Planet implemented an improved normalization approach for surface reflectance basemaps that use Sentinel-2 as a normalization target.
Why is this important?
PlanetScope imagery is collected using a constellation of different satellites—including different satellite generations—that have different relative spectral responses (RSRs). The different RSRs are due to the different sensors, and the variability in PS surface reflectance (SR) data. SR correction for PS data relies on MODIS estimates of the atmospheric optical thickness, which is collected at a different time of day and at a coarser spatial resolution when compared with PS imagery. This leads to significant variability in PS SR data even for images that have the same RSR. For example, same satellite generation and sensor. Therefore, we apply normalization to correct for the variability of SR data leading to more consistent time series.
What did we change?
We processed the entire Sentinel-2 imagery catalog between 2017 and 2023 using the Framework for Operational Radiometric Correction for Environmental monitoring (FORCE) method then created monthly-cadence, multi-year composites by estimating the 30th percentile for each pixel location using only cloud-free data.
How does this improve the product?
- Sentinel-2 normalized basemaps will have better correlation to FORCE making them more consistent and accurate for time series use.
- The improvements were intentionally made to be fully backwards compatible with our current normalization models so existing customer and archive data is still 100% valid and usable.
Are there any possible impacts?
- The new approach works best when there are a high number of observations to build the model. In areas where observations are limited due to clouds or other conditions, the algorithm may produce inaccurate results that manifest as artifacts in the final product. Examples of this are shown in the technical datasheet.
- If you notice this or any other artifacts please file a ticket and the customer support team will address the issue. This is not expected to be widespread so there is minimal risk of impact.