You are here: > ESRI Forums > arcgis desktop discussion forums > Thread Replies

ArcGIS Desktop Discussion Forums

ArcGIS Desktop - Extension - Spatial Analyst forum

Interpreting pixel values and LSCV   Drew Henry Dec 05, 2008
Re: Interpreting pixel values and LSCV   William Huber Dec 05, 2008
Re: Interpreting pixel values and LSCV   Drew Henry Dec 08, 2008
Re: Interpreting pixel values and LSCV   Cory Overton Jul 13, 2009
Re: Interpreting pixel values and LSCV   Mary von der Porten Jul 14, 2009
Re: Interpreting pixel values and LSCV   William Huber Jul 14, 2009
Re: Interpreting pixel values and LSCV   Mary von der Porten Jul 14, 2009
Re: Interpreting pixel values and LSCV   William Huber Jul 14, 2009
Re: Interpreting pixel values and LSCV   James Casey Oct 28, 2009
Re: Interpreting pixel values and LSCV   James Casey Oct 28, 2009
Re: Interpreting pixel values and LSCV   Mary von der Porten Oct 28, 2009
Re: Interpreting pixel values and LSCV   James Casey Oct 28, 2009
Re: Interpreting pixel values and LSCV   John C. Dec 09, 2009
Re: Interpreting pixel values and LSCV   Mary von der Porten Dec 15, 2009
Re: Interpreting pixel values and LSCV   John C. Dec 16, 2009
Re: Interpreting pixel values and LSCV   John C. Dec 16, 2009
Re: Interpreting pixel values and LSCV   Mary von der Porten Dec 16, 2009
Re: Interpreting pixel values and LSCV   John C. Dec 16, 2009
Re: Interpreting pixel values and LSCV   Mary von der Porten Dec 16, 2009
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Interpreting pixel values and LSCV 
Author Drew Henry 
Date Dec 05, 2008 
Message I am currently using the Fixed Kernel Density Estimator in ArcMap to create utilization distributions for elk data that I am working with for my Masters work.
Can anyone tell me how to interpret the the "pixel values" that are created in the Kernel Raster? I specified a 10 meter raster cell size and also used a scaling factor of 100000000, which gave me pixel values from 0 to 21.2241. I also used a single smoothing parameter which was calculated using LSCV.
Am I incorrect to interpret these pixel cell values as a percentage probability? If I am correct, then the highest probability of use that I calculated within this particular home range is only 21%, which seems odd to me. I also feel that you should know that I am using GPS data, so I have a large amount of locations.
Finally, can anyone recommend an extension that calculates the Least Squares Cross Validation (smoothing parameter)? I am currently using the Animal Movements extension in ArcView to calculate that parameter (very time consuming), so I can input that value into the Hawth's Tools Kernel Estimator in ArcMap. Interestingly, when I compare these two rasters (ArcMap and ArcView), each with the same specified cell size and smoothing parameter, they do not seem to match in their probability distributions, so I am not sure if one is superior to the other in calculating these kernel densities.
Thanks so much for any help you may be able to give me.
Drew 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author William Huber 
Date Dec 05, 2008 
Message A kernel density produces estimates of population per unit area, Drew, not probability. You are probably obtaining numbers of elk per square meter, multiplied by your scaling factor of 10^8. Equivalently, you get elk per 100 square kilometers. A value of 21 strikes me as low, but maybe your radius is large and you have small isolated sightings. 
  --Bill Huber
Quantitative Decisions (http://www.quantdec.com )
More GIS Q&A at http://gis.stackexchange.com/q/3083/664 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Drew Henry 
Date Dec 08, 2008 
Message Thanks for the reply Bill. You are right, a value of 21 is low, which is why I am still confused. I have been trying to use the kernels created in ArcView, which is where I have been calculating my LSCV parameter, but these kernels only correspond to the percentage home range area. To try and be more clear, ArcView creates my % home range countours (just a shapefile) and it also creates my Kernel Raster. This raster does not have pixelated values related to it they way they do in ArcMap; the values do not vary within the contour. Can you get estimates of population per unit area in ArcView using the Animal Movements kernel home range extension?
Thanks,
Drew 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Cory Overton 
Date Jul 13, 2009 
Message Animal movement extension does not use the accepted form from kernel estimation. Additionally, it does not calculate LSCV "properly" and should not be used to develop "H" parameters for use in another program. John Horne's program (Animal Space Use) and Rodgers and Carr's Homre Range Extension (and Home range tools) calculate H values consistently. (See mitchel 2007. Comparison of programs for fixed kernel home range analysis).

Interestingly, Animal Movement extension home ranges calculated using LSCV for smoothing parameters are qualitatively very similar to John Horne's Animal Space Use home ranges calculated using likelihood cross validation derived smoothing parameters....despite very different values. 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Mary von der Porten 
Date Jul 14, 2009 
Message Hi There
I am doing the exact same thing with bear movement data. I calculated LSCV in Animal Space Use, then I used Hawth's tools to create my kernels using the given h.

However I am still stuck on how to get the pixel value from an exact point. I can't seem to find a way to intersect the raster with the point. Do I need to convert the kernel output from a raster?
thanks
 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author William Huber 
Date Jul 14, 2009 
Message Use the "Extract Values to Points" tool, Mary. 
  --Bill Huber
Quantitative Decisions (http://www.quantdec.com )
More GIS Q&A at http://gis.stackexchange.com/q/3083/664 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Mary von der Porten 
Date Jul 14, 2009 
Message Thanks for your reply.
I couldn't locate the Extract tool so I used Hawth's tools "Intersect point tool". I think this does the same thing an produces a new field in my point layer containing the pixel value.
If you could let me know where the Extract Values to POints tool is (can't find it in spatial analyst) I'd appreciate it.
thanks again
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author William Huber 
Date Jul 14, 2009 
Message It's in ArcToolbox. 
  --Bill Huber
Quantitative Decisions (http://www.quantdec.com )
More GIS Q&A at http://gis.stackexchange.com/q/3083/664 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author James Casey 
Date Oct 28, 2009 
Message Hi Mary,

Did you need to multiply the smoothing parameter that you obtained using Animal Space Use to use it for kernel estimates in Hawth Tools? I have been playing around with Hawth Tools and my smoothing parameter for a good fit is 11000. Interestingly, the LSCVh is 3.02 E-2. Not sure why this number is so small and what to do with it in Hawth Tools. Any advice would be greatly appreciated.
Thanks.
james 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author James Casey 
Date Oct 28, 2009 
Message sorry, need to update the check box to receive emails when this thread is updated! 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Mary von der Porten 
Date Oct 28, 2009 
Message Hey There
I am not sure why your LSCV is so small. Are your points extremely close together? I would advise checking how your x,y vals are set up, the settings required for animal space use (assuming that is the program that you are using), etc.
And to answer your question-no I didn't need to use a multiplier. My h vals were between 10-100 which matched my data well.
Best of Luck!
 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author James Casey 
Date Oct 28, 2009 
Message Hi Mary,

Thanks for the quick response and advice. Strangely the h parameter I obtain for the same shape file in Home Range Tools is ~13500 and it fits very well.

Thanks again for you advice and response--I'll be checking on things.

Best,
James 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author John C. 
Date Dec 09, 2009 
Message Here is a link to an ArcGIS extension that converts Animal Space Use text files into vector and raster formats.

https://gce-lter.marsci.uga.edu/public/app/resource_details.asp?id=355

A note worth mentioning, when generating small kernel home ranges (i.e. for animals that don't move very far) constructed with fewer than 50 locations and using a Likelihood Cross-validation smoothing parameter, the home range may result in broken contour lines. This error is, however, an artifact of the ASU software and is scheduled to be resolved.

Anyways, for the limited data set that I used to test it, it has performed reasonably well and thought it might be helpful. 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Mary von der Porten 
Date Dec 15, 2009 
Message Does anyone know if Hawths tools calculates an asymmetric kernel (ie the x and y can fluctuate)? Or is it symmetrical?
Apparently that it how it should be, but I can't find any info on the website about it.
Any insights would be greatly appreciated.
Best,
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author John C. 
Date Dec 16, 2009 
Message As I understand it, adaptive kernels have a variable smoothing parameter and therefore size, whereas fixed kernels remain constant. Hawth's Tools only creates fixed kernels.

 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author John C. 
Date Dec 16, 2009 
Message FYI from www.spatialecology.com:

"[Sept 09] HawthsTools has not been updated for a considerable time and is soon to be formally discontinued. It will be replaced with an entirely new software package that has much greater functionality. Issues have been found when using these tools with recent versions of ArcGIS and I recommend that you cease to use HawthsTools. These problems arise because the ArcGIS programming interface (ArcObjects) tends to change with each update of ArcGIS. This is particularly true for major releases. This breaks backward compatibility, and means that bugs can be introduced into tools that were compiled against a previous version of ArcGIS. Of course this creates a great deal of work for developers who have to constantly keep up with changes in the object model and the lack of backwards compatibility - a highly unsatisfactory state of affairs. I suggest you watch www.spatialecology.com for details of the new product, due for release before the end of the year." 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Mary von der Porten 
Date Dec 16, 2009 
Message Thanks John.
Yes the note on the website is a bit concerning.

I think you are correct on the fixed kernel.

Does anyone have thoughts on using an adaptive kernel and if there is a better program for it?
best,
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author John C. 
Date Dec 16, 2009 
Message I've read arguments for both and I'm sure you could justify using either. Having said that, fixed kernel has been used the most.

I know some have argued that selection of the smoothing parameter is actually most important. Horne & Garton (2006) say LSCV for sample sizes > 50, CVh for samples < 50.

I've found that the Animal Space Use software (http://www.cnr.uidaho.edu/population_ecology/animal_space_use.htm) for most purposes is ideal because it provides an information-theoretic approach to model selection and a way to calculate core areas that is more intuitive than simply assigning the 50% contour. It's main limitation is that it's output file (.txt) is not very GIS-friendly, and the reason I created the ArcGIS extension mentioned in an earlier post. I believe a newer version of ASU (1.4?) is in the works.

Another choice is Home Range Tools (http://blue.lakeheadu.ca/hre/) which calculates both Fixed and Adaptive kernels and is very fast. It doesn't offer CVh as a smoothing parameter though.

Lastly, if there are any hard edges (e.g. a lake) present in your study area/home ranges then you might be interested in a non-kernel approach using the LoCoH software (http://locoh.cnr.berkeley.edu/arctutorial). 
   
Report Inappropriate Content • Top • Print • This Forum is closed for replies.    
Subject Re: Interpreting pixel values and LSCV 
Author Mary von der Porten 
Date Dec 16, 2009 
Message John,
Thanks again for your help. That helps a lot.
kind regard,
mary