the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Long-term trends in aerosol properties derived from AERONET measurements
Abstract. Over the past two decades, remarkable changes in aerosol compositions have been observed worldwide, especially over developing countries, potentially resulting in considerable changes in aerosol properties. The Aerosol Robotic Network (AERONET) offers high precision measurements of aerosol optical parameters over about 1700 stations globally, many of which have long-term measurements for one or more decades. Here we use AERONET Level 2.0 quality assured measurements to investigate long-term trends for aerosol optical depth (AOD) and Ångström exponent (AE) trends, and quality-controlled Level 1.5 inversion products to analyze trends of absorption aerosol optical depth (AAOD) and single scattering albedo (SSA) at stations with long-term records. We also classify the aerosol properties in these sites into 6 types, and analyze the trends of each type. Results reveal decreases in AOD over the majority of the stations, except for North India and the Arabian Peninsula, where AOD increased. AE also decreased in Europe, eastern North America, and the Middle East, but increased over South Asia and East Asia. The decreased AE over Europe and eastern North America is likely due to decreased fine-mode anthropogenic aerosols, whereas that over the Arabian Peninsula is attributed to increased dust activities. Conversely, increased AE over North India is probably attributed to increased anthropogenic emissions and decreased dust loading. Most stations in Europe, North America, East Asia, and South Asia exhibit negative trends in AAOD, whereas Solar_Village in the Arabian Peninsula has positive trends. SSA at most stations increases and exhibits opposite trends to AAOD, but with several stations in central Europe and North America showing decreased SSA values. Trend analysis of different aerosol types further reveals the changes of different aerosol components that are related to AOD, AE, AAOD, and SSA trends. Stronger reductions in fine-mode absorbing species than that of non-absorbing aerosols are found over Europe and East Asia, whereas in eastern North America the reductions of aerosols are dominated by non-absorbing species. Increased aerosols in Kanpur over North India should be mainly comprised of scattering species, whereas those in Solar_Village over the Arabian Peninsula are mainly dust. Weak seasonality is found in the trends of all aerosol parameters analyzed in this work.
- Preprint
(3673 KB) - Metadata XML
-
Supplement
(5997 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-2533', Anonymous Referee #1, 06 Sep 2024
Please find attached the Comment on egusphere-2024-2533
-
AC1: 'Reply on RC1', Prof Li, 15 Nov 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-2533/egusphere-2024-2533-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Prof Li, 15 Nov 2024
-
RC2: 'Comment on egusphere-2024-2533', Martine Collaud Coen, 19 Sep 2024
Review of « Long-term trends in aerosol properties derived from AERONET measurements” by Zhang et al.
This paper described the long-term trends of several compounds from the AERONET network that extends over 5 continents. The trends of the measured AOD for at least 8 years in the 2000-2022 period are first considered and discussed. Trends of the AOD Ångström exponent as well as the SSA and AAOD inversion products. Seasonality of the trends are also mentioned. A further analysis classified the aerosol into 6 different types (dust, mixture, and aerosol fine fraction with various absorbing capacities) as a function of the fine mode fraction and the SSA.
This long-term trend analysis of AOD and related parameters over the very dense AERONET network provide very interesting results on changes in aerosol concentration and composition. Several methodological points have however to be better described and, if necessary, improved.
General comments:
- Methodology for trend analysis:
The authors correctly chose Mann-Kendall test associated to the Sen’s slope, which are both non-parametric methods. The Mann-Kendall (MK) test giving the statistically significance (ss) is however not described in the methodology section. The following points have to be clarified:
- MK test has to be applied on serially independent data. This means that MK test without prewhitening can only be applied on time series without ss auto-correlation. In case of ss auto-correlation, prewhitening methods have to be applied. In this study, no mention of auto-correlation is found. I then require from the authors that either to report no ss auto-correlation in all the time series or to use a prewhitening method to minimize the artifacts bounded to serially dependent data. Since the authors cite Collaud Coen et al., ACP, 2020, they should also be aware of the companion paper Collaud Coen et al., AMT, 2020 (https://amt.copernicus.org/articles/13/6945/2020/ ) on MK methodology and the associated github repository ((https://github.com/mannkendall) giving access to a complete MK and Sen’s slope routines with prewhitening methods in R, Python and Matlab.
