Images are exported as solid colour blocks

Question

My goal is to download RGB imagery over a selection of regions and times. To do this, I am using the geemap python library to locally export a filtered image collection. However, the images are exported as almost solid colour blocks, except for the last two image rows. After investigating the metadata of faulty images, I noticed that all the faulty images have MGRS_TILE=32UMG. At first, I thought I could filter the collection to only download images with MGRS_TILE=32VMH, as those look fine for this region, using the following command:

col = ee.ImageCollection('COPERNICUS/S2_SR') \
    .filterBounds(bbox) \
    .filterDate('2018-01-01', '2018-12-31') \
    .select(band_list) \
    .filter(filterCloudCoverage) \
    .filterMetadata('MGRS_TILE','equals','32VMH')

but that filters out good looking images with the 32UMG code for different regions I need to cover.

Is there a way to filter these solid colour images, or do I have to somehow find and delete these images later on?

I do not care if the exported image is 32UMG or 32VMH, but I do care about the overall number of images exported as I am using this number for some statistics.

If such a thing is not possible, could you explain to me why is this happening?

The attached picture shows how the same image is downloaded twice, once with MGRS_TILE=32UMG and faulty, and then again with MGRS_TILE=32VMH.

Here is my code:

import ee
ee.Initialize()
import geemap
import os

filterCloudCoverage = ee.Filter.lt('CLOUD_COVERAGE_PERCENTAGE', 50)
filterNoData = ee.Filter.lt('NODATA_PIXEL_PERCENTAGE', 10)
proj=ee.Projection('EPSG:25832')
bbox = ee.Geometry.Rectangle([440000, 6200000, 450000, 6210000],proj, True, False)
band_list = ['TCI_R','TCI_G','TCI_B']
out_dir_samples = os.path.join(os.path.expanduser('~'), 'Downloads/EE/RGB')
filename = os.path.join(out_dir_samples, 'metadata.csv')

col = ee.ImageCollection('COPERNICUS/S2_SR') \
    .filterBounds(bbox) \
    .filterDate('2018-01-01', '2018-12-31') \
    .select(band_list) \
    .filter(filterCloudCoverage)

geemap.ee_export_image_collection(col, out_dir=out_dir_samples, scale=10, crs='EPSG:25832', region=bbox)
feature_col = ee.FeatureCollection(col)
geemap.ee_to_csv(feature_col, filename=filename)

Answer 1

I did not manage to find a solution that could filter these faulty images during a download process. However, I did remove the faulty images by filtering them after the download. Because the faulty images were mostly solid colour blocks, they were significantly smaller in size than the valid samples. Using this property, I managed to filter out bad images regardless of their MGRS code.