I’m working with the Recreation model to calculate photo user days and predictor information across the whole country of Colombia. While playing with this, two things have happened that seem odd, and I haven’t found mention of elsewhere in this forum.
1/ The model ran successfully and quickly on both 10km and 5km resolution. However, the 5km version has areas of missing data (not grid cells with 0 values, but no grid cell at all), and interestingly they’re located right in the middle of the highest-visitation areas.
Here’s an example. The first image is the 10km PUD result, which shows complete coverage and high PUD values in yellow:
The second image is the 5km PUD result, which shows the areas of missing data in turquoise. (Note: I do understand why there’s no data around the outer edge of the polygon, that’s not the issue here, I’m interested in the holes within the yellow areas).
2/ One of the predictors I tried was “polygon_percent_coverage” of a national park shapefile. The predictor_data values for this should be 0-100%, but I’m getting values up to 152%. Any idea why?
Hey @swolny , this is super interesting. For the question of the holes, could you share your AOI so I can reproduce that? I can’t begin to explain it otherwise.
For the percentage values over 100, I wonder if the predictor layer could have overlapping polygons such that their computed area sums to greater than the area of a grid square. If so I’d say that’s a bug. So it would be useful to get that predictor layer from you as well. Thanks!
I also confirmed that the Parks layer has some overlapping polygons and the area of overlap is being counted once for each polygon, inflating those area metrics. I’ve filed an issue to fix that bug. In the future we can “union” those predictor polygons before computing the area that intersects with the AOI cell.
As for the holes in the PUD result, I’m not able to reproduce that. I think it comes down to the map symbology. Those missing areas appear to have PUD values greater than the upper bound in your legend.
Thanks a lot for looking at this Dave. Especially since the missing data turned out to be ArcPro/user error. Since you didn’t see the same issue with the missing PUD cells, I brought that result into QGIS and ArcDesktop, to compare with ArcPro, which I was originally using, and found out what was going on.
The cells that were missing in ArcPro were present in QGIS. In ArcDesktop, when I was classifying the value bins while symbolizing, it informed me that the “maximum sample size had been reached”, then by default it actually clipped the data values, and along with it, clipped those grid polygons from the map entirely! ArcPro appears to have done the same thing, but did not inform me that the sample size had been reached, it simply clipped the data and polygons silently. Argh! But maybe this will help someone later.
