Interactive comment on “ Comparison of aerosol optical depth from satellite ( MODIS ) , Sun photometer and pyrheliometer ground-based measurements in Cuba

Antuña-Marrero et al. have compared aerosol optical depth (AOD) retrievals from the spaceborne MODIS instruments to ground-based observations done in Cuba. The ground-based observations include sun photometer (AERONET) and pyrheliometer measurements from several sites. The authors conclude that both MODIS instruments produce AOD data with corresponding accuracy, the Dark target retrievals are in better agreement with the ground-based observations than the Deep blue retrievals, and the pyrheliometer measurements could be used to construct reliable time-series of broad-

-Secondly, I do not see the value of doing the comparison between the MODIS and AERONET observations using single observations.Single observation pairs may not represent the same air mass thus, they might have different values for the right reasons.In addition, the uncertainty/noise in single observations is larger than in spatially or temporally averaged values.As the comparisons using single observations and the so-called "daily means" produced comparable statistics I do not see any reason to use single observations in the analysis.Therefore, I suggest that the authors leave out the discussion/results regarding single observations.Answer: Precisely, the purpose of the comparison in this case is to test if single observations could be used for the determination of the aerosols climatology over land in Cuba because of the mixing of water and land areas in our area of study.A new version of Figure 1 has been included to highlight the reasons for using L2 MODIS data instead of L3.The sentence "The grid cell of 1° in latitude and longitude shown in red in figure 1 is an example of the limitations of the MODIS L3 products to represent land areas in the case of Cuba." has been included on the 1st paragraph of the Section "2.4 Coincidence criteria for MODIS and Sun photometer measurements" A priori we cannot ensure that both criteria give similar results, we must test this in our area, taking into account the different number of data and the characteristics of the land/water surface.After that, obviously we analyzed only the results of one criterion.We think that this position is correct.Furthermore in the case of solar direct radiation, only single data are available.
We also rewrote part of this paragraph and unified the description of both criteria in section 2.4 Otherwise, we cannot understand why single observation may not represent the same air mass and median values do.Single observation or median values during a time overpass are currently used by the satellite community to do this type of studies.Despite the variability of aerosols, the time between the pairs of observations of MODIS and sunphotometer is generally smaller than the time of air mass changes.Bear in mind that these measurements are of columnar type, and the AOD parameter represents the total load content of particles including whole atmospheric column and they are not influenced by winds in the sense of particle concentration at surface.
-Thirdly, not all the methods are described accurately enough.For example, the calculation of broadband AOD (BAOD) or monthly averages are not described at all.

Answer:
The section 2.3 was implemented in order to describe in detail the main retrieval equation and the parameterized variables.In addition, the main assumptions are described.Anyway, the reader is referred for further details to the original paper for where all the assumptions are described in detail.
The monthly average was described by the new sentence: "based on the mean of each month for every year of the measured period".
-Fourthly, the manuscript lacks discussion on the results.What do the results mean and how do they compare with are studies done in this region?
Answer: We think that we analyze and discuss the results of the comparison between sunphotometer and MODIS with an extend number of statistics and the linear correlation methodology in an extensive and correct way.We have experience in this type of studies/analysis as can be seen by other published works for aerosol studies (Bennouna et al., 2011;2013)  In spite of the differences found between the different areas of study, we have carried out the comparison with our results and this has been incorporated in the text, at the end of section 3.1.
-Lastly, the comparison of AERONET and MODIS AODs is a routine task thus, the results may not be that interesting to a wider audience.The most interesting part of the paper is the broadband AOD thus, the authors should discuss it in more detail.For example, it would be interesting to see the time series of BAOD from the four sites: how they compare with each other and with AERONET and MODIS.And if there are clear differences during some periods, it would be interesting to read what is causing the differences.

