Assessment of Land Surface Temperature Estimates from Landsat 8-TIRS in A High-Contrast Semiarid Agroecosystem. Algorithms Intercomparison

Round 1

Reviewer 1 Report

This manuscript used Landsat 8 TIRS data to compare seven LST retrieval algorithms in semiarid regions, and all of the SC algorithms and SW algorithms performed well at the study regions. It is useful for the research of LST retrieval, but the structure and description of the manuscript should be adjusted and supplemented in order to its greater influence. Therefore, several problems should be improved by the following suggestions:

• Please introduce the new SW_CLAR algorithm proposed in this manuscript in detail, such as the process of the algorithm, and it is not represented just with Eq. (8). Besides, it is necessary to explain the radiative or stray-light correction of band 11 and make the uncertainty analysis if it is used to retrieve LST with a new algorithm. Table 3, the results of SC algorithms are slightly more accurate than SW algorithms, whether it is related to the existing error in band 11.
• Page 6, lines 216-217, L8_ST product is available with a 16-day period.
• The Landsat 9 is not relevant in this work, and please explain why it is always involved in this manuscript. (Page 2, line 52; Page 6, lines 217-218..) In contrast, it is the Sentinel 2A RGB bands but not the Landsat 8 RGB images are used in this study to show the region, please explain it.
• Please summarize the similarities and differences of the seven algorithms, and discuss the reasons they are chosen. As table 3, the SW_CLAR algorithm proposed in this work performed without obvious superiority, why it is included in this study?
• The Landsat 8 TIRS has already been resampled to 30 m, and the Landsat C2 LST product is released with 30 m spatial resolution, but the manuscript introduced the Landsat 8 data with 100 m (Page 2, lines 50-52), what spatial resolution has been used in this study?
• It is not suitable to only introduce the atmospheric correction in section 2.1 about the retrieval algorithms, please describe them in more detail.
• Page 4, lines 171-172, and Fig.1, the unit of water vapor content “cm” should be corrected.
• Page 8, lines 283-284, “…a patchy flat cropland area…”, Page 14, lines 459-461, “…a heterogeneous cropland area…”, please check the description which is accurate for the study region.
• Section 4.3, a sensitivity analysis of emissivity should be further showed that how it affects the retrieval LST.
• Some related references could be cited in this manuscript, e.g., Duan et al. (2019), Land-surface temperature retrieval from Landsat 8 single channel thermal infrared data in combination with NCEP reanalysis data and ASTER GED product, International Journal of Remote Sensing, 40(5-6), 1763-1778.

Author Response

Reviewer 1

Comments from the reviewer are in Italic and our replies are straight below in normal font.

The authors would like to thank this reviewer for his/her comments/suggestions. His/her constructive revision has contributed to an improvement of the manuscript in many aspects. See below point-by-point detailed responses, with indication of the changes made.

This manuscript used Landsat 8 TIRS data to compare seven LST retrieval algorithms in semiarid regions, and all of the SC algorithms and SW algorithms performed well at the study regions. It is useful for the research of LST retrieval, but the structure and description of the manuscript should be adjusted and supplemented in order to its greater influence. Therefore, several problems should be improved by the following suggestions:

• Please introduce the new SW_CLAR algorithm proposed in this manuscript in detail, such as the process of the algorithm, and it is not represented just with Eq. (8).

Following this referee comment, a more comprehensive description of the SW_CLAR algorithm has been added (see new Section 2.1.6): “This database contains 382 radiosounding profiles over land, with nearly uniform distribution of precipitable water between 0.02 and 5.5 cm. The sondes are distributed in three latitude ranges, with around 40% of radiosoundings placed at low latitudes (0◦–30◦), an-other 40% at middle latitudes (30◦–60◦), and 20% at high latitudes (> 60◦). The temperature of the first layer of the radiosoundings (T0) ranges from −20 ◦C to 40 ◦C. The setup of the simulations performed consider the simulated ground T following a Gaussian distribution as T0 − 6, T0 −2, T0 + 1, T0 + 3, T0 + 5, T0 + 8, and T0 + 12. Radiative transfer calculations were performed with the MODTRAN 5 code for two viewing angles in this case, nadir and the Gaussian angle 11.6 ̊ [46], sufficient to account for the 15 ̊ field of view of L8. The resulting radiance spectra were convoluted with the response filter functions of L8-TIRS bands 10 and 11, and coefficients in Eq. (8) were derived”

Besides, it is necessary to explain the radiative or stray-light correction of band 11 and make the uncertainty analysis if it is used to retrieve LST with a new algorithm. Table 3, the results of SC algorithms are slightly more accurate than SW algorithms, whether it is related to the existing error in band 11.

