Latest Contributions by EricKrause

The best tool we have for this is the Crosscovariance cloud tool.  You can also do cokriging with the two variables of interest, and look at the crosscovariance between the two variables.  This is done on the semivariogram screen of the Geostatistical Wizard by changing "Between: Var1 - Var1" to "Between: Var1 - Var2". ... View more

* First: I want to compare the relationship between modeled values and radar detected values in the two datasets. When using OLS is it correct to use the modelled values as "Dependent variables" and the radar detected values as "Explanatory variables"? It doesn't matter which you call the dependent and explanatory variables.  The p-value of the coefficient should be the same either way. * Seccond: I also want to perform a binary test to find the spatial relationship between modelled clutter (1= Present, 0= Non- present) and the radar detected values  (1= Present, 0= Non- present). Can I use OLS also for this purpose or should I use a different tool? You can use the same tool to get the correlation, but you won't be able to trust the p-value. We don't have a specific tool for it, but if you need to test the correlation of two binary variables, use the Phi Coefficient of Correlation.  You can probably write a Python script to calculate and test it for you, but it won't be simple. ... View more

You could also convert the rasters to points, join the points by location, and run the Ordinary Least Squares tool in Spatial statistics.  Testing the significance of the coefficient in the linear model is the same as testing the correlation. ... View more

I think you want the Band Collection Statistics tool in Spatial Analyst.  It takes rasters as input and can output the correlation matrix (among other statistics).  Just make sure to check the "Compute covariance and correlation matrices" box. http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#//009z000000p5000000.htm ... View more

Remember that the inclusion probability is relative to the other inclusion probabilities.  For example, cutting all probabilities in half will not affect the output points.  The inclusion probability is often used to prevent suggested locations that aren't feasible or are too costly (such as in a lake or on a mountain).  You can use the probabilities to make sure that building the sampling network is not too expensive or difficult. ... View more

I need to automate the production of my xml files (I have 120 xml files to produce, which will then be applied to cokrig more files using the code discussed above). The only parameter I am changing in the cokriging process is to press the "optimize model" button. Can I cokrig two shapefiles & "optimize" the model using Python, then alter the script to loop through my 120 files? Thank you, Brenna You only need to create a single xml.  Run the wizard with cokriging, and save the xml at the end.  Open the xml with a text editor, and the second line of code should read: <model name="Kriging"> Add the following optimize tag to that line (which is case-sensitive): <model name="Kriging" optimize = "ByCrossvalidation"> Save the xml, and use it as the model source in the Create Geostatistical Layer tool.  Then provide two new cokriging datasets, and the output will be a geostatistical layer that will be the same as if you had pressed the optimize button on the new data. Just keep looping the CGL tool and providing it the original altered xml, and all the outputs will behave as if you had pressed the optimize button in the Wizard with the new datasets. More information can be found in these links: http://forums.arcgis.com/threads/11981-Spatial-Analyst-Interpolation-versus-Geostatistical-Analyst-Wizard. http://blogs.esri.com/Dev/blogs/geoprocessing/archive/2010/06/18/Automating-geostatistical-interpolation-using-template-layers.aspx NOTE: The optimize button was added in ArcGIS 10, so this process won't work for any versions before 10. ... View more

Your surface is covering all the points.  The default extent of a geostatistical layer is the rectangular extent of the input points.  The layer isn't covering the entire picture of some of the points because you've used large symbols, but the layer is still covering the location of the points (the centers of the circles). ... View more

As for the differences between the different standard errors, the prediction standard error map that can be created in the Wizard refers to the standard deviation of the prediction for any individual point (formulas are in the pdf I just posted). The cross-validation statistics are calculated for each input location (suppose there are n points).  To calculate the error for some point x, the software removes the point and recalculates the kriging weights based on the on the remaining (n-1) points and generates a prediction at point x.  The error is the difference between the prediction at point x and its actual value.  The software does this for all n points.  The root mean square, for example, is the average squared error for the n points. Be careful not to confuse "standard error" with "standardized error."  They aren't the same. Page 35 of the pdf talks more about it. ... View more

http://resources.arcgis.com/content/product-documentation?fa=viewDoc&PID=48&MetaID=778 Appendix A contains the mathematical details of methods available at ArcGIS 9. ... View more

Did you use any ESDA tools to check the kriging assumptions?  If the predictions are vastly different than the input points, it may be that your data is violating the assumptions for kriging, or it might be that your fitted semivariogram model isn't well specified. By default in ArcGIS 10, the filled contours map is a surface for the predictions, and the map will display predictions at the input points, not the value of the input points themselves.  If you want Ordinary Kriging to give exact predictions at the input point locations, you can set Measurement Error to 0.  However, this will cause a jump in the surface at the input point locations. ... View more

