the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 16 Feb 2026
Generation of angular-normalized, cloud-filled, 0.01°-downscaled land surface temperature from 2018 to 2023 based on official FY-4A dataset
Abstract. Land surface temperature (LST) is an essential climate variable in geophysical, ecological, and environmental researches. Remote sensing provides a unique observation approach for obtaining large-scale LST products. However, current official LST datasets (such as FY-4A) are limited by the unaddressed thermal radiation directionality effect, and suffer the spatial discontinuities due to the pervasive presence of clouds. What’s more, the geostationary LST products have relatively coarser resolution than those of polar-orbiting satellites due to trade-off between spatial and temporal resolutions. Based on the official hourly FY-4A LST dataset, this study proposes a novel framework for generating angular-normalized, cloud-filled, and 0.01°-downscaled LST (ANCFDS-LST) product, encompassing directional (Tdir), nadir (Tnadir), and hemispherical (Themi) LST layers. First, the angular-normalized Tnadir and Themi were generated using a time-evolving kernel driven model (TEKDM) with the inputs of multi-temporal FY-4A Tdir. Subsequently, hypothetical clear-sky LST were predicted using a CatBoost model optimized via Bayesian methods. The cloudy-sky LST values were then derived through a cloud radiation force (CRF) correction. Finally, the 0.05° all-weather Tdir, Tnadir, and Themi values were downscaled to 0.01° resolution using an improved hybrid downscaling algorithm (IHDA) combining fusion and kernel-based methods. Taking the daytime clear-sky near-nadir VNP21A1 LST as reference, the 0.05° Tdir before angular-normalization has a root mean squared difference (RMSD) of 6.21 K and a mean bias difference (MBD) of -4.04 K, whereas the angularly normalized Tnadir has a much smaller RMSD of 3.48 K and a better MBD of -2.13 K. For the all-weather Themi, temperature-based validation over 15 sites in the Heihe River Basin and the Tibetan Plateau shows a root mean squared error (RMSE) and mean bias error (MBE) of 2.99 K and -0.77 K under clear-sky conditions, 4.56 K and -1.56 K under cloudy-sky conditions. After the spatial downscaling, the 0.01° all-weather Themi with abundant texture details exhibits an RMSE (MBE) of 3.99 K (-1.32 K) over 15 sites. The generated LST products from 2018 to 2023 over the FY-4A disk exhibit enhanced angular consistency, spatial continuity, and finer resolution, offering valuable support for subsequent LST-related applications. The ANCFDS-LST data is freely available at https://doi.org/10.11888/RemoteSen.tpdc.303249 (last access: 30 January 2026; Na et al., 2026).
- Preprint
(5184 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on essd-2026-74', Anonymous Referee #1, 09 Mar 2026
-
RC2: 'Comment on essd-2026-74', Anonymous Referee #2, 10 Mar 2026
The manuscript mainly assembles previously published approaches for angular correction, cloud-gap filling, and downscaling into a production workflow for FY-4A LST. The methodological novelty is limited, several key assumptions are questionable, and the writing quality is low. The stacked machine-learning workflow likely propagates errors, while validation remains geographically limited.
Major comments
- Incorrect use of a sinusoidal annual temperature cycle
The method assumes a sinusoidal annual temperature cycle to generate gap-free LST fields and to support downscaling. This assumption may work at the higher latitude sites selected in the study, but it is fundamentally flawed at large spatial scales. Many regions, particularly in monsoon and low-latitude climates, exhibit multiple seasonal temperature peaks or asymmetric annual cycles. Previous studies have shown that it is difficult to design a universally valid ATC model across broad latitudinal gradients. Using a simple sinusoidal ATC as the backbone of cloudy reconstruction and downscaling is therefore physically incorrect.
- Temporal coverage and model stability are not demonstrated
The product only spans the period up to 2023, and the training samples appear to be drawn from the same time span. The manuscript does not demonstrate whether the model can maintain stable performance for future years or under changing climatic conditions. Without testing temporal extrapolation or independent-year validation, it is difficult to assess whether the framework can support sustained long-term product generation. The dataset currently stops in 2023 and appears more like a test dataset, which cannot meet the standard of ESSD.
