A Land-Corrected ASCAT Coastal Wind Product

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

Review: A land-corrected ASCAT coastal wind product

Summary

ASCAT σ⁰ measurements near the coast combine signal contributions from the ocean and from the land surface. Since the contribution of the land backscatter is much larger than the contribution from the ocean, this will impact the scatterometer wind retrieval close to the coast. Since November 2022, a new parameter estimating the land contribution (LCR) is available in the ASCAT products. The paper introduces an empirical correction for the land contribution which is based on the LCR. With this correction, it is possible to derive wind vectors closer to the coast. The method is cross-checked against ECMWF forecasts and validated using buoy data.

General concept comments

The manuscript presents the research topic in a concise and well-structured manner. It describes the background very clearly, describes the correction method, and provides the validation.

There is space for improvement in the description of the methodology: the reader has to implicitly assume that the regression is done on each individual WVC. This should be explicitly stated. In some places, empirical parameters (thresholds etc) are introduced without describing how they were derived.

The central point that needs clarification is the assumption of a Gaussian distribution of the data. If this is not the case (i.e. if the data is processed in linear units), then the core assumptions on the regression are likely not applicable. The authors are requested to carefully assess this part of the manuscript and confirm that the approach is valid.

Specific comments

 

Line	Comment
51	“The land correction algorithm is based on the assumption […]”: This sentence sounds like it refers to the LCR mentioned in the previous sentence. Can you please re-phrase it to make the distinction clear?
54	“The optimal value lies between 0.2 and 0.5.” Where do those numbers come from?
67	It would be helpful for the reader to know how many measurements fall into a WVC (approximately).
72	Is the analysis done in dB or in linear units? This information should appear somewhere.
82	How is the linear regression derived – individually for each WVC, or using a larger data set? This should be stated explicitly. Also, a>>b so there is a very high sensitivity to the land fraction. How can you confirm that the dependency is linear?
110	If this is in linear units (see comment above), then we will have a distribution function that is closer to a logarithmic distribution rather than a Gaussian. This implies that the mean value / standard deviation you calculate are not meaningful parameters for the description of the distribution. Please discuss the impact of the distribution function on the derived parameters.
118	“A crucial parameter is the maximum land fraction, 𝑓𝑚𝑎𝑥, that is allowed in the regression analysis”. Given the comments above (line 82), is there a saturation effect?
120	“In this work a value of 0.5 is adopted.” Please elaborate where this value comes from.
125	“Further analysis showed that this is mainly caused by the noise estimate 𝐾𝑝 exceeding its threshold of 10.0 %. This may be due to increased wind variability near the coast […]” Another explanation for high Kp values is that the mix of land and ocean in the footprint results in a high backscatter variability.
156	“800,000 winds per 10-km bin”: is this number referring to the full year of data?

 

Author Response

Reply to reviewer #1
We thank the reviewer for the constructive remarks that helped to improve readability of the paper. Below we repeat her/his comments in normal font and our replies in italic.

51    “The land correction algorithm is based on the assumption […]”:
This sentence sounds like it refers to the LCR mentioned in the previous sentence. Can you please re-phrase it to make the distinction clear?

We changed the presentation slightly to distinguish between the Land Contamination Ratio (LCR) from {5} and the new land fraction developed by EUMETSAT. In particular, we dropped the term LCR for the EUMETSAT land fraction and instead refer to it as land fraction. 

54    “The optimal value lies between 0.2 and 0.5.” Where do those numbers come from?

From visual inspection of results. This is added in the text, also after equation (7). Further, for clearness we stick to the value of 0.5 as the optimal compromise between quality and quantity.

67    It would be helpful for the reader to know how many measurements fall into a WVC (approximately).

Is added in the text: between about 25 to about 45, depending on beam and incidence angle.

72    Is the analysis done in dB or in linear units? This information should appear somewhere.

In linear units, added in the text.

82    How is the linear regression derived – individually for each WVC, or using a larger data set? This should be stated explicitly. Also, a>>b so there is a very high sensitivity to the land fraction. How can you confirm that the dependency is linear?

