A New Approach to Defining Uncertainties for MODIS Land Surface Temperature

Round 1

Reviewer 1 Report

Journal = Remote Sensing (MDPI)

Manuscript ID = 465893

Title:  A new approach to defining uncertainties for MODIS land surface temperature.

 

Review Completion Date :  Mar 18, 2019

Review comments:

In this manuscript, authors described a method of estimating LST product uncertainty from MODIS, which may promote validation of satellite LST product and therefore further impacts LST applications , such as  data assimilation of weather forecasting models. The paper is well organized and presented, and is recommended for publication, under condition of the following issues being responded:

 Generic comments:

1.      What is the data source of the water vapor used in the LST retrieval algorithm and what about its uncertainty and its impact to LST retrieval? One additional uncertainty is coefficients misused due to the inaccurate water vapor input, how do you evaluate such uncertainty?

2.      In the locally correlated surface uncertainty, equation (13) is surface type dependent, which makes sense for the surface type based emissivity; however, the emissivity dataset from University of Wisconsin is used in this paper, which is physically derived pixel level value, thus, alternative uncertainty estimation is required here.

3.      For the systematic uncertainty, as the author said, the calibration and RTM model accuracy is the two possible sources, how do you assign this uncertainty in this study?

4.      In term of the emissivity random uncertainty, the author uses 100m ASTER GED to perform a sub-pixel statistics, how do you deal with it for each channel (e11 and e12)?

5.      According to the uncertainty budget, there would no significant uncertainty difference between day and night. In Figure 5, the daytime uncertainty of level 3 LST is obviously larger than nighttime, does this mainly attribute to the LST heterogeneity issue introduced by the last term in equation (16)?

Some minor issues:

Line 249, more explanations is required for Equation (3), such as, what does R and F mean?

The equation should be consistent with the text, item e_wv in line247, equation index in line276, line287, Line 525

Line 404, I think it should be 70*70 m² rather than 702m

Line 408, actually, only 7 sites are used, same for Line 457

End 

Comments for author File: Comments.pdf

Author Response

Referee 1:

Comments and Suggestions for Authors

In this manuscript, authors described a method of estimating LST product uncertainty from MODIS, which may promote validation of satellite LST product and therefore further impacts LST applications, such as data assimilation of weather forecasting models. The paper is well organized and presented, and is recommended for publication, under condition of the following issues being responded:

We would like to thank the referee for the very helpful comments, to which we reply in detail below.

Generic comments:

1. What is the data source of the water vapor used in the LST retrieval algorithm and what about its uncertainty and its impact to LST retrieval? One additional uncertainty is coefficients misused due to the inaccurate water vapor input, how do you evaluate such uncertainty?

ECMWF ERA-Interim daily fields are used for the source of the water vapour in the LST retrieval. These data are not used explicitly in the retrieval Equation 1, but instead to determine which band of coefficients to apply. Equation 12 presents the uncertainties from water vapour error effects categorised by each band of coefficients. This uncertainty is the standard deviation of the difference between the input surface temperatures into the model-fitting and the simulated-retrieved output. What is missing in the original text is clarification that the simulated-retrieved output is actually calculated by using perturbations to the water vapour to quantify the uncertainty as a result of imperfect knowledge on the water vapour field. These perturbations are performed within error ranges of water vapour determined from comparison of the input ECMWF water vapour with respect to in situ measurements from ARM sites. A new Figure 1 is included to show the error ranges used in the perturbations. Thus the impact of locally correlated error effects from this source on the selection of retrieval coefficients is wrapped up in this existing Equation 12. Attempting to add an additional component here at the retrieval level would be double-counting these uncertainties. We have added a better explanation of this in the text.

2. In the locally correlated surface uncertainty, equation (13) is surface type dependent, which makes sense for the surface type based emissivity; however, the emissivity dataset from University of Wisconsin is used in this paper, which is physically derived pixel level value, thus, alternative uncertainty estimation is required here.

We thank the reviewer for noting the difference between a surface type based emissivity estimate and a physically derived pixel level emissivity estimate. However, it should be understood that it is not the emissivity estimates themselves that are assumed to be correlated but the error effects. This is a reasonable assumption because a source of error in the estimation of emissivity will be the response of the retrieval to different materials. Surface types sharing similar materials, as can be categorised by land cover classes, present the emissivity retrieval with similar error structures. An improved description is provided in the text.

