Pith. sign in

REVIEW 4 major objections 6 minor 59 references

Fast Fourier Convolutional GAN for 30 m Clear-Sky Land Surface Temperature Gap-Free Reconstruction

T0 review · 4 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A fast-Fourier-convolution GAN reconstructs 30 m Landsat surface temperature even where clouds hide more than 70% of a scene, with scene-averaged RMSE interquartile range 0.8–1.8 K.

desk verdict Solid multimodal adaptation of LaMa for 30 m LST gap-filling, but the >70% gap recovery claim is asserted, not demonstrated quantitatively. read the letter →

arxiv 2607.22734 v1 pith:4O2EBDNY submitted 2026-07-22 cs.CV eess.IVphysics.geo-ph

Marwa Alfouly , Smajil Halilovic , Nils Bochow , Thomas Hamacher , Niklas Boers , Konrad Schindler This is my paper
classification cs.CVeess.IVphysics.geo-ph
keywords landsurfacetemperaturegapfillingcloudcoverFastFourierConvolutiongenerativeadversarialnetworkLandsat30mSyntheticApertureRadarclear-skyreconstruction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that cloud-blocked pixels in 30 m Landsat land surface temperature (LST) images can be reconstructed not only for small holes but for scenes where clouds cover more than 70% of the area. The method is a generative adversarial network built around fast Fourier convolutions, which give the model a global view of the scene rather than only local context. An unmasked stack of globally available auxiliary data—land cover, elevation and terrain indices, vegetation index, solar hillshade, and cloud-penetrating synthetic aperture radar—guides the reconstruction and needs no prior gap-filling. Across all LST quantiles the interquartile range of scene-averaged RMSE on reconstructed pixels is reported as 0.8–1.8 K. If true, this is a scalable path to gap-free, fine-resolution clear-sky LST products for climate, urban heat, agriculture, and health applications.

What carries the argument

The load-bearing component is the Fast Fourier Convolution (FFC) block—a residual block that splits features into a spatial path and a spectral path and performs convolutions in the frequency domain, giving the model an image-wide receptive field. FFC blocks sit in the bottleneck of an encoder-decoder generator that is trained against a local patch discriminator with a hybrid loss: adversarial, L1, gradient matching, discriminator feature matching, and perceptual terms. Full 6000×6000 scenes are processed by a wave inpainting pass that sorts 256×256 tiles by fraction of valid data, reconstructs the most confident tiles first, and lets newly filled pixels steer overlapping neighbours, skippin

What would settle it

Run the released model on held-out Landsat scenes with real cloud masks from the QA_PIXEL band, and compare reconstructed pixels to clear-sky observations from a temporally adjacent overpass or a colocated radiometer; if the interquartile range of scene-averaged RMSE exceeds 1.8 K, or errors concentrate in contiguous gaps over 70%, the synthetic-mask transfer claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that a multimodal, frequency-domain GAN can propagate thermal information across very large missing regions of 30 m Landsat scenes and recover a clear-sky LST field with errors comparable to or below typical satellite retrieval uncertainty. The authors show error falling as physical guidance is added: land-cover alone leaves large errors within crop and forest classes; adding topography and vegetation narrows them; adding hillshade and SAR backscatter yields the tightest distributions and stabilises the hottest quartile. With the full stack, a scene with 77% missing data is reconstructed, and the scene-averaged RMSE interquartile range is consistently 0.8–1.8 K across al

Load-bearing premise

Accuracy measured on clear-sky scenes with artificial cloud masks is assumed to transfer to real cloud-contaminated acquisitions, and the full-scene protocol skips tiles with less than 5% valid data—so the hardest, most extreme gaps are not actually tested.

