This page documents the application of the Complete Data Fusion (CDF) algorithm to the GOME-2 / AC SAF ozone profile product. It reports the FAIR references of the dataset, a summary of its main characteristics, and the results of the completeness and auto-consistency tests performed on a representative test file.

1. Overview

Provider AC SAF — Atmospheric Composition Satellite Application Facility (EUMETSAT SAF network)
Instrument GOME-2 (Global Ozone Monitoring Experiment–2), nadir scanning
Platforms Metop-A, Metop-B, Metop-C
Product Offline high-resolution vertical ozone profile (OHP)
Measured constituents O3
File format HDF5
CDF status Tested — completeness and auto-consistency tests passed

2. FAIR References

The following references allow the dataset to be uniquely identified, accessed, and cited according to FAIR (Findable, Accessible, Interoperable, Reusable) principles. Verified on 2026-04-11 against the AC SAF and EUMETSAT official portals.

Persistent identifier EO:EUM:DAT:0646 (EUMETSAT product code)
DOI Not assigned by AC SAF for this product (verified 2026-04-11).
Product version v2.20 — operational since 30/04/2024 for Metop-B and Metop-C
Product landing page https://acsaf.org/products/ohp.php
Data access (offline) https://acsaf.org/offline_access.php
EUMETSAT catalogue record https://user.eumetsat.int/catalogue/EO:EUM:DAT:0646
Algorithm Theoretical Basis Document (ATBD) ATBD NHP/OHP/O3Tropo — Nov 2022
Product User Manual (PUM) PUM NHP/OHP/O3Tropo — Nov 2022
License / access conditions Free of charge. Data are owned by EUMETSAT and distributed under the AC SAF data policy.
Required acknowledgement (scientific publications) “We thank the AC SAF project of the EUMETSAT for providing data and/or products used in this paper.”

3. Dataset Description

The AC SAF (Atmospheric Composition Satellite Application Facility) is a project under the EUMETSAT Satellite Application Facility network. It focuses on processing, archiving, and disseminating satellite data related to the Earth’s atmospheric composition. AC SAF consortium members develop radiative transfer calculation methods and other algorithms for creating atmospheric remote sensing data from the polar-orbiting satellites Metop-A, Metop-B and Metop-C.

The analysed dataset is the offline high-resolution vertical ozone profile product, which provides, from the GOME-2 nadir scanning mode:

  • The retrieved ozone profile
  • The full error covariance matrix
  • The retrieval noise covariance matrix
  • The a priori profile
  • The averaging kernels
  • Retrieved auxiliary parameters (e.g. surface albedo, cloud albedo)

The availability of the dataset depends on the platform:

Platform Temporal coverage
Metop-A 23/01/2007 – 15/11/2021
Metop-B 12/12/2012 – ongoing
Metop-C 14/09/2020 – ongoing

The nominal spatial resolution is 40 × 80 km2, except for Metop-A between July 2013 and November 2021, when it was 40 × 40 km2.

4. State Vector

The state vector is composed by 42 elements:

  • Partial columns profile of O3 in Dobson Units — 40 elements
  • Albedo — 1 element
  • Calibration error — 1 element

For the specific product selected as reference, the ratio between the largest and the smallest element of the state vector is approximately 1.707 × 105, indicating a wide dynamic range that is relevant for numerical aspects of the CDF computation.

Vertical grid of the GOME-2/AC-SAF ozone profile product under test
Figure 1. Vertical grid of the product under test.

5. Test File

The test file used for the CDF tests on this dataset is a Metop-B HDF5 file (≈ 250 MB) with name:

S-O3M_GOME_OHP_02_M01_20191226083256Z_20191226092956Z_N_O_20191226162125Z.hdf5

Quality filtering: only products with DATA/QualityProcessing[0,:] == 1 have been considered, resulting in 11 299 valid products. The average distance between each pixel and its nearest neighbour is 39.54 km (across track), with an along-track distance of approximately 80 km — consistent with the nominal 40 × 80 km2 resolution.

Map of all valid products of the GOME-2/AC-SAF test file
Figure 2. Map of all the products of the test file passing the quality filter DATA/QualityProcessing[0,:] == 1.
Zoom on the test-file products with the product under test highlighted
Figure 3. Detail of the test-file products in the neighbourhood of the selected product under test (red circle). The bottom-left arrows are 100 km long.

6. Completeness Test Results

All the completeness tests are passed. In particular, all the quantities needed for the CDF are stored in the HDF5 file without compression.

# Requirement Result
1 Optimal estimation retrieval ✓
2 Full state vector available ✓
3 Vertical profiles in terms of concentrations or partial columns ✓
4 Vertical grid availability ✓
5 Full a priori state vector availability ✓
6a Averaging kernel availability (full state vector) ✓
6b Total errors covariance matrix availability ✓
6c A priori covariance matrix availability ✓

7. Auto-consistency Test Results

The auto-consistency test has been applied to the product under test in both the CDF(2022) and CDF(2015) formulations. The results are summarized in the table below and represented in Figure 4 for CDF(2022). In the CDF(2015) formulation, the pseudo-inversion of mathbf{S}_{ni} considered only the eigenvalues larger than fifteen orders of magnitude below the largest one.

# GOME-2 Auto-consistency test results max(|Delta x|)% (|Deltamathrm{DOFs}|)%
1 CDF(2022) total error VCM inversion 0.002504 0.000000
2 CDF(2015) noise error VCM inversion 0.456911 0.000029
Detailed results of the auto-consistency test CDF(2022) for GOME-2/AC-SAF ozone
Figure 4. Detailed results of the auto-consistency test in the CDF(2022) formulation. Panels (b) and (c) show that differences between the original state vector and the one reconstructed by CDF are negligible both in absolute and relative terms; panels (d), (e) and (f) show the same for the AKM and the VCM.
Detailed results of the auto-consistency test CDF(2015) for GOME-2/AC-SAF ozone
Figure 5. Detailed results of the auto-consistency test in the CDF(2015) formulation, applied to the same product under test.

8. CDF Variables Mapping

The following table lists the quantities strictly required by the CDF algorithm and their exact path inside the GOME-2/AC-SAF HDF5 file, as implemented by the GOMEhdf5Reader class used in the test scripts.

CDF quantity Symbol Reader attribute HDF5 variable path
Retrieved state vector
(40 O3 partial columns + albedo + calibration error)		 widehat{mathbf{x}} p.o3 DATA/StateRetrieved
A priori state vector mathbf{x}_{a} p.o3_apriori DATA/Apriori
A priori covariance matrix mathbf{S}_{a} p.vcm_a DATA/AprioriErrorCovariance
Total error covariance matrix
(used as mathbf{S}_{n} in the GOME-2 CDF test)		 mathbf{S}_{n} p.vcm_total DATA/ErrorCovarianceTotal
Averaging kernel matrix mathbf{A} p.ak DATA/AveragingKernel
Vertical grid (pressure levels) — p.pressure DATA/OutputPressureGrid

Note: the HDF5 file also provides DATA/ErrorCovarianceNoise, which could be used as mathbf{S}_{n}. The current GOME-2 test script uses DATA/ErrorCovarianceTotal instead, consistent with the treatment of the total retrieval error as the effective noise term in the auto-consistency test.

9. Notes and Open Issues

  • The wide dynamic range of the state vector (≈ 105) requires careful handling of matrix inversions in the CDF equations — see the Extensions page for a discussion of numerical aspects.
  • All characterisation matrices (AKM, noise VCM, a priori VCM) are stored uncompressed in the HDF5 file and are directly usable by the CDF code without pre-processing.

Back to: Tested Datasets overview.