- MK test also requires a homogeneous distribution, namely no seasonal cycles. The presence of seasonality in the used time series is clearly visible (e.g. Fig. 4a, b, c, d, f, Fig. 6 b, c, d, e, g and h, Fig. 10 d, e, f, Fig. 12c). Figs. 7, 8 13 and 14 clearly present the trend results for meteorological seasons. The methodology is however not described so that it is not clear if the homogeneity test between season is performed or not. The paper however describes different trends directions for different seasons (e.g. L 200s). This has to be clarified
- Concerning seasonality, the climate specificities has also to be taken into account. The applied seasons correspond to mid-latitude climate but not e.g. to lands with a monsoon seasonality. The different seasonality has to be taken into account in the analysis.
- As specified in Collaud Coen et al., AMT, 2020, the use of a lower time granularity (e.g. daily) than month could also help to increase MK test’s power
- Finally, confidence limits can also be computed and help the interpretation of the results.
2. Homogeneity of the time series
Long-term trend analysis can only be performed on homogeneous time series. The authors reported the case of Birdsville, where false results were reported due to false data filtering. Which procedure was applied to check the homogeneity of the time series ? I do really appreciate to have all time series in supplement. It’s worth to have a look if we are interested at one particular station.
Generally, I would really have a look at all time series and remove too high or too low values (e.g. SSA below 0.6), to see if too few data are present in the first of end years so that the time series should be shortened and if there is evident ruptures.
Here some comments on the time series:
- Ames AOD: global decrease but an increase in maxima: to check
- Amsterdam: strange high values in 201072011 and 2014
- Anmyon and Arica AOD: is there a rupture due to the long missing period ?
- Bozeman: are high AOD in 2017 and 2021 due to e.g. biomass burning ?
- CabauwAOD: I would not consider the 2 data in 2003
- Canberra: I would not take the too high 1-3 first data
- Cartel: increasing until 2006 and decreasing after a missing period in 2008-2022: to check
- Ceilap: value > 0.15 in 2012 is doubtful
- Chen-Kung: AOD: I would not use the first two months in 2002, even if MK accept missing data, having a full first and end years remains important. AE: idem
- Davos AOD: I would not take the 2001 Data
- Egbert: do you have an explanation for the high maxima after 2014?
- Fort-McMurry: I would not take 2005 data
- Hamburg: AOD I would only use 2003-2016 since there is few data otherwise
- Morin: strange AOD>4 in 2003
- Issyk : AOD seems very high in 2021
- Shiraham: I would stop in 2016
- Osaka: AOD: I would not use the first two months. The maxima are in 2000-2007 are much higher than thereafter. Is there a change in 2006-2007? AAOD: the very high data (> 0.1) should probably be invalidated and the low data end of 2017 to mid-2019 are also strange. SSA : similar comment as for AAOD (but inverse dependence)
- Solar village: AOD: seems ok, SSA: the mid 2000-2002 data seems strange and too high and max in 2010 as well as min in 2012 should be checked.
- Gandhi college: the max at 2.5 is very strange and should be analysed. The four last months are also much higher after a missing period. Is there a rupture in the time series?
- Carpentras AOD: the first ~6 months are much higher. AAOD: the 2002-2005 data seems too high and a rupture in the time series in the missing period (2005-2006) is probable. SSA: the three high values in 2006 should be checked
- Mexico city: AOD: the three low data in jan-feb 2010 are strange.
- Missoula AE: I would not use the data before 2004
- Beijing: AAOD and SSA: I would not use the first isolated 2-3 months
- GSFC: AAOD: the low minima in 2010 and 2011 should perhaps be investigated
- MD Science Center SSA: the first high value until 2002 and the high values in2016-2028 should be checked as well as the very low values in 2019.
- Concerning the AAOD, I just looked at some station: Lille has too few data before 2007, Rome data are really too low in 2012, there is a problem, White Sands: the increase after 2019 is so rapid that it is doubtful
- Concerning SSA. I have the impression that SSA time series are the more uncertain. For example, SSA at Granada is not homogeneous at all, 2012-2026 are too low. There si various station with very low SSA (e.g. 0.5 at IMAA-Potenza, 0.3 at OHP) that should be removed before the trend analysis. I also have the impression that there is rupture in some dataset such as Tucson, Trelew (in 2017?), Toulon in mid-2006, Palencia in 2007-2008.
Please have a look at all time series to improve their relevance for long-term trend analysis.
3. Results reported in a map:
The representation in a map is very useful to have an overview of the trends around the world. I have however some remarks:
- the very small trends (e.g. with AOD slopes in [-0.02, 0.02] (Fig. 3)) are in white but still sometimes ss. Since no table with all results are given, it’s not easy to know if the trend are positive or negative. Moreover it means that not ss trend does not appears on the map since there is no dark circle.