3. For the systematic uncertainty, as the author said, the calibration and RTM model accuracy is the two possible sources, how do you assign this uncertainty in this study?

We assign this uncertainty as                                                in Equation 4 as a systematic addition. This means it does not reduce as LST data is propagated to higher level gridded products. Of the two sources of error, the Level -1 instrument effect is quantified by Absolute Radiometric Accuracy (ARA). This would be expected to be removed through an instrument calibration model. Calibration results from the literature [1, 2, 3] indicate the systematic error to be very small for the Split-Window bands 31 and 32; with the Aqua bands 31 and 32 in particular being stable with very low systematic errors. The second source is the RTM accuracy, and has been estimated for RTTOV in [4] for AATSR. We have assumed an equivalent magnitude for the RTM when applied to MODIS filter functions. In future studies we aim to test the RTM accuracy for different sensors which may have a small change on these estimates, although we do not expect significant differences. We have added more explanation in the text.

4. In term of the emissivity random uncertainty, the author uses 100m ASTER GED to perform a sub-pixel statistics, how do you deal with it for each channel (e11 and e12)?

We use band 14 of ASTER, centred at 11.3µm, to determine our statistics. For many materials the emissivity spectra is much flatter in the traditional split-window windows around 11µm and 12µm than in the 8-9µm region. We have exploited this by assuming the coefficient of variability at 11.3µm would be similar to that for both the MODIS bands centred at 11µm and 12µm. We have now explained this in the text. In the absence of higher spatial resolution emissivity data with corresponding spectral response functions to MODIS we deemed the use of the ASTER GED at 100m to be the most suitable source of data to analyse the sub-pixel variability of emissivity for moderate resolution data.

5. According to the uncertainty budget, there would no significant uncertainty difference between day and night. In Figure 5, the daytime uncertainty of level 3 LST is obviously larger than nighttime, does this mainly attribute to the LST heterogeneity issue introduced by the last term in equation (16)?

The reviewer is correct that the differences between day and night should not be significant given the components of the uncertainty budget as applied to the Level-2 retrieval. On propagation to Level-3 however, the final term in equation 16 related to the sub-sampling of clear-sky pixels introduces divergence. There are two factors relevant here: i) LST heterogeneity is generally higher during the day within each corresponding grid-cell; and ii) there is on average a difference between the proportion of pixels masked as cloudy between day and night since the cloud masking algorithms utilise different channel combinations. We have added such an explanation to the text.

Some minor issues:

Line 249, more explanations is required for Equation (3), such as, what does R and F mean?

The notation in Equation (3) is now better defined in the text.

The equation should be consistent with the text, item ewv in line247, equation index in line276, line287, Line 525

These inconsistencies have now all been corrected.

Line 404, I think it should be 70*70 m2 rather than 702m

The reviewer is correct and this has been amended in the text.

Line 408, actually, only 7 sites are used, same for Line 457

These have been corrected in the text.

 

 

References

[1]          Liu, R.G., Liu, J.Y., & Liang, S. Estimation of Systematic Errors of MODIS Thermal Infrared Bands. IEEE Geoscience and Remote Sensing Letters, 2006, 3, 541-545

[2]          Xiong, X., Chiang, K.-F., Wu, A., Barnes, W., Guenther, B., & Salomonson, V. Multiyear On-orbit Calibration and Performance of Terra MODIS Thermal Emissive Bands. Technical Report available at https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20080040164.pdf, 2007.

[3]          Xiong, X., Wu, A., & Cao, C. On‐orbit calibration and inter‐comparison of Terra and Aqua MODIS surface temperature spectral bands. International Journal of Remote Sensing, 2008, 29, 5347-5359.

[4]          Embury, O., Merchant, C. J., & Corlett, G. K. A reprocessing for climate of sea surface temperature from the along-track scanning radiometers: Initial validation, accounting for skin and diurnal variability effects. Remote Sensing of Environment, 2012, 116, 62–78. https://doi.org/10.1016/j.rse.2011.02.028.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper presents a methodology to derive a complete uncertainty budget, together with the respective in situ validation. The topic is of high relevance for the remote sensing comunity.

