Latest Contributions by EricKrause

Hi Carine, What you are seeing is to be expected.  The Gaussian Geostatistical Simulations tool works by creating many different rasters that are all simulated from the same geostatistical model.  When you output the mean raster, you are creating a raster where each cell is the average of all the simulated values in that raster cell.  This average value should not change drastically as you increase the number of simulations because the mean value will stabilize fairly quickly.  When you output a maximum raster, you are creating a raster where each cell is the maximum of all the simulated values in that raster cell.  Unlike the mean, the maximum will keep getting bigger and bigger as you increase the number of simulations. Please let me know if you need more clarification, but what you are seeing is what is supposed to happen. ... View more

Doing any kind of kriging (other than fancy things like double kriging, as jevans noted) is going to be very questionable with high degrees of zero-inflation.  Technically, zero-inflation does not mean that using kriging is "invalid," but the issue is pretty complicated.  In theory, kriging can be performed on just about any data distribution, even ones with zero inflation.  However, to do this correctly, you need to be able to accurately estimate the semivariogram with the typical covariance functions that are provided.  In practice, fitting an accurate semivariogram for strange data distributions is not something that I would recommend because I don't know of a reliable methodology to do it.  And the covariance models that are supported in Geostatistical Analyst are very, very unlikely to be able to estimate an accurate semivariogram for data with large zero-inflation. ... View more

I don't have any good solutions for the zero-inflation problem (other than, potentially, the one posted by jevans), but I can shed some light on the NaN issue.  Empirical Bayesian Kriging computes local models based on subsets of the data.  However, if a subset is composed entirely of a constant value (in this case, zero), the kriging equations cannot actually be solved (the algorithm will attempt to invert something that cannot be inverted).  We put in a special exception when this happens; the subset will predict a constant value everywhere in the subset with a standard error value of zero.  This alleviates the problem of a single subset preventing you from getting any output at all, but it has the size effect of making it impossible to calculate many crossvalidation statistics.  This is why many of them will report NaN (Not a Number) when this happens. ... View more

I'm definitely not the best person to ask about ModelBuilder variables.  Hopefully someone with more direct experience can chime in here, but your best bet is probably to ask this on the ModelBuilder forum. ... View more

Hi Stacy, The raster that comes out of GA Layer to Grid will nearly always have a different max/min than the geostatistical layer.  This is because they build their classification intervals based on different things.  The geostatistical layer classifies based on the min/max of the input data field, and the raster classifies based on the histogram of the output raster.  This can get confusing when some of the predicted values are larger than the max or lower than the min (which can happen with Radial Basis Functions).  The geostatistical layer will always set the minimum class break to the minimum of the input data and the maximum class break to the maximum of the data even if some of the predicted values are outside of that range.  In your case, the minimum of the raster is less than the minimum of the input data, and the maximum of the raster is less than the maximum of the input data. The reason that they classify differently is that geostatistical layers are basically functions.  They contain references to the input data and all of the interpolation parameters, and all predictions are done on the fly.  In this sense, they have basically infinite resolution and are not constrained by a cell size like a raster.  Every unique (x,y) location will receive a unique predicted value.  This is very convenient for post-processing because the geostatistical layer itself contains all of the information about the interpolation model, so it can be directly passed into post-processing tools without needing to explicitly reference the data or the interpolation parameters (the layer itself already contains them).  However, the downside is that you cannot create a histogram of a geostatistical layer (since it has an infinite number of possible predictions), so we must fall back to the input data to build the classification intervals.  At the end of the day, the raster is a much better representation of what the surface actually looks like.  As a rule, we generally suggest using geostatistical layers throughout your workflow, then export to a raster at the very end.  Geostatistical layers are very useful for long workflows, but their visual representation will never be as accurate as the raster. Let me know if you need any clarifications or have any other questions. ... View more

As far as I can tell from the graphic, everything looks good.  And, yes, you can safely uncheck the "Always reset input..." option. ... View more

Since you seem to have a good understanding of kriging, I'll try to explain why you originally got the error that you encountered.  Densify Sampling Network was originally designed only to add new locations to an existing network.  It was only later that we realized that you could also use it to decommission stations by using the original points as the candidate locations.  However, when you implement this workflow, the software must keep all of the original points in order to calculate the standard errors when each new point is added.  So, when you add the first point, it actually creates a coincident point because you still have all of the original points.  When this happens, if your kriging model does not have a nugget or if the nugget is interpreted entirely as microscale variation (ie, not measurement error), then the software does not know how to reconcile this contradiction, and you receive the error. This limitation is frustrating, but unfortunately I don't think there is any way to get around it. ... View more

