Seeking Guidance on Processing Methodologies for Sentinel-2-C1-L2A Collection Dataset

[jaiindimple]

Overview

As outlined in the readme, the Sentinel-2 Collection 1 L2A (sentinel-2-c1-l2a) is designed to eventually replace the existing Sentinel-2 L2A collection (sentinel-2-l2a). Items in this new collection include assets referring to COGs of data processed to at least baseline 5.0. After conducting a comparative analysis of both collections, I am seeking to understand the specific processing methodologies that have been applied to each.

Current Understanding and Observations

• The Sentinel-2-L2A collection the offset has been applied to some of the Items.
• The Sentinel-2-C1-L2A collection, intended to replace the Sentinel-2-L2A collection, does not apply the offsets but includes scale offsets in the raster metadata. This feature has proven beneficial when working with QGIS and Titiler.

Comparative Analysis of Both Collections

To better understand the impact of these processing differences, I conducted a comparative analysis using the same image from both collections on the same date. Here are the results:

2020 Dataset

• AWS Sentinel-2-C1-L2A: 7820
• AWS Sentinel-2-L2A: 4520

2024 Dataset

• AWS Sentinel-2-C1-L2A: 1752
• AWS Sentinel-2-L2A: 752

The analysis indicates that despite post-baseline changes, the datasets are consistent across all collections except in the application of offsets. Notably, the 2020 datasets show significant discrepancies in pixel values.

Request for Information and Guidance

• Could you provide details on the specific processing methodologies applied to each collection?
• What solutions are available or recommended for standardizing these values to ensure consistency across datasets?

Attachment

Link to the Jupyter notebook and dataset for reconducting the analysis.

I appreciate any guidance or recommendations you can provide to help make an informed decision.