I have only some minor suggestions:

lines 214-220: I feel this paragraph would be more apropriate for the introduction or conclusions section

section 2.5.1 Do you have an estimate of the impact of such assumption on the overall uncertainty? Is this kind of information available for other sensor? how much can the uncertainty vary?

lines 235-236 there seems to a be typo here?

lines 239-241: this sentence is confusing, there's maybe a typo on "primarly water vapour"?

line 284: typo "are Gaussian"

line 308: typo "to take account for the..."

line 359: reference format needs correcting

line 382: typo "larger errors atmospheric water"

line 462: typo "pyrgeometersRM"

line 476: typo "becoming remaining"

line 488: typo "wherein"

Author Response

Referee 2:

Comments and Suggestions for Authors

This paper presents a methodology to derive a complete uncertainty budget, together with the respective in situ validation. The topic is of high relevance for the remote sensing comunity.

We would like to thank the referee for the very helpful comments, to which we reply in detail below.

Minor suggestions:

lines 214-220: I feel this paragraph would be more appropriate for the introduction or conclusions section

We agree that much of this paragraph is more appropriate for the introduction and have therefore moved lines 214-218.

section 2.5.1 Do you have an estimate of the impact of such assumption on the overall uncertainty? Is this kind of information available for other sensor? how much can the uncertainty vary?

For many operational sensors this information is not available, and so an assumption is generally made that random effects can be categorised through NEdT estimates, and systematic locally-correlated effects are removed through calibration models. Propagation of random effects through levels reduces their impact, while locally correlated effects do not reduce. For future mission design the concept of better quantifying the impact of Level-1 uncertainties is now being addressed, for example in the science studies for the prospective Copernicus LSTM mission. The impact for any sensor will depend both on the instrument design and the effectiveness of the instrument calibration model. The FIDUCEO project (http://www.fiduceo.eu/) is attempting to confront this for the AVHRR series of instruments. It is not deemed appropriate though in this paper to try to estimate such an impact for MODIS using insufficient information on the instrument or misapplying information from other sensors. Further clarification is provided in the text.

lines 235-236 there seems to a be typo here?

Having read carefully we are unable to see what the typo is here.

lines 239-241: this sentence is confusing, there's maybe a typo on "primarly water vapour"?

We have corrected this sentence to remove the confusion.

line 284: typo "are Gaussian"

Corrected.

line 308: typo "to take account for the..."

Corrected to “to take into account the …”

line 359: reference format needs correcting

Corrected.

line 382: typo "larger errors atmospheric water"

Corrected to “larger errors in estimating atmospheric water vapour”.

line 462: typo "pyrgeometersRM"

Corrected.

line 476: typo "becoming remaining"

Corrected to “remaining”.

line 488: typo "wherein"

Corrected to “where”.

Author Response File: Author Response.docx

Reviewer 3 Report

This paper introduces a comprehensive and consistent approach to determining an uncertainty budget for LST products. The results are quite novel since they show a significant contribution. However, the manuscript needs minor revision. The details comments as below mention:

1.      The introduction part needs to revise and elaborated. The authors need to better justify the methods described. The other study related to an uncertainty budget for LST products needs to be addressed.

2.    Literature review part is needed to be elaborated more

3.    Figure 1 and 5. Please insert the scale bar or scale information

4.    Conclusion part need to be revised, recommendations for future      studies are not significant as per the findings of this study, so rewrite      the future works

5.      Keywords should not be the repetitions of the title words, please find such words which are not in the title, this way search engines of the web will find your manuscript with higher probability.

6.      Recheck the cited references and also references list is not formatted according to journal criteria. 

Author Response

Referee 3:

Comments and Suggestions for Authors

This paper introduces a comprehensive and consistent approach to determining an uncertainty budget for LST products. The results are quite novel since they show a significant contribution. However, the manuscript needs minor revision. The details comments as below mention:

We would like to thank the referee for the very helpful comments, to which we reply in detail below.

1. The introduction part needs to revise and elaborated. The authors need to better justify the methods described. The other study related to an uncertainty budget for LST products needs to be addressed.

We have updated the introduction to point towards other studies on LST uncertainty estimation and frame our work within the context of these past studies.

2. Literature review part is needed to be elaborated more

We have added some additional text in the introduction to highlight better previous studies on uncertainty models for LST and validation of MODIS LST products.

3. Figure 1 and 5. Please insert the scale bar or scale information

Scale bars are included in both Figures 1 and 5.

4. Conclusion part need to be revised, recommendations for future studies are not significant as per the findings of this study, so rewrite the future works

We have updated the Conclusions with further recommendations for future studies.

5. Keywords should not be the repetitions of the title words, please find such words which are not in the title, this way search engines of the web will find your manuscript with higher probability.

