Review Reports - Benchmarking Landsat-8 Collection 2 Level-2 Land Surface Temperature Accuracy Using SURFRAD Stations: Effects of Seasonality and Atmospheric Water Vapor

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

It is valuable for validating Landsat-8 C2L2 LST product using ground measurements, however some comments need pay attention.
1.In L43-44, the keywords should aviod repeat with title words. Please rewrite it.
2.In L98, what is the meaning of SMW? please give explantion of full words.
3.In whole manuscript, the correlation coefficient R, determination coefficient of R2 and LST et al, should maintain two decimal.
4.How select 7 stations in the validation? what is it of select principle?
5.In L321, the equation lack of number (7)
6.In fig 2, it does not demonstrate the correlation coefficient between L367-368, please add it.
7.In L200 of 139 ground measurements for seven staitons, whle appear 385 measurements in L399, what the relationship bewteen them?
8.In section 3.4, the text has not list the figure 5, please add it in available sentence.
9.In L512-514 and fig 6, please check the number of positive and negative measurements, because 127 and 258 respectively in text, but the figure legends are oppsitive.

Author Response

Response to Reviewer (1)

Comments and Suggestions for Authors:

It is valuable for validating Landsat-8 C2L2 LST product using ground measurements, however some comments need pay attention.

Comment 1: In L43-44, the keywords should aviod repeat with title words. Please rewrite it.

Response: We fully agree with the reviewer’s comment. The keyword list has been revised to complement the manuscript title rather than repeat terms already included in it. In addition, several keywords have been added to better reflect the main scientific themes of the study, including thermal infrared remote sensing, in-situ validation, atmospheric water vapor, seasonal bias, surface emissivity, and Landsat thermal Band 10.

Action Taken: The following keywords were added to the revised manuscript: Thermal infrared remote sensing; in-situ validation; atmospheric water vapor; seasonal bias; surface emissivity; thermal Band 10.

Comment 2: In L98, what is the meaning of SMW? please give explantion of full words.

Response: We agree with the reviewer that the acronym SMW should be clearly defined and its origin properly described. The full name of the algorithm, Statistical Mono-Window (SMW) algorithm, has been added and its background has been clarified in greater detail. Specifically, the SMW algorithm was originally developed by the Satellite Application Facility on Climate Monitoring (CM-SAF) to generate climate data records of land surface temperature from first- and second-generation Meteosat sensors (MFG and MSG).

Action Taken: A detailed explanatory paragraph was added in Lines 105–108 of the revised manuscript. In addition, the reference to Duguay-Tetzlaff et al. (2015) has been included in the reference. list.
Comment 3: In whole manuscript, the correlation coefficient R, determination coefficient of R2 and LST et al, should maintain two decimal.

Response: We agree with the reviewer that reporting most numerical results to two decimal places is sufficient and provides a more realistic representation of the actual measurement uncertainty. Accordingly, numerical values throughout the Abstract, Conclusion, and figure annotations have been rounded to two decimal places where appropriate (e.g., r = 0.98 instead of 0.979; R² = 0.91 instead of 0.914; RMSE = 4.20 °C instead of 4.200).

However, three decimal places were retained in the detailed statistical tables to support scientific reproducibility. The additional decimal place may be useful for researchers who wish to perform precise comparisons, replicate the analysis, or conduct follow-up calculations based on the reported statistics.

Action Taken: Numerical values were updated to two decimal places in the Abstract, Conclusion, and figure labels. Three decimal places were retained only in the detailed statistical tables for reproducibility purposes.

Comment 4: How select 7 stations in the validation? what is it of select principle?

Response: We appreciate the reviewer’s important comment regarding the selection of SURFRAD stations. We fully agree that the selection criteria should be explicitly stated to ensure transparency and reproducibility of the analysis. Accordingly, the seven stations used in this study were selected from the eight currently active SURFRAD stations based on three main criteria: (1) balanced geographic distribution to ensure representation of the major climatic and physiographic regions of the United States (including humid coastal, humid continental, cold continental, high-elevation grassland, and arid desert climates); (2) sufficient availability of Landsat-8 clear-sky matchups at each station to ensure statistical robustness; and (3) long-term data quality and continuity of thermal radiation measurements.

Action Taken: A dedicated paragraph describing the station selection criteria has been added in Lines 223–231 of the revised manuscript. In addition, each station has been explicitly classified according to its corresponding climatic zone.

