Level 2 reflective atmospheric algorithms
This document describes the algorithms used to produce a Level 2 product from an Earth observation sensor in FarEarth.
These algorithms apply to both reflective (visible to near-infrared) and emissive (thermal-infrared) imagery. This document is intended as a summary of the algorithms used.
The following terminology is specific to FarEarth
| Term | Description |
|---|---|
| Radiometric Parameter File (RPF) | File containing information on the radiometric sensor properties, band spectral ranges, and the radiometric calibration coefficients |
| Detectors vs pixels | "Detector" refers to the physical single element of a sensor array. In contrast, "pixel" is used when referring to a subgroup of detectors, which can be achieved by binning and/or time delay integration (TDI) |
Level 2 algorithms apply atmospheric corrections to top-of-atmosphere (TOA) Level 1 products. This generates bottom-of-atmosphere (BOA) Level 2 products that provide information about properties of the Earth's surface. Level 2 products are essential for remote sensing applications due to the significant effect the atmosphere has on Earth observation measurements.
The following are generated during image processing
- cloud masks
- surface classification masks
- atmospheric conditions
For information about Level 2 product generation of emissive bands, see Level 2 emissive bands.
Reflective bands
Level 2 products generated from reflective bands are surface reflectance. The algorithm for the generation of a Level 2 surface reflectance product includes
- extraction of Level 1C (L1C) reflective bands and metadata
- cloud mask evaluation and determination of suitability for Level 2 (L2) processing
- extraction of atmospheric, surface, and satellite conditions
- application of atmospheric correction using a radiative transfer model
- generation of a Level 2 surface reflectance product
L1C reflective bands and metadata
The first step is to extract the reflective bands and metadata from the Level 1C input products for L1-to-L2 processor.
Cloud masking
The Level 1C (L1C) reflective bands are used as input to a cloud detection algorithm, which determines which pixels are likely cloudy. Given a certain cloud percentage threshold, the scene is either marked as suitable for Level 2 (L2) processing or not. For example, if a 50% cloud-free threshold is set, but 60% of the image is marked as likely cloudy, the L2 processing will not continue.
Atmospheric, surface, and satellite conditions
To determine the atmospheric conditions at the time of acquisition, atmospheric measurements are retrieved from ancillary databases (for example, VIIRS). If a satellite has sensor bands such as coastal aerosol or water vapour, these can be used to improve atmospheric condition estimates. The aerosol optical depth (AOD) can be estimated using an aerosol detection algorithm if the required spectral bands are available.
Surface conditions and satellite configuration conditions, such as altitude, satellite viewing angle, and solar zenith angle, are extracted from the L1C product metadata.
There is often a latency of a few hours to a few days for the ancillary atmospheric condition databases. Historical values for the area of interest can be used to generate an initial estimate of a Level 2 product, which can later be reprocessed when more accurate data become available.
Atmospheric correction
The atmospheric, surface, and satellite conditions are used as inputs to a radiative transfer model (RTM) to determine the atmospheric correction required to convert the L1C pixel data to surface reflectance.
Product generation
After converting the Level 1 pixel data to surface reflectance, it is saved as a Level 2 surface reflectance product, including its corresponding metadata and quality information. This may include cloud masks, ancillary atmospheric conditions, and other quality metrics.
Validation
During validation, the generated surface data of the satellite image is compared to a reference surface image. The reference image should be a cloud-free image that is temporally close to the satellite image (for example, within a few hours) and from a reference satellite (for example, Sentinel L2A or Landsat L2SR products). The comparison is more suitable for satellite sensors that have bands with spectral responses matching those of the reference satellite, but can also be used for other sensors to monitor relative performance over time.
Performance can also be evaluated over time by comparing the results from acquisitions over reference sites that provide reliable data.
Metrological traceability
Satellites with an absolute radiometric calibration against a radiometric reference site are traceable. The traceability is through the calibration of ground instruments and validation of the radiative transfer models, using the International System of Units (SI units).
Satellites that have been calibrated to a reference sensor will have metrological traceability through the reference sensor's calibration. A FarEarth In-orbit Calibration Report documents the calibration process.
Uncertainty characterisation
Although a complete uncertainty evaluation is not performed, the validation process indicates the accuracy of the radiometric performance by comparing it to a reference sensor or reference site.
Contributions that influence the radiometric uncertainty include
- uncertainty from the Level 1 product generation
- uncertainty from the atmospheric, surface, and satellite conditions, for example, AOD, water vapour, viewing angle, and interpolation of these values
- uncertainty from the application of the radiative transfer model
- uncertainty from the emissivity estimation and optimisation for surface temperature and surface emissivity