Open TonioF opened 6 years ago
The inference engine only provides an output where (i) the state mask is true (spatial domain) and (ii) where the temporal grid is defined (time domain). So both choices affect the output.
Currently, the AtCor algorithm do not rely on the temporal information, but spatial information, since the atmospheric parameters are highly spatially correlated up to 10s kms. Applying them on smaller area is possible to an extend, but if it's for doing them in parallel, then it may be unnecessary as I have used multiprocessing on any part if it is possible.
Sorry, maybe I was not specific enough when I asked the question. Spatial / temporal correlation are the important issues here. So, this question is very tightly coupled to #3 .
For SAR preprocessing: The spatial extent has no effect on the results itself only on the processing time The temporal extent will have an impact on the results (preprocessing step: multi-temporal speckle filter). But therefore I need to do an analysis how many scenes from different time steps are really needed for good processing results. Right now I think that maybe 5 time steps (10 days) are enough. But that is only a guess.
Does the definition of the temporal or spatial region affect the outcome? I.e., will the result be different if I choose another spatial subset or start and ending times in my configuration? I would like everyone in the project to answer this question. I know that the inference engine is not affected by the choice of the spatial region, but heavily by the temporal. Also, I know that the opposite is true for the atmospheric correction (though, @MarcYin, I would like to know why). What about the pre-processing steps and the prior engine?