Comment 5: In L321, the equation lack of number (7).

Response: We agree with the reviewer’s comment regarding equation numbering consistency. The missing equation numbering has been corrected, and all equations in the manuscript have been carefully reviewed to ensure a consistent and sequential numbering scheme.

Action Taken: Equation numbering has been corrected (Equation 7 has been added in its appropriate position), and all remaining equations have been checked and renumbered where necessary to ensure full consistency throughout the manuscript.

Comment 6: In fig 2, it does not demonstrate the correlation coefficient between L367-368, please add it.

Response: We thank the reviewer for identifying this omission. The correlation coefficient has now been explicitly included in the figure caption to provide a more complete statistical description of the relationship. In addition to Pearson’s correlation coefficient (r), we have also reported Spearman’s rank correlation coefficient (ρ), the coefficient of determination (R²), and the statistical significance level (p-value), ensuring a more comprehensive representation of the relationship between Landsat-8 land surface temperature and ground-based measurements.

Action Taken: The figure caption has been updated accordingly to:

“Figure 2: Scatter plot of the relationship between Landsat-8 surface temperatures and ground measurements from SURFRAD stations with correlation coefficients (Pearson r = 0.98; Spearman ρ = 0.97; R² = 0.91; p < 0.001).”

Comment 7: In L200 of 139 ground measurements for seven staitons, whle appear 385 measurements in L399, what the relationship bewteen them?

Response: We thank the reviewer for this careful observation and for highlighting the apparent discrepancy. The difference between the two reported numbers arises from the level of aggregation and the data filtering process.

The value reported in Line 200 (139 ground measurements) refers to the number of temporally matched Landsat–SURFRAD observations after initial strict filtering criteria at the station level. In contrast, the value reported in Line L399 (previously 385, and revised to 382 after additional quality control) represents the total number of valid Landsat–SURFRAD matchup pairs used in the final statistical analysis across all seven stations at the minute-scale temporal matching level.

To ensure consistency and eliminate any ambiguity, an additional quality-control step was applied to remove a small number of duplicate or incomplete observations. As a result, the final dataset used for statistical evaluation consists of 382 valid matchup pairs.

Action Taken: The final number of valid Landsat–SURFRAD matchup pairs has been unified as 382 across all sections of the manuscript, including the abstract, methodology, results, and conclusion.

Comment 8: In section 3.4, the text has not list the figure 5, please add it in available sentence.

Response: We thank the reviewer for this observation. We agree that explicit referencing of figures at the beginning of relevant sections improves clarity and readability. Accordingly, a direct reference to Figure 5 has been added in the opening sentence of Section 3.4 to ensure proper linkage between the text and the corresponding visual representation.

Action Taken: The phrase “(Figure 5)” has been added in Line 476 of the revised manuscript in the opening sentence of Section 3.4.

Comment 9: In L512-514 and fig 6, please check the number of positive and negative measurements, because 127 and 258 respectively in text, but the figure legends are oppsitive.

Response: We thank the reviewer for identifying this inconsistency. We carefully re-checked the statistical outputs against the original processing script and confirmed that the positive bias (Landsat-8 > SURFRAD) is the dominant feature in the dataset. After reprocessing and quality control, the correct distribution consists of 261 positive cases and 121 negative cases out of a total of 382 valid matchup pairs.

We further ensured consistency between these values and the legend of Figure 6 to avoid any ambiguity or misinterpretation.

Action Taken: The numbers have been corrected and consistently reported in Lines 480–484 of the revised manuscript. In addition, the legend of Figure 6 has been updated to match the corrected values (Landsat-8 > SURFRAD: 261; Landsat-8 < SURFRAD: 121).

Quality of English Language: The English is fine and does not require any improvement.

Response: Thank you for confirming that the English language quality in the manuscript meets the publication standards. We are confident that the manuscript now adheres to the publication standards. Your constructive feedback during the major revision process was invaluable, and we thank you again for your time and dedication in helping us improve our work for publication in the Atmosphere.

Reviewer 2 Report

Comments and Suggestions for Authors

This study conducted an accuracy validation of the Landsat-8 Collection 2 Level-2 LST product, using in-situ observations from SURFRAD stations across the United States. It analyzed the impacts of seasonality, atmospheric water vapor content, surface temperature ranges, and site characteristics on retrieval accuracy. The research topic has clear application value, filling part of the gap in the systematic validation of this product against SURFRAD station observations. The data processing workflow of the study is relatively standardized, the statistical analysis indicators are comprehensive, and the results are clearly presented, which can provide certain references for users of this product. However, the manuscript still has several critical issues.