We are aware of this referee concern. Nevertheless, note the new Collection 2 Level 1 and Level 2 L8 products were used in this work. This new Landsat Collection 2 already implements updated radiometric calibration of both Landsat-8 TIRS bands, besides the stray light correction (https://www.usgs.gov/landsat-missions/landsat-8-oli-and-tirs-calibration-notices). Previous studies (Guo et al., 2020; Gerace et al., 2017; Meng et al., 2019; Yu et al., 2014) evaluated the change in TIRS TOA radiances after the stray-light correction in the 2017 reprocessing. In a more recent paper, Niclòs et al. (2021), evaluated the performance of successive re-calibrations implemented in L8-TIRS since launch, using a robust and accurate multi-year set of reference ground data. Negligible biases and RMSD<1.5 K were observed when using the TIRS data in previous Collection 1 (i.e., data after the 2017 reprocessing), and results even improved when using the calibration updates included in new Collection 2, using both SC and SW algorithms.

Meng, X., Cheng, J., Zhao, S., Liu, S., Yao, Y., 2019. Estimating Land Surface Temperature from Landsat-8 Data using the NOAA JPSS Enterprise Algorithm. Remote Sens. 11, 155.

Yu, X., Guo, X., Wu, Z., 2014. Land surface temperature retrieval from Landsat 8 TIRS—Comparison between radiative transfer equation-based method, split window algorithm and single channel method. Remote Sens. 6, 9829–9852.

As stated at the end of Section 5.3: “Overall, estimation errors using SW formulation do not differ significantly from those applying SC algorithms in terms of accuracy, although a slight increase in the BIAS is noticed”. However, we believe a more robust and accurate ground dataset would be necessary to evaluate the relation between this bias and eventual remaining errors in band 11, and this is beyond the scope of this research focused on high-contrast agroecosystems.

• Page 6, lines 216-217, L8_ST product is available with a 16-day period.

This information referred to the processing time for the L8_ST product to be delivered after acquisition, not to the L8 revisit time. However, thanks to this comment we realized this might be confusing and has been modified in the revised version as follows: “The L8_ST product is available with a 16-day period…”

• The Landsat 9 is not relevant in this work, and please explain why it is always involved in this manuscript. (Page 2, line 52; Page 6, lines 217-218..) In contrast, it is the Sentinel 2A RGB bands but not the Landsat 8 RGB images are used in this study to show the region, please explain it.

We agree with the referee that L9 is not relevant in this work. So, references to L9 have been removed from the text in page 2 and 6 to avoid distraction.

• Please summarize the similarities and differences of the seven algorithms, and discuss the reasons they are chosen. As table 3, the SW_CLAR algorithm proposed in this work performed without obvious superiority, why it is included in this study?

Following this comment, a new paragraph has been added at the end of section 2.1 [225-243] to  summarize the differences and similarities of the algorithms, and justify their selection: “Beyond the obviousness that some of the algorithms described above work with a single or both Landsat 8 thermal bands, there are some common features, and also differences between them. For instances, the three SW algorithms adopt a second-order polynomial for the temperature difference between bands although different coefficients. SC_JM algorithm also adopts a quadratic relation with the brightness temperatures, whereas the SBAC and L-SBAC follow the radiative transfer equation. Another similarity between models is that they all depend on the surface emissivity. While emissivity information for a single band (10 or 11) is required for SC algorithms, SW approaches need the emissivity in both spectral ranges to run. The main differentiating feature between the algorithms is the way the atmospheric correction is conducted. All the algorithms, except SW_Du, need the water vapor content as an input, but the way this is estimated and implemented varies. SW_JM considers a linear dependence with w whereas SC_JM and SW_CLAR opt for a second-order polynomial. The strength of SBAC is its capability to account for the spatiotemporal and orography variability within a scene, although its operational application may result very computational demanding. A shortcut is introduced with L-SBAC with direct linear dependence again on w data. In summary, the six LST algorithms selected for this study range a variety of approaches, some of them well-known and widely used (SC_JM, SW_JM or SW_Du) and other more recently introduced (SBAC), besides the new L-SBAC or SW_CLAR explored in this work.”

It is truth that performance of the SW_CLAR is very similar to the other algorithms. However, we decided to include it for consistency since CLAR is the database used to derive the coefficients in SC_SBAC.