- The presented results for all parameters does not correspond neither to the same time period nor to the same length (e.g. AOD at GSFC corresponds to the 23 y trend ending in mid-2022, whereas result from Ghandi-College correspond to 17 y trend ending in 2021 and result from Solar in 13 y results ending in 2013) (+ Fig. 1). My opinion is that trends with up to 10 y differences for the end point or with large differences in the length of the time period should not be represented in a similar way in the same figure. For example, the high positive AOD trend for Solar Village cannot be compare with the Ghandi or Kampur trends since there is almost one decade difference of the end time.
4. Results with low AOD value and consequently larger uncertainties:
As well explained in the manuscript, low AOD values leads to high uncertainties for the derived parameters. I think that the trends with high uncertainties should appears differently in the map. I don’t know what is the best solution. Perhaps by representing only trends with 95% confidence level and different size as a function of the uncertainty ?
5. Data used
It is not easy to understand which data are used. AERONET Solar Level 2 and AERONET almucantar Level 1.5 data are both used, the 1.5 ones for the inversion products. L. 87-88 says that L 1.5 are similar to L 2.0 but for the AOD threshold ? meaning that no AOD threshold are used ? It would be very helpful to have a more precise description with eventually the mention of the level in the figures’ captions.
Minor comments:
- Are all the average done with median? Are first daily medians computed and then monthly medians or is the monthly medians computed from hourly data ?
- L1: there is changes in aerosol composition but also in their concentration.
- L 10: I would specify that AE correspond to the wavelength dependence of AOD, since AAOD and SSA also depend on the wavelength.
- L17-19: long sentence, please rephrase.
- L 34: “which mainly located in …”: please check the language
- L35: It is not possible to consider SSA as representative of the scattering. Please rephrase
- L84-85: Considerations on the uncertainties of the various parameters are explained at various places in the manuscript. Please sample them at the same place so that the reader can have a direct overview.
- L 100 and Figs 1 and 2: Figs 1a and b could perhaps be merged with different color for Level 2 and 1.5? A map (perhaps divided into continents) with all stations’name could appears in the supplement and/or a table with the stations’coordinates.
- L102: does the AE corresponds to a fit including all the wavelengths between 440 and 870 nm?
- Eck 1999
- L 123: what do you mean by “all-point”?
- Table 1 and L 121: Why Uncertain is not called sea salt ?
- L125: it means that the trend results for the various aerosol types are computed from time series with three time less data points due to the seasonal median? How is the seasons defined for monsoon climate ?
- L130-131: This is not the right causality: negative AOD trends demonstrate the global reduction of aerosol loading.
- L135: Higher slope in Li et al. 2014 can also be due to the shortest time series leading to larger slopes due to a much lower number of data.
- L139-140: In this case, it is important to know the length and end year of the time series. Do the larger slopes correspond to the shorter time series ? or to earlier end year ?
- L141-144: please rephrase
- L 147-148: does both time series have the same end year ?
- L 150 and L161-162. The special case of Birdsville should be reported only once in the paper.
- L159-160: are all these trends ss ?
- L176: which time series and seasons are less robust due to low AOD ? A map with AOD values (or seasonal AOD) could perhaps help
- L179: From the map I see 2/4 stations in western North America have positive AE trends.
- L198-199: I have the impression that no ss AE trends is just an indicator of no modification of the size distribution. Is it right ?
- L200-201: As mentioned in the general comments, is the homogeneity between the seasonal trends computed ?
- L204-205: Are AOD higher in spring and lower in winter for all stations in the Northern Hemisphere? Here too a map of AOD for the various seasons could help.
- L 229: please rephrase: AAOD does not characterizes the scattering.
- L234-239: this should be discussed in the method/data section.
- L244: increases in either the concentration of absorbing aerosol or in the composition (higher imaginary part of the refractive index)
- L262: absorbing (b missing)
- L271-272: Is there not change in BC or BrC concentrations in middle East ?
- L 310: I have the impression that, e.g. SSA and AE in western North America, AOD in India or AAOD in Africa have different seasonal trends (Fig. 14).
Citation: https://doi.org/10.5194/egusphere-2024-2533-RC2 -
AC2: 'Reply on RC2', Prof Li, 15 Nov 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-2533/egusphere-2024-2533-AC2-supplement.pdf
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
418 | 100 | 36 | 554 | 24 | 6 | 9 |
- HTML: 418
- PDF: 100
- XML: 36
- Total: 554
- Supplement: 24
- BibTeX: 6
- EndNote: 9
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1