Error during interpolation caused by np.nans in noiseRangeVector

[hfisser]

Thanks for the great work and the neat package.

Summary
The noise range lookup-table returned only 0s that were replaced by np.nans in get_noise_range_vectors, which led to an error in InterpolatedUnivariateSpline.

Setup
Machine: Kubuntu 22.04, 8 x 11th Gen Intel Core i5, 15.4 GiB of RAM.
Program: Python 3.8.15

Processing

• S1A_EW_GRDM_1SDH_20160912T132507_20160912T132607_013020_0149F1_B83A.SAFE (ascending)
• HV polarization
• Tested remove_thermal_noise and remove_texture_noise

Error
An error was raised in remove_texture_noise and remove_thermal_noise.
The investigation implies:

• The error was caused by the InterpolatedUnivariateSpline method called in get_shifted_noise_vectors.
• InterpolatedUnivariateSpline failed because empty vectors were passed to it.
• The vectors were empty because the noise vector at this azimuth and range had only np.nans.
• The noise vector had only np.nans because the noiseRangeVector had zeros that had been replaced by np.nans in get_noise_range_vectors. The corresponding line is 153 in sentinel1image: n[n == 0] = np.nan.
• Among 16673 pixels, 176 were np.nan in the noise vector filled in get_noise_range_vectors.

Side note

• Processing HH of this product did not cause any error
• Processing HV and HH of a different product on the same date and in the same geographic region did not cause any error S1A_EW_GRDM_1SDH_20160912T033838_20160912T033942_013014_0149C3_BCC8.SAFE (descending)

Suggestions
A. Quick fix: Introduce an error handling that catches the exception raised by InterpolatedUnivariateSpline.
B. Root cause: Find out why noiseRangeVector['noiseRangeLut'] has 0s and either address it and/or do A.

[akorosov]

I'm afraid this image is too old and ESA doesn't provide noise range values in the auxiliary metadata.