Suitability Modeler extremely slow

703
9
03-01-2024 07:23 PM
MK13
by
New Contributor III

Is it normal for the suitability modeler to perform so slowly? Each time I select a variable(criterion) to transform, it takes about 5 minutes to apply the tramsformation. This happens each time I select a different transformation function making it an excruciating exercise to go through to perform all the analysis that I need.  I suspect that the fact that I am utilising a Mask layer is not helping matters. Am I doing something wrong or is this expected behaviour for the modeler? I am utilising ArcGIS Pro 3.2.2

@MariannaGnedin @AndreasHall @ChrisLandvogt @DavidPike @RyanDeBruyn @SarmisthaChatterjee @KevinMJohnston @SteveLynch @DanPatterson 

0 Kudos
9 Replies
RonaldHaug
Occasional Contributor II

Hi MK 13!,

Spatial Analyst processes oftentimes use more memory, cpu and gpu than other processes. Have you checked out what you've got under the hood on your computer lately? https://pro.arcgis.com/en/pro-app/latest/get-started/arcgis-pro-system-requirements.htm

Do you have an NVIDEA GPU which you can set up and use as a coprocessor?

I'm sure having a raster file in the mix is slowing things down, too. Will your analysis still be meaningful if you convert it using raster to vector geoprocessing tool?

Closing tabs and programs you're not using will free up memory, too.

Let us know how you're doing with a screenshot of your results.

0 Kudos
MK13
by
New Contributor III

MK13_0-1709521645423.png

 

MK13_1-1709521683634.png

@RonaldHaug I have a pretty good system as you can see from the specs above.

I am performing a suitability analysis that is utilizing slope, precipitation, wind and temperature rasters to give me an output suitability raster from which I can extract suitable locations. I can't imagine that converting these rasters to vector for analysis would give me good results.

0 Kudos
RonaldHaug
Occasional Contributor II

Hi MK 13,

Looks like you have enough under the hood.

You didn't mention if you are using the coprocessing capability of your graphics card. Are you able to use it for the math you're doing?

https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/gpu-processing-with-spatial-...

All considering, 5 minutes doesn't seem like an inordinate amount of time.

Your slope file alone has 8 billion points on it. If you rerun slope and make it less sensitive, the file  will be smaller, and will take less time to process subsequent math. If your topography is fairly regular, that will work fine. If it's complex you may want to leave it or dial the sensitivity in so you get what you came for with your model.

I'd love to see your finished product. Very exciting.

MK13
by
New Contributor III

I believe the tool utilises the graphics card by default??..... 

Actually it was more like 10 minutes per change of transformation function which can really add up. I retrieved an elevation layer with lower resolution and that definitely performed better, which is a shame that I have to downgrade my data quality just to be able to perform analysis... 

This is my initial result for a sample service area but I am far from done. I still have to read about the rescale functions and make them work for my data.

MK13_0-1709602553710.png

 

0 Kudos
DanPatterson
MVP Esteemed Contributor

Questions

locally stored data?

coordinate system ?  (all the same?  projections required?)

size of the area? (cell size, rows columns etc)

 


... sort of retired...
0 Kudos
MK13
by
New Contributor III

- All data is locally stored.

- All data is projected to the North American Equidistant Conic projection

For 3 of my rasters:

columns - 1022
rows - 989
cell size - 870

For my slope raster:

columns - 91853
rows - 88902
cell size - 9.68

I have to say that I've noticed that the modeler performs well in terms of time to complete transformation until I transform my slope raster. That's probably where my issue lies but I am unsure of how to resolve it.

@DanPatterson 

0 Kudos
SteveLynch
Esri Regular Contributor

Some useful comments and questions above.

Here are some more
1) is your data stored locally or on a network (local is better)
2) turn off the visibility of the layers after they are transformed and also other layers in your Contents
3) uncheck Auto Calculate while modifying the transformation
4) "Save transformed dataset when model is run" needs only to be enabled if the transformed layers are to be used outside the Suitability Modeler
5) size of your data (number of columns and rows)
6) are all the data in the same coordinate system and have the same cell size?
7) how many criteria are you using?

0 Kudos
MK13
by
New Contributor III

- Data is locally stored

- Data size: For 3 of my rasters:

columns - 1022
rows - 989
cell size - 870

For my slope raster:

columns - 91853
rows - 88902
cell size - 9.68

- All data is projected to the North American Equidistant Conic projection

- I am using 4 criteria (slope, precipitation, wind and temperature)

All the other tips are duly noted although I have noticed that I get a performance hit when I attempt to transform the slope raster since it's more detailed. Transforming the other 3 rasters takes seconds versus 10 minutes for the slope raster.

@SteveLynch 

0 Kudos
DanPatterson
MVP Esteemed Contributor

The slope raster has a much smaller cell size than the others, so to standardize the inputs to the coarser scale a lot of work has to go on.  This type of work would be best done prior to using the suitability modeler.


... sort of retired...