The abstract lacks relevant quantitative descriptions.

The author adopted the MCD19A2 product for water vapor analysis. Why was this particular product selected? The overpass time of MODIS does not fully align with that of Landsat in many cases. Why not directly use reanalysis data or the GEOS-5 FP Product adopted in official Landsat products instead?

There is considerable redundancy in the error evaluation metrics adopted, including RMSE and MSE.

The author only revealed that Landsat temperature tends to be overestimated under high atmospheric water vapor conditions. However, the underlying causes were not adequately explained. Is this bias attributed to inaccuracies in the adopted reanalysis products, limitations of the algorithm, or issues with the validation sites?

Others:

Page 3, Line 110: "(Author 110 et al., 2024)". Many similar formatting errors remain throughout the manuscript. The authors are required to carefully check the full text thoroughly.

ST and LST are used interchangeably throughout the manuscript, as can be seen in Figure 7.

Author Response

Response to Reviewer (2)

Comments and Suggestions for Authors

Comment 1: The abstract lacks relevant quantitative descriptions.

Response: We fully agree with the reviewer’s suggestion regarding the inclusion of key quantitative indicators in the Abstract. The abstract has therefore been substantially revised to improve its informativeness and to provide a clearer quantitative summary of the main findings.

The revised version now includes the principal performance metrics, including RMSE, positive bias, Pearson correlation coefficient, Spearman correlation coefficient, coefficient of determination (R²), total number of matched pairs, seasonal performance differences, and representative station-level relationships, including atmospheric water vapor effects.

Action Taken: The Abstract has been updated to include the main numerical results as follows: n = 382 matchup pairs, RMSE = 4.20 °C, R² = 0.91, r = 0.98, bias = +1.75 °C, winter RMSE = 2.17 °C, summer RMSE = 5.84 °C, and vapor-related correlations (FPK: r = 0.78; DRA: r = 0.75), in addition to station-level statistics.

Comment 2: The author adopted the MCD19A2 product for water vapor analysis. Why was this particular product selected? The overpass time of MODIS does not fully align with that of Landsat in many cases. Why not directly use reanalysis data or the GEOS-5 FP Product adopted in official Landsat products instead?

Response: We thank the reviewer for raising this important methodological point and for allowing us to clarify the rationale behind the selection of MCD19A2. We fully agree that the choice of atmospheric water vapor product is critical for ensuring methodological consistency and scientific validity.

Our selection of MCD19A2 (MAIAC algorithm) is based on several scientific considerations. First, MCD19A2 provides column water vapor derived from MODIS observations at a relatively high spatial resolution (1 km), whereas reanalysis products such as NCEP/NCAR and ERA5, as well as GEOS-5 FP (which is embedded in the official Landsat Collection 2 Level-2 retrieval chain), operate at substantially coarser spatial scales and are inherently integrated within the retrieval algorithm itself. Second, the objective of this study is to evaluate the relationship between independently observed atmospheric water vapor and Landsat LST errors; therefore, using GEOS-5 FP would introduce a circular dependency, as it is the same auxiliary input used in the generation of the product under evaluation, thereby compromising the independence of the analysis.

Third, although MODIS overpass timing does not perfectly coincide with Landsat-8 acquisitions, a strict temporal matching window of ±2 hours was applied, resulting in an average temporal difference of only 11 minutes, which is considered acceptable given the relatively slow variability of atmospheric column water vapor at this temporal scale. Finally, MCD19A2 is currently the closest satellite-based product to Landsat-8 in terms of overpass timing, as both systems acquire observations in the early afternoon overpass window.

Action Taken: A dedicated justification for the use of MCD19A2 has been added in Section 2.3.3 of the revised manuscript, including explicit reporting of the mean temporal offset (11 minutes). It is also clarified that comparisons with reanalysis-based water vapor products or Landsat-derived atmospheric inputs represent a valuable direction for future research, and this has been explicitly stated in the Conclusion as part of the future work outlook.

Comment 3: There is considerable redundancy in the error evaluation metrics adopted, including RMSE and MSE.

