Re: AI-GEOSTATS: co-kriging on data with spatial trends
I think that you can try to avoid cokriging. Clearly the relation is between k and the mean values of clay content. Instead of using a pure cokriging you could use a kriging with external drift in which the mean value of k is related to the map with clay content. Goovaert'book Geostatistic for Natural Resources Evaluation chapter 6 could give you an hand. Gstat code permit you easily to perform kriging with extrernal drift. Sincerely Sebastiano Trevisani At 04:35 PM 11/27/2003 +0100, Sigrun kvarno wrote: Dear AI-GEOSTATS members, I tried co-kriging for the first time yesterday, and now I have some questions: I have a dataset with 42 measurements of saturated hydraulic conductivity (Kfs) within a grid where I have measured soil texture in 256 points with 10 m spacing. Kfs measurements are irregularly spaced. Kfs is approximately log-normal, and there is a significant correlation between log(Kfs) and log(clay content), R2 = 0.61. I know that there is a significant spatial trend in clay content. Earlier, when using geostatistics on the clay content alone, I computed the variogram and kriged on the residuals, and then added back the trend function to the kriged estimates. Considering co-kriging, I need a variogram for both the primary variate (Kfs) and the covariate (log(clay content)), but due to the trend in clay content I have to compute the variogram on the residuals (this is important - a variogram on the original data gives a range of about 600 m (when using the GS+ program), whilst a variogram on the residuals reduces the range to about 55 m). But the correlation between residuals and logKfs is not so good. It is significant, and R2 = 0.20, which is not too good compared to 0.61 for the original data. What is the correct way to proceed in this case? Are there any rules of thumb for such problems? I'd also like to know which search radius should be used - the range of the logKfs data is about 40 m, the effective range (exponential model) of logclay is 600 m for original data (55 if residuals can be used), and the range of the cross-variogram is 53 m. I've learnt that the search radius should be approximately the same as the (effective) range when performing kriging. Is it the range of the cross-variogram I should use in this case? I hope there are some simple solutions out there! Kind regards, Sigrun ** Miss Sigrun H. Kværnø Arealressursavd./Dept. of Land Resources Jordforsk - Norwegian Centre for Soil and Environmental Research Frederik A. Dahls vei 20 N-1432 Ås NORWAY phone: +47 64948159 fax: +47 64948110 **
AI-GEOSTATS: gstat in R doubts
Dear Collegues, I have some particular doubts about gstat for R (most probabily very basic for what I apoligise): 1. Does it compute indicator kriging? How to define it? It should be I=.5, but where? I could not find the argument in 'variogram' 2. I could not manage to make it do the backtransformation of the kriging predictions when using log data. Is it possible and how? 3. How exactly we specify universal kriging? I have found different ways in different places: kk=krige(abu~x+y, ~x+y, dat, bl, model=wls) # I think this one indicates a quadractic internal trend :-) or kk=krige(abu~(covariate), ~x+y, dat, bl, model=wls) or kk=krige(abu~x+y+(covariate), ~x+y, dat, bl, model=wls) And the variogram of the residuals for the universal kriging... how do we do it? the same way? 4. The function plot.variogram as the same name as in geoR. This causes bugs and problems runing it, and does not allow us to run both packages which would be important. Could this be solved by any of the authors, simply by renaming the function... probabily there are more like this, although I did not proceed because I got stuck. (To solve this the only way I found is to attach and detach the packages, ...) 5. The variogram produced by both packages (gstat and geoR) are different. Why? It was excelent and extremely usefull for me the implementation of gstat in R!!! Thank you so much! Thank you very much in advance and sory to disturb you, Marta Rufino -- * To post a message to the list, send it to [EMAIL PROTECTED] * 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 [EMAIL PROTECTED] 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
AI-GEOSTATS: Re: co-kriging on data with spatial trends
Sigrun Calculating of cross semi-variograms has to be done on residuals if there is a trend, just like any single variable semi-variogram. You might find our MUCK papers useful, as we were co-kriging two variables both with trend back when we first started in the 1980s. You can find our papers at http://uk.geocities.com/drisobelclark/resume/Publications.html (note the capital P). Your other question about search radii. When trend is present, you need to enlarge your search radius past the range of influence to get enough samples to condition the equations properly - that is, to solve for the trend as well as the weights. You can see how this works with uor kriging game, free to all at http://geoecosse.bizland.com/softwares Hope this helps Isobel [Clark] Download Yahoo! Messenger now for a chance to win Live At Knebworth DVDs http://www.yahoo.co.uk/robbiewilliams -- * To post a message to the list, send it to [EMAIL PROTECTED] * 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 [EMAIL PROTECTED] 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
AI-GEOSTATS: About gstat and binomial negative family data
Dear members I'm newer to geostatistics analysis. So, my work now is probe to my lab.chief that geostat. anal. is better than other analysis to create probability maps from fishery acoustic data of surveys in Brazil. Then, generate kriging maps of probability distribution of Maurolicus stehmanni (fish sp). 1)First step: using GMT, convert long, lat to linear projection [EMAIL PROTECTED] trab_R]$ mapproject file1.dat -R-53/-38/-35/-23 -Fn -Jm0/0/1c > file2.dat where R is region Fn to nautic miles JM is mercartor proj. 2) using "R" gstats: compute variogram..., and points show no spatial dependence! Range is 0.5 nautical miles! 3) selects limited areas, repeat step 1 and 2 and result is equal! 4) remove nule values, repeat steps 1 and 2, and result is equal! 5) change scale of projection, "-Jm0/0/10c", repeat steps 1 and 2, result is equal. Whats wrong? Someone could help me? The family of distribution of M. stehmanni is binomial negative, is possible define these prior to variogram and then result better variograms? My apologies for newbie questions, i'm very gratefully for this list! Marcelo Ps:someone research could contact me in PVT. = ## ~~~ Oceanólogo ~~~ ## # Marcelo Alexandre Bruno # Linux User: 124592 # Pós-graduação Oceanografia Biológica # FUNDACAO UNIV. FEDERAL do RIO GRANDE # Departamento de Oceanografia # Lab. de Tecnologia Pesqueira e Hidroacústica # AV. ITÁLIA km 8 s/n - CARREIROS # 96201-900 (0xx53) 2336528 # Rio Grande - RS - BRAZIL ## ## __ Yahoo! Mail: 6MB, anti-spam e antivírus gratuito! Crie sua conta agora: http://mail.yahoo.com.br -- * To post a message to the list, send it to [EMAIL PROTECTED] * 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 [EMAIL PROTECTED] 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
Re: AI-GEOSTATS: co-kriging on data with spatial trends
Hi, It looks like you have a large-scale trend and I believe that ordinary cokriging with moving search windows would be appropriate. Remember that with ordinary (co)kriging the stationarity assumption is restricted to the search window and this could be rather small in your case given the sampling density you have for clay measurements. Instead of specifying a search window, I usually specify a maximum number of primary and secondary data and let the window be as big as necessary. You could do the same and given the high sampling density of clay, then the effective search radius wouldn't be too big and the stationary assumption be realistic. Hope it helps, Pierre <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> Dr. Pierre Goovaerts President of PGeostat, LLC Chief Scientist with Biomedware Inc. 710 Ridgemont Lane Ann Arbor, Michigan, 48103-1535, U.S.A. E-mail: [EMAIL PROTECTED] Phone: (734) 668-9900 Fax: (734) 668-7788 http://alumni.engin.umich.edu/~goovaert/ <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> On Thu, 27 Nov 2003, Sigrun kvarno wrote: > Dear AI-GEOSTATS members, > > I tried co-kriging for the first time yesterday, and now I have some > questions: > > I have a dataset with 42 measurements of saturated hydraulic > conductivity (Kfs) within a grid where I have measured soil texture in > 256 points with 10 m spacing. Kfs measurements are irregularly spaced. > Kfs is approximately log-normal, and there is a significant correlation > between log(Kfs) and log(clay content), R2 = 0.61. > > I know that there is a significant spatial trend in clay content. > Earlier, when using geostatistics on the clay content alone, I computed > the variogram and kriged on the residuals, and then added back the trend > function to the kriged estimates. > > Considering co-kriging, I need a variogram for both the primary variate > (Kfs) and tReceived: from SJMJF-MTA by mail.jordfohe covariate (log(clay content)), > but due to the trend in > clay content I have to compute the variogram on the residuals (this is > important - a variogram on the original data gives a range of about 600 > m (when using the GS+ program), whilst a variogram on the residuals > reduces the range to about 55 m). But the correlation between residuals > and logKfs is not so good. It is significant, and R2 = 0.20, which is > not too good compared to 0.61 for the original data. What is the correct > way to proceed in this case? Are there any rules of thumb for such > problems? > > I'd also like to know which search radius should be used - the range of > the logKfs data is about 40 m, the effective range (exponential model) > of logclay is 600 m for original data (55 if residuals can be used), and > the range of the cross-variogram is 53 m. I've learnt that the search > radius should be approximately the same as the (effective) range when > performing kriging. Is it the range of the cross-variogram I should use > in this case? > > I hope there are some simple solutions out there! > > Kind regards, > Sigrun > > > > > > ** > Miss Sigrun H. Kværnø > Arealressursavd./Dept. of Land Resources > Jordforsk - Norwegian Centre for Soil and Environmental Research > Frederik A. Dahls vei 20 > N-1432 Ås > NORWAY > phone: +47 64948159 > fax: +47 64948110 > ** > -- * To post a message to the list, send it to [EMAIL PROTECTED] * 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 [EMAIL PROTECTED] 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
AI-GEOSTATS: co-kriging on data with spatial trends
Dear AI-GEOSTATS members, I tried co-kriging for the first time yesterday, and now I have some questions: I have a dataset with 42 measurements of saturated hydraulic conductivity (Kfs) within a grid where I have measured soil texture in 256 points with 10 m spacing. Kfs measurements are irregularly spaced. Kfs is approximately log-normal, and there is a significant correlation between log(Kfs) and log(clay content), R2 = 0.61. I know that there is a significant spatial trend in clay content. Earlier, when using geostatistics on the clay content alone, I computed the variogram and kriged on the residuals, and then added back the trend function to the kriged estimates. Considering co-kriging, I need a variogram for both the primary variate (Kfs) and the covariate (log(clay content)), but due to the trend in clay content I have to compute the variogram on the residuals (this is important - a variogram on the original data gives a range of about 600 m (when using the GS+ program), whilst a variogram on the residuals reduces the range to about 55 m). But the correlation between residuals and logKfs is not so good. It is significant, and R2 = 0.20, which is not too good compared to 0.61 for the original data. What is the correct way to proceed in this case? Are there any rules of thumb for such problems? I'd also like to know which search radius should be used - the range of the logKfs data is about 40 m, the effective range (exponential model) of logclay is 600 m for original data (55 if residuals can be used), and the range of the cross-variogram is 53 m. I've learnt that the search radius should be approximately the same as the (effective) range when performing kriging. Is it the range of the cross-variogram I should use in this case? I hope there are some simple solutions out there! Kind regards, Sigrun **Miss Sigrun H. KværnøArealressursavd./Dept. of Land ResourcesJordforsk - Norwegian Centre for Soil and Environmental ResearchFrederik A. Dahls vei 20N-1432 ÅsNORWAYphone: +47 64948159fax: +47 64948110**