27 January 2023
16 May 2022
The IASI original radiance (OR) and Principal Component Compression (PCC) products are distributed operationally by EUMETSAT and the latter are ingested into the Level 2 (L2) product processing. The significant reduction in data volume is one of the main motivations for the PCC. The other advantage is that it leads to a reduction of noise in the reconstructed radiances (RR). However, a concern with the RR spectra is the potential loss of infrequent spectral signatures which might not be adequately represented by the PC basis, in particular for Atmospheric Composition (AC) and Air Quality (AQ) applications.
The main aim of this study is to provide a comprehensive comparison between IASI original radiances and PCC reconstructed radiances and derived products for AC and AQ. Unlike for IASI, principal components compression will be the only representation of MTG-IRS observations for near-real dissemination. This study is hence intended to increase user preparedness with PC products from hyperspectral sounding and to inform the debate with actual results from using RR for atmospheric composition purposes.
Objectives
Two alternative PCC methods have been investigated:
- The standard PCA approach based on static eigenvectors, also referred to as PC-global
- The PC-hybrid approach which uses a combination of global and local eigenvectors.
The results of two parallel studies by ULB and LATMOS on one side (ULB/LATMOS study), and by SPASCIA and UniBas on the other side (SPA/UniBas study), are reported here.
The main objectives of these studies were to:
- Evaluate how well the IASI Level 1C can be reconstructed from the eigenvectors produced at EUMETSAT, and to document the added value of PC-hybrid over PC-global for rare or extreme situations.
- Compare the performance of atmospheric composition products retrieved from original and reconstructed radiances obtained from PC-global and PC-hybrid methods, applied to background, extreme or rare AC/AQ observations.
- Evaluate the performance against independent reference data and with self-consistency analysis.
- Establish a catalogue of rare but important spectral signatures, which are not well represented by the leading eigenvectors of the big global training set, to test the possibility of their inclusion as supplementary directions in the global eigenvector basis.
Overview
The selection of study cases has been based on datasets and experience from the two study teams. This covers:
- ULB/LATMOS : O3, CO, HNO3, NH3 in standard situations, O3 profile retrievals and comparison with sondes, CO retrievals in severe fire conditions, SO2 in volcanic eruptions, and rare features (e.g dust minerals, emissivity effects)
- SPA/UniBas : SO2, HCl, HCN, HNO3, OCS, H2S from volcanic situations (including Hawaii region), CO, C2H2, C2H4, HCOOH, OCS, HCN, HCl from wild fires (including 2017 Mediterranean fires and 2019 Australian fires), SO2, OCS, NH3, HNO3 from a selection of IASI measurements for polluted situations over Europe and China, and rare features specifically identified by the use of the IASI-PCA detection tool dedicated to the detection of extreme atmospheric events from the analysis of reconstruction outliers.
The analysis has been performed on Level 1 IASI measurements showing anomalies (outliers) in the reconstructed radiances based on statistical analysis of the spectral residuals OR-RR.
The analysis of Level 2 retrievals encompassed long time-series of IASI soundings in order to check the capability of reconstructed radiances to provide retrieval results similar or comparable to that based on original radiances, as far as the detection and retrieval of the seasonal cycle of a given gas are concerned. In addition, large spatial 2-D IASI sounding fields have also been studied to get insight into the capability of RR to recover and reproduce patterns over significantly polluted regions.
Selected results are presented and commented in Figures 1, 2, 3, 4, 5 and 6.
Summary of results/conclusion
For Level 2 retrievals, we cannot see any problematic loss of sensitivity and information content in Level 2 retrieved products with RR compared to OR. RR is improving the retrieval convergence rate because of reduced noise, which is important for retrieval over land surfaces with complex emissivity. RR is seemingly slightly improving the retrieval of trace gases; this has been shown especially in the case of NH3.
The level 1 analysis of outliers in reconstructed radiance also showed that the hybrid PC approach is generally well-suited to atmospheric chemistry applications in an operational framework. For most of the studied cases, the hybrid RR corrects the global RR limitations. However, in a few specific cases, there is room for improvement of the hybrid approach when too large residuals remain, in particular for:
- Some large pollution episodes.
- Individual outliers.
For the very specific and rare cases where RR failed to capture significant signatures of trace gases in the IASI spectra, these signatures have been identified in the measurements and delivered to EUMETSAT through a dedicated catalogue.
At the end of this study, an improved hybrid method was successfully tested, where most of those outlying spectra are now reconstructed correctly.
In addition, recommendations have been provided for the monitoring of reconstructed radiances in an operational framework. Dedicated, spectrally-focused reconstruction scores (associated to spectral structures of geophysical interest) have been extensively used for the detection and the analysis of geophysical events leading to potential difficulties in the compression/reconstruction process. The use of such dedicated reconstruction scores should be studied as quality criteria for flagging the cases where EUMETSAT-PCC compression process potentially fails to capture spectral structure of geophysical interest.