Answer:
The opinion of the authors is that for the present paper time series does not provide better information than the one provided by the statistics already reported, but this is just a result, because as mentioned this Caribbean area has not been analyzed before in detail (the mixing of land and ocean water areas is a big challenge for the retrieval algorithms) and this area is very interesting for regional climatological studies.Our current national research project is to determine the climatology of AOD for Cuba from MODIS: the AOD and AE climatology for Camaguey and the BAOD climatology for Camagüey, Topes de Collantes, Jovellanos and La Fe.It includes the analysis of the respective AOD, AE and BAOD time series and their trends.Those results will be reported in future publication and here we report the climatology given by MODIS and its comparison with the others two series of BAOD and Photometer.The article is already long enough and it is focus on MODIS data, not on the characteristics of the other time series of data.But they serve for an interesting and necessary comparison.
Consequently, the manuscript should be thoroughly revised to clarify the content and to make it more interesting to the readers.
Corrected.The sentence is now: "The cloud-free condition in the line of sight to the sun is satisfied selecting DNI measurements with a clear line of sight between the pyrheliometer and a region of 5° around the sun (GOAC, 2010)." P7, 180: Why is the monthly mean PW calculated differently for Camagüey than for the other sites?How large difference in the BAOD could this change cause?It would be clearer and more robust to use the same method for each site.
Answer: According to table 3 in Gueymard, (1998) for PW = 5 cm with an error of ± 20 % in the PW value, the magnitude of the possible absolute error in BAOD is between 0.0145 and 0.0325.The first value is estimated for an instrumental error of ± 0.5 % and the second for ± 3 %.
We used PW from sunphotometer at Camagüey because of its lower error than PW from reanalysis data.At the other 3 sites we have only PW from reanalysis data.The differences between monthly mean PW from sun photometer and reanalysis at Camagüey (estimated for 2008 to 2015) are in the order of 1% for the wet season (November to April) and 8% in the rainy season (May to October).Then the magnitudes of the possible error in BAOD produced by the PW differences between the sun photometer and the reanalysis are inside the estimated total error in the determination of the BAOD, 10 -2 .
P7, 186: Why is Cuba different from the other regions?Are there more clouds or something else?
Corrected: The difference we are referring to is the areal extension of Cuba (the case of the insular countries) compared with continental regions.Two sentences were included after the first sentence of this paragraph: "The reason in general is the little areal extension of islands.In addition, in the case of Cuba its particular narrow latitudinal, elongated longitudinal extensions and the vicinity of the sea makes the MODIS L3 product not suitable for climatological studies." P7, 187: To my knowledge, 2 data is typically used when comparing with AERONET observations.L3 is used in model comparisons and climatological analysis.
Corrected: We modified the sentence to express clearly that we are referring to the climatology.The following sentence have been included after the former one: "In that sense is absolutely necessary to validate the single MODIS L2 with the single sun photometer measurements." P7, 188: Which methodology are you referring to?
Answer: We are referring to the methodology to maximize the number of pairs of single observations without any repeated observation.In addition, the sentence was modified: "We designed and applied a methodology for maximizing the available pairs of MODIS L2 and sun photometer AOD and AE measurements coincident in space and time avoiding the duplicate use of any of them.
P8, 215: As I mentioned in the general comments, the exclusion of the analysis regarding individual measurements would make the manuscript easier to follow.

Answer:
We do not agree.The comparison of the individual measurements is necessary to determine the real differences between the single MODIS L2 products and the sun photometer measurements in Cuba.
P9, 243: As you mention in the text, "daily mean" is not the best term for the calculated values.
Maybe collocated mean values or something like that would be better.

Corrected:
The term "daily mean" was replaced by "collocated daily mean" Corrected: Now reads "We used the EEDT expression in equation ( 2) for estimating also the uncertainty when the DB algorithm is applied.The purpose is to allow comparing the performance of DB and DT algorithms directly (Sayer et al., 2014)."