 

• The Landsat 8 TIRS has already been resampled to 30 m, and the Landsat C2 LST product is released with 30 m spatial resolution, but the manuscript introduced the Landsat 8 data with 100 m (Page 2, lines 50-52), what spatial resolution has been used in this study?

In this study the 30-m data have been used for all radiances and LST products. Although the original spatial resolution of the thermal bands is 100 m, the products are released with 30-m pixel size. A sentence has been added to clarify this point. ([368-369]).

• It is not suitable to only introduce the atmospheric correction in section 2.1 about the retrieval algorithms, please describe them in more detail.

This section 2.1 has been completed with additional material. Details on the CLAR database, and a summary of similarities and differences between models have been added (see response above). For further details on the previously published approaches, the reader is referred to the literature.

• Page 4, lines 171-172, and Fig.1, the unit of water vapor content “cm” should be corrected.

Traditionally, “cm” is the unit used for the water vapor content in works dealing with radiosounding datasets or atmospheric correction algorithms (see for instance Galve et al., 2008; Galve et al., 2018; Coll et al., 2010…). For this reason, the authors prefer to maintain this unit here.

• Page 8, lines 283-284, “…a patchy flat cropland area…”, Page 14, lines 459-461, “…a heterogeneous cropland area…”, please check the description which is accurate for the study region.

The referee is right the word “patchy” is not accurate for our study region and has been removed from the text. The terms “high-contrast” or “heterogeneous” describe more accurate the crop types variability in the area.

• Section 4.3, a sensitivity analysis of emissivity should be further showed that how it affects the retrieval LST.

Following this referee suggestion, a sensitivity analysis of the effect of emissivity on LST retrieval was conducted, and summarized at the end of new Section 5.2: “It is well known that minor differences in emissivity of 1% may lead to significant deviations in LST up to 2 K. A sensitivity analysis was carried out to quantify the effect of emissivity deviations in the LST retrieval. A discrepancy of 0.5% in emissivity yields an LST uncertainty of ± 0.5K for L8_ST and ± 0.4K for T10_LS. However, if the discrepancy increases up to 2%, as reported above for the emissivity values in the L8_ST product, deviation in LST reaches ± 2K for L8_ST and ± 1.8 K for T10_LS. Therefore, this underestimation in the emissivity values justifies the overestimation in LST values of the L8_ST product showed above. A recalculation of these LST values using our emissivity estimates, reduces the underestimation of the L8_ST product to 0.2 K, resulting in a final MAE of ±1.5 K”.

• Some related references could be cited in this manuscript, e.g., Duan et al. (2019), Land-surface temperature retrieval from Landsat 8 single channel thermal infrared data in combination with NCEP reanalysis data and ASTER GED product, International Journal of Remote Sensing, 40(5-6), 1763-1778.

Thanks for this missing reference. This has been added to the reference list and referred in the text.

Author Response File: Author Response.doc

Reviewer 2 Report

This paper evaluates several algorithms for LST retrieval from Landsat 8 in an agricultural area. The paper includes multiple information of interest however I think there are two many objectives that the authors would like to achieve in this paper, which makes it confusing. At the same time none of the objectives is fully achieved because there aren’t enough details on each. I will try to explain my view in the following comments:

1) one of the purposes of the paper is to inter-compare different algorithms. However, this in situ dataset is not appropriate for such assessment since the data only covers very dry atmospheres (column water vapour between 1.5 and 2.3 cm as mentioned in line 332). A complete inter-comparison would have to incorporate different atmospheric and surface conditions in order to provide better conclusions on the relative quality of each algorithm. A full sensitivity analysis to the algorithm inputs would also be recommended in this case. The authors obtained similar statistics for the SC algorithms using band 10 and 11, but I suspect that under higher water vapour concentrations the ones using band 11 would necessarily have lower accuracies. In fact, such assessment could easily be performed using the training database used to calibrate the models. From my understanding, the authors also did not calibrate all algorithms with the same training database, which does not yield a fair comparison.

2) On the other hand, the dataset of ground measurements is very interesting. It is a very unique dataset since there is rarely data available for the derivation of both LST and emissivity on validation sites. However, the information the authors provide on this dataset is very limited. I don’t see that there is any reference to a paper where the data might be better described so I think it would be most appropriate to do it in this paper. It would have been interesting to have more details on how the emissivities where derived for each land cover, specially for tree cover that is much more complex. It would also have be very interesting to see how the Landsat derived LSTs perform over each one of these land covers.

