Ryan Grout

A quick benchmark using some datasets from your blog @perrygeo (https://www.perrygeo.com/zonal-stats-with-postgis-rasters-part-2.html) ```python import time import rasterstats print(rasterstats) from rasterstats import zonal_stats start = time.perf_counter() stats = zonal_stats( "ne_50m_admin_0_countries/ne_50m_admin_0_countries.shp", "wc2.1_2.5m_tavg/wc2.1_2.5m_tavg_07.tif", stats=["unique",...

@perrygeo I will need to dig up my real-world test case today. IIRC, the changes in the PR yielded a dramatic performance increase.

> @groutr Would you like to split the percentile computation into a separate PR? That seems like a clear win to me and I don't want to hold that part...

@perrygeo My real-world test case was flawed. The performance difference was largely accounted for by a change in formats that escaped my notice. However, I've been thinking about this further...

@markwang0 I think the issue here might be filling depressions. Is TauDEM filling depressions? The depression filling step fills in a large portion of the DEM. On my machine, if...

@sgillies After digging into this, I think the root cause is how the destination bounds are determined. Here are some of the lines from merge that are relevant to triggering...

I don't think I've tried merging rotated rasters. What happens? Is it something that can be fixed?

@sgillies good point. Would a good test be to rotate some rasters and merge them? If that is representative situation, I might try that in some spare time to check...

How is your tag name derived? Perhaps the update in gdal changed how it is represented? ``` > src.tags(bidx=1) {'valid_max': '0.5', 'valid_min': '0.02', '_FillValue': '-9999'} > src.get_tag_item("valid_max", bidx=1) '0.5' ```...

I am also hitting this issue with latest release of xarray. :(