Editorial extensions

If this is right

  • Fine-resolution clear-sky LST no longer needs a cloud-free scene or a temporally adjacent clear observation: gaps beyond 70% of a Landsat scene become fillable from a single overpass plus global auxiliary data.
  • Gap-free 30 m LST would make urban heat island, drought, permafrost, wildfire, and disease-vector studies usable at city-block scale instead of only where skies happen to be clear.
  • Because the guidance stack is cloud-free and globally available, the pipeline can be retrained for other regions and biomes without an optical gap-filling preprocessing stage.
  • Clear-sky reconstructions are a baseline thermal state, not the actual below-cloud temperature; the paper notes that all-weather LST needs energy-balance or meteorological inputs, which the generated cloud-free fields can support.
  • Ablation results indicate that every added modality helps, with hillshade and SAR giving the largest stabilisation in the warmest quartile; land-cover alone cannot represent heterogeneous vegetation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Validation uses synthetic masks painted on clear-sky scenes; the transfer to real cloud fields—different shape, fragmentation, and relation to surface state—is untested. A natural next step is to validate against real cloud masks and temporally close clear-sky overpasses.
  • The full-scene protocol skips tiles with less than 5% valid data, so the >70% claim applies to gaps of a particular shape; sparse, highly fragmented slivers of valid data may still be unreconstructable. Testing contiguous versus fragmented masks would map that boundary.
  • Training is confined to one region; the near-global claim rests on input availability, not demonstrated transfer. Zero-shot or fine-tuned evaluation in a different climate zone would show whether the learned thermal–terrain–SAR relationships generalise.
  • The same multimodal inpainting recipe could be applied to other gap-prone geophysical fields with globally available proxies, such as soil moisture or snow cover, wherever a clear-sky or clear-condition analogue exists.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes a multimodal Fast Fourier Convolutional GAN, adapted from LaMa, to inpaint cloud-contaminated pixels in 30 m Landsat LST imagery. The generator takes masked LST plus LULC, DOY, NDVI, terrain, hillshade, and Sentinel-1 SAR inputs; a hybrid loss combines adversarial, perceptual, feature-matching, L1, gradient, and R1 terms. The model is trained on ~32,000 clear-sky tiles from one Bavarian WRS-2 path/row and tested on 19 held-out clear-sky scenes with synthetic masks at 30–35%, 35–50%, and 50–70% missingness, using RMSE per LST quantile. The authors report consistent IQR RMSE 0.8–1.8 K across quantiles with the full multimodal stack and claim ability to recover scenes with more than 70% gaps. The paper includes ablations and baselines (original LaMa, SwinUNet-GAN, adapted SDX-LST).

Significance. If the claims were fully supported, the method would be a practically useful tool for generating gap-free 30 m LST products. The quantitative evaluation on 19 held-out scenes with synthetic masks is a serious strength, as are the staged ablations and the comparisons against several baselines; reporting error distributions by LST quantile is better than averaging-only metrics. The open-source pipeline is a further plus. However, the central >70% gap-recovery claim currently rests on visual evidence alone, and the >5%-valid-tile skip rule means the reported errors may exclude the hardest areas. These are fixable in revision, but until addressed the headline claims exceed the evidence.

