the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Validation of Aeolus L2B products over the tropical Atlantic using radiosondes
Peter Knippertz
Martin Weissmann
Benjamin Witschas
Cyrille Flamant
Rosimar Rios-Berrios
Peter Veals
Abstract. Since its launch by the European Space Agency in 2018, the Aeolus satellite has been using the first Doppler wind lidar in space to acquire three-dimensional atmospheric wind profiles around the globe. Especially in the tropics, these measurements compensate for the currently limited number of other wind observations, making an assessment of the quality of Aeolus wind products in this region crucial for numerical weather prediction. To evaluate the quality of the Aeolus L2B wind products across the tropical Atlantic Ocean, 20 radiosondes corresponding to Aeolus overpasses were launched from the islands of Sal, Saint Croix and Puerto Rico during August–September 2021 as part of the Joint Aeolus Tropical Atlantic Campaign. During this period, Aeolus sampled winds within a complex environment with a variety of cloud types in the vicinity of the Inter-tropical Convergence Zone and aerosol particles from Saharan dust outbreaks. On average, the validation for Aeolus Raleigh-clear revealed a random error of 3.8–4.3 ms−1 between 2–16 km and 4.3–4.8 ms−1 between 16–20 km, with a systematic error of −0.5 ± 0.2 ms−1. For Mie-cloudy, the random error between 2–16 km is 1.1–2.3 ms−1 and the systematic error is -0.9 ± 0.3 ms−1. Below clouds or within dust layers, the quality of Rayleigh-clear measurements can be degraded when the useful signal is reduced. In these conditions, we also noticed an underestimation of the L2B estimated error. Gross outliers which we define with large deviations from the radiosonde but low error estimates account for less than 5 % of the data. These outliers appear at all altitudes and under all environmental conditions; however, their root-cause remains unknown. Finally, we confirm the presence of an orbital-dependent bias of up to 2.5 ms−1 observed with both radiosondes and European Centre for Medium-Range Weather Forecasts model equivalents. The results of this study contribute to a better characterization of the Aeolus wind product in different atmospheric conditions and provide valuable information for further improvement of the wind retrieval algorithm.
Maurus Borne et al.
Status: open (until 08 Jul 2023)
-
RC1: 'Comment on egusphere-2023-742', Anonymous Referee #1, 02 Jun 2023
reply
The authors present statistics of the validation of Aeolus winds against independent ECMWF model fields and radiosondes. This is important work to gain knowledge on the errors of Aeolus winds. The region used for validation is limited to the tropics, which on the other hand is a very interesting region because of the challenging weather conditions with dust events and convective clouds and because of limited other Aeolus related Cal/Val campaigns in this region.
Major comments
==============
G1) At many places in the paper, the authors compare MADI against EEtot. This is a fundamental mistake as MADI is not a metric related to standard deviation such as EEtot (and SMAD). The authors can confirm this by taking a sample of random numbers, with normal (Gaussian) distribution, and compare the MADI value with the input standard deviation value.G2) The authors should be more strong on their main conclusion in the abstract, by ending e.g. with: "Based on the data used in this study Aeolus Rayleigh winds do not meet the mission random error requirement and Mie winds do most likely not meet the mission bias requirement."
G3) The classes discussed in lines 171 to 174 are very unclear. For instance, what is meant with "below 3 km (very high, high, mid-level, low, very low and fractional cloud types)"? How can you have very high clouds below 3 km?
Also, why not using the useful signal at measurement level to identify clouds within the profile?G4) line 195. Did you check this statement, e.g., using spectra following Skamarock (2008). They show that the area below the kinetic energy spectrum (which is actually the atmospheric variability over the integrated scales) can be quite substantial when starting at 340 km (or truncation wavenumber 60).
General comments
================
line 11; measurements -> observations (Note that for Aeolus an observation is the result of accumulated measurements; mixing these terms in the text is confusing. Please correct everywhere in the text accordingly)
line 15; the orbital-dependent bias of up to 2.5 m/s applies to only some parts of the atmosphere. This nuance should be made here.
line 33; "..... along the LOS of the instrument, which is directed perpendicular to the direction of satellite propagation. Please add the last part.
line 54: Replace: "..... that still needs to be explored ..... potentially affecting ....." => .... that needs further exploration .... which impact ....
line 117: replace ".... some SRs ....." by ".... small SR values, which are dominated by instrument noise, ...."
line 120; I do not understand what you mean with ".... and distances between the instruments and the height bins"? Please explain or rephrase.
line 123: ".... especially in the case of strong Mie returns, which are not detected by the classification procedure, ...." The addition is important because in principle measurements with strong returns should be classified as "cloudy" and not enter the Rayleigh-clear wind.
line 138-139; vertical resolution is not in m/s. Probably you mean that the balloon ascending speed is 5 m/s, then measuring every 2 seconds gives a vertical resolution of 10 m. Please correct.
line 216; this a surrogate for the standard error, Right? Please use this more well-known terminology in statistics, rather than "uncertainty of the mean bias".
line 227; I guess the representativeness error is different for Mie and Rayleigh winds as they sample the atmosphere along different length scales, i.e., about 10-15 km for Mie and and about 90 km for Rayleigh, along the satellite track? Can the authors please comment on this?
line 239; with EE you mean EE_Aeolus as in Eq. (9), right? Please be consistent in the text
line 243; what do the authors mean with: "noise related to atmospheric temperature and pressure"? Do errors in these parameters lead to wind random errors or biases?Figure 1. red and orange are hard to discriminate. Please use a different color for orange (Rayleigh-cloudy).
