Eo-Learn Bandwise Median & Percentiles over Time

EO Products / Scripts
1

Hey there,
I am currently playing around with the excellent eo-learn library and I am wondering if there is any EOTask or workflow that extracts bandwise mean and percentile values over the entire timeperiod?

I found MapFeatureTask, but I am struggling with accessing the single bands from my data set inside the function. So what I am currently doing is something like a nested loop in which I loop over all bands first and execute the workflow from new for each patch. Is this the way to go? Or how would you go on about this?

BAND_NAMES = ["B02","B03,"B04","B05","B06"]

#loop over band bands, and define workflow from new for each band
for b in range(len(BAND_NAMES)-1):

    print(BAND_NAMES[b])
    #extract band from my S2 Band Stack
    move_band = ExtractBandsTask((FeatureType.DATA, 'S2_L2A_BANDS'),
                                 (FeatureType.DATA, BAND_NAMES[b]),
                                 [b])
    #calculate mean
    calc_mean = MapFeatureTask((FeatureType.DATA, BAND_NAMES[b]),  # input features
                               (FeatureType.DATA_TIMELESS, '{}_mean'.format(BAND_NAMES[b])),  # output feature
                               np.mean,  # a function to apply to each feature
                               axis=0)  # function's kwargs

    #remove extracted temp. band again
    remove_band = RemoveFeature((FeatureType.DATA, BAND_NAMES[b]))

    # define the workflow
    workflow = LinearWorkflow(load,
                              move_band,
                              calc_mean,
                              remove_band,
                              save)

    for idx, patch_name in enumerate(patch_list):
            print(patch_name)

            extra_param = {load: {'eopatch_folder': patch_name},
                           save: {'eopatch_folder': patch_name}}

            workflow.execute(extra_param)
2

Hi @siklar,

Your approach of extracting mean values for each band seems to be ok. However, I have a couple of suggestions for improvements:

  • Instead of creating and running a separate linear workflow for each band you can have one workflow with multiple tasks that extract bands. You can even chain these tasks one after another. The advantage of having a single workflow is that you save and load data only once.

  • Your task that calculates mean values also works on all bands together. Therefore you can switch the order with the ExtractBandsTask. This way you’ll first have a single task that calculates mean and afterward, you have multiple ExtractBandsTask tasks, one for each band.

1 Like
3

Thank you for quick response, @maleksandrov! :partying_face: