Running AcoIite in default settings highly overestimated the Rrs values in all the bands of MSI and OLI (collected over the Barents Sea). The MAPE range from 19 to 65%. Following your suggestion, I have used a fixed AOT value (higher than the estimated one) for all the matchup scenes. The results have improved considerably. The MAPE across all the visible bands is < 15% (>300 m matchups). My question is can I report this (tunning of Acolite DSF method over the study region) in a scientific paper? I will appreciate your feedback.
Which tuning did you do? Did you add a fixed offset to the AOT, or optimise the AOT with the in situ measurements you have?
I guess this could be documented as a method in your paper, but if it is not generally applicable it should be stressed it is for that specific application only.
Did you do a comparison of ACOLITE derived AOT and the optimised AOT? I would be interested in seeing those results to understand the required change in AOT.
S2A_MSIL1C_20210626T130711_N0300_R081_T33XWH_20210626T151136.SAFE AOT 0.08 to 0.09 (for different matchup locations)
S2B_MSIL1C_20210621T112629_N0300_R080_T33XXF_20210621T121350.SAFE AOT 0.08 to 0.14
S2B_MSIL1C_20210825T103619_N0301_R008_T33WYV_20210825T130130.SAFE AOT 0.13 to 0.14
....
S2B_MSIL1C_20210605T110619_N0300_R137_T35XLC_20210605T121017.SAFE AOT 0.10 to 0.12
I have used AOT 0.15. The offset method did not work because for some scenes estimated AOT was too low (0.08) and even an offset of (+0.05) overestimated Rrs.
Yes, I have applied fixed AOT= 0.15 to all scenes.
I processed the scene two times.
First run:
aerosol_correction=dark_spectrum
dsf_path_reflectance=tiled / fixed
dsf_write_aot_550 = True
The Rrs data was overestimated on the first run. The estimated AOT from the first run was 0.07 to 0.14 for all scenes.
Second Run:
dsf_aot_estimate=fixed
dsf_fixed_aot=0.15 (optimized)
dsf_fixed_lut=LUT name (from first run)
Please note that the fixed AOT (0.15) may not work in other areas. In my case, the Rrs data were overestimated in all matchup scenes so I thought using user-defined AOT value may decrease the overestimation of Rrs values. Please let me know if it is not clear.
For the matchups we have, it looks like a reasonable value. What do you suggest? An offset of (0.05) sometimes overcorrects and produces negative results. In some cases, the estimated AOT is too low (0.07) and highly underestimates Rrs.
I have also tried ICOR. It fails to estimate AOT from all the scenes, but the default AOT value of 0.1 has yielded good results.