The revised manuscript has been substantially improved after the first-round review. The authors have added clearer positioning of the work as a system-level LST product-generation framework rather than as a set of new individual algorithms; they have expanded the validation with OzFlux sites, added representativeness screening for in-situ sites, introduced bias correction for the VNP21A1 cross-validation, and provided additional diagnostics for cloud filling and downscaling uncertainties. These changes address many of the main concerns raised in the first round. The revised manuscript now presents a more coherent and better supported dataset paper for an hourly, angular-normalized, cloud-filled, and 0.01° downscaled FY-4A LST product from 2018 to 2023. The dataset is potentially valuable for the LST, surface radiation, evapotranspiration, and land–atmosphere interaction communities, especially because the product provides directional, nadir, and hemispherical LST layers rather than only a conventional directional LST field. The authors also now explicitly state that the novelty is mainly in the unified processing chain and angular consistency of the final product, not in each individual methodological component.
I appreciate that the authors have responded carefully to most comments. The response document shows that they revised the DTC equation, added Appendix A for the hemispherical integration approximation, expanded site validation, added cloud-filling and downscaling uncertainty maps, added texture metrics, clarified the role of reanalysis data, and discussed limitations related to nighttime TRD, cloudy-sky TRD, ATC modeling, and reanalysis dependence. In the revised manuscript, the angular normalization performance is now more convincing: the comparison against near-nadir VNP21A1 shows a reduction in RMSD from 4.53 K for FY-4A directional LST to 2.56 K for normalized nadir LST, with the improvement especially evident at larger VZA. The all-weather hemispherical LST validation is also improved, with reported RMSE values of 2.48 K under clear sky and 3.65 K under cloudy sky over the selected eight sites.
However, several points still require clarification before publication.
Major comments
1. The final 0.01° product is not independently validated.
The manuscript validates the angular-normalized nadir LST mainly at 0.05° against bias-corrected near-nadir VNP21A1, and validates 0.05° all-weather hemispherical LST against in-situ pyrgeometer-derived LST. The downscaling step is evaluated mainly through model fitting RMSE and texture similarity metrics using ATC-derived texture as reference. This is useful, but it does not constitute an independent accuracy assessment of the final 0.01° product. Since the released dataset is the 0.01° ANCFDS-LST, the authors should either provide a direct validation of the final 0.01° product at the in-situ sites, or clearly state that the absolute temperature accuracy assessment is primarily for the 0.05° product and that the 0.01° assessment is limited to spatial texture consistency. A comparison with Landsat/ECOSTRESS clear-sky LST for selected regions and dates would further strengthen the paper, but at minimum the existing validation claims should be carefully qualified.
2. The downscaling texture metrics are partly circular.
The authors use local variance ratio and SSIM to show that the 0.01° product better matches ATC-simulated texture than bilinear interpolation does. However, the downscaling method itself uses ATC-simulated 0.01° LST as the initial high-resolution thermal texture field. Therefore, the LVR and SSIM results demonstrate consistency with the imposed ATC texture prior, but they do not independently prove that the downscaled thermal heterogeneity is physically correct. The authors should explicitly acknowledge this in Sect. 4.3 and avoid over-interpreting the texture metrics. The manuscript should state that the metrics evaluate preservation of the prescribed ATC-based texture rather than independent sub-pixel LST accuracy.
3. Clarify the temperature scale after FY-4A–MODIS and FY-4A–VIIRS bias corrections.
The manuscript states that FY-4A LST was linearly adjusted to match MODIS MxD11A1 before TEKDM calibration, with slope 1.0240 and intercept −5.7378. It also states that VNP21A1 was bias-corrected to FY-4A using nighttime matchups before cross-validation. This is methodologically reasonable for reducing inter-sensor systematic differences, but it creates ambiguity about the final product’s radiometric reference scale. Are the released Tdir, Tnadir, and Themi values on the original FY-4A official LST scale, the MODIS-adjusted FY-4A scale, or a hybrid scale? The authors should state this explicitly. In addition, the VNP21A1 cross-validation results should preferably include uncorrected and bias-corrected metrics, at least in a supplement, so that readers can separate angular-normalization improvement from inter-sensor bias removal.