The regression is done for each beam in each WVC. This has been added in the text. If the radar cross sections of land and sea (in linear units) are constant, the dependency is linear a-priori from equation (1). The problem is of course that these radar cross sections are not constant, and extreme deviations must be filtered out by quality control.

110    If this is in linear units (see comment above), then we will have a distribution function that is closer to a logarithmic distribution rather than a Gaussian. This implies that the mean value / standard deviation you calculate are not meaningful parameters for the description of the distribution. Please discuss the impact of the distribution function on the derived parameters.

The regression analysis results in equations (2) to (4) are independent of any assumption of the distribution. The standard deviations in (5) to (7) are candidates for quality control. Only σ_b^2 showed any skill. This has been added in the text.

118    “A crucial parameter is the maximum land fraction, ????, that is allowed in the regression analysis”. Given the comments above (line 82), is there a saturation effect?

No, we found no saturation effect. Especially in urban and harbor areas the radar cross section can be very high due to corner reflectors formed by buildings and other man-made structures.

120    “In this work a value of 0.5 is adopted.” Please elaborate where this value comes from.

From visual inspection. This has been added in the text.

125    “Further analysis showed that this is mainly caused by the noise estimate ?? exceeding its threshold of 10.0 %. This may be due to increased wind variability near the coast […]”
Another explanation for high Kp values is that the mix of land and ocean in the footprint results in a high backscatter variability.

Yes, variation in σ^0 over land may also play a role. We added this in the text.

Further note that the regression assumes a cross section partially based on land coverage and partly on sea coverage. The NRCS values from land and sea can be very different, but where ideally the land fraction determines the contributions from land and sea for each NRCS value. Where the summed contributions can show large variability, they ideally constitute in our model a separate sea and land contribution of signal and noise. As in many other simplified retrieval procedures, we do not explicitly model the noise contributions on an advanced level (e.g., Bayesian). We believe this is not adding much skill for the smaller land fractions.

 156    “800,000 winds per 10-km bin”: is this number referring to the full year of data?

This refers to one month of data (January 2017). This was stated in the heading of table 1, but we added it also in the text.

 

Reviewer 2 Report

Comments and Suggestions for Authors

Summary:

The manuscript describes the development and implementation of a new approach to correcting scatterometer backscatter measurements near land. The approach makes use of an assumption on constant backscattering of the over-ocean and land contribution to solve a linear regression-based formulation for the correction. The land contribution is removed allowing for application of the scatterometer geophysical model function (GMF). They provide an assessment of the winds against ERA5 and buoys as a function of distance from the coast and explore the development and ramifications of using a standard error estimate on the bias parameter to apply quality control.

 

Comments:

Overall, I found this manuscript to be well-conceived and well-written with a tight focus on the land correction development. I found the figures to be appropriate and convey the information necessary to support the development of the land correction methodology and its validation.  Nevertheless, some clarifications could be useful.

 

1)        For developing the linear regression, was this done on a per-pixel basis or were all WVC pixels across all global coastlines included as part of a single regression?  If all pixels were pooled, is there any reason to suspect that the land element of the signal should be constant? If not, then it seems likely that there may be some geographic/spatially-dependent element of residuals that could result in certain regions being flagged disproportionately as being “outliers”.

2)        The authors provided a couple of tables showing changes in sampling and a few figures showing “close-up” images of the approach in action. However, I think a global-scale figure and metric (e.g., maybe a zonal average of changes in counts, perhaps normalized to account for differences in number of coastal cells) is warranted. It would help give a better understanding of the full scope and may also highlight if there are any specific regional disparities/issues that are lost in the aggregated statistics that are shown.

3)        Equation 9 and the use of the weights is not well described. How exactly is the weight factor applied and used as part of either the retrieval or QC effort?