We thank the reviewer for the helpful suggestion and have changed the keywords accordingly

6. Recheck the cited references and also references list is not formatted according to journal criteria.

We have checked through the manuscript and changed any references which do not meet the journal criteria

Author Response File: Author Response.docx

Reviewer 4 Report

Dear authors,

It was a pleasure reading your manuscript. Please see my review in the attached doc. I sincerely hope it contributes to improve the manuscript. I still think it needs a bit of work before being accepted for publication.

Best regards

Comments for author File: Comments.pdf

Author Response

Referee 4:

Comments and Suggestions for Authors

This manuscript describes an alternative way to derive LST from MODIS (with respect to the official product) as well as a framework to describe its uncertainty. The latter is extended from established previous works in the literature and also systematically separated into random, locally correlated and systematic components. The major advance proposed here is a methodology to validate these uncertainty estimates. There is an urgent need for this kind of methodologies in the remote sensing community, since most satellite derived products are distributed with an associated error bar but there is no well-established way to assess the validity of those error bar estimates. This is therefore a much appreciated effort from the authors to attempt fill in this gap in the literature. However, I have a few remarks that I would like to be addressed before the manuscript is acceptable for publication.

We would like to thank the referee for the very helpful comments, to which we reply in detail below.

Comments:

The emissivity database that is used is not updated in NRT with changes in vegetation. Is this considered in the uncertainty scheme?

We thank the reviewer for noting this point. While the emissivity database is not updated in NRT we perform temporal interpolation between monthly emissivity estimates to better represent the emissivity on any given date. We acknowledge that there will be differences to “real” data on a given date. These are likely to manifest themselves more at the sub-pixel level. This is important because these are random components that reduce when propagated. Nevertheless, there may be some additional unknown error that is not accounted due to the temporal estimation. This will be investigated in future work.

Line 173 – I understand the advantages of increasing the number of cases used for calibration, which is most likely why RTTOV was used instead of a line-by-line reference forward model. You claim that the biases between both approaches are low, based in [35]. However, that study focuses on AATSR, so I wonder if this assumption holds when a wider swath width instrument such as MODIS (and hence with larger viewing angles) is used. In fact, you don’t explain how you take into account viewing angle in the calibration database, since you refer that RTTOV is only “nadir-viewing”.

We use the term “nadir-viewing” as opposed to sensors which are limb sounding for example. For AATSR on Envisat and SLSTR on Sentinel-3, both of which have dual views, there is one view which is defined as “nadir” but with viewing angles varying between 0 and ~22 for AATSR, and for a wider swath between 0 and ~55° for SLSTR. RTTOV is used to determine both retrieval coefficients and cloud probabilities for the wider swath operational SLSTR LST with good validation results [1] and has been successfully applied to LST retrieval for VIIRS with good validation results [2]. In RTTOV radiances can be modelled for varying satellite zenith angles, and in our derivation of the coefficients we stratify based both on water vapour and satellite zenith angle. In other words coefficients are determined by setting different viewing angles in the RTTOV configuration up to 60-65° for the end of a MODIS scan line. We have added more explanation of how we use the database for large viewing angles in the text.

Line 240 – double use of “primarily” sounds weird.

We have corrected this sentence to remove the repetition.

Equation 3 – F needs to be defined and explained in the text, as well as ?11 and ?12.

The notation in Equation (3) is now better defined in the text.

Line 276: correct reference to equations to (7) and (8).

The equations have now been correctly referenced in the text.

Line 287 – Replace equation numbers to (10) and (11)

The equations have now been correctly referenced in the text.

I think there is something wrong with the derivation of (10) and (11), I am getting different results. Could you provide a derivation offline or even in an annex to the manuscript? please double-check those factors of “2”. For example, for me, the first term in (10) yields −?24(?11+?12)/(?11+?12)2. I wonder what would be the effect of an error here in the following results.

Co-authors have re-derived the equations and we indeed find the e11 and e12 terms in the numerators for the second and fourth terms in Equation 10 should be swapped over in the manuscript. This has now been amended. We find Equation 11 to be correctly formulated in the manuscript. We have subsequently gone back to the software in which the equations are coded and find the code is correct with the updated formulae in the manuscript. Thus there is no impact on the results and we conclude that this must have been an inaccurate writing of the formulae from the code.