Response: We agree with the reviewer that reporting both MSE and RMSE is redundant, since RMSE is simply the square root of MSE and provides a more interpretable error metric in geophysical validation studies. Accordingly, MSE has been removed from the set of primary evaluation metrics to avoid unnecessary duplication and to align the analysis with standard practices in land surface temperature validation literature.

To further improve methodological clarity, the set of statistical metrics has been streamlined to include only functionally distinct and complementary indicators, ensuring a more robust and non-redundant evaluation framework.

Action Taken: Section 2.3.4 has been updated to explicitly state that MSE was removed from the main metrics. RMSE has been retained as the primary error metric, alongside the following set of complementary statistics: MAD, MaxAD, Bias, Median Absolute Difference, 95th Percentile Error, Willmott’s d, R², Pearson’s r, and Spearman’s ρ.

Comment 4: The author only revealed that Landsat temperature tends to be overestimated under high atmospheric water vapor conditions. However, the underlying causes were not adequately explained. Is this bias attributed to inaccuracies in the adopted reanalysis products, limitations of the algorithm, or issues with the validation sites?

Response: We thank the reviewer for this important and insightful comment. We fully agree that the observed positive bias under high atmospheric water vapor conditions likely arises from multiple interacting sources rather than a single factor. Accordingly, the discussion section has been substantially revised to explicitly disentangle three primary and physically consistent contributors to this behavior.

First, uncertainties in the GEOS-5 FP water vapor product used within the Landsat Collection 2 Level-2 retrieval chain may propagate into the atmospheric correction process. Underestimation of actual column water vapor can lead to insufficient atmospheric compensation, resulting in an overestimation of land surface temperature. Second, inherent limitations of the Single-Channel Algorithm, including its simplified radiative transfer assumptions, reduce retrieval accuracy under high-temperature and high-humidity conditions where atmospheric absorption effects are more pronounced. Third, spatial heterogeneity effects, particularly in elevated stations such as TBL and FPK, introduce additional complexity due to strong vertical stratification of water vapor and sub-pixel variability in surface emissivity, leading to mismatches between satellite-retrieved atmospheric state and ground-level conditions.

To strengthen this interpretation, station-level correlations between water vapor and LST error (Table 6) have been explicitly integrated into the discussion to provide empirical support for these physically based explanations.

Action Taken: The first and second paragraphs of the Discussion section have been expanded to explicitly describe the three contributing mechanisms. In addition, station-specific results (e.g., FPK: r = 0.78; BND: r = 0.04) have been directly linked to their corresponding physical interpretations within the revised manuscript.

Others: Comment 5: Page 3, Line 110: "(Author 110 et al., 2024)". Many similar formatting errors remain throughout the manuscript. The authors are required to carefully check the full text thoroughly.

Response: We apologize for this formatting error. We fully agree that incorrect or placeholder citations should be avoided to ensure reference integrity and clarity.

Action Taken: The incorrect citation “Author 108 et al., 2024” has been removed and replaced with (Duguay-Tetzlaff et al. 2015) in the revised manuscript.

Comment 6: ST and LST are used interchangeably throughout the manuscript, as can be seen in Figure 7.

Response: We thank the reviewer for this precise and important clarification. We fully agree that a clear distinction between USGS product nomenclature and the physical variable terminology is necessary to avoid ambiguity.

In the USGS Landsat Collection 2 Level-2 product, the term Surface Temperature (ST) is used for the official product layers (e.g., ST_B10, ST_TRAD, ST_EMIS, ST_ATRAN), which are technically defined outputs of the processing chain. In contrast, Land Surface Temperature (LST) is the standard term used in the remote sensing literature to refer to the physical geophysical variable representing the radiative skin temperature of the Earth's surface.

To ensure conceptual clarity, we explicitly clarified in the manuscript that both terms refer to the same physical variable (land surface temperature). The term LST has been adopted consistently throughout the text as the reference terminology, while the ST_ prefix has been retained only when referring to the official, fixed product band names that cannot be altered (e.g., ST_B10, ST_TRAD, ST_EMIS, ST_ATRAN). Additionally, terminology in Figure 7 has been standardized to LST to avoid inconsistency.

Action Taken: LST has been adopted as the unified term for the geophysical variable throughout the manuscript. The ST_ prefix is retained only for official USGS product band names. Terminology in Figure 7 has been revised to ensure consistent use of LST.

Quality of English Language: The English is fine and does not require any improvement.

