[azlinszkysinergise]

Adds sentinelhub/clms_lst_landsat_sharpening.ipynb. This is an end to end use case, It sharpens the hourly 3 km CLMS LST V3 to 100 m with a Landsat 8/9 residual field, using the Sentinel Hub Process API on CDSE. It then evaluates the method:

• temporal stability of the residual pattern (Budapest, 2022–2024);
• portability of the Landsat–CLMS bias across ten cities in different climate zones;
• adequacy of the additive model;
• an error budget from a leave-one-date-out validation.

Notes

• Executed with outputs on the CDSE JupyterLab sentinelhub kernel; formatted with black 23.3.0; kernelspec sentinelhub.
• Credentials: uses SHConfig(), and asks with getpass only when no credentials are configured. None are stored.
• Not added to sentinelhub/.tests: a full run makes several hundred Process API requests and takes a long time.
• Size ~7.4 MB (embedded figures), within the range of existing notebooks.
• Archived on Zenodo: https://doi.org/10.5281/zenodo.22944002. This file is identical to the archived version.
• Companion custom script: sentinel-hub/custom-scripts, data-fusion/clms_lst_sharpening_with_landsat.

🤖 Generated with Claude Code - fully checked and creation guided by me

Description:

PR Checklist:

• Code is formatted and linted (Python linting action)
• Kernel spec metadata was correctly set (check_kernelspec action)
• Tests pass (test-samples action)
• If applicable, documentation has been updated to reflect changes

[azlinszkysinergise]

this is the pr for the custom script sentinel-hub/custom-scripts#375

[mariiaShevchuk]

Really nice notebook, well thought-out!

I just found one thing that affects the city results in Part 3:
night -time Landsat images have cloud cover -1, so they get through the eo:cloud_cover < max_cloud filter. The notebook can't use them because B04 and B05 are empty at night, so there's no NDVI and every pixel comes back invalid.
I ran the same search and it looks like 3 cities are affected

• London: 7 of 20 are night images
• Brasilia: 7 of 20
• Alicante: 3 of 20

Could you change the filter to eo:cloud_cover >= 0 AND eo:cloud_cover < {max_cloud} & rerun Part 3?
It would probably take quite some effort to update the city numbers in the text, and maybe conclusion, but I think it is worth it and London might make it into the comparison too then 🙂

[azlinszkysinergise]

Thanks for catching the night-time scenes as an error source, I was not aware they existed! Now we have change the catalog filter, to eo:cloud_cover >= 0 AND eo:cloud_cover < {max_cloud} and re-ran the whole notebook on the CDSE Jupyter Lab.

• Budapest (Parts 1-2) is unchanged: its 14 night scenes were already being dropped as empty, so the same 28 daytime pairs were used.
• London now passes with 13 pairs, so 9 of 10 cities are in the comparison (only Novosibirsk is excluded). Brasilia and Alicante also gained pairs.
• The bias conclusion holds (between-city spread 1.55 K vs 1.52 K within cities).
• New finding: in London and Alicante the sharpened maps are on average worse than plain CLMS (skill -0.20 and -0.10). The text now discusses why, and the recommendations include checking the skill on a second same-season Landsat date before relying on the method for a new area.

All Result paragraphs and the conclusions are updated, and the notebook is now version 1.1 (also on Zenodo).

[mariiaShevchuk]

Thank you, looks great!

One small thing I noticed: the notebook is outside the sentinelhub/use_cases/ folder -is that intentional?
Otherwise looks ready to merge. (the failing checks aren't from this PR)

[azlinszkysinergise]

No, it should actually merge to the sentinelhub/use_cases folder - do I have to change something on my side?

[azlinszkysinergise]

PR updated, content moved to sentinelhub/use_cases/ and added to the list in sentinelhub/README.md. The notebook content itself is unchanged.
ready to be merged on my side - if this is published by 4 November I can also show it at the Climate Visualization Conference

[mariiaShevchuk]

Thanks, the latest commit covers it!

[mariiaShevchuk]

Merged.
Since the notebook moved to use_cases/, the notebook link in the custom script README is now broken. I'll fix it, and update the citation to v1.1