Sentinel-5p Machine Learning-- Interpolating Missing Values

I am feeding Sentinel-5p data into a prediction model.

Because I am using daily images, I have quite a few gaps do to clouds/qa values.

Is there a recommended method for interpolating missing values (and is this possible through harp)? Is it not advised to lower qa values to minimize the gaps? Seems like both lowering the qa value or using an interpolation method will increase uncertainty, so I’m not sure which is advised.

Filling gaps is usually done by making use of atmospheric models, but this requires quite some advanced knowledge.

If you want global data without gaps, you might want to look into using CAMS data. This already provides forecasts (i.e. predictions for the next 5 days) at both global and regional (Europe) scale.