Application of spectral analysis techniques to the intercomparison of aerosol data – Part 4: Synthesized analysis of multisensor satellite and ground-based AOD measurements using combined maximum covariance analysis

Abstract

Abstract. In this paper, we introduce the usage of a newly developed spectral decomposition technique -combined maximum covariance analysis (CMCA) -in the spatiotemporal comparison of four satellite data sets and groundbased observations of aerosol optical depth (AOD). This technique is based on commonly used principal component analysis (PCA) and maximum covariance analysis (MCA). By decomposing the cross-covariance matrix between the joint satellite data field and Aerosol Robotic Network (AERONET) station data, both parallel comparison across different satellite data sets and the evaluation of the satellite data against the AERONET measurements are simultaneously realized. We show that this new method not only confirms the seasonal and interannual variability of aerosol optical depth, aerosol-source regions and events represented by different satellite data sets, but also identifies the strengths and weaknesses of each data set in capturing the variability associated with sources, events or aerosol types. Furthermore, by examining the spread of the spatial modes of different satellite fields, regions with the largest uncertainties in aerosol observation are identified. We also present two regional case studies that respectively demonstrate the capability of the CMCA technique in assessing the representation of an extreme event in different data sets, and in evaluating the performance of different data sets on seasonal and interannual timescales. Global results indicate that different data sets agree qualitatively for major aerosol-source regions. Discrepancies are mostly found over the Sahel, India, eastern and southeastern Asia. Results for eastern Europe suggest that the intense wildfire event in Russia during summer 2010 was less well-represented by SeaWiFS (Sea-viewing Wide Field-of-view Sensor) and OMI (Ozone Monitoring Instrument), which might be due to misclassification of smoke plumes as clouds. Analysis for the Indian subcontinent shows that here SeaWiFS agrees best with AERONET in terms of seasonality for both the Gangetic Basin and southern India, while on interannual timescales it has the poorest agreement.

Merging regional and global AOD records from 15 available satellite products

“…Ground-based measurements such as from the AERONET (cloud screened and quality assured Version 3 Level 2.0, Giles et al, 2019) provide highly accurate measures of AOD that are widely used as ground truth for the validation of L2 satellite AOD data. Although extensive L2 AOD validation has been performed for different aerosol products, only a few attempts have been made to evaluate AOD monthly aggregates retrieved from satellites (e.g., Li et al, 2014b, Wei et al, 2018.…”

“…Retrieval assumptions may work well in certain conditions, but not globally. Thus, regional differences in the consistency between AOD products exist (Li et al, 2014b). An important factor behind the differences could be related to the strictness of cloud masking, affecting which pixels are processed by retrieval algorithms, propagating into differing levels of clear-sky bias in daily and monthly aggregates (Sogacheva et al, 2017;Zhao et al, 2013;Li et al, 2009).…”

“…However, different techniques applied to reveal the spatial and temporal differences between AOD monthly products, e.g. principal component analysis (Li et al, 2013;Li et al, 2014b) or maximum covariance analysis (Li et al, 2014a, b), show that there are key similarities among the AOD products tested, including MODIS, MISR and SeaWiFS.…”

Dataset Reduction Techniques to Speed Up SVD Analyses on Big Geo-Datasets

“…A combination of CCA with Minimum Noise Fraction (MNF) has proven successful at filtering out noise from images [9]. Finally, the SVD has been used to compare ground data and measurements done by Earth observation satellites [10,11]. This is because SVD offers an effective method to find frequent and simultaneous patterns between two spatio-temporal fields [12,13].…”

Reducing multisensor satellite monthly mean aerosol optical depth uncertainty: 1. Objective assessment of current AERONET locations

“…Even in the monthly mean AOD, systematic differences due to instrument viewing geometry and sampling strategies, cloud screening, surface parameterization, and retrieval assumptions can still be significant. Previous comparison studies have shown noticeable differences in the monthly mean AOD products among various satellite sensors and ground‐based measurements [e.g., Liu et al, ; Li et al, ; Popp et al, ]. Even for coincident satellite measurements such as those from Terra Moderate Resolution Imaging Spectroradiometer (MODIS) and Multiangle Imaging Spectroradiometer (MISR), which are exempt from differences attributable to sampling issues, monthly mean AODs still do not always agree [e.g., Kahn et al, , ; Shi et al, ].…”