In ee_to_df, non-geometry features are created using toDictionary with the default setting, which excludes non-system properties. Later, when the dataframe is subset by the col_names parameter, if the default column names (None) is used, column names are fetched from a representative feature using propertyNames, which includes system properties. The two lists should be the same to avoid errors.

The fix is to get propertyNames and use the variable for both creating features and later subsetting dataframe columns.

This code demonstrates the issue:

import geemap
import ee
ee.Authenticate()
ee.Initialize()

img = ee.Image('LANDSAT/LC08/C02/T1_L2/LC08_044034_20210508')

# sample image and copy image properties to each feature
samp = img.sample(numPixels=10).map(
    lambda feature: feature.copyProperties(img, img.propertyNames()))

df = geemap.ee_to_df(samp)
df