major comments (4)
  1. [§3.3, Fig. 7, Abstract] The headline claim — recovery of scenes with more than 70% cloud-induced gaps — is not quantitatively tested. Section 3.3 defines only three synthetic mask categories: Low (30–35%), Medium (35–50%), and High (50–70%). No >70% synthetic mask is evaluated. The only >70% example, Fig. 7, is a real scene with 77% missing data caused by ASTER GED gaps, not clouds, and its caption states that black regions are 'original cloud-free pixels or missing data lacking ground truth'; no error metric is reported over those pixels. I request either (i) adding a fourth synthetic mask category (e.g., 70–80%) and reporting scene-averaged RMSE for all 19 scenes, or (ii) revising the abstract and conclusions to state that >70% recovery is demonstrated only as a case study, not quantitatively validated.
  2. [§3.3, adaptive wave inpainting] The skip rule for tiles with <5% valid data is load-bearing for the gap-free claim. When a large contiguous gap exceeds 70% of a scene, many 256×256 tiles will fall below this threshold; those tiles are not reconstructed and are presumably excluded from the reported RMSE. This makes the output not truly gap-free and biases the error statistics toward easier tiles. Please report, for each mask category, the fraction of missing pixels that fall in skipped tiles, and either (a) implement a second pass that fills skipped tiles using already-inpainted context and report RMSE over all missing pixels, or (b) explicitly qualify all claims as applying only to tiles with ≥5% valid data. As written, Section 6's statement that the framework enables 'consistent thermal inpainting over ... partially observed satellite scenes' overstates the evaluation.
  3. [§3.3, synthetic cloud masks] The procedure for generating 'artificial cloud interference' is not specified. The paper does not state how masks are drawn (random ellipses, realistic cloud fields from other scenes, or something else), what range of gap sizes and shapes is used within each category, how many masks are applied per scene, or whether masks are generated independently for each scene. Since the central validation is on synthetic masks, this is both a reproducibility problem and a threat to external validity: without these details the reader cannot judge whether the synthetic mask statistics resemble real cloud fields, especially large contiguous cloudy regions. Please document the mask generation algorithm and, ideally, validate on real cloud masks (e.g., using the QA_PIXEL band on cloudy scenes) to show transfer.
  4. [§2.1.2, §3.3] All training and test data come from a single WRS-2 path/row (193/26) in Bavaria. The paper argues for geographic transferability and 'near-global' applicability because the auxiliary data are globally available, but no experiment outside this region is reported. The single-region design is acceptable for a methods paper, but the generality claims in the Introduction and Discussion should be tempered, or a cross-region evaluation (even a small one) should be added.
minor comments (6)
  1. [§2.2.1] The text says Stage 2 'reduces the spatial resolution from 30 m to 7.5 m'; given two stride-2 convolutions on a 256×256 tile, the resolution should be 120 m (see Figure 3). Please correct the typo.
  2. [Eq. (1)] The hillshade formula should be clamped to [0, 255] (negative values set to 0) as in the standard ArcGIS implementation cited; otherwise self-shadowed slopes receive negative illumination values.
  3. [Fig. 7 caption] 'Black regions denote original cloud-free pixels or missing data lacking ground truth' conflates two different exclusions. Please state explicitly which pixels enter the error map (d), and whether the ground-truth gaps in (a) are excluded from all error metrics.
  4. [Figures 5 and 8] Figure 5 uses a linear RMSE axis to 7 K while Figure 8 uses a log axis from 0.5 to 20 K. This makes visual comparisons across the two figures misleading. Use consistent axis scaling or state the log scale explicitly in the caption.
  5. [§3.1] The input-channel counts (13, 18, 21) are introduced before the full input list is given in §2.3.1. Consider moving the channel-count definitions to §2.1 or adding a short table for clarity.
  6. [§5] The sentence 'extending this framework to predict actual all-weather LST. This can be done by integrating surface energy balance models...' is grammatically incomplete. Please rephrase.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the core validation is an empirical benchmark on held-out pixels, not a derivation that reduces to its own assumptions.

full rationale

The paper is an empirical machine-learning benchmarking study rather than a formal derivation chain, so the circularity patterns (self-definition, fitted-input-called-prediction, self-citation load-bearing, imported uniqueness, ansatz-by-citation, renaming) do not apply to the central claim. The model is trained on ~32,000 clear-sky tiles and evaluated on 19 held-out clear-sky Landsat scenes under synthetic masks (30–35%, 35–50%, 50–70%); the reported RMSE is computed over reconstructed pixels against true observations, not against training data or fitted targets. Loss weights and learning rates were tuned by sensitivity experiments, but those choices are not used to fabricate the held-out test metrics. The self-citations present (Alfouly et al. review of LST reconstruction; Bochow et al. on LaMa for climate data) are contextual/supporting statements and are not load-bearing for the method's validity. The more serious concerns—that the >70% recovery claim is not quantitatively supported by the experimental protocol, that tiles with <5% valid data are skipped rather than filled, and that synthetic masks may not match real cloud spatial statistics—are limitations of evidence and external validity, not circularity. No equation or result in the paper is shown to be equivalent to its own inputs by construction, so the appropriate circularity score is 0.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim rests on data-quality and evaluation premises rather than on derived physics. No new physical entities are postulated. The principal unstated costs are the synthetic-mask equivalence assumption, the single-region training/evaluation, and the skipped low-information tiles; the paper acknowledges some of these in §5.