4)        For the buoy comparisons, line 210 seems to indicate that the “latter” IS TAC and MARS data have more data points than NDBC and so are preferred. If so, then why is NDBC mentioned at all? I was able to find a list of IS TAC buoys that are easily accessible. I’ve also found easy and direct access to NDBC buoys. However, it was not clear that any of the MARS buoys are openly and directly accessible. If these are not open data, then perhaps they should not be included. For example, the MARS catalogue noted at: https://www.ecmwf.int/en/forecasts/access-forecasts/access-archive-datasets is locked. Further, there appear to be many potential sub-products. Please clarify the accessibility of the MARS buoys, how they were accessed if they are openly available, and the specific MARS data/product collection/id used to obtain buoy data. Further, please identify the source of all metadata information used to identify instrument height information especially if it is not found directly within the observation files. This information is essential to implementing the standardization.

5)        Figure 7 and the last paragraph of the Conclusions seem to be tacked on to the end of the study and feel out of place. If the authors really want to provide more discussion on this, they should incorporate it prior to the conclusions section.

 

Author Response

Reply to reviewer #2

We thank the reviewer for the constructive remarks that helped to improve the readability of the paper. Below we repeat the reviewer’s remarks in normal font and add our reply in italic.

1)        For developing the linear regression, was this done on a per-pixel basis or were all WVC pixels across all global coastlines included as part of a single regression? 

The regression is done for each WVC and each beam separately. This has been stated in the text more clearly.

 If all pixels were pooled, is there any reason to suspect that the land element of the signal should be constant? If not, then it seems likely that there may be some geographic/spatially-dependent element of residuals that could result in certain regions being flagged disproportionately as being “outliers”.

The regression analysis is strictly local, and so is the flagging of outliers.

2)        The authors provided a couple of tables showing changes in sampling and a few figures showing “close-up” images of the approach in action. However, I think a global-scale figure and metric (e.g., maybe a zonal average of changes in counts, perhaps normalized to account for differences in number of coastal cells) is warranted. It would help give a better understanding of the full scope and may also highlight if there are any specific regional disparities/issues that are lost in the aggregated statistics that are shown.

We do not think global and zonal metrics will add new information, as the land correction is calculated separately for each individual WVC. Local features, especially cities and harbors will certainly lead to increased rejection by quality control, but to our opinion this is rather trivial.

Further note that Figure 7 provides an example of the issues that we encounter in the coastal processing in some specific regions of the globe.

3)        Equation 9 and the use of the weights is not well described. How exactly is the weight factor applied and used as part of either the retrieval or QC effort?

We clarified the use of weights in the text. The weights are only used in the calculation of K_p for those WVCs where regression is applied.

4)        For the buoy comparisons, line 210 seems to indicate that the “latter” IS TAC and MARS data have more data points than NDBC and so are preferred. If so, then why is NDBC mentioned at all?

Line 210 also states that this happens in a number of cases. In the other cases the NDBC data contain more measurements. We changed the formulation a bit to make this more clear.

 I was able to find a list of IS TAC buoys that are easily accessible. I’ve also found easy and direct access to NDBC buoys. However, it was not clear that any of the MARS buoys are openly and directly accessible. If these are not open data, then perhaps they should not be included. For example, the MARS catalogue noted at: https://www.ecmwf.int/en/forecasts/access-forecasts/access-archive-datasets is locked. Further, there appear to be many potential sub-products. Please clarify the accessibility of the MARS buoys, how they were accessed if they are openly available, and the specific MARS data/product collection/id used to obtain buoy data.

The MARS data are indeed only available for ECMWF members, but non-members may request ECMWF data for research purposes. Leaving out the MARS data would decrease the size of the data set considerably. Therefore we keep the MARS buoy data in our analysis. This is indicated in the text.

The online documentation at the ECMWF site suffices to find the data code and contains also example scripts. We don’t think that a scientific journal is the place for this information. For assistance on these points one can contact the corresponding author.

Further, please identify the source of all metadata information used to identify instrument height information especially if it is not found directly within the observation files. This information is essential to implementing the standardization.

All metadata were found in the data files or in separate metadata files supplied by the data provider. Unfortunately, the availability of metadata is a problem and a large number of buoys could not be used due to the lack of metadata.

5)        Figure 7 and the last paragraph of the Conclusions seem to be tacked on to the end of the study and feel out of place. If the authors really want to provide more discussion on this, they should incorporate it prior to the conclusions section.

We changed the title of section 3 to “Conclusions and outlook”.