I am currently using ArcPro3.6.0, and have the Advanced License with the Spatial Analyst Extension. I have been tasked with computing various landscape metrics (including Shannon Diversity, Edge Density and Interspersion and Juxtaposition) across a custom land cover raster we built that covers Canada's agricultural landscape. We typically use 5kmX5km grids and use the Split Raster tool to export each grid as a .tif, and then use a batch import file to bring everything into FragStats where the necessary metrics are calculated.
We are now exploring the possibility/viability of moving to a 1kmX1km grid, which is proving difficult. To "experiment", we clipped our area of interest to the province of Ontario, which still leaves us with over 348,000 grids. When trying to run the Split Raster tool, each .tif was taking about 4min, which immediately makes that a not very efficient process. So then we tried running this Split Raster tool to which exported rasters were being put into a fgdb, which did speed up the time to about 30sec per grid. For curiosities sake, I took another subset of my grids (about 2000 units) and tried the same split raster approach, and it went back to processing 200grids in less then a minute, so it seems like the more grids its splitting, the longer each individual grid takes (which I don't really understand, but that just may be the reality of the situation).
I'm looking for advice/suggestions on what we can do to try and build efficiencies in our process. I explored using model builder to iterate through my grids and employ the Extract by Mask tool, which seemed to work well but we're still looking at ~7days (estimate) to just get the grids out for Ontario, and this is to be used across the country. I've looked into some downloadable toolboxes and explored ATtILA Landscape metrics toolbox, but it throws an error when trying to create their Land Cover Classification Editor, which pretty much puts me dead in the water with the remaining tools.
Additional context that may be of value: I am working off of two local 2TB-SSDs, one of which is setup as my scratch workplace. My computer has 20 cores, which I've maxed out in my Environments setting. The raster I'm working with has 56 land cover classes (not sure if that matters) and has a resolution of 10m. I'm hoping there is some alternative way to EITHER get my raster split by the 1kmX1km grids OR find a way to extract the metrics I need directly in ArcPro. Any suggestions would be appreciated!! I would be willing to explore R or Python approaches, but I'm not necessarily convinced that this would improve processing times...
