search

nasa-jpl / autoRIFT

A Python module of a fast and intelligent algorithm for finding the pixel displacement between two images
Apache License 2.0
214 stars 52 forks source link

Add additional filter to support Landsat 7 SLC-off imagery #59

Closed jhkennedy closed 2 years ago

jhkennedy commented 2 years ago

fixes #52

alex-s-gardner commented 2 years ago

@jhkennedy would you be able to test of these 2 images [band 8]? I1 = LE07_L2SP_062018_20210717_20210812_02_T1 I2 = LE07_L2SP_062018_20210701_20210728_02_T1

leiyangleon commented 2 years ago

Looks good to me.

jhkennedy commented 2 years ago

Since we're processing Level-1 now, the example pair would be: LE07_L1TP_062018_20210717_20210812_02_T1 LE07_L1TP_062018_20210701_20210728_02_T1

jhkennedy commented 2 years ago

In order to better test Landsat-7, we can compare directly to version 1 ITS_LIVE products:

I've also attached a CSV of these scene name. Importantly, these are Collection 1 scene names, so we'll need to convert them to collection 2 names (I think just processing date has changed) to run them.

L7_testpairs.csv

jhkennedy commented 2 years ago

Here's the Landsat-7 naming convention, which holds across collections: https://www.usgs.gov/faqs/what-naming-convention-landsat-collections-level-1-scenes

So it looks like just the collection number and possibly the processing date needs to change. You can use this function to confirm you've got collection 2 scene names: https://github.com/ASFHyP3/hyp3-autorift/blob/develop/hyp3_autorift/process.py#L51

jhkennedy commented 2 years ago

Alright, I believe L7 processing using the Wallis-no-data-fill preprocessor is working as intended at this point. Here is an example product with outputs, including the source scene and the associated ITS_LIVE V1 data: https://drive.google.com/file/d/1FTtd8kS5Tz2OLtAzL1P2uXiH2pWsv0Fx/view?usp=sharing

And the browse image for this example: LE07_L1TP_063018_20040810_20200915_02_T1_X_LE07_L1TP_063018_20040911_20200915_02_T1_G0120V02_P041_IL_ASF_OD

alex-s-gardner commented 2 years ago

@jhkennedy this looks really good. The striping in V02 is due to a less aggressive interpolation for a 120m grid. The loss of data over land in the right of the image is a bit concerning but difficult to trace. The V02 input images have been reprocessed since v01... that might explain some of the difference. I also noticed the the image order in the file name is swapped. I vaguely remember discussing this previously by have now forgotten what we decided on that.

jhkennedy commented 2 years ago

@alex-s-gardner In HyP3-autoRIFT we enforce "earliest scene first" as the reference image, and we build the file name with reference x secondary.