- Limited methodological novelty
The three core components—angular normalization (Na et al., 2024), cloudy-sky reconstruction (Zhang et al., 2024), and spatial downscaling (Liang et al., 2025)—are from existing studies the authors recently published. The manuscript mainly integrates these methods into a processing chain rather than introducing a simplified or integrated conceptual framework or breakthrough. The claimed novelty is therefore overstated.
- Fragmented pipeline and potential error accumulation
The workflow relies on multiple sequential ML-based steps (bias correction, angular normalization, gap-filling, and downscaling). Such a chained structure inevitably accumulates errors, yet the manuscript provides no uncertainty propagation analysis or ablation study to quantify how each module contributes to the final product accuracy.
- Problematic nighttime assumption
The study assumes that nighttime LST has negligible angular effects. This assumption may hold for deep-night polar-orbiting observations, but hourly geostationary observations shortly after sunset can still exhibit directional thermal anisotropy due to soil–vegetation temperature contrasts. The validity of this assumption is not demonstrated.
- Weak physical justification for cloudy-sky directional consistency
The framework extends clear-sky directional normalization to cloudy-sky conditions without demonstrating whether directional effects persist, weaken, or change under cloud cover. The manuscript effectively transfers the clear-sky logic to cloudy conditions without physical justification.
- Heavy dependence on reanalysis and auxiliary variables
The models rely extensively on reanalysis radiation, air temperature, and other auxiliary variables from reanalysis (even dewpoint T?). This raises the question of whether the final product genuinely reflects satellite observations or largely reproduces reanalysis-driven temperature fields. A semi-physical framework can, in principle, leverage multiple data sources and physical constraints, but this advantage is not fully utilized here. In practice, the model appears to be dominated by reanalysis inputs while still relying on an oversimplified SEB parameterization, making the choice of a semi-physical approach difficult to justify.
- Questionable value of the downscaling step
The product is downscaled from 0.05° (~4 km) to 0.01° (~1 km), but the resulting improvement is minimal. The downscaling is not helpful but may introduce additional uncertainty. Air temperature is one of the dominant predictors in the downscaling model, yet it is obtained from simply interpolated reanalysis fields. If the downscaled LST strongly depends on interpolated air temperature rather than satellite observations, the added scientific value of the product becomes questionable.
- Limited and geographically biased validation
The validation relies heavily on stations concentrated in the Heihe River Basin. Such spatially clustered sites cannot represent the accuracy of a product intended to cover the full FY-4 disk. In addition, the spatial representativeness of point measurements at the 4-km scale remains uncertain.
- Overall accuracy improvement is not convincing
The reported errors remain relatively large, and in some cases are lower than those reported for other products. Given the complexity of the workflow and the number of assumptions introduced, the demonstrated improvement in accuracy does not convincingly justify the methodological complexity.
Minor comments
- researches -> research
- suffer the spatial discontinuities -> suffer from
- “What’s more” is informal in academic English
- hypothetical clear-sky LST were -> “was”
- combining fusion and kernel-based -> “combining fusion- and kernel-based”
- “jointly estimation” is wrong
- “resampled to a same spatial resolution” -> resampled to the same spatial resolution
- “with the conditions of” -> “under the condition”
- “The VNP21 LST values … was employed” -> “were”
- “as did by” -> done
- “cross evaluation” -> “cross- evaluation”
- “in southern hemisphere” -> “the southern hemisphere”
- “To abundant the angular information” ?
- “ML-based LST reconstructing have been widely used” wrong sentence
- “over fitting” -> “overfitting”
- data-sets -> datasets
- in the right of -> “on the right-hand side of”
- widely-used -> widely used
Citation: https://doi.org/10.5194/essd-2026-74-RC2
Data sets
Angular-normalized, cloud-filled, 0.01°-downscaled land surface temperature from 2018 to 2023 based on official FY-4A dataset Qiang Na, Biao Cao, Boxiong Qin, Tian Hu, Hua Li, Lixin Dong, Huanyu Zhang, Wenfeng Zhan, Qinhuo Liu https://data.tpdc.ac.cn/en/data/4adbc070-afb3-4e9c-85a0-2ce68d1388ad
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 247 | 187 | 19 | 453 | 24 | 25 |
- HTML: 247
- PDF: 187
- XML: 19
- Total: 453
- BibTeX: 24
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
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: