated values
cross validation to have surfer generate a grid file)?
3. My variogram shows completely different behaviour from distance A to
for distance B to C. I've tried fitting various nested models, but I havent
able to come out with a good fit for the whole variogram. Is it possible to
analise A-B data separately from B-C data?
I'm new at this too, so have an aswer to your first question only. :)
Basically, you can't use latitude and longitude to indicate the location of
your samples in a geostatistical analysis because the distance between
locations is not euclidean. When you calculate variograms, you are
that you are on a plane X-Y coordinate system. Lat and log account for the
curvature of the earth. So unless you are dealing with very short distances
you should first convert your lat and long locations into UTM coordinates
(or other plane projection). I use a program called geotrans to convert lat
and long to UTM, you can download it at
Once you have your locations in UTM, the variogram distance are
Julian M. Burgos
University of Washington
School of Aquatic and Fishery Sciences
1122 NE Boat St. (Box 355020)
Seattle, WA 98195-5020
1- You need to transform your latitude longitude geographical data into
UTM coordinates (or any equivalent transformation) because you need to
work in metres or km (or cm!)
> 2. Surfer doesnt understand the grid file created by GeoEAS KRIGE.How do
> transfer my results to surfer (until now, I've used the estimated valuesfrom
> cross validation to have surfer generate a grid file)?If you want to krige witn Geo-Eas you have to use its CONREC program to
have your map. Otherwise, if you like to use Surfer, you just need to
krige there, using its own grid utility.
> 3. My variogram shows completely different behaviour from distance A toB,
> for distance B to C. I've tried fitting various nested models, but I haventbeen
> able to come out with a good fit for the whole variogram. Is it possibleto
> analise A-B data separately from B-C data?What are A,B,C? any kind of different variables, or just some different
If there are just different points you need to know the ir reliability
before doing any assunption on them. If there are different variables,
you do not need to suppose they behave in the same way.
I think there is something wrong with your variogram. If variables are
quite the same for small distances (flat variogram) they cannot start
being different from one each other at large distances.
Maybe you have chosen a bad lag (h), say a lag too short. It is very
important for you to know the precision of your data (it doesnt make any
sense to say temperature is 10.5 if your thermometer measures
Hope this helps,
>Most geostatistics programs are going to treat the coordinates as though
>1. When I use latitude and longitude in my DAT file, in what units is the
>variogram x-axis (h)?
they are "x", "y", i.e., rectangular. If you are using Lat,
are not exactly rectangular coordinates. That is, one minute of Lat is
one nautical mile but in general one minute of Long is not one nautical
mile, the distortion can be very large at high latitudes. You need to
change to UTM coordinates instead.
As a further comment, most packages will only accept numbers in
decimal form so if you have Lat, Long in degrees, minutes and seconds
the program will either give you an error message or completely
mis-interpret the entries. I let my students use Lat , Long in decimal
form in my geostatistics class but ONLY in class and I warn them that
there will be a distortion. The only reason I do this is that they are
learning about how to do geostatistics and not using the end results for
anything outside of the class.
>2. Surfer doesnt understand the grid file created by GeoEAS KRIGE. How
>transfer my results to surfer (until now, I've used the estimated valuesfrom
>cross validation to have surfer generate a grid file)?I don't use Surfer so I can't completely answer your question. I assume
by "grid file" you mean the output file from KRIGE. This is an ASCII
file and easily edited in any text editor. The format is exactly the
same as the input file (different column names of course). If you only
estimated one variable then there will be four columns; two for
coordinates, one for the kriged estimate and one for the kriging
standard deviation. The header identifies the columns. You might have to
strip off the header for use in Surfer.
>3. My variogram shows completely different behaviour from distance A toB,
>for distance B to C. I've tried fitting various nested models, but I haventbeen
>able to come out with a good fit for the whole variogram. Is it possible toI assume that A, B, C are points?? If so, your statement doesn't quite
>analise A-B data separately from B-C data?
make sense. The sample variogram is not defined or dependent on a line
segment. The plotted values of the sample variogram (or the computed
values of the model variogram) are determined by lag distances. In that
case, there is no reason to expect that the value for two different lag
distances would be the same. Perhaps you are referring to different
directions? In that case you would be describing an anisotropy, this
would show up (assuming that you are using PREVAR and VARIO or
VARIOWIN) when you compute and plot directional sample variograms.
geometric anisotropy, this would show as a change in the range of the
variogram with respect to distance.
Finally is it possible you are asking whether the region of interest can
be split into two separate parts?? This is certainly possible, you
would have to split the data set into two different parts estimate and
model the variograms separately. Krige the two subregions separately
using the respective variograms and data sets. It is likely that the
grids (generated by the software) would overlap and you would have to
check to see which grid nodes should be associated with which
and use the appropriate kriged value. If you were going to do this, you
would want to ask if there is some property that changes between the
regions to justify/motivate splitting the region into two parts.
Donald E. Myers
Carolina Garcia Imhof
Marine Mammal Research Group
Marine Science Department
310 Castle Street, PO Box 56
University of Otago
Dunedin, New Zealand
Fax: 64 3 479 8336
Phone: 64 3 479 5476
* To post a message to the list, send it to ai-geostats@...
* As a general service to the users, please remember to post a summary of any useful responses to your questions.
* To unsubscribe, send an email to majordomo@... with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list
* Support to the list is provided at http://www.ai-geostats.org