Here's an implementation of the cloud/ice detection that is not susceptible to the problems in #163 and #181. It looks at the bit-packed data directly rather than masking on a list of blacklisted values individually and adding them (potentially double-counting). Based on the deductions of @haoliangyu.
Right now, this only masks pixels if they are detected to be cloud or snow/ice with high (>67%) confidence. I don't know whether this is the best threshold for sensitive change detection or NDVI algorithms.
Here's an implementation of the cloud/ice detection that is not susceptible to the problems in #163 and #181. It looks at the bit-packed data directly rather than masking on a list of blacklisted values individually and adding them (potentially double-counting). Based on the deductions of @haoliangyu.
Right now, this only masks pixels if they are detected to be cloud or snow/ice with high (>67%) confidence. I don't know whether this is the best threshold for sensitive change detection or NDVI algorithms.