P11, 278-279: I wasn't able to follow this sentence
Corrected: All this part of results section has been modified substantially.
P11, 284: Can you really say that the monthly values will also be good because the daily values are good?In addition to the accuracy of the daily means, the quality of the monthly means depends on the temporal sampling within the months.For example, if you have 5 accurate daily means from a month but all 5 values are from the first week of the month, will the monthly mean be representative?When you calculate monthly averages you should also consider the distribution of the daily means within the months.If the temporal coverage is poor, the monthly mean will not be that reliable.Consequently, the authors should explain in text in detail how they calculated the monthly means and they should use some kind of a lower limit for the daily means before monthly means are calculated.

Answer:
The sentence on P11, Line 284 says: "From the results described above it is evident that the monthly means AODt and AODa derived using the DT algorithm agree better with the AODSP than the ones derived using the DB algorithm".We do not say that the monthly values will also be good because the daily values are good.
We agree that it will better to apply strict climatological procedures for conducting the AOD comparisons.However, the available sun photometer and the pyrheliometers measurements in Cuba do not have homogeneous time distribution allowing complying with those climatological procedures for the comparisons.To deal with these data limitations we are deriving simultaneously the climatology the AOD from the sun photometer for Camaguey and the BAOD and MODIS climatology for the four pyrheliometric stations.
P11, 290: This section could able be omitted.

Answer:
We do not agree.The comparison of the individual measurements is necessary to determine the real differences between the single MODIS L2 products and the sun photometer measurements in Cuba.

Answer:
The number of points could not explain the fact that for both DB and DT the magnitude of R is equal or lower than 0.5 on December and January.From December to March the number of cases remain over 150 for both DB and DT, while the rest of the year the number of cases are equal or lower than 90 (except for DT in August).Then if the number of cases is the cause we should expect for February and March values of R lower than 0.5 and that did not happen both for DB and for DT.
P12, 331: This is surprising result as the correlation coefficients are the lowest during the months with the highest fractions.What could explain this contradiction?
Answer: We agree this is a surprising result.We have no explanation for it for now.
P13, 351: I think this analysis should also be done using daily means instead of individual observations.
Answer: We followed the reviewer suggestion.Answer: That is an option.However, we decided to remain the one hour time window.
P14, 368: The combination of the sites works only if all the sites have similar aerosol populations.Otherwise the combination might mask some site specific features and, in the worst case, ead to erroneous conclusions.Are the aerosols the same at each site?
Answer: Very few aerosols studies exist in Cuba.The unique aerosol characterization among the four stations used in the present research has been conducted in Camaguey.In addition, the statistics for the individual stations will not be robust because of the few pyrheliometer and MODIS coincident cases at the individual stations.We will do that in the near future.We are at the beginning of the data rescue of the actinometrical observations conducted before 2010 in Jovellanos and before 2011 at La Fe, both extending far before Terra's record.
P14, 370: Why did you leave out the days with high AOD?Are they cloud contaminated?
Answer: Yes.We changed the sentence to: "In addition, we did not considered the very few cases with values of BAOD > 0.5, around 1 %, of all the cases, to avoid the possibility of an inadvertent cloud contamination." P14, 376: Why the DB retrievals match better with BAOD than DT retrievals.It was the opposite with the AERONET data.What about monthly comparisons between MODIS and BAOD?

