Landsat datasets don't support Level-1 products

[landscapelc]

The part of Geospatial datasets tells me that if I have downloaded Landsat8 images，I can use the code ：
landsat8 = Landsat8(root="...", bands=["B2", "B3", "B4", "B5", "B6", "B7", "B8", "B9"]).
But when I run the code, the warning will appear:
FileNotFoundError: No Landsat data was found in ...
I make sure that my root is correct and I have downloaded Landsat8 image from USGS, but the warning also exist.

[adamjstewart]

It could be that the filename format of the files you downloaded is different than what the dataset class expects. What do the filenames look like?

[landscapelc]

I have downloaded Landsat8 images from USGS, including Landsat 8OLI/TIRS C1 Level-1 products and Landsat 8-9 OLI/TIRS C2 L1 products. I don't know wheather they are suitable for torchgeo.
Sincerely for your answer!

[adamjstewart]

Can you share one of the filenames?

[landscapelc]

LC08_L1TP_020038_20211023_20211103_02_T1

[adamjstewart]

Is there no file extension or spectral band in the filename? Are all bands in a single file?

It looks like our Landsat dataset is only designed to handle level 2 surface reflectance products, it doesn't know how to handle level 1 products. For now, a quick workaround would be:

from torchgeo.datasets import Landsat8

Landsat8.filename_regex = r"""
    ^L                                                                               
    (?P<sensor>[COTEM])                                                              
    (?P<satellite>\d{2})                                                             
    _(?P<processing_correction_level>[A-Z0-9]{4})                                    
    _(?P<wrs_path>\d{3})                                                             
    (?P<wrs_row>\d{3})                                                               
    _(?P<date>\d{8})                                                                 
    _(?P<processing_date>\d{8})                                                      
    _(?P<collection_number>\d{2})                                                    
    _(?P<collection_category>[A-Z0-9]{2})                                                                                                            
"""
Landsat8.separate_files = False

landsat8 = Landsat8(root="...")

Let me know if the above code works. I'll need to think about how to handle both level 1 and 2 products. We may need separate datasets for both.

[landscapelc]

LC08_L1TP_020038_20211023_20211103_02_T1 is the name of folder and all relative files exist. All bands files in the folder. I don't change the format which is provided by USGS.

[adamjstewart]

What are the names of the files in this folder? TorchGeo doesn't look at folder names at all.

[landscapelc]

Including LC08_L1TP_020038_20211023_20211103_02_T1_B1.TIF ,LC08_L1TP_020038_20211023_20211103_02_T1_B2.TIF and so on. And I try to fuse all bands to get a new tif file, but also can't be recognized. Therefore, I can make a concise conclusion that TorchGeo is only designed to handle level 2 surface reflectance products which need to include all bands?

[adamjstewart]

Ah thanks, this is much simpler to handle than I expected. At some point in the near future I'll submit a PR to fix this. In the meantime, the following should work:

from torchgeo.datasets import Landsat8

Landsat8.filename_regex = Landsat8.filename_regex.replace("_SR", "")

landsat8 = Landsat8(root="...")

[adamjstewart]

Can you check to see if this is fixed by #504? With this PR, you should be able to simply say:

Landsat8(root="...", bands=["B1", ..., "B11"])

and TorchGeo will pick up all Level-1 data for you.