Reviewer 3 Report

Comments and Suggestions for Authors

Please see attached document

Comments for author File: Comments.pdf

Author Response

Response to Reviewer (3)

The manuscript presents an assessment of the accuracy of the Landsat-8 Collection 2 Level-2 Land Surface Temperature (LST) product by comparison with in situ data from SURFRAD stations distributed in different climatic regions of the United States, also analyzing the influence of seasonality, thermal ranges, and atmospheric water vapor on estimation errors. Although the topic addressed is relevant and potentially interesting for the scientific community, the manuscript has important methodological, analytical and conceptual limitations that significantly reduce its real scientific contribution. The work is excessively descriptive, with limited statistical analysis, a shallow discussion and a scientific novelty that is insufficiently justified for an international journal of this level. In its current state, I consider that the manuscript requires major revisions before it can be considered for publication.

We express our sincere appreciation to Reviewer 3 for the thorough and critical reading of the manuscript, which has significantly contributed to strengthening the scientific framing of the study.

Before addressing individual comments, we would like to emphasize a fundamental point that defines the scope and methodological positioning of this work. This study is classified as an operational product benchmarking and independent validation assessment, rather than an algorithm development or multivariate modeling study. Its primary objective, as stated in the Introduction, is to provide the first systematic and independent evaluation of the Landsat-8 Collection 2 Level-2 land surface temperature product using high-quality ground-based observations from SURFRAD stations.

The study further aims to characterize the error behavior of this operational product across seasonal cycles, temperature ranges, atmospheric conditions, and site-specific characteristics, thereby providing end-users with a reliable evidence base for understanding when and where the product can be confidently applied. The analytical framework has been intentionally kept transparent and relatively parsimonious to allow future researchers to extend it using more advanced techniques such as machine learning, multivariate explanatory models, or hybrid approaches in dedicated follow-up studies.

We also note that expanding the current study to incorporate all suggested advanced statistical or modeling approaches would fundamentally change its nature, transforming it from a focused benchmarking assessment into a multi-objective synthesis that would compromise its scientific coherence and interpretability.

My specific comments are as follows:

Comment 1: Introduction: One of the main problems of the work is the limited scientific novelty. Although the authors indicate that this is the first validation of the Landsat-8 C2L2 product using SURFRAD stations, there are numerous previous studies focused on the validation of Landsat LST products and the analysis of errors associated with atmospheric and thermal factors. The manuscript does not clearly establish what the specific scientific contribution is with respect to previous research or what methodological or conceptual advance the study really provides. It would be necessary to define the existing scientific gap much more precisely and adequately justify the originality of the work.

Response: We fully agree with the reviewer that the contribution of this study should be stated more explicitly. In response, a substantive paragraph has been added to the end of the Introduction to clarify the original scientific contribution and to position the work more clearly within the existing literature.

Specifically, we emphasize that although the Landsat-8 Collection 2 Level-2 (C2L2) land surface temperature product has been officially released by USGS since 2018 and widely used in applied studies, it has not yet been comprehensively evaluated using an extended multi-year SURFRAD network. Existing studies such as Malakar et al. (2018) are limited to Landsat 5 and 7, while Ermida et al. (2020) focuses on an alternative algorithm (SMW) rather than the official Landsat-8 C2L2 product.

Accordingly, the contribution of this work is threefold: (i) the first independent multi-station evaluation of the official Landsat-8 C2L2 product over the 2023–2025 period; (ii) a systematic decomposition of error sources across seasons, atmospheric water vapor conditions (using independent MCD19A2 data), surface temperature ranges, and site characteristics; and (iii) the development of a user-oriented reliability framework that indicates the conditions under which the product can be considered reliable or should be used with caution.

Action Taken: A detailed explanation of the study’s scientific novelty, research gap, and original contributions has been incorporated into the revised manuscript (Lines 147–177).

Comment 2: Statistical analysis: The authors use only basic descriptive metrics such as RMSE, MAD, Bias, R2 and correlation coefficients, without incorporating more advanced or robust statistical analyses that allow a solid scientific interpretation of the results. The work lacks statistical significance analysis between seasons or seasons of the year, multivariate analysis, sensitivity analysis or explanatory models that allow adequately identifying the factors responsible for the observed errors. Given the focus of the study, it would have been highly recommended to incorporate more advanced techniques, such as Mann-Kendall-type trend analysis, robust regressions, ANOVA/Kruskal-Wallis or multivariate explanatory models.

