Pixel value extraction with rasterio retrieves different value than QGIS

Question

I am extracting the pixel values from a raster using two different methods: one is extracting the values to points (centroids of polygon grids)

folder = (...)
S2_green = "band_green.tif"
OS_green = os.path.join(folder, S2_green)
Open the raster and store metadata
src = rasterio.open(OS_green)
centroids['lon'] = centroids['geometry'].x
centroids['lat'] = centroids['geometry'].y
coords = [(x,y) for x, y in zip(centroids.lon, centroids.lat)]
Sample the raster at every point location and store values in DataFrame
centroids['Raster Value'] = [x for x in src.sample(coords)]
centroids['Raster Value'] = centroids.apply(lambda x: x['Raster Value'][0], axis=1)

(seen in Python - Extract raster values at point locations)

The other method is using the zonal_stats() function in rasterstats python library (as I have only one value, the "mean" should be the same value as the pixel)

Then, what I see is a small variation of value (see image took in QGIS pixel info). Should I consider this as normal?

Important: all layers are in the same projection.

Agreed with BERA. I suspect 0.0678 was stored as an integer at some point and scaled by 0.0001. When it was converted to float there was precision errors which QGIS is probably rounding off. — Shawn, Oct 21 '21 at 13:20
What is the accuracy of the values ? Demystifying Floating Point Precision and What Every Computer Scientist Should Know About Floating-Point Arithmetic — gene, Oct 21 '21 at 13:36
Thank you, I did not know about this. That makes sense, as the raster has a pixel depth of 2 bytes unsigned. — GoldenTomato, Oct 21 '21 at 13:56

Pixel value extraction with rasterio retrieves different value than QGIS

Open the raster and store metadata

Sample the raster at every point location and store values in DataFrame

0 Answers0