4. The representativeness-based site selection improves validation quality but may also introduce selection bias.
The revised validation uses eight sites selected from HiWATER, OzFlux, and Tibetan Plateau candidates based on Landsat LST spatial STD and nighttime FY-4A–in-situ RMSE. This is a reasonable way to ensure spatial representativeness, but because one of the screening criteria uses the agreement between FY-4A and in-situ LST, the final reported validation accuracy may be optimistic relative to the full range of surface conditions within the FY-4A disk. I recommend that the authors briefly report the validation statistics for all candidate sites, or at least state how much the RMSE changes before and after site screening. The response document mentions that the original clear/cloudy RMSE values were 2.99 K and 4.56 K and improved to 2.48 K and 3.65 K after site replacement/screening; this useful information should be included in the manuscript or supplement.
5. Cloudy-sky Tdir and Tnadir need clearer interpretation.
The CRF correction is formulated for hemispherical LST, but the manuscript states that cloudy-sky Tdir, Tnadir, and Themi are obtained by adding the same CRF correction to CatBoost-estimated hypothetical clear-sky LST. Since cloudy-sky thermal directional anisotropy is not directly observed or validated, the physical meaning of cloudy-sky Tdir and Tnadir should be stated more cautiously. I suggest explicitly noting that cloudy-sky Tdir and Tnadir are model-derived layers inferred under the adopted CRF and angular-normalization assumptions, whereas the physically most defensible all-weather quantity is Themi. This is important for downstream users who may otherwise treat all three layers as equally validated.
6. The product should include, or at least document, quality flags.
The data availability section states that each GeoTIFF contains three sequential bands: Tdir, Tnadir, and Themi. For an all-weather, angular-normalized, and downscaled LST product, users need to know whether each pixel is original clear-sky FY-4A, angular-normalized clear-sky, ML gap-filled cloudy-sky, nighttime uncorrected, or downscaled from a high-uncertainty region. I strongly recommend adding a QA layer or separate flag file indicating clear/cloudy status, day/night status, whether TEKDM was directly solved or temporally composited, and whether the value was reconstructed. If the current dataset cannot be changed, the authors should at least provide a detailed QA table and uncertainty guidance in the documentation.
Minor comments and editorial corrections
Several wording and consistency issues remain. These are minor but should be corrected because the manuscript is still not fully polished.
In the abstract, the sentence “After the spatial downscaling, the 0.01° all-weather Themi with abundant texture details” is incomplete and should be rewritten, for example: “After spatial downscaling, the 0.01° all-weather Themi exhibits abundant spatial texture details.”
“the Heihe River Basin and the Australia” should be changed to “the Heihe River Basin and Australia.”
Line 344 still says “introduced to abundant the spatial texture”; this should be “introduced to enrich the spatial texture” or “introduced to provide spatial information not captured by satellite observations.”
The conclusion says that the in-situ hemispherical LST reference is from “the Heihe River Basin and the Tibetan Plateau,” but the revised validation sites are from HiWATER and OzFlux, i.e., the Heihe River Basin and Australia. This inconsistency should be corrected.
The sentence “The generated all-weather Themi generates comparable accuracy” should be revised to “The generated all-weather Themi shows comparable accuracy.”
The manuscript should consistently use “cloudy-sky” rather than alternating among “cloud-sky,” “cloudy-sky,” and “cloud-covered.” Similar consistency is needed for “clear-sky,” “all-weather,” “T-based,” and “R-based.”
The wording “the normalized 0.05° Themi” and “0.05°FY-4A” should be formatted consistently with spaces between values and units/symbols.
The authors should check subject–verb agreement throughout, for example “ML-based LST reconstruction methods have been widely used because it could…” should be “because they can…”.
Final assessment
The revised manuscript is much stronger than the original version and, in my view, is close to being suitable for publication as an ESSD dataset paper. The dataset is valuable, the angular-normalized LST concept is important, and the authors have made substantial revisions in response to the first-round comments. The remaining issues are mainly about clarification, avoiding overstatement, documenting uncertainty and QA information, and improving language. I therefore recommend minor revision before acceptance. |
This study adopted three well-established methods to generate an angular-normalized, cloud-filled, and 0.01°-enhanced LST dataset covering the spatial extent of the FY-4A disk. The methodology is sound, and the validation is comprehensive, which includes both in situ and cross-validation approaches. I have the following suggestions to improve the readability of the manuscript: