ACOLITE forum

Hi Zhigang Do you mean negatives are retrieved in the NIR band? For the scene you sent (I got the collection 2 equivalent LC08_L1TP_042034_20200523_20200820_02_T1) I get some slight NIR negatives in part of the lake, marked in red. From the spectrum view you can see they are very close to 0. zhigang...

Hi Bryan At least using interpn there is no attempt at creating a 28GB array? :lol: 20 GB is still a huge amount for the computations that are done. Maybe there is still a simpler/faster/lighter method for the tile interpolation. I now also tried using the RegularGridInterpolator which uses slightly...

Hi Bryan Well, it also depends a bit on the size of your region of interest whether I would recommend a subset with dsf_aot_estimate=fixed. It also depends on the magnitude of the water signal and spatial variability of aerosol concentration. Processing separately and merging afterwards will lead to...

Hi Bryan Were you able to process the full 10 metre scene on Linux? I usually recommend processing a region of interest subset of a scene, ideally using dsf_aot_estimate=fixed. It of course depends on your application whether this is feasible. The yaml file from your export is way too long, and I wo...

Hi Bryan Thanks for the updates! I am a bit reluctant to pin versions to the packages, but it is mainly a matter of laziness. :-) There is an open issue on the GitHub . Something I need to take into account in the future... I believe the issue lies with my use of scipy.interpolate.griddata in tiles_...

Hi Bryan With normal settings for a S2 bundle there should be 19x19 tiles, and datasets are interpolated to 10980x10980 pixels. At double precision this is ~965MB (10980 * 10980 * 8 bytes), so nowhere near what it is trying to allocate in your case. I presume something is going wrong with the constr...

Hi Bryan Yes processing MSI data at 10 m resolution currently uses a lot of memory. For processing on a laptop I would suggest adapting the resolution of the MSI output products, e.g. s2_target_res=20 or s2_target_res=60. It is slightly strange that creation of a 28GB array is attempted. I don't kno...

Hi Sidney Your setting of dsf_exclude_bands=10,11,12 had no effect since the default for MSI is to use only bands in the range 400-900 nm as set by dsf_wave_range=400,900. So running either of the following setting combinations would have the same effect: dsf_exclude_bands=10,11,12 dsf_wave_range=40...

Hi Sidney The current defaults for S2 already exclude the SWIR bands by the default dsf_wave_range=400,900 for MSI. If you do a run with dsf_wave_range=400,2500 you can include the SWIR to evaluate this impact. (Make sure to remove your dsf_exclude_bands setting if you want to try this!) We are look...

Hi Marco

Indeed all the masking is applied post-correction! At the moment the interface reflectance correction is either on or off based on the setting dsf_interface_reflectance.

Quinten