3) I’m not sure that the introduction of a new methodology to perform the atmospheric correction is appropriate here as well. To fully assess the quality of such new method the paper would have required a much more comprehensive study, again including multiple atmospheric conditions and appropriate sensitivity studies.

Based on these two comments, I would recommend restructuring the paper to focus on the validation over this unique site. Include a detailed description of the derivation of the ground LSTs and emissivities and then focus the validation exercise on 1 or 2 algorithms (maybe one SC and one SW plus the operational product). Focus also on the different approaches used to derive the emissivity in the different products. It would also be interesting to understand how the SC compares to the SW approaches in terms of sensitivity to emissivity errors. The authors could highlight the problem of emissivity estimation and validation and that would be by itself a significant contribution. I will suggest rejection just because I don’t think a major revision would give the authors enough time to apply these changes.

I think that the introduction of the linear SBAC is also of interest since it speeds up the SBAC approach. I think it should be presented in a different paper, with more focus on the derivation of the approach, quality over different atmospheres, with complete sensitivity analysis and using a reference database derived with a radiative transfer model such as MODTRAN or RTTOV.

Minor comments:

Section 4: please limit the statistics to the bias, std and RMSE. Having so many statistics is confusing and has no added value in the assessment.

Table 3: I find it strange that T11_LS and L8_ST have such different biases but the same std and RMSE. Please confirm your computations.

Author Response

Reviewer 2

Comments from the reviewer are in Italic and our replies are straight below in normal font.

The authors would like to thank this reviewer for his/her comments/suggestions. His/her constructive revision has contributed to an improvement of the manuscript in many aspects. See below point-by-point detailed responses, with indication of the changes made.

This paper evaluates several algorithms for LST retrieval from Landsat 8 in an agricultural area. The paper includes multiple information of interest however I think there are two many objectives that the authors would like to achieve in this paper, which makes it confusing. At the same time none of the objectives is fully achieved because there aren’t enough details on each. I will try to explain my view in the following comments:

1) one of the purposes of the paper is to inter-compare different algorithms. However, this in situ dataset is not appropriate for such assessment since the data only covers very dry atmospheres (column water vapour between 1.5 and 2.3 cm as mentioned in line 332). A complete inter-comparison would have to incorporate different atmospheric and surface conditions in order to provide better conclusions on the relative quality of each algorithm. A full sensitivity analysis to the algorithm inputs would also be recommended in this case. The authors obtained similar statistics for the SC algorithms using band 10 and 11, but I suspect that under higher water vapour concentrations the ones using band 11 would necessarily have lower accuracies. In fact, such assessment could easily be performed using the training database used to calibrate the models. From my understanding, the authors also did not calibrate all algorithms with the same training database, which does not yield a fair comparison.

We understand this referee concern. However, note the intercomparison was not among the main aims of this work. As stated at the end of the Introduction the objectives are:

“- To assess the performance of traditional LST algorithms applied to L8/TIRS after the recalibration implemented in new Collection 2 under the conditions of a semiarid agroecosystem.

- Explore the feasibility of the new LST Level 2 operational product for agricultural applications in our site.

- Introduce a new procedure to derive atmospheric correction parameters, as part of a novel algorithm to estimate LST from L8/TIRS, reducing processing time and improving accuracy.”

This paper aims at showing the performance of the LST estimates from L8 under the specific conditions of our semiarid site, which are quite common for vast areas in the Southeastern Spain, and also focusing on agricultural targets for the further implication of LST in water management tasks. In this context we had the need to report the results we are obtaining using the algorithms derived from the CLAR dataset and the SBAC technique (as previously done with L7 (Galve et al. 2018)). Nevertheless, we considered the interest of the manuscript would be reinforced by testing also a set of other well-known SC and SW algorithms, using the coefficients originally proposed by the respective authors.

We agree with the referee that a paper focused on a pure intercomparison of algorithms would require a larger variety of atmospheric conditions and even more homogeneous surfaces (as recently done in Niclòs et al., 2021, for instances). But, as mentioned above, this is not the primary aim of this study.

To avoid misunderstanding and clarify this point, several sentences have been inserted in the manuscript, at the beginning of the Discussion section: “…This study does not intends to be a pure intercomparison of algorithms that would require a larger variety of atmospheric conditions and even more homogeneous surfaces…”, or at the end of the Discussion section: “…The performance of L-SBAC must be further evaluated under a larger variety of environmental and surface conditions in future works.”

