Hi Suzanne, Did you test in any way that your data does not violate the stationarity assumption??? maybe your data set is actually a random draw of two (or more) different populations that result from maybe 2 different phenomena. Maybe you have local outliers not only extreme outliers .... These local outliers are not extreme values but are "suspicious" in the context of neighboring values. What type of data do you have? In my experience pollution data can result sometimes form 2 (or more) pollution processes superimposed on the same area - like a diffuse pollution process and a point-source process. These 2 processes will have data with different mean and variances, so a mixed dataset certainly will violate the stationarity assumption. Take a look at LISA (Anselin, L., 1995, Local indicators of spatial association ? LISA, Geographical Analysis vol. 27, no. 2, 93-115) - this may help you in understanding your data better and make more informed decisions on how to proceed further.
In my experience kriging is not always the answer, although it is always nice when it is ;-) I hope this helps a little, Monica ==================================== Monica Palaseanu-Lovejoy ETI / US Geological Survey Florida Integrated Science Center 600 4th Street South St. Petersburg, FL 33701 Ph: 727-803-8747 x 3068 Fx: 727-803-2031 email: [EMAIL PROTECTED] ==================================== Suzanne <[EMAIL PROTECTED]> Sent by: [EMAIL PROTECTED] 10/15/2007 08:46 AM Please respond to Suzanne <[EMAIL PROTECTED]> To Mailing list Geostatistics <[email protected]> cc Subject RE: AI-GEOSTATS: Smoothness of indicator kriging over ordinary kriging Thank you very much Sebastiano and Piere Goovaerts for your suggestions. First of all I did not see any special trend. As Piere Goovaerts said I took log of data and semivariogram of logarithm still show a moderate spatial correlation however the nugget effect is higher. I have a sparse sampling of data values with areas of high values located mostly in the north and south of the area. I tried to divide the area to three more homogenous sub-areas. But the semivariograms for these sub-areas show less spatial correlation than for whole area. What can I do now? By the way I already removed a few very suspicious values from the data sat. Should I stick with ordinary kriging only? Regards Suzaneh Pierre Goovaerts <[EMAIL PROTECTED]> wrote: Hi Suzanne, I am surprised that you don't obtain a well-structured indicator variogram for the median threshold at least. This might indicate that the structure you see in the variogram of raw data is caused by a cluster of extreme values. These data are distinguished only for extreme quantile thresholds, which should explain why you don't see any correlation for middle thresholds. I would suspect that taking the log of the data would also reduce the structure you see on your variogram. Hope it helps, Pierre Pierre Goovaerts Chief Scientist at BioMedware Inc. Courtesy Associate Professor, University of Florida President of PGeostat LLC Office address: 516 North State Street Ann Arbor, MI 48104 Voice: (734) 913-1098 (ext. 8) Fax: (734) 913-2201 http://home.comcast.net/~goovaerts/ ________________________________ From: [EMAIL PROTECTED] on behalf of Suzanne Sent: Mon 10/15/2007 4:03 AM To: [email protected] Subject: AI-GEOSTATS: Smoothness of indicator kriging over ordinary kriging Dear list I have a data set of highly positively skewed. I tried to use indicator kriging to improve the estimation accuracy over OK. But I found out some difficulties: 1- The omnidirectional semivariogram show a strong to moderate spatial correlation whereas indicator semivariograms except for 0.1 and 0.8 quantiles do not show any spatial correlation. 2- I tried to use some quantiles, which their indicator kriging show a weak spatial correlation. I run IK with 5 possible cutoffs. The estimation accuracy goes a little bit higher however the map produced using IK is much smoother than OK. I do not know why this happen? And what should I do now? I really need help. Please let me know your opinion about that. Best regards Suzaneh ________________________________ Check out the hottest 2008 models today at Yahoo! Autos. Yahoo! oneSearch: Finally, mobile search that gives answers, not web links.
