the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Vertical Retrieval of AOD using SEVIRI data, Case Study: European Continent
Abstract. Accurately determining Aerosol Optical Depth (AOD) across various altitudes with sufficient spatial and temporal resolution is crucial for effective aerosol monitoring, given the significant variations over time and space. While ground-based observations provide detailed vertical profiles, satellite data are crucial for addressing spatial and temporal gaps. This study utilizes profiles from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) and data from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) to estimate vertical AOD values at 1.5, 3, 5, and 10 km layers. These estimations are achieved with spatial and temporal resolutions of 3 km × 3 km and 15 minutes, respectively, over Europe. We employed machine learning models—XGBoost (XGB) and Random Forest (RF)—trained on SEVIRI data from 2017 to 2019 for the estimations. Validation using CALIOP AOD retrievals in 2020 confirmed the reliability of our findings, emphasizing the importance of wind speed (Ws) and wind direction (Wd) in improving AOD estimation accuracy. A comparison between seasonal and annual models revealed slight variations in accuracy, leading to the selection of annual models as the preferred approach for estimating SEVIRI AOD profiles. Among the annual models, the RF model demonstrated superior performance over the XGB model at higher layers, yielding more reliable AOD estimations. Further validation using data from EARLINET stations across Europe in 2020 indicated that the XGB model achieved better agreement with EARLINET AOD profiles, with R2 values of 0.81, 0.77, 0.71, and 0.56, and RMSE values of 0.03, 0.01, 0.02, and 0.005, respectively.
- Preprint
(2024 KB) - Metadata XML
-
Supplement
(1726 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on amt-2024-105', Anonymous Referee #1, 08 Oct 2024
This is a review for “Vertical Retrieval of AOD using SEVIRI data, Case Study: European Continent” by Pashayi et al. In this paper, AOD is determined at several altitudes from SEVIRI/MSG data using machine learning (ML) methods trained on vertical aerosol profiles from the CALIOP lidar satellite. Validation is performed using CALIOP and independent EARLINET ground lidar data. The article is well written and generally clear. However, I find the scientific interest of the study to be very limited, at least in the way findings are presented. Indeed, results are limited to a large number of scores calculated by validating the vertical variation of the AOD retrieved at certain single pixel locations, thus showing no maps or vertical profiles estimated by the presented methodology. Furthermore, the temporal variation (daily and diurnal) of the retrieved aerosol variables is not discussed, nor shown, in the paper. I find this highly surprising, mainly because the authors state in the abstract that "These estimations are achieved with spatial and temporal resolutions of 3 km × 3 km and 15 minutes, respectively, over Europe”. In my opinion, this statement needs to be proven in order to properly evaluate the contribution of this paper, as SEVIRI's main asset is that it is a high temporal resolution imager that provides images of the entire disk of the Earth every 15 minutes. I suggest including examples of maps and time series of vertical AOD variations across space and time, as well as their validation with reference data, as this is in my eyes the main interest of the proposed methodology for the scientific community. In addition, I strongly suggest going beyond the statistical analysis, which is extensively presented in this paper, to discuss more the physical interpretation of the results. This applies not only to the results in general, but also to other parts of the paper, such as the analysis of the importance of different variables in vertical AOD retrieval. To avoid writing a lengthy paper, some details about the methodology can be omitted (for example, I would only consider the RF method if it is the best of the two).
Citation: https://doi.org/10.5194/amt-2024-105-RC1 -
RC2: 'Comment on amt-2024-105', Anonymous Referee #2, 22 Oct 2024
This paper presents an awkward title, "Vertical Retrieval of AOD...". There seems to be a misunderstanding regarding the term "vertical retrieval." AOD (Aerosol Optical Depth) is generally understood as a columnar quantity, representing the total extinction of light from the surface to the top of the atmosphere. The concept of "vertical retrieval" is unclear. Upon reading the paper, it appears that the authors are referring to the retrieval of AOD in layers at specific altitudes, with each layer having a thickness of 60 m, which corresponds to the vertical resolution of CALIOP data. If this is the case, a more appropriate title would be something like "Retrieval of Extinction at Four Layers Using...". From what I gather, the paper does not conduct a true retrieval of the aerosol extinction profile, as provided by CALIOP. If they are only retrieving AOD for a few layers, it raises the question—what is the purpose of this retrieval? Who will use such data? most climate models have vertical resolution coarser than 500m. so, layered AOD at 60 m at a few altitudes seems not useful.
Due to significant flaws in the methodology and validation, I recommend rejecting this paper. Below are my major comments:
-
Introduction: The introduction fails to acknowledge recent advancements in aerosol layer height retrieval from instruments like EPIC and TROPOMI. I recommend referencing recent literature, such as https://doi.org/10.1016/j.rse.2021.112674 and the references therein.
-
Methodology: The method has significant flaws, particularly in the use of machine learning as a "black box." The authors fail to explain the underlying physics of how SEVIRI would contain information about the aerosol vertical distribution—specifically, at what wavelengths and why? The method pairs CALIOP layered AOD with meteorological data and SEVIRI radiance for training but lacks a clear justification for this approach.
-
Cross-validation: The cross-validation approach is problematic. The data used for training and validation likely have significant auto-correlation in space or time. I suggest separating the datasets by year, for example, using the first two years of data for training and the third year for validation.
-
Validation and Results: The paper should provide a map of retrieved AOD at each hour and validate these results with AERONET sites. Additionally, the paper should present an extinction profile from their retrieval and compare it with either ground-based lidar measurements or other aerosol profile measurements, such as those from aircraft.
Citation: https://doi.org/10.5194/amt-2024-105-RC2 -
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
280 | 69 | 12 | 361 | 44 | 6 | 4 |
- HTML: 280
- PDF: 69
- XML: 12
- Total: 361
- Supplement: 44
- BibTeX: 6
- EndNote: 4
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1