Compute area-weighted statistics #640
Conversation
| "minority": float(keys[counts.tolist().index(counts.min())].tolist()) | ||
| if valid_pixels | ||
| else numpy.nan, | ||
| "mean": float(array.mean()), |
There was a problem hiding this comment.
we don't use the weighted array for the min and max (like https://github.com/isciences/exactextract#supported-statistics)
| "std": float(array.std()), | ||
| "median": float(numpy.ma.median(array)), | ||
| "majority": majority, | ||
| "minority": minority, | ||
| "unique": float(counts.size), | ||
| **dict(zip(percentiles_names, percentiles_values)), |
There was a problem hiding this comment.
🤔 maybe the percentiles should use the weighted array?
| # 3, 4 | ||
| data = np.ma.array((1, 2, 3, 4)).reshape((1, 2, 2)) | ||
|
|
||
| # Coverage Array |
There was a problem hiding this comment.
Maybe we should call this the weight array instead of coverage? coverage doesn't seem to make the most sense to me here.
There was a problem hiding this comment.
in https://github.com/isciences/exactextract, weight are really weight, while the coverage is called cell coverage fractions. I didn't want to confuse people 🤷
There was a problem hiding this comment.
I see, coverage came from exactextract because their weights are usually derived from the coverage of each polygon in the cell. I think weight is more general than coverage though. There could be use cases for weighted zonal stats that aren't partial-coverage
There was a problem hiding this comment.
coverage_weights ?
There was a problem hiding this comment.
Would we ever have both coverage and weights?
There was a problem hiding this comment.
It looks like exactextract does support both coverage and weight in effect, because it allows weight as a parameter and it computes coverage itself
There was a problem hiding this comment.
Would we ever have both coverage and weights?
Not planned right now but I don't want to be blocked in the future. I agree that coverage is not a perfect name but I don't want to use weights because it's not it's not weights but spatial fraction. I'm open to change to better name if you have ideas :D
| (self.height, cover_scale, self.width, cover_scale) | ||
| ).astype("float32") | ||
|
|
||
| return cover_array.sum(-1).sum(1) / (cover_scale**2) |
There was a problem hiding this comment.
slightly modified version of perrygeo/python-rasterstats#136
@sgoodm I know ☝️ is a 7 years old PR (😅) but I hope you don't mind that I reused some of the code here 🙏
There was a problem hiding this comment.
Not at all, @jacobwhall and I have been pushing use of COGs in our work at @aiddata and are fans of these projects, so happy to share some code to support 👍 👍
|
Note: we don't have a BaseReader method that returns statistics for a GeoJSON Feature but this is how it could look like in the with Reader(path) as src:
data = src_dst.feature(
shape,
shape_crs=WGS84_CRS,
)
coverage_array = data.get_coverage_array(
shape, shape_crs=WGS84_CRS
)
stats = data.statistics(coverage=coverage_array) |
ref NASA-IMPACT/veda-backend#226
This PR adds a
coverageoptions to theget_array_statisticsfunction to enablearea weightedstatistics. This PR do not take care of thecoveragearray creation which should be done by the client application (maybe with some helper in rio-tiler)cc @kylebarron @j08lue
To Do
coverageutility functions