SDM locations overlap with adjacent HUCs

[hpoulos]

In looking at the plot locations for the next SDM modeling element, all of the hydrologic units that were included in the original masking of the DEMs for each range have plot data with tree species in them except for CM and two microwatersheds in DM. This means that we should likely not clip the grids into smaller chunks to save computational time. Making them smaller will nave no effect on the model results themselves, since the model fits are based on 10 fold CV of the actual data, and not on the prediction surfaces themselves (i.e. tmin, tmax). Here are some screen grabs for each range to show you what I mean:

For all screen grabs blue dots are IV plot locations and purble dots are your sensor locations GM

DM

CM

[dschwilk]

It is really just a question of weather we predict climate values across this wide elevation range when our plots cover much less of it. We could always just run the spatial predictions across the tree distribution plot locations since we have presence/absence data for occurrence. Right now my code runs the tmin/tmax predictions across the whole dang surface, then summarizes by year, so it is very computationally intensive, especially the big matrix multiplication to get the time x space data back from the individual loading and scores predictions.

[dschwilk]

Is this a response to issue #37?

[hpoulos]

I have now downloaded the national land cover dataset of canopy cover. Here are the DM plots overlain on top of the canopy cover. I couldn't get the legend to display correctly but numbers go from 0 (black) to 100 (dark green) and represent % canopy cover.

If I drape the DEM that was clipped to all watersheds where plots fall and adjacent watersheds to plots I get this image:

These don't display that well here, but there are issues with using either approach. Using the DEM HUC watershed approach makes the grids very large. Using the random forest-generated canopy cover NLCD dataset would necessitate the recreation of all of the grids to a new extent, and the canopy cover falls well outside the range of where the plot data are into areas with other juniper species that we did not encounter in the plots, etc.

I am wondering if a combined approach might be useful. Perhaps clip existing grids that are at the HUC level to locations with canopy cover > 0?

[dschwilk]

Yes, intersection of both should work, right? I think that sounds good.

[dschwilk]

Moving this back to original issue, #37.