free parameters (4)
  • Loss weights κ, α, β, λ, δ, γ = κ=10, α=30, β=100, λ=1.0, δ=0.1, γ=0.001
    Selected via sensitivity experiments (§2.3) to balance adversarial, perceptual, feature-matching, L1, gradient, and R1 terms; they directly define the optimized objective.
  • Generator/discriminator learning rates and scheduler = 2e-4 (G), 5e-5 (D), ReduceLROnPlateau factor 0.5, patience 5
    Set after preliminary experiments (§3.2) to stabilize adversarial training; performance depends on these values.
  • Discriminator configuration = N=2 layers, ndf=32
    Chosen because a deeper N=5 discriminator overpowered the generator (§2.2.2); this changes the training dynamics and final accuracy.
  • Tile minimum valid-data threshold = 5% valid pixels
    Tiles with less than 5% valid data are skipped during full-scene inpainting (§3.3), so this threshold defines which gaps the model actually attempts to reconstruct.
assumptions (6)
  • domain assumption Synthetic cloud masks applied to clear-sky test scenes are representative of real cloud-induced gaps.
    Evaluation (§3.3, §5) validates only against synthetic cloud interference; real clouds have different spatial structure and under-cloud thermal states.
  • domain assumption The multimodal auxiliary stack (LULC, NDVI, DEM, hillshade, SAR) carries enough physical signal to constrain LST where the LST channel is masked.
    Core of the method (§2.1, Fig. 2-3); if these predictors are only weakly correlated with LST in unseen regions, reconstruction error would rise.
  • domain assumption Clear-sky LST is the correct reconstruction target, and under-cloud kinetic temperature is deliberately not modeled.
    Acknowledged in §5; limits the claim to clear-sky products rather than all-weather LST.
  • domain assumption Landsat Collection 2 Level-2 LST and QA_PIXEL cloud masks are accurate at 30 m and the ASTER GED-derived missing data pattern is acceptable for training/validation.
    Ground truth and training tiles rely on these products (§2.1.2, §4).
  • domain assumption GEE Sentinel-1 GRD data are sufficiently radiometrically calibrated and terrain-corrected to serve as an unmasked auxiliary guide.
    Relied on in §2.1.2; the authors note residual topographic-radiometric distortions in the Alpine fringe.
  • standard math Fast Fourier Convolutions implement global receptive fields as described.
    The architecture depends on FFT-based spectral mixing to propagate context across the image (§2.2.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast Fourier Convolutional GAN for 30 m Clear-Sky Land Surface Temperature Gap-Free Reconstruction." pith.science (2026). https://pith.science/paper/4O2EBDNY

@misc{pith2026260722734,
  author       = {Pith},
  title        = {Pith review of: Fast Fourier Convolutional GAN for 30 m Clear-Sky Land Surface Temperature Gap-Free Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4O2EBDNY}},
  note         = {Machine review of arXiv:2607.22734}
}
read the original abstract