Caption of figure 1, please mention explicitly that you used model equivalents from the model background (which did not (yet) use the radiosonde), see also line 129. This is important, obviously and good to mention again.line 275 mentions a STD of 2.1 m/s for sqrt(<HLOS_ECMWF-HLOS_RS>^2) at Rayleigh-clear locations. The same metric shows a value of 2.93 m/s at Mie-cloudy locations in line 279. That is quite a large difference for parameters with quite consistent and well-known error characteristics. Assuming that the quality of radiosonde observations is rather constant for the complete profile this suggests that ECMWF performs substantially worse at locations where Mie winds are found (lower troposphere) than at locations of Rayleigh-clear winds (upper troposphere, lower stratosphere). Or is this discrepancy simply a statistical effect due to the limited data set? Can the authors please comment?
line 283; "as most of the systematic and random errors seem to be specific to the Aeolus Rayleigh-clear winds". But in the text above you show that Mie-cloudy biases are larger than for Rayleigh-clear. Please correct.
line 284: "This stresses the need to identify the underlying potential error sources of Rayleigh clear observations with respect to the presence of clouds and dust aerosols ......"
Given the larger systematic errors in Mie-cloudy I would think that these are more sensitive to clouds and aerosols. The fact that random errors are larger for Rayleigh than Mie is pretty clear. Please comment.Table 2. sigma_mu is not defined in section 3.2. Please do.
line 304; "For Mie-cloudy, the systematic difference indicates a bias of 0.9±0.3 ms1, which is within the uncertainty range of the
ESA’s specification ..."
No, it is not, see major comment G2. Please correct.line 306; how do you arrive at 1.1-2.3 m/s? Following Eq.8 with sigma_rep = 1.5-2.5, sigma_RS=0.7 and sigma_tot=2.9, I end up with sigma_Aeolus in the range 1.3-2.4. Where do I go wrong? See also table 3.
line 315. I think AVATAR-T carries a 2 micron lidar, so measuring particles only. How can you compare these with Rayleigh-clear, measured in clean air conditions?
Figure 2. "Differences (dots) and average differences (lines)"
I cannot conclude from the plot that the line is the average value. For instance in the left panel at 17500 m, all blue dots are on the right hand side of the line. Similar issues appear at all altitudes.In the caption of Figure 2, mention Aeolus Rayleigh-clear winds.
line 351; How is it possible to have '+'s with EE values > 5 m/s in figure 3a?
line 356; "discrepancy". This discrepancy is expected, see major comment G1.Figure 3. The binning of the stepwise solid lines is not explained. Why does it go up to 8 m/s in 3a, while you have much less than 40 data points at this value.
line 407; "Table 4 describes the error dependency of the Rayleigh-clear observations with respect to the presence of clouds and dust"
This classification is not clear. Do the authors mean presence inside the bin or from bins aloft or both?line 415-417. MADI compared against EE is invalid, see major comment G1. The conclusion that: "EEtot in clear sky conditions is well calibrated, while it is becoming gradually too low with the increasing presence of clouds and dust." is not well explained.
Table 4. In the caption replace 25% by 50%
Figure 7a. use a different x-axis scaling to better visualize the differences between the curves, e.g., x in [-25,35]. Also for fig 8a and 9a.
Figure 7b, how do you arrive at the blue curve? And how the grey curves? Are the latter obtained from Eq.(9) with EE_Aeolus from the L2B product?
line 483; "with a minimum of 3.5 m/s ...". This value does not follow from fig 7b. Please correct.
Figure 8; the grey lines in b/c/d in look the same as in figure 7. Same for Figure 9. Despite the completely different scenes. Where do these curves come from?
Minor comments
==============
line 9; Raleigh -> Rayleigh
line 11; can be degraded -> are degraded
line 87; add "off-nadir" and "in the tropics" in "it points at 35 degrees off-nadir with an angle of ~10 degrees from the zonal direction in the tropics".
line 109; processing chain -> mission
line 112; L2bP 3.50 -> L2Bp version 3.50 (the rest of the paper uses L2B with all capitals)
line 113; "should is" - remove "should"
line 134; "Between the 7 and 28th ...". Correct to either: "Between 7 and 28 ...." or "Between the 7th and 28th of ....".
line 183; "in the presence of ...". Add "the"
line 200; remove "bin-to-bin"
line 209; Eq. (5) misses the index (i) between the brackets. Please correct.
line 232; "the the". Please correct
line 244; In contrary -> In contrast or Contrary? Please check.
line 473; black lines -> black lineCitation: https://doi.org/10.5194/egusphere-2023-742-RC1
Maurus Borne et al.
Maurus Borne et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
73 | 29 | 6 | 108 | 2 | 2 |
- HTML: 73
- PDF: 29
- XML: 6
- Total: 108
- BibTeX: 2
- EndNote: 2
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1