Better results with previous version

[cdivittorio]

I have been using ACOLITE to process Landsat 7 imagery over a very turbid inland lake. I then developed regression models with in-situ measurements of turbidity and TSS. I had a student working on this last summer, prior to the 2021 update and we managed to get some pretty high R^2 values for our fitted models. However, now that we have updated our data with the latest version of the software on REMSEM (Jan 2021), the performance of our regression models has decreased drastically. I am wondering if anyone here has had a similar experience. I am also curious whether I could obtain the older version of the software, or if I can somehow adjust the settings in the newer version to replicate the older version. Thank you!

[quinten]

Could you provide your settings files for the old and new versions?

There are some defaults that have changed between the 20190326 and 20210114 releases that may provide different results. Especially if you are using a ROI and have images near the principal sun glint spot.

Quinten

[cdivittorio]

Hello Quinten, thank you for the prompt reply! I have attached both settings files using the default parameters. I did actually look into the differences between them and tried reverting the "sky_correction_option" back to "all" instead of "rsky_new" since this seemed to be one of the biggest updates. I have not done a thorough analysis yet, but I don't think it helped us out too much. I also reduced my tile to the following sub-area: limit=35.51,-80.47,35.79,-80.12 for our water body of interest.

[quinten]

I will take a look at your scenes! Probably the "dsf_path_reflectance=tiled" default in the new version makes a large difference, perhaps setting it back to "dsf_path_reflectance=fixed" works better for you?

Quinten

[cdivittorio]

great, thank you! I will test out the fixed path modification as well. I look forward to hearing your advice once you take a look at the scenes - I appreciate your time.

[quinten]

I processed with the latest release (20210802.0) and the results look quite good for your scenes using the following settings:

Code: Select all

limit=35.51,-80.47,35.79,-80.12
dsf_aot_estimate=fixed
resolved_geometry=False
dsf_residual_glint_correction=True
dsf_residual_glint_correction_method=alternative

Note that these options are not compatible with the 20210114.0 release as the new release uses the generic code base. There is quite some glint on the surface, but the residual glint correction seems to be able to take care of that:

Screenshot 2021-08-03 at 09.37.56.png (954.85 KiB) Viewed 119199 times

What do you think?

Quinten

[cdivittorio]

Thank you for testing this out and recommending some alternate settings! I have adjusted the l2r settings file as you recommended and will see how the revised images turn out. Could you please tell me which L7 date you used so I can double check that I get similar results?

Do you think the same settings be applied to Landsat 8 and Sentinel 2 for my area? Or is it more of a trial-and-error process to identify the most appropriate settings?

[quinten]

These settings should be good for L5/7/8 and S2 A/B as well! The optimal settings depend more on the study area and topic than on the sensor. Since you are processing a rather small area of inland waters for retrieval of turbidity the fixed aot over the study area seems appropriate. The residual glint correction seems necessary in your region.

Edit: The date of the L7 scene was 2009-04-16

Quinten

[cdivittorio]

Great, thank you for the follow up. For our Landsat 8 and Sentinel 2 analysis we will also be looking at Chl-a and CDOM - do you recommend reverting back to the tiled aot settings?

We are getting much better regression models using the revised dataset with the settings you suggested. We really appreciate your support!

[quinten]

For the same study area I would recommend keeping the "fixed" aot estimate. The tiling seems to be more important for very large areas (or e.g. a Sentinel-3 scene).

CDOM and Chl-a may be quite tough to do, especially with the L8/OLI band set. Good luck!

Quinten