Lines 315-317: The treatment of the uncertainty due to atmospheric profile is rather simplistic. Actually, you do not provide the source of the data you use as input to decide which GSW class to use in the operational retrieval. You say you used ERA-Interim profiles for model calibration (Line 178), but nothing about the retrieval itself. If this is also the source to input on the retrieval, then some sort of measure of uncertainty of TCWV in ERA-Interim should be used to account for the uncertainty due to this input.

ECMWF ERA-Interim daily fields are used for the source of the water vapour in the LST retrieval. These data are not used explicitly in the retrieval Equation 1, but instead to determine which band of coefficients to apply. Equation 12 presents the uncertainties from water vapour error effects categorised by each band of coefficients. This uncertainty is the standard deviation of the difference between the input surface temperatures into the model-fitting and the simulated-retrieved output. What is missing in the original text is clarification that the simulated-retrieved output is actually calculated by using perturbations to the water vapour to quantify the uncertainty as a result of imperfect knowledge on the water vapour field. These perturbations are performed within error ranges of water vapour determined from comparison of the input ECMWF water vapour with respect to in situ measurements from ARM sites. A new Figure 1 is included to show the error ranges used in the perturbations. Thus the impact of locally correlated error effects from this source on the selection of retrieval coefficients is wrapped up in this existing Equation 12. Attempting to add an additional component here at the retrieval level would be double-counting these uncertainties. We have added a better explanation of this in the text.

Line 360 – reference do not conform to the journal template

This has now been corrected.

Figure 1- Is the cloud mask applied to these figures?

The cloud mask is not applied to the data in Figure 1. In the output Level-2 product the retrieval is applied to all pixels and the cloud mask is stored as a separate field to be applied by the user. Clarification is now provided in the Figure caption. We apply the cloud mask when producing the Level-3 data as displayed in Figure 5.

Please include the equation to estimate in-situ LST and how you estimate broadband emissivity for each site.

LST from SURFRAD is derived from measurements of broadband radiances, and the associated broadband emissivities (BBE). We estimate the BBE using the corresponding monthly emissivities of the CIMSS Baseline Fit Emissivity Database (http://cimss.ssec.wisc.edu/iremis/) applied in the linear equation given by [3]:

                                             

We then use the BBE following the Stefan-Boltzmann law to convert the measured upwelling and downwelling radiances to in situ LST using the following formula:

Where  is the upwelling radiance,  is the upwelling radiance, and  is the Stefan-Boltzmann constant. This is the same method as used in previous validation studies with SURFRAD such as [4, 5]. We have added these derivations to the text.

Figure 3 – Did you use any kind of outlier filter for these scatterplots? If that is the case, please specify which filter. Also the sampling for the GT products seems much higher, could you include the number of observations in each comparison?

We used a 3-sigma Hampel filter using robust statistics [6] to limit the number of outliers primarily caused by undetected clouds. This typically reduces number of matches by < 10% and has been used is previous LST validation studies such as [7, 8]. We have updated the text to indicate this. We apply a user-defined selection of the individual cloud flags in the MOD35_L2 and MYD35_L2 products. Experience indicates these maintain conservative masking where required but reduces over-masking. This is the reason for the difference in number of matchups for the GlobTemperature product compared with the operational product. Clarification together with the number of matchups are provided in the text and table respectively.

Line 461 – there seems to be a typo at the end of the line /paragraph.

This has now been corrected.

Line 476 – becoming remaining?

Corrected to “remaining”.

What happens for ?????? values between 0 and 1 (i.e. why did you draw horizontal dashed lines in this interval? Also, if your assumptions are correct, shouldn’t the sizes of the red bars increase towards higher ??????? This only happens in Penn State University.

The horizontal dashed lines represent the component of σ_total that cannot be reduced further, specifically the uncertainty due to the calibration on the in situ instrumentation σ_ground. This is estimated to be 0.3 K for the instruments at the SURFRAD sites. We have therefore adjusted Figure 4 to better represent this buy reducing the horizontal line to between 0 and 0.3 K. The equivalent change is made to the vertical axis also. We have now explained this in the text. For many sites the bars tend to flatten out as σ_total increases. For different conditions there is a change in σ_total as a result of both changes in σ_sat and σ_space. As such, we see σ_total can increase towards 2 K. However, the standard deviation of the difference between the satellite LST and in situ LST (σ_sat-ground) does not increase much above 1.5 K. The implication here is that there is an overestimate in σ_total, although such an overestimate is only for higher bars with much lower frequencies and for the main part of the pdf there is a good fit between σ_sat-ground and σ_sat. More explanation is provided in the text.