Satellite-derived Land Surface Temperature (LST) provides spatially comprehensive data that ground stations cannot match. However, its utility is frequently limited by severe data gaps due to the presence of clouds. As LST is essential for understanding land-atmosphere interactions, numerous methods have been proposed to address this challenge. Yet, the development of a scalable and adaptable pipeline for generating gap-free LST datasets and reconstructing cloud-contaminated pixels remains challenging. Moreover, the reconstruction of extensive missing regions in fine-spatial-resolution observations is particularly difficult. To address this challenge, we propose a Multimodal Fast Fourier Convolutional GAN for reconstructing cloud-contaminated pixels in fine-resolution (30 m) Landsat imagery to generate gap-free clear-sky LST products. The method leverages Fast Fourier Convolution to enable a global receptive field across the image, and is guided by a stack of data consisting of satellite observations and Synthetic Aperture Radar (SAR) data. Across all LST quantiles, the interquartile range of scene-averaged RMSE (computed over reconstructed pixels) is consistently between 0.8 K and 1.8 K. The proposed approach enables the recovery of extensive missing regions, including scenes with more than 70% cloud-induced gaps, while relying on auxiliary data that are readily available at a near-global scale.

Figures

Figures reproduced from arXiv: 2607.22734 by the authors.

Figure 1
Figure 1. Geographical overview of the Bavaria study area [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Spatial distribution of the unnormalized input channels for the August 11, 2022 scene. The multimodal data stack [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. The overall architecture of the LST reconstruction model. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Iterative inpainting progression for the masked regions on August 11, 2022 [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]
Figure 5
Figure 5. Figure 5: Comparison of RMSE across LST quantiles for successive guidance strategies. Performance consistently improves [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]
Figure 6
Figure 6. Figure 6: Visual and statistical evaluation of LST reconstruction performance over a predominantly agricultural (crop) land [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: Clear-Sky LST Reconstruction on August 11, 2022, using the proposed Full Multi-Modal architecture [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: Comparative performance analysis of LST reconstruction methods. [PITH_FULL_IMAGE:figures/full_fig_p023_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 26 canonical work pages

  1. [1]

    Opportunities and chal- lenges of spaceborne sensors in delineating land surface temperature trends: A review

    Ahmed, M.R., Ghaderpour, E., Gupta, A., Dewan, A., Hassan, Q.K., 2023. Opportunities and chal- lenges of spaceborne sensors in delineating land surface temperature trends: A review. IEEE Sensors Journal 23, 6460–6472. doi:10.1109/JSEN.2023.3246842

    arXiv 2023

  2. [2]

    Identification of geothermal potential zone associated with land surface temperature derived from landsat 8 data using split-window algorithm

    Akhyar, Sary, C.A., 2024. Identification of geothermal potential zone associated with land surface temperature derived from landsat 8 data using split-window algorithm. Journal of Applied Research and Technology 22, 125–137. URL:https://jart.icat.unam.mx/index.php/jart/article/view/2091, doi:10.22201/icat.24486736e.2024.22.1.2091

    work page doi:10.22201/icat.24486736e.2024.22.1.2091 2024

  3. [3]

    Reconstructing land surface temperature for cloud-covered regions: A review of methods

    Alfouly, M., Halilovic, S., Boers, N., Hamacher, T., 2026. Reconstructing land surface temperature for cloud-covered regions: A review of methods. EarthArXiv URL:https://doi.org/10.31223/X5NX91, doi:10.31223/X5NX91

    work page doi:10.31223/x5nx91 2026

  4. [4]

    Toward the detection of permafrost using land-surface temperature mapping

    Batbaatar, J., Gillespie, A.R., Sletten, R.S., Mushkin, A., Amit, R., Trombotto Liaudat, D., Liu, L., Petrie, G., 2020. Toward the detection of permafrost using land-surface temperature mapping. Remote Sensing 12. URL:https://www.mdpi.com/2072-4292/12/4/695, doi:10.3390/rs12040695

    work page doi:10.3390/rs12040695 2020

  5. [5]

    Pan-sharpening with color-aware perceptual loss and guided re-colorization, in: 2020 IEEE International Conference on Image Processing (ICIP), pp