2) On the other hand, the dataset of ground measurements is very interesting. It is a very unique dataset since there is rarely data available for the derivation of both LST and emissivity on validation sites. However, the information the authors provide on this dataset is very limited. I don’t see that there is any reference to a paper where the data might be better described so I think it would be most appropriate to do it in this paper. It would have been interesting to have more details on how the emissivities where derived for each land cover, specially for tree cover that is much more complex. It would also have be very interesting to see how the Landsat derived LSTs perform over each one of these land covers.

We agree with the referee ground data registered in this experimental site provides a unique dataset for the evaluation of LST estimates from Landsat. We have been involved in several experimental campaigns to collect these ground LST datasets in the Barrax site in the last decade. Some of these datasets (2015-2016) have been used in previously published works such as Bisquert et al. (2016) in RSE, or Galve et al. (2018) in RS, where an in-depth description of the measure protocol or the IRTs instrumentation was included. Since the ground dataset used in the present work (2018 and 2019) is new and has not been published before, and following this referee suggestion, a new Table 3 has been inserted listing detailed information on these ground LST measurements.

Also, thanks to this referee comment, detailed results on the performance of L8 derived LST over the different land covers have been added. See new Table 5 and related discussion in the text.

3) I’m not sure that the introduction of a new methodology to perform the atmospheric correction is appropriate here as well. To fully assess the quality of such new method the paper would have required a much more comprehensive study, again including multiple atmospheric conditions and appropriate sensitivity studies.

As mentioned above, we are aware of this referee concern, and we do agree a full assessment of the performance of an algorithm would require a more comprehensive analysis and to complete the dataset. However, our main purpose is the evaluation of the LST estimates for the specific conditions of our study site (high-contrast semiarid agroecosystem), and this is the context in which we introduce and test the new methodology (L-SBAC). We are aware of the necessity to further evaluate the performance of the model under a wider variety of environmental and surface conditions before it can be applied to other areas or atmospheric conditions. This has been clearly stated at the end of the Discussion: “The performance of L-SBAC must be further evaluated under a larger variety of environmental and surface conditions in future works.”

Based on these two comments, I would recommend restructuring the paper to focus on the validation over this unique site. Include a detailed description of the derivation of the ground LSTs and emissivities and then focus the validation exercise on 1 or 2 algorithms (maybe one SC and one SW plus the operational product). Focus also on the different approaches used to derive the emissivity in the different products. It would also be interesting to understand how the SC compares to the SW approaches in terms of sensitivity to emissivity errors. The authors could highlight the problem of emissivity estimation and validation and that would be by itself a significant contribution. I will suggest rejection just because I don’t think a major revision would give the authors enough time to apply these changes.

We appreciate your suggestions, but as mentioned above this would be another paper. In a further work we will focus on the emissivity issue, looking into details and evaluating emissivity estimates using ground measurements with the multispectral thermal radiometer CIMEL CE-312. But, now for this work we prefer to reserve this discussion so not to distract the reader from the main goal. Nevertheless, some more details about emissivity have been added, extracted values for each date/site have been listed in a new Table 3, and a sensitivity analysis of the effect of deviations in emissivity on LST retrievals has been conducted (see last paragraphs in Section 5.2).

I think that the introduction of the linear SBAC is also of interest since it speeds up the SBAC approach. I think it should be presented in a different paper, with more focus on the derivation of the approach, quality over different atmospheres, with complete sensitivity analysis and using a reference database derived with a radiative transfer model such as MODTRAN or RTTOV.

Again, we thank the referee for his/her valuable suggestion, but still we consider worth it maintaining the evaluation of the L-SBAC in this work, being aware of the specific conditions in which it has been applied and tested. This has been clearly stated in the text throughout the manuscript, for instances at the end of the Discussions “…the results using the simplified and faster version of the SBAC are promising. L-SBAC could be then implemented to generate time series of LST images from L8 in semiarid agroecosystems at a near Real-Time, saving computational resources. The performance of L-SBAC must be further evaluated under a larger variety of environmental and surface conditions in future works”. Of course, in a future work L-SBAC should be evaluated under a larger variety of environmental and surface conditions.

Minor comments:

Section 4: please limit the statistics to the bias, std and RMSE. Having so many statistics is confusing and has no added value in the assessment.

We totally agree with the referee at this point, and statistics have been reduced to the bias, std, RMSE and MAE. No added value in the assessment for the rest. Slope and interception for the different linear regressions have been added instead for completeness.

Table 3: I find it strange that T11_LS and L8_ST have such different biases but the same std and RMSE. Please confirm your computations.