Response: We appreciate the reviewer’s scientifically grounded suggestion. However, we respectfully note that its implementation within the present study requires careful methodological consideration due to the design and scope of the work.

First, as clarified in the framing of this study, the objective is operational product benchmarking rather than explanatory or predictive model development. The selected statistical metrics (RMSE, MAD, Bias, R², Willmott’s d, etc.) are the standard and widely accepted framework in Land Surface Temperature (LST) validation studies (e.g., Malakar et al., 2018; Ermida et al., 2020; Cristóbal et al., 2018), and they ensure direct comparability with the established literature.

Second, the Mann–Kendall trend test is primarily designed for long, continuous time series (typically ≥30 years) to detect monotonic trends. In contrast, our dataset spans only three years (2023–2025) and consists of irregular Landsat overpass matchups rather than continuous temporal series, making the application of Mann–Kendall statistically inappropriate in this context.

Third, while ANOVA or Kruskal–Wallis tests can formally assess seasonal differences, the seasonal signal in this study is already strongly and unambiguously expressed through the magnitude of RMSE differences (2.17 °C in winter versus 5.84 °C in summer), which constitutes a large effect size that is clearly interpretable without additional hypothesis testing.

Fourth, multivariate explanatory modeling was intentionally excluded from the scope of this work. Such models require broader temporal coverage, denser station networks, and additional independent predictors (e.g., emissivity fields, vertical water vapor structure, high-resolution topography), which fall beyond the design of this benchmarking study and are better suited for dedicated follow-up research.

Action Taken: The manuscript continues to rely on established LST validation metrics (RMSE, MAD, Bias, MaxAD, Median Absolute Difference, 95th Percentile Error, Willmott’s d, R², Pearson’s r, and Spearman’s ρ) as mentioned in section “2.3.4. Statistical Analysis”, complemented by station-level vapor–error correlation analysis. The conclusion explicitly highlights advanced statistical and multivariate modeling approaches as recommended directions for future work, including robust, multivariate, and sensitivity-based frameworks, see lines 713-728.

Comment 3: The effective sample size is relatively limited for some of the conclusions drawn. Although 385 valid observations are initially presented, certain analyses are perfor with very small datasets, especially in the case of atmospheric water vapour analysis, where only 106 observations are used. In some specific cases, such as the PSU station, the number of observations is extremely low, which considerably limits the statistical robustness of the conclusions obtained. Despite this, the manuscript makes relatively general statements that do not always seem sufficiently supported by the available evidence.

Response: We thank the reviewer for this insightful comment. We completely agree that some of the sub-analyses are based on relatively small sample sizes. To address this, a dedicated cautionary paragraph has been added to the Conclusion section to explicitly distinguish between our primary findings (which are based on a statistically robust sample size of n = 382) and the sub-analyses. Specifically, the water vapor analysis (n = 106) and the PSU station statistics (n = 26 overall, and n = 3 for the water vapor correlation) are now explicitly characterized as "indicative" rather than "conclusive." We have noted that these sub-findings require further validation in future studies utilizing longer time-series data and expanded ground-based networks. Furthermore, we have carefully revised the wording of all conclusions throughout the manuscript to avoid unwarranted generalizations and to strictly align our assertions with the limitations of the available dataset.

Action Taken: A clear cautionary paragraph has been incorporated at the very end of the Conclusion section (lines 729-741). Additionally, an explicit warning note has been appended to Table 6, and a corresponding clarifying discussion has been integrated into the PSU station analysis subsection.

Comment 4: There are problems of coherence in the interpretation of error bias. In some sections of the manuscript it is stated that the Landsat-8 product shows a general tendency to overestimate the surface temperature, while later it is indicated that the product tends to underestimate temperatures in approximately two thirds of the cases. This apparent contradiction should be carefully clarified, specifying more precisely the definition and sign of the errors used, as well as the differences between mean bias and relative frequency of positive and negative errors.

Response: We appreciate the reviewer’s diligence in spotting this ambiguity, which stemmed from a phrasing oversight that conflated the overall mean bias analysis with the detailed temperature-interval analysis. To clarify: on an overall average level, the product exhibits a positive bias of +1.75 °C (overestimation) in approximately two-thirds of the matches (261 out of 382). This reflects the dominant trend of Landsat-8 C2L2 according to our defined sign convention:

Bias = T_Landsat - T_ SURFRAD.