    Bello, J.L.G., Seo, S., Kim, M., 2020. Pan-sharpening with color-aware perceptual loss and guided re-colorization, in: 2020 IEEE International Conference on Image Processing (ICIP), pp. 908–912. doi:10.1109/ICIP40778.2020.9190785

    arXiv 2020

  6. [6]

    Reconstructing historical climate fields with deep learning

    Bochow, N., Poltronieri, A., Rypdal, M., Boers, N., 2025. Reconstructing historical climate fields with deep learning. Atmospheric science URL:https://www.science.org/doi/10.1126/sciadv.adp0558

    doi:10.1126/sciadv.adp0558 2025

  7. [7]

    Principles of Geographical Information Systems

    Burrough, P.A., McDonnell, R.A., 1998. Principles of Geographical Information Systems. Oxford University Press, Oxford, UK

    1998

  8. [8]

    Swin-unet: Unet- like pure transformer for medical image segmentation

    Cao, H., Wang, Y., Chen, J., Jiang, D., Zhang, X., Tian, Q., Wang, M., 2021. Swin-unet: Unet- like pure transformer for medical image segmentation. URL:https://arxiv.org/abs/2105.05537, arXiv:2105.05537

    arXiv 2021

Show all 59 references
  1. [9]

    Missing pixel reconstruction on landsat 8 analysisreadydatalandsurfacetemperatureimagepatchesusingsource-augmentedpartialconvolution

    Chen, M., Sun, Z., Newell, B.H., Corr, C.A., Gao, W., 2020. Missing pixel reconstruction on landsat 8 analysisreadydatalandsurfacetemperatureimagepatchesusingsource-augmentedpartialconvolution. Remote Sensing 12. URL:https://www.mdpi.com/2072-4292/12/19/3143, doi:10.3390/rs12193143

    2020 doi

  2. [10]

    Role of soil thermal inertia in surface temperature and soil moisture-temperature feedback

    Cheruy, F., Dufresne, J.L., Aït Mesbah, S., Grandpeix, J.Y., Wang, F., 2017. Role of soil thermal inertia in surface temperature and soil moisture-temperature feedback. Journal of Ad- vances in Modeling Earth Systems 9, 2906–2919. URL:https://agupubs.onlinelibrary. 27 wiley.co...

    2017 doi

  3. [11]

    Enhancing remote sensing image super-resolution guided by bicubic-downsampled low-resolution image

    Chung, M., Jung, M., Kim, Y., 2023. Enhancing remote sensing image super-resolution guided by bicubic-downsampled low-resolution image. Remote Sensing 15, 3309. doi:10.3390/rs15133309

    2023 doi

  4. [12]

    Remotely sensed land surface temperature- based water stress index for wetland habitats

    Cieężkowski, W., Szporak-Wasilewska, S., Kleniewska, M., Jóźwiak, J., Gnatowski, T., Dąbrowski, P., Góraj, M., Szatyłowicz, J., Ignar, S., Chormański, J., 2020. Remotely sensed land surface temperature- based water stress index for wetland habitats. Remote Sensing 12, 631. URL...

    2020 doi

  5. [13]

    2024 – a second record-breaking year, following the exceptional 2023

    Copernicus Climate Change Service (C3S), 2025. 2024 – a second record-breaking year, following the exceptional 2023. URL:https://climate.copernicus.eu/global-climate-highlights-2024. updated 2025-01-17

    2025

  6. [14]

    Land surface temperature and emissivity estimation from passive sensor data: Theory and practice-current trends

    Dash, P., Göttsche, F.M., Olesen, F.S., Fischer, H., 2002. Land surface temperature and emissivity estimation from passive sensor data: Theory and practice-current trends. International Journal of Remote Sensing 23, 2563–2594. URL:https://doi.org/10.1080/01431160110115041, doi...

    2002 doi

  7. [15]

    Landsat 8-9 oli/tirs collection 2 level-