Sorry for the delay in response.  You caught us on a particularly busy week. I think you may have a few misconceptions about what Geostatistical Analyst is and what it can do.  I'll do my best to clear up any confusions. What I seem to have stumbled across so far is that you can use Geostatistical Analyst/Explore Data to get an insight into the spatial distribution of point vector data and you can convert raster layers into geostatistical layers using the Geostatistical Wizard in Geostatistical Analyst. Then you can use these Geostatistical Layers as input to the Geostatistical Tools in Arc ToolBox. Am I right so far? Geostatistical Analyst deals almost exclusively with interpolation methods, with special emphasis given to kriging.  Many variables change continuously across space (like temperature and elevation).  The goal of interpolation is to predict the value of the variable at any location based on measurements taken at particular locations.  The general workflow is to start with point vector data, where each point contains a measurement of the variable at that location.  Before proceeding with interpolation, the Exploratory Spatial Data Analysis (ESDA) tools allow you to explore the structure of your data.  You can see histograms, apply transformation, explore trends, etc.  Note that these "tools" are not geoprocessing tools (they are not in ArcToolbox); they are GUI-based and interactive, and they're found in the Geostatistical Analyst toolbar.  Once you are satisfied that your data is appropriate for interpolation, the points are used as input for the Geostatistical Wizard.  Inside the Wizard, the GUI environment allows you to visually fit an interpolation model (interpolation methods like kriging have many parameters, and the quality of your predictions depends on the quality of your model).  The Wizard provides many interactive diagnostics for fitting an appropriate model.  When you are satisfied with the model, the output of the Wizard is a geostatistical layer.  This layer is a smooth surface of predictions for the variable.  It is visualized as a contour map, but the predicted value of the variable can be calculated at any point on the surface. Geostatistical layers are different than raster layers in that they are dynamic.  A geostatistical layer contains all the information it needs to make predictions at any specified location, but it doesn't actually store the predictions themselves.  Instead, it stores the mathematical model needed to make the predictions, and then it calculates them when you ask it to.  Geostatistical layers can be converted to raster, and this is done by simply applying the mathematical model to the center of each cell in the raster.  Ignoring memory and time constraints, you can make the cell size of the raster as small as you want. You cannot convert a raster layer into a geostatistical analyst layer because the resolution of the raster is limited by the cell size, whereas geostatistical layers have infinite resolution (limited only by the precision of the computer).  You could, however, convert the raster to points, then use these points to build an interpolation model in the Wizard.  The Wizard would then create a geostatistical layer from the model (though I would be very cautious before doing this).  The tools in the Geostatistical Analyst toolbox are generally used for post-processing after a geostatistical layer is created in the Wizard (such as converting a geostatistical layer to a raster or getting predictions for a set of points).  In addition, the interpolation methods that do not require a high level of interactive model fitting are offered as tools.  Like the Wizard, they take point vector data as input and produce geostatistical layers of predictions. All that being said, it doesn't sound like interpolation is what you need.  I wish I could be more help, but my specialty is interpolation, and I'm just not sure how to go about solving your particular tasks.  You may want to look into Geographically Weighted Regression, which is available in the Spatial Statistics toolbox. Let me know if you have any other questions, and I'll do my best to help. ... View more

After reading your post again, it sounds like you are trying to convert a layer into a geostatistical layer.  That can't be done.  A geostatistical layer can be exported to a raster but not the other way around. ... View more

Geostatistical layers are fundamentally tied to interpolation methods and their associated parameters.  Interpolation is generally done in the Geostatistical Wizard, and the output will be a geostatistical layer. The Create Geostatistical Layer tool is used for creating a new geostatistical layer from an existing one.  It requires a model source because the tool needs to know which interpolation method and parameters you are using.  It sounds like you don't have an existing geostatistical layer, so the tool isn't going to work for you until you make one in the Geostatistical Wizard. If you tell me what exactly you're trying to do (some kind of interpolation, I assume), I can provide more help. ... View more

If you're on the semivariogram screen, you can edit the Major Range by clicking the calculator button directly on its right.  In order for Minor Range to be exposed, you need to set Anisotropy to True. If you're on the Search Neighborhood screen (where you can preview the prediction surface), set "Copy from Variogram" to False.  Then you should be able to alter the Major and Minor Semiaxes. ... View more

If the semivariogram was different for two exponential models compared to one exponential model, then they must have had different Major Semiaxes (or Minor Semiaxes if you corrected for anisotropy).  When you supply more than one model, the semivariogram curve is generated by a weighted average of the multiple models (where the weights are determined by least-squares).  Using multiple models is beneficial when your data is the result of multiple underlying processes (wind patterns and temperature, for example).  Each model can be used to account for a different underlying process.  Sometimes these underlying processes are known, and sometimes they aren't.  Using multiple models when you don't know the processes that generated your data can be suspect, but if the cross-covariance cloud and empirical semivariogram suggest that different distances follow different semivariograms (for example, one model fits the data up to 1000 meters, and another model fits beyond that), it may be justified to use two models with different search neighborhoods. ... View more