Performance considerations regarding polygon vector input

[WCMC-vblanque]

Hi ExactExtract community,
Hi @dbaston,

We are very very happy with the use of exactextract, the speed and the quality of stats produced.

1. Is there a performance benchmarking existing for Zonal Stats calculated with different level of simplification of a polygon for a same raster? If not I will run considering the original polygon to compare it with different tolerance: 0.0001,dd 0.001dd, 0.01dd and 0.1dd and then share it with the community. Also considering how much it can affect the final stats (it might depend the size of the polygon and the entropy of the raster.
2. Do you recommend a specific vectorial format (GeoJSON, GPKG, (geo)parquet, Shapefile...) to get the best performance of exactextract?

Thanks in advance.
Best regards,

Vincent

[dbaston]

1. Unfortunately the best answer I have is "it depends." In my experience processing time is usually governed by raster I/O rather than polygon complexity. But, all things being equal, polygons with more vertices will take longer to process. If you haven't already, it's probably worth looking at how raster compression settings (in particular, chunk size) affect your processing time (if you have any control over these.). In some cases you may find that the processing strategy ("feature-sequential"/"raster-sequential") has a large impact. There is also a new "raster-parallel" strategy that works quite well. Please feel free to share any results you come up with.

2. I don't expect the choice of vector format to be very significant (and this is really a GDAL question, not specific to exactextract.) I would expect a text format like GeoJSON to be slower, and Shapefle may be slow for polygons that have many holes.

[WCMC-vblanque]

Thanks @dbaston for your answer. I will give a try with the new strategy "raster-parallel" and with Shapefile rather than GeoJSON.