Batch Iso Cluster Unsupervised Classification

Question

Having trouble getting the tool to work as a batch. When run individually with same parameters it seems to work fine. When run as a batch it informs me there is an error with number of classes being less than one when I have designated 16 classes for all. I know the script would be helpful but can't seem to figure out how to view it right now.

Secondly, why is it that the Iso Cluster Unsupervised Classification only works when you browse to/add all the individual raster band files and not the main raster that contains the different bands already?

The overall intent is to run numerous iso cluster classifications on about 20 different rasters for the purpose of vegetation mapping with 16 different classes, a minimum class size of 1300 at a sample interval of 20. Am I asking the tool to do much or something?

paulagarcia_llamas · Answer

Hi, sorry for my delay reply...
i havent normalized my data because PCA proces already normalized the data, otherwise the band range used in the cluster continue being very different...yeah the ArcGis help use a normalized acuation, the point is that my vegetaion variables range between 1 and 100% cause it is percentage of vegetation while topographic variables for example solar radiation ranges between 1800 an 7 millions so....i have to get a new máximum and mimimum acording to ArcGis help
 Z=(((x-oldmin)*(nexmax-newmin))/(oldmax-oldmin))+newmin
so my question is what new values getting in order to not distorting data...

curtvprice · Answer

[noparse]y[/noparse] only other question is why the tool cannot work with certain parameters, ie:
The smallest area of interest (AOI) is a raster consisting of about 1500 cells but even on that raster the tool cannot find a cell cluster of at least 10-50 (I have tried all) cells when all of the cells are almost identical??? I know each cell has its own values but with a resolution (cell size) of about .36 meters it cannot locate 10 cells to group??? Even when they are all a slab of concrete for example on the raster......ahhhhhhhhhhh!

This is the downside of using an unsupervised technique like IsoCluster - your data may not support the number of classes you are requesting. The results (and whether you get a result at all) are highly dependent on the parameters you provide and the distribution of your data. Normalizing data often helps (see my last post) -- if the parameters have extremely different distributions, parametric techniques like IsoCluster are unlikely to produce very helpful classifications.

curtvprice · Answer

The point is that with the same PCA and asking for the same amount of classes i get tow different maps.

IsoCluster algorithms usually start self-organizing your data by generating random "seeds" and iterating until those means converge. This means if you run the same tool twice with the same data you can get slightly different cluster results. I'm pretty sure these tools follow that procedure, but it isn't clear to me from the help. It is as if more weight is given to some components (the vegetation what i really want to) from the PCA sometimes and to different ones other time (the topographic)..

Have you normalized your data? Better results will be obtained if all input bands have the same data ranges. If the bands have vastly different data ranges, the data ranges can be transformed to the same range using Map Algebra to perform the equation. (Iso Cluster help doc)