We have double-checked our computations and can confirm results are correct. The reason is that L8_ST shows a lower scatter compared to T11_LS and average uncertainty (std or RMSE) becomes similar although the systematic deviation (bias) is larger.

Author Response File: Author Response.doc

Reviewer 3 Report

In this paper, the authors assessed various sets of SC and SW algorithms in the context of L8/TIRS sensors at the given site.  The study also includes a new SC algorithm that seems to do well at the agro-ecosystem site (table 2). The study is well thought out, and hard work has gone into creating the paper. However, there are a few concerns that I have.The paper is confused between proposing the new algorithm versus comparing 10 different ones. Given that the author is a respected name in the community, this presents a confusing perspective to the readers as to the utility of the paper and the new Landsat algorithms. Shall we just adapt equation 8 in general for various parts of the world? How well does it fair? Or is it only applicable for the site in Spain?  How does the retrieval behave for various humidity conditions? Is it possible that the emissivity retrievals (figs 5 and 6) are influenced by the water vapor as the major bias is due to the L(down) component? The answer to this question can be tested by using one of the reliable surfrad sites. Figure 2: since the NDVI may not fully represent the at-site conditions, depict the time series using emissivity for the relevant bands (T10/T11). Please indicate the dates so that it is easier to navigate for the readers. Also, since the agricultural site may not be fully homogeneous, did you take the average of the field or the single-pixel value? Also, I am not sure that the high-contrast semiarid agro-ecosystem poses any challenge to the problem at hand. Please include the LST images (if the paper becomes too long, use the supplemental info section). An LST paper without a single LST image makes me nervous. Thank you.

Author Response

Reviewer 3

Comments from the reviewer are in Italic and our replies are straight below in normal font.

The authors would like to thank this reviewer for his/her comments/suggestions. His/her constructive revision has contributed to an improvement of the manuscript in many aspects. See below point-by-point detailed responses, with indication of the changes made.

 

In this paper, the authors assessed various sets of SC and SW algorithms in the context of L8/TIRS sensors at the given site.  The study also includes a new SC algorithm that seems to do well at the agro-ecosystem site (table 2). The study is well thought out, and hard work has gone into creating the paper. However, there are a few concerns that I have. The paper is confused between proposing the new algorithm versus comparing 10 different ones. Given that the author is a respected name in the community, this presents a confusing perspective to the readers as to the utility of the paper and the new Landsat algorithms. Shall we just adapt equation 8 in general for various parts of the world? How well does it fair? Or is it only applicable for the site in Spain?  How does the retrieval behave for various humidity conditions?

A new paragraph with more details on the procedure to derive Eq. (8) has been added to Section 2.1.6. Equation (8) was derived using the full CLAR database (Galve et al. 2008), with values of W ranging 0-5 cm. So, the coefficients in Eq.(8) are valid for a wide variety of conditions all around the world. Another question is whether this performs equal for other sites or different humidity conditions, and this is a good point. To answer this question, ground data under a larger variety of atmospheric conditions and even more homogeneous surfaces (as recently done in Niclòs et al., 2021, for instances) would be necessary. But note the objective of this work aims at exploring the performance of LST retrievals from L8 under the specific conditions of our semiarid agroecosystem which are quite common for vast areas in the Southeastern Spain, and also focusing on agricultural targets for the further implication of LST in water management tasks. In this context we had the need to report the results we are obtaining here using the algorithms derived (as previously done with L7 (Galve et al. 2018)).

Following this referee comment, several sentences have been inserted in the manuscript to clarify this point; for instances, at the beginning of the Discussion section: “…This study does not intends to be a pure intercomparison of algorithms that would require a larger variety of atmospheric conditions and even more homogeneous surfaces…”, or at the end of the Discussion section: “…The performance of L-SBAC must be further evaluated under a larger variety of environmental and surface conditions in future works.”

 

 Is it possible that the emissivity retrievals (figs 5 and 6) are influenced by the water vapor as the major bias is due to the L(down) component? The answer to this question can be tested by using one of the reliable surfrad sites. 

The emissivity retrieval is not really affected by the atmospheric correction nor the systematic deviations in the downwelling sky radiances (check Landsat 8 Collection 2 (C2) Level 2 Science Product (L2SP) Guide for details on the USGS L8 emissivity product).

In a future work we will focus on the emissivity issue, looking into details and evaluating emissivity estimates using ground measurements with the multispectral thermal radiometer CIMEL CE-312. We might also include SURFRAD sites for evaluation. But, now for this work we prefer to reserve this discussion, and focus on our dataset, so not to distract the reader from the main goal.

Figure 2: since the NDVI may not fully represent the at-site conditions, depict the time series using emissivity for the relevant bands (T10/T11). Please indicate the dates so that it is easier to navigate for the readers.

Following this referee comment, emissivity values for both bands 10 and 11 have been listed in new Table 3. Also, as suggested by the referee, dates have been used instead of the Julian day in Figure 2.

Also, since the agricultural site may not be fully homogeneous, did you take the average of the field or the single-pixel value?

3x3 pixel average values were calculated, representative of an area of 90 x 90 m², to avoid pixel singularity.

Also, I am not sure that the high-contrast semiarid agro-ecosystem poses any challenge to the problem at hand.

With this term “challenge” we meant that the assessments of algorithms usually are conducted in more homogeneous areas. Thanks to this referee comment we realized maybe this term was not appropriate and decided to remove it from the text in the revised version.

Please include the LST images (if the paper becomes too long, use the supplemental info section). An LST paper without a single LST image makes me nervous. Thank you.

We agree with the referee at least some examples of full LST images were necessary in this work to illustrate the spatial distribution of this parameter. A new Figure 5 has been added showing an example (corresponding to date 07/24/2018) of the 10 LST products for comparison and discussion. Honestly, we believe that including images for all dates would not add any value to the message of this study. However, this could be done in case the referee considers it necessary.

Author Response File: Author Response.doc

Reviewer 4 Report

This paper assessed the SC and SW algorithm for LST retrieval under semiarid conditions. My comments are below.

1. Improve the quality of figure3. Clearly indicate the geographic information of the study area and accurately describe the location of the corresponding squares (yellow and blue boxes). I also suggested to include crop type information in figure3.
2. What is the criteria for choosing a total set of 11 L8/TIRS scenes ? There are only clear-sky 11 images during growing seasons of 2018-2019?
3. How to obtain NDVI data from Landsat and Sentinel? Did you apply the atmospheric correction for reflectance values? Please describe more information about NDVI in the data section.
4. Line335-343: Move these contents to the method sections. This is the evaluation criteria, not the results.
5. Could you explain the way to match the ‘ground-measured LST’ and satellite LST in more detail? For example, how to deal with the difference of the spatial scale? there could be some bias coming from this difference.
6. Please highlight the novelty of this paper in the conclusion section more.
7. Minor comments

• Citation needed in Line 435-436
• Define ‘growing season’ for line 289-290

Author Response

Reviewer 4

This paper assessed the SC and SW algorithm for LST retrieval under semiarid conditions. My comments are below.

1. Improve the quality of figure3. Clearly indicate the geographic information of the study area and accurately describe the location of the corresponding squares (yellow and blue boxes). I also suggested to include crop type information in figure3.

We agree with the referee Figure 3 needed improvement. Lat/lon. coordiantes have been added, and an RGB image from Landsat 8/OLCI has been used instead to illustrate the location of the different cropfields (following the suggestion of another reviewer). Crop type changes from 2018 to 2019 in some fields. For this reason the authors believe is better to list this information in Table 2, together with the each date. 

1. What is the criteria for choosing a total set of 11 L8/TIRS scenes? There are only clear-sky 11 images during growing seasons of 2018-2019?

The limitation factor is the availability of ground LST measurements for the evaluation since most data are ground transects. There were a few more clear-sky images during the study period, but the dates with ground sampling available were those 11 used. This has been clarified in the text: “A total set of 11 scenes L8/TIRS were selected for this work. All these are cloud-free images with available ground data for the evaluation.”

1. How to obtain NDVI data from Landsat and Sentinel? Did you apply the atmospheric correction for reflectance values? Please describe more information about NDVI in the data section.

Thanks to this referee comment we noticed this information about NDVI was confusing. The methodology we introduce stands on NDVI values obtained from the L8-OLI Level-1 reflectance data (TOA), i.e. non-corrected reflectance values in the VNIR bands. This information has been clarified now in Section 4.1.: “using NDVI values obtained using reflectance data from L8/Operational Land Imager (OLI) red and infrared data (bands 4 and 5, respectively) provided in the Level-1 product. Note Level-1 data must be used at this step to achieve near Real-Time LST estimation.”

Although NDVI values are slightly affected by the atmospheric correction of the VNIR reflectivity values, the impact on the emissivity is very minor and can be neglected here in terms of LST deviations.

Also, additional details have been added to new version of the manuscript in Section 2.3.1: “Figure 2 shows the temporal evolution of the mean NDVI values for the selected land covers, accessible from the WebGIS platform (www.spiderwebgis.org, accessed on 10 February 2022).”  