I intended to use a cross-validation procedure by sequentially removing each of the 137 station to register the increase of the mean kriging standard error of all prediction point locations (nearly 21,000 points on a regular grid). The station leading to the smallest increase of the mean kriging standard error would be decommissioned and I would restart the procedure with the remaining 136 stations based on the new Geostatistical Layer. However, the manual procedure is too time consuming. If you are comfortable using ModelBuilder or making Python script tools, this workflow can be automated without too many headaches.  And you can use the kriging model with no nugget without any problems. I've never done this workflow before, but I think this general workflow will do what you're trying to do: You'll need to iterate over all 137 points.  For each point: Create a selection of the other 136 points. Use "Create Geostatistical Layer" geoprocessing tool to create a geostatistical layer of the 136 selected points.  For the model source, use the kriging layer that was created with all 137 points.  This will apply all the kriging parameters from the kriging model to the 136 points.  Create Geostatistical Layer is a little tricky to use, but I can help if you run into trouble.  Look at the scripting examples in the documentation, and you should be able to figure it out. Use "GA Layer to Points" to predict to the 21000 points.  Use the geostatistical layer created in step 2.  This will create a feature class with the kriging predictions and standard errors at the 21000 points.  These predictions and standard errors are calculated only from the 136 selected points. Calculate the average of the Standard Error field in the feature class created in step 3. Save the average standard error value. After iterating through all 137 points, you can identify the point whose removal resulted in the smallest average standard error value, and that point can be most safely decommissioned.  You would then repeat the above steps for the remaining 136 points to identify the next point that can be most safely decommissioned.  Then you would do it for the remaining 135 points, etc. I'm not an expert in ModelBuilder or Python, but I'm happy to try to help if you run into any problems implementing this automation. ... View more

Hi Maurice, Unfortunately, the workflow for removing points from an existing network requires that the nugget parameter in kriging be interpreted at least partially as measurement error.  I will add this limitation to the documentation for Densify Sampling Network. To resolve the error, you need to alter one parameter of your kriging layer using the following steps: Optionally, make a copy of your kriging layer in case you make a mistake. Right-click the kriging layer in Table of Contents and choose "Method Properties".  This will open the Geostatistical Wizard and allow you to change parameters for that kriging layer. Click Next until you get to the Semivariogram/Covariance modeling page. In the parameters on the right, find "Measurement Error" (it is under the "Model Nugget" menu).  To encounter Error 040043, this parameter must have been set to 0 (if it is any other value, please let me know). Change the Measurement Error value to anything larger than 0 but no larger than 100. Click Finish, then OK on the Method Report screen.  This will update the kriging layer, but it will probably look identical in the map. Re-run Densify Sampling Network with the altered kriging layer, and it should now run successfully. Please let me know if you have any other questions or run into any other problems. -Eric Krause, Geostatistical Analyst Product Engineer PS, if you intended to remove 5 points from your network, you should actually use 132 for the number_output_points rather than 5.  The five points that are not created by the tool are the stations that can be most safely removed; the 132 points that are created from the tool are the ones that help the most in the interpolation. ... View more

If you want to not include some points in the interpolation, you should deselect them instead of using the Extent environment In Geostatistical Analyst, you can't use the Extent environment to control which points are used in the interpolation.  In nearly every ArcGIS tool, Extent will operate on the input and the output datasets, but Geostatistical Analyst will always use every point to perform the interpolation, and the Extent environment only controls the extent of the output surface.  We use all points regardless of the Extent environment because including input points outside of the Extent will improve the predictions within the Extent, particularly along the perimeter. ... View more

Update:  This information has been added to the Interpolate Points REST API documentation, and it can be viewed here: Interpolate Points—ArcGIS REST API: Spatial Analysis Service | ArcGIS for Developers ... View more

To export a geostatistical layer to a raster, use the GA Layer to Grid geoprocessing tool.  You must first create the geostatistical layer from the Geostatistical Wizard.  Once it is in the Table of Contents, use GA Layer to Grid to export it to a raster.  ... View more