Invest habitat quality model

Hi, I am running the invest habitat quality model.
The pixel size of the rasters I made is 15.15. My study area is very wide and because of this I have divided it into four areas. The model runs very slowly and may take a few days. Also, c drive memory runs out quickly and the software gives an error. Even when it has 65 GB of memory, can you help?

1 Like

@fatima Thanks for posting here!

The Habitat Quality model does a lot of convolutions which is an expensive operation and if you have really large data it would not surprise me if the model took days. Can you say what the Max Distances are for your threats? Those values should be in Kilometers (KM) and NOT Meters. The larger those distances the larger the kernel that is used for the convolutions. 15 x 15 meter inputs should be okay, but depending on the size of the rasters and the distances to decay threats over, will determine how long it takes the model to run.

Busting 65 GB of memory is worrisome and reminds me that we recently fixed a memory leak with these convolutions! You can look for that fix in InVEST 3.9. With InVEST 3.9 Habitat Quality has been reworked such that some of the inputs will be a little different. We will post to the forums when InVEST 3.9 is released, which hopefully is tomorrow!

Best,

Doug

1 Like

Dear Doug
Thank you for your early reply.
My study area is very large and I had to divide the area into 4 parts.
The maximum distance I set for threats is 9 km.
I was very eager to choose the size of the raster cells to be 10 meters, but I gave up because it slowed down the model and chose the size of 15 meters.
The memory loss problem was solved and I got output for two parts of the study range.
Finally, my question is, if I use a PC instead of a laptop, will the model run faster?
My English is not very good and I hope I was able to get it.
best regards
Fatima

Hi Fatima,

Glad to hear you have some things working! I understand it can be a tough choice to trade off resolution for model performance and time. Certainly if you have a Desktop PC with a solid state drive (SSD) and a faster processor you could see some improvements in run time. But you’ll only get linear speedups (as a function of processor speed and disk I/O). Unfortunately with this model the convolutions are the bottle neck, so reducing cell size and the max distance will have a much bigger impact on the time it takes to run.

Best,

Doug

1 Like