1. Line335-343: Move these contents to the method sections. This is the evaluation criteria, not the results.

We agree with the referee at this point. See new Section “2.4. Evaluation statistics”

1. Could you explain the way to match the ‘ground-measured LST’ and satellite LST in more detail? For example, how to deal with the difference of the spatial scale? there could be some bias coming from this difference.

The matching between ground-measurements and satellite LST was explained in Section 4.1, although it is true some important details for a full understanding were indicated in Section 3. In the new version of the manuscript we have merged some information in a new sentence to clarify: “…Mean emissivity and brightness temperature values for the 3x3 pixels centered in the exact location of the ground transects were extracted from the images, to guaranty the spatial matching between ground and satellite LST data. Table 3 lists these average values together with its their standard deviation…”

As stated in Section 3: “Ground transects were conducted by carrying the radiometer back and forth,…, covering 30-50 m/min, and a total area of several hectares in each plot.” This is the traditional protocol followed in this type of studies (see for instances, Coll et al., 2010; Galve et al., 2018; Niclòs et al., 2021), and also note the standard deviation of the data (in new Table 3) accounts for the temporal (10-min) and spatial (several hectares) variability of the LST at the test site for the ground measurements, and for the spatial variability within the 3x3 pixels for the L8 data.

1. Please highlight the novelty of this paper in the conclusion section more.

Following this referee suggestion, a few sentences have been added to reinforce the newness of this work and highlight its main findings.

1. Minor comments

• Citation needed in Line 435-436

At this point, one of the referees suggested to run a sensitivity analysis to report our own findings instead. Thus, a new paragraph has been added here: “A sensitivity analysis was carried out to quantify the effect of emissivity deviations in the LST retrieval. A discrepancy of 0.5% in emissivity yields an LST uncertainty of ± 0.5K for L8_ST and ± 0.4K for T10_LS. However, if the discrepancy increases up to 2%, as reported above for the emissivity values in the L8_ST product, deviation in LST reaches ± 2.0 K for L8_ST and ± 1.8 K for T10_LS.”

• Define ‘growing season’ for line 289-290

“Growing season” is an agronomic term referring to the period of time in which most crops are grown in an area, i.e, summer season in our study site. Nevertheless, “summer” has been used instead in the new version to avoid confusion.

 

Author Response File: Author Response.doc

Round 2

Reviewer 1 Report

The authors have addressed all my concerns. I have no more comments. I would like to recommend this manuscript for publication in RS.

Author Response

The authors would like to thank this reviewer for his/her appreciated contribution 

Reviewer 2 Report

The authors have addressed all my comments. Although I still think this work would have benefited from a larger re-structuration of the manuscripts, I accept the author’s arguments. I have only some minor suggestions to the text:

Lines 225-226: Sounds a bit rude. Suggestion: “Beyond the difference in the number of thermal bands used by the different algorithms, there some…”

Line 228: suggestion: “…second‐order polynomial for the temperature difference between bands, although with different coefficients.

Line 239: “… application may be computationally demanding.”

Line 367: “A total set of 11 L8/TIRS scenes were selected…”

Line 372: “… pointing to the surface…”

Line 435: “colormap”

Line 449: “… with a standard deviation ±1.9 K, …”

Line 480: “… does not intend to be…”

Line 488: “…global long time-series of Landsat LST…”

Author Response

Reviewer 2

Comments from the reviewer are in Italic and our replies are straight below in normal font.

The authors would like to thank this reviewer for his/her edits to the text. They are very much appreciated.

The authors have addressed all my comments. Although I still think this work would have benefited from a larger re-structuration of the manuscripts, I accept the author’s arguments. I have only some minor suggestions to the text:

• Lines 225-226: Sounds a bit rude. Suggestion: “Beyond the difference in the number of thermal bands used by the different algorithms, there some…”

Changed. Thanks for this suggestion.

• Line 228: suggestion: “…second‐order polynomial for the temperature difference between bands, although with different coefficients.

Changed. Thanks for this suggestion.

• Line 239: “… application may be computationally demanding.”

Modified.

• Line 367: “A total set of 11 L8/TIRS scenes were selected…”

Corrected

• Line 372: “… pointing to the surface…”

Revised

• Line 435: “colormap”

Corrected

• Line 449: “… with a standard deviation ±1.9 K, …”

Revised

• Line 480: “… does not intend to be…”

Corrected

• Line 488: “…global long time-series of Landsat LST…”

Revised