improve maskCover #349
improve maskCover #349
Conversation
… a geometry. maskCoverRegions does the same but inside a FeatureCollection.
| region: ee.Geometry, | ||
| scale: Optional[int] = None, | ||
| band: Optional[str] = None, | ||
| proxy_value: int = -999, |
There was a problem hiding this comment.
cannot we use a default value instead ? like int max ? As a user that's definietely a parameter that i would not like to touch.
| def maskCoverRegion( | ||
| self, | ||
| region: ee.Geometry, | ||
| scale: Optional[int] = None, |
There was a problem hiding this comment.
The reducer related parameters should be regrouped under a "kwargs" parameter with a direct reference to the reducer documentation so we don't need to support them all, the user can decide. Also the reason why I used "bestEffort" in my initial implementation was to avoid to force the user to take descision for scales before the "final" reduction. I usually try to avoid setting it in earlier computation and this function will force me to.
| mask, geometry = self._obj.select(0).mask(), self._obj.geometry() | ||
| cover = mask.reduceRegion(ee.Reducer.frequencyHistogram(), geometry, bestEffort=True) | ||
|
|
||
| band = band or 0 |
There was a problem hiding this comment.
You could drop this line directly into the select, it's never used elsewhere and release 1 extra line
| # we want to display this resutl as a 1 digit float | ||
| ratio = ratio.multiply(1000).toInt().divide(10) | ||
| def compute_percentage(feat): | ||
| """function to map over the resulting table and compute the percentage from the reducer's output.""" |
There was a problem hiding this comment.
no need for docstring in nested function as it will never be reached by the documentation. On the other hand you could use hints to make mypy happier.
| image = self.image.geetools.maskCover() | ||
| assert isclose(image.get("mask_cover").getInfo(), 99.2) | ||
|
|
||
| def test_deprecated_mask_cover(self): |
There was a problem hiding this comment.
it's too soon to remove this one I deprecated it less than 6 month ago
| @@ -1315,11 +1317,27 @@ def distance(self, other: ee.image) -> ee.Image: | |||
|
|
|||
| return ee.Image(distance) | |||
|
|
|||
| def maskCover(self) -> ee.Image: | |||
There was a problem hiding this comment.
I used this function very recently so I remain convinced it's very useful to have it available. It's specifically fun to use in a simple map in an ImageCollection.
The only constraints is to have bounded images but that's exactly the error raised by GEE console so I'm happy with it.
As a rule of thumb I think the GEE errors are now well handling unbounded images and we should not spend too much time trying to work around it. When you start doing geometric stuff one need to clip and that's all.
Originally, this function was designed to compute the percentage of masked pixels inside a geometry. This allows the user to compute a similar value to
"CLOUD_COVER"property in Landsat scenes or"CLOUD_COVERAGE_ASSESSMENT"in Sentinel 2 scenes, but in a specific region instead of the whole scene. When creating composites we could discard "good" pixels if we filter the collection using the cloud cover property of the scene, but using this method we can discard pixels of regions with high percentage of masked pixels.The method has been divided in two methods:
maskCoverRegion: computes the value using aee.GeometrymaskCoverRegions: computes the value in eachee.Featureof aee.FeatureCollectionThis method can also be helpful in the creation of composites using unbounded images, like MODIS.
The param
scaleis to avoid using too much resources to compute this value when a rough computation is enough.