SCL band not relaible to filter cloudy images at all?
@arashmad found the following AOI, which is obviously cloudy for more than 20%, gets downloaded even with a aoi_min coverage of 99%. To reproduce use:
config = loadConfiguration(path="data/default_config.json")
config["user_settings"]["aoi_settings"]["bounding_box"] = [12.958365076017566, 52.3548652317379,
13.033381097746084, 52.42725408033372]
config["user_settings"]["aoi_settings"]["date_range"] = ["2020-06-22"]
config["user_settings"]["tile_settings"]["s2:nodata_pixel_percentage"] = {"lte": 100}
config["user_settings"]["aoi_settings"]["apply_SCL_band_mask"] = False
config["user_settings"]["aoi_settings"]["aoi_min_coverage"] = 80
config["user_settings"]["tile_settings"]["bands"] = ["blue", "green", "red"]
config["user_settings"]["result_settings"]["results_dir"] = "x/y/z"
# start download
s2Downloader(config_dict=config)
I downloaded the same extent and date with the s2Downloader and found the following SCL band for that scene:
And indeed the problem here is the SCL Band. I looked into the values and calculated the to be filtered values (0.25%), which fit to the log of the s2Downloader (99.25% aoi_min_coverage).
Of course I was aware, that the SCL filtering is has not 100% accuracy, but this example from above makes me slightly unsure if we can relay on it in any way.
Because MiSa.C depends on that more or less.