Conversely, the negative sign observed in the boxplots for temperature intervals above 20 °C captures a distinct physical behavior at higher temperatures; specifically, the median shifts toward negative values (underestimation) under warm conditions, even though the cumulative overall mean remains positive due to extreme values during cold and transitional seasons. This distinction has now been made explicitly clear in the text. Furthermore, the sign convention has been formally defined in Section 2.3.4, and a clear differentiation between the Mean Bias and the relative frequency of positive/negative errors has been integrated into Figure 6 and its accompanying discussion.

Action Taken: The mathematical definition of bias (Bias = T_Landsat - TSURFRAD) has been explicitly stated. The text in both the Results and Discussion sections has been thoroughly rewritten to differentiate clearly between the overall Mean Bias and the interval-specific Median. Additionally, the corresponding numerical figures have been thoroughly cross-checked and synchronized throughout the manuscript to consistently show 261 positive and 121 negative instances.

Comment 5: Results: In relation to the presentation of results, the manuscript is excessively long and repetitive. Much of the results section consists of detailed narrative descriptions of values that are already clearly reflected in tables and figures. This significantly reduces the ability to synthesize and makes it difficult to read the work. It would be advisable to condense some of this information and focus the discussion on the findings that are really relevant from a scientific point of view.

Response: We sincerely thank the reviewer for this constructive feedback aimed at improving the readability of the manuscript. We completely agree with the importance of conciseness and focus in presenting scientific data. In response, we have thoroughly reviewed the Results section and carefully streamlined the narrative. Redundant verbal descriptions of numerical values that are already explicitly presented in the tables and figures have been minimized. The text has been revised to focus more sharply on the physical interpretations and the key relationships between variables, thereby enhancing the overall synthesis and flow of the section while preserving the necessary rigorous documentation.

Action Taken: The Results section (Section 3) was carefully revised and streamlined to minimize repetitive descriptions and enhance the presentation of key scientific findings, resulting in a reduction in length from approximately 3,650 to 1870 words.

Comment 6: Discussion: This section is limited and excessively descriptive. The authors focus mainly on listing results and numerical values obtained for each season or season of the year, but they do not delve sufficiently into the physical or methodological causes that could explain the observed patterns. Relevant aspects such as the role of topography, surface heterogeneity, emissivity variations, atmospheric complexity or the inherent limitations of the single-channel algorithm used by Landsat are hardly discussed in depth. Similarly, critical comparisons with results obtained in previous international studies are lacking.

Response: We thank the reviewer for this important comment highlighting the need to strengthen the conceptual depth and physical interpretation of the Discussion section. In response, the manuscript has been substantially revised to shift the focus from descriptive reporting toward a more mechanistic and literature-supported interpretation of the observed patterns.

Specifically, the Results section has been streamlined by reducing repetitive numerical descriptions, allowing the Discussion to focus more clearly on the physical and methodological drivers of Landsat-8 LST retrieval performance. The revised Discussion is now organized into three integrated analytical axes: (i) seasonal variability and atmospheric water vapor controls, (ii) temperature-dependent algorithmic sensitivity and systematic bias, and (iii) station-scale spatial heterogeneity driven by topography and land surface complexity.

In the revised version, we explicitly link seasonal degradation patterns to physically based processes, including increased atmospheric water vapor loading during summer, enhanced radiative attenuation, and limitations in reanalysis-based atmospheric correction, as demonstrated in previous studies (e.g., Single-Channel Algorithm performance constraints reported by Li et al., 2013; Jiménez-Muñoz et al., 2014; Duan et al., 2021). We further incorporate recent validation evidence showing consistent seasonal RMSE amplification in Landsat-based LST products under humid atmospheric conditions (Ermida et al., 2020; Wang et al., 2020).

Moreover, the revised Discussion introduces a dedicated mechanistic explanation for station-specific differences, explicitly addressing the roles of topographic elevation, surface heterogeneity, and emissivity uncertainty derived from ASTER GED products. In particular, we highlight how high-elevation and structurally complex sites (e.g., TBL and FPK) amplify atmospheric correction errors due to vertical stratification of water vapor and rapid microclimatic variability, consistent with findings from Ermida et al. (2020) and Meng & Cheng (2018). In contrast, homogeneous low-elevation sites (e.g., GWN and SXF) demonstrate improved retrieval stability due to reduced scale mismatch and more stable surface emissivity conditions, as also reported in Ma et al. (2021).