Answer:
We have no answer for this fact.We verified the calculations and found no errors.
Because the hourly time step of the DNI measurements the BAOD have a low level of coincidence with MODIS observations as is shown in table 8.In addition, between May and October the clear sky conditions are less than the 10% of the available solar radiation measurements.Hence, the number of cases for monthly statistics of the coincident BAOD and MODIS measurements is very small for the statistics between May and October.
P14, 381: I would suggest to change the places of the sections 3.3 and 3.4.It would be clearer if the BAODs would be compared firs with ground-based and then with space borne measurements.
Corrected, but we have removed the comparison between BAOD and sun-photometer, because the paper is focused on the comparison of MODIS and ground-based instruments.
P15, 401: There isn't much discussion regarding the results.How do these findings compare with other studies done in this region/with similar methods?
Answer: The authors have not found any other comparison between MODIS and sun photometers in the Caribbean Basin and only two conducted on islands at different latitudes and regions.A comparison with former regional results for North and South America, have been added.
I would also like to see the long BAOD time-series from these sites and how they compare with the AERONET and MODIS time-series.Those results would make the manuscript more interesting to a wider audience.
Answer: The opinion of the authors is that for the present paper time series does not provide better information than the one provided by the statistics already reported.The current research project is producing the climatology of the AOD for Cuba from MODIS; the AOD and AE climatology for Camaguey and the BAOD climatology for Camagüey, Topes de Collantes, Jovellanos and La Fe.It includes the analysis of the respective AOD, AE and BAOD time series and their trends.Those results will be reported in a publication.
P15, 402: Please include numerical values in the conclusions to make it more robust and clear.
Answer: Numerical results have been included in the conclusions.
P26, Fig 3: Please include the error envelopes in the plots.Density plots would make it easier to see where most of the observations are (see for example Fig. 5 in Petrenko et al. (2012)) and you should limit the axis range to 0.0-0.6 to remove unnecessary empty space.Corrected:The new figure 3 contains density plots, with axes ranges between 0 and 0.6.In addition, figure6is now a density plot.In both figures the least squares linear fit is also shown and drawn, together with the number of cases.P28, Fig 5: Shouldn't these values be collocated?Now the AERONET data seems to have over two times more points.Collocated values would enable a more meaningful comparison.Corrected: The new figure 5 shows the frequency distribution of Ångström Exponents only for coincident sun photometer and Terra and Aqua values.The same distribution but for sun photometer and Terra and sun photometer and Aqua separately are shown in figure S4, added in the supplements , 247: Is there a minimum number requirement for theMODIS and AERONET observations?Sayer et al.. (2014)required only single observations from both instruments P9, 248: Did you limit the AE comparison to cases with moderate or high AOD?I think you should because the MODIS AE's are only usable in those cases.In figure 3 may be appreciated that low AOD values predominate, at least for the spatio -temporal coincident MODIS and sun photometer observations.In the case of the coincident measurements of the AE, after eliminating the cases with 1.5 and 1.8 AE values (columns 4 to 6 on table 6) and also eliminating the cases with AOD <= 0.3 we ended with 29 cases for Terra and 10 cases for AQUA.The statistics are below: This is a confusing sentence.You should explain here that EE_DT is defined relative to AERONET AOD and is therefore independent of the MODIS retrievals.
but other studies have used lower limits ranging from 2 to 5 (e.g.Petrenko et al. (2012)).I would prefer the usage of some lower limit (e.g. at least 3 observations from MODIS and 2 from AERONET).Of course the selection of these limits affects the number of overpasses available for analysis so you have to select in a way that you do not throw away too much data but at the same time, you only compare representative observations.referenceswere added in the new version of the manuscript.Answer: The results for the Collocated daily means, excluding MODIS AE values of 1.5 & 1.8 were added to table 6.Taking into account the statistics for this new set of results we re-wrote the discussion of table 6, which reads now: "Statistics on table 6 for the single observations, both considering and excluding AEt and AEa equal to 1.5 or 1.8 show high values of RMSE, MAE and BIAS.These results in addition to the values of R, below 0.5, evidence big differences between the AE from both instruments.Similar results are in the case of the collocated daily mean both considering and excluding AEt and AEa equal to 1.5 or 1.8.The comparison showed the low quantitative skill of the AEt and AEa for this site providing numeric magnitudes of it.One factor contributing to this result is that the AE from MODIS has large uncertainty in low-AOD conditions, because the AE is a gradient between two small numbers(Wagner and Silva, 2008).Another factor could be the poor performance that the DB algorithm showed in the comparison with AODSP."As you have DNI measurements only once an hour, you could modify the coincidence criteria to average a couple of measurements even though both of them are not within the one-hour time window.That might provide you with more comparable observations.