Eq 16- correct “afc” to “sfc”

The equation has been corrected.

Line 525 – correct equation 15 to 16.

The equations have now been correctly referenced in the text.

Figure 5 – The uncertainty due to emissivity seems rather low, otherwise the Sahara region would show higher values.

The uncertainty due to emissivity is split into two separate components: i) random effects which relate to the sub-pixel variations in emissivity; and ii) locally correlated effects on the scales of variability in land cover. Figure 5 illustrates the Level 3 uncertainty estimates, in this case at a spatial resolution of 0.05°. In the propagation from Level 2 to Level 3 the random component of the uncertainty due to emissivity is reduced by 1/√N where N is the number of 1 km pixels averaged in a 0.05° grid cell. Thus one of the two emissivity components reduces towards zero in the propagation. Over much of the land surface, for the example date in Figure 5, the reduction in random components from the Level 2 is offset by the additional sampling term in Equation 16. For the Sahara which is principally cloud free this additional term is also therefore near zero for most grid cells.

Line 576 – This sentence seems too optimistic to me. Figure 4 actually shows that at least a fraction of the cases classified as having low total uncertainty should have higher uncertainties, so something must be missing or some parameter assumption must be incorrect in the calculation of ??????.

The authors recognise that not every possible error source will have been captured, which is why we discuss other possible sources in Section 2.5. In addition, the overestimation at some sites, although with lower frequencies, could be the result of inaccurate upscaling. Both these points are to be investigated in future studies. While overall across the sites agreements are good, the sentence in question has been reworded to account for these points.

 

References

[1]          Ghent, D. S3 Validation Report – SLSTR. Technical Report S3MPC.UOL.VR.029 - i1r0 - SLSTR L2 Land Validation Report.docx available at https://sentinels.copernicus.eu/web/sentinel/technical-guides/sentinel-3-slstr/cal-val-activities/validation. 2017.

[2]          Islam, T., Hulley, G.C., Malakar, N.K., Radocinski, R.G., Guillevic, P.C., & Hook, S.J. A Physics-Based Algorithm for the Simultaneous Retrieval of Land Surface Temperature and Emissivity From VIIRS Thermal Infrared Data. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55, 563-576.

[3]          Cheng, J.; Liang, S.; Yao, Y.; Zhang, X. Estimating the Optimal Broadband Emissivity Spectral Range for Calculating Surface Longwave Net Radiation. IEEE Geosci. Remote Sensing Lett. 2013, 10, 401–405. doi:blackhttps://doi.org/10.1109/lgrs.2012.220636710.1109/lgrs.2012.2206367.

[4]          Li, S.; Yu, Y.; Sun, D.; Tarpley, D.; Zhan, X.; Chiu, L. Evaluation of 10 year AQUA/MODIS land surface temperature with SURFRAD observations. Int. J. Remote Sens. 2014, 35, 830–856.

[5]          Martin, M.A.; Ghent, D.; Pires, A.C.; Göttsche, F.-M.; Cermak, J.; Remedios, J.J. Comprehensive In Situ Validation of Five Satellite Land Surface Temperature Data Sets over Multiple Stations and Years. Remote Sens. 2019, 11, 479.

[6]          Pearson, R.K. Outliers in Process Modeling and Identification. IEEE Trans. Control Syst. Technol. 2002, 10, 55–63.

[7]          Göttsche, F.M.; Olesen, F.S.; Bork-Unkelbach, A. Validation of land surface temperature derived from MSG/SEVIRI with in situ measurements at Gobabeb, Namibia. Int. J. Remote Sens. 2013, 34, 3069–3083.

[8]          Göttsche, F.-M.; Olesen, F.-S.; Trigo, I.F.; Bork-Unkelbach, A.; Martin, M.A. Long Term Validation of Land Surface Temperature Retrieved from MSG/SEVIRI with Continuous in-Situ Measurements in Africa. Remote Sens. 2016, 8, 410.

Author Response File: Author Response.docx

Round 2

Reviewer 4 Report

Dear authors,

Please find my new comments in the attached document. I am afraid the manuscript still needs some before it is acceptable for publication.

Comments for author File: Comments.pdf

Author Response

All the authors would like to thank the reviewer for taking the time to work through in detail all our responses to the first review, we appreciate the very useful feedback. We have added our new responses to all the points where a response is required in blue in the attached PDF.

Author Response File: Author Response.pdf