Importantly, we now provide a more critical interpretation of the systematic positive bias in Landsat-8 C2L2 LST estimates, linking it to combined uncertainties in atmospheric water vapor estimation and emissivity parameterization, particularly the limitations of NDVI-adjusted ASTER GED emissivity in rapidly changing vegetation conditions. This interpretation is now explicitly supported by prior radiative transfer and emissivity modeling studies (Hulley et al., 2015; Malakar et al., 2018; Duan et al., 2021).

Overall, the revised Discussion no longer focuses on numerical repetition of results, but instead provides a physically grounded explanation of observed spatial, seasonal, and thermal patterns, supported by recent high-impact literature on Landsat LST validation and radiative transfer modeling. These revisions significantly improve the interpretability, scientific depth, and international comparability of the manuscript in line with the expectations of Remote Sensing and related journals.

Action Taken: Accordingly, Section 4 (Discussion) has been substantially expanded in terms of interpretative depth and scientific rigor, with a deliberate shift from purely descriptive reporting of results toward a more mechanistic explanation of the observed patterns, consistent with the standards of high-impact remote sensing journals.

Comment 7: Figures: Several figures are excessively basic for a high-impact international journal and some provide limited information regarding the space they occupy in the manuscript. In addition, the graphical representation lacks more advanced statistical elements, such as confidence intervals, uncertainty analysis, or more robust comparative representations.

Especially the time series in Figure 7 are visually inefficient and difficult to interpret comparatively between seasons.

Response: We sincerely appreciate the reviewer’s constructive feedback regarding the clarity and scientific quality of the figures. In response, we have comprehensively revised and enhanced all graphical elements throughout the manuscript, including Figure 7. Improvements include higher resolution, improved contrast, optimized line weights, and enhanced font clarity to ensure compliance with the journal’s graphical standards. Regarding the manuscript space allocated to figures, we respectfully emphasize that these visualizations are essential for effectively communicating the study’s scientific outcomes. In particular, Figure 7 presents multi-station time-series data that is critical for illustrating and comparing seasonal variations in Landsat-8 land surface temperature (LST) errors under different environmental conditions. While the figure occupies substantial space, representing such detailed temporal variability through text or tabular formats would significantly reduce interpretability and obscure key scientific patterns.

Action Taken: Figure 7 has been fully redrawn with improved graphical clarity and now includes additional statistical performance indicators, including R², RMSE, and bias values for each SURFRAD station, thereby strengthening the robustness and interpretability of the results.

Comment 8: The general writing of the manuscript requires an important work of synthesis and stylistic revision. In many sections, similar ideas are repeated, excessively long descriptions are used and numerical results already visible in tables and figures are reiterated.

Response: We sincerely appreciate this insightful comment regarding the overall clarity and structure of the manuscript. We fully acknowledge that scientific writing should be concise, coherent, and free from redundancy to ensure effective communication. In response, the manuscript has been thoroughly revised and carefully edited to improve its stylistic quality and narrative focus. Redundant and excessively detailed descriptions have been reduced, overlapping content has been consolidated, and repeated reporting of numerical values already presented in tables and figures has been removed. The revised version emphasizes synthesis, interpretation of results, and the broader scientific implications, resulting in a clearer, more coherent, and better-structured manuscript..

Action Taken: The manuscript has undergone comprehensive language polishing and structural refinement. Redundancies have been eliminated across all major sections, including the Introduction, Methodology, Results, and Discussion. Dataset descriptions and analytical explanations have been streamlined to enhance clarity, coherence, and scientific focus throughout the paper..

Quality of English Language: The English could be improved to more clearly express the research..

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have revised the muanuscript carefully, I propose to accept it for the Journal.

Reviewer 2 Report

Comments and Suggestions for Authors

I have carefully reviewed the revised manuscript and the authors’ response letter. The authors provided detailed revisions and thorough explanations addressing all my concerns. I have no additional questions.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have adequately addressed the main concerns raised during the previous review round. The manuscript has been substantially improved, particularly regarding the clarification of its scientific contribution, the discussion of physical mechanisms, the interpretation of retrieval errors and the acknowledgement of limitations. Although some methodological suggestions regarding advanced statistical analyses were not implemented, the authors provided a reasonable justification consistent with the scope of the study. Therefore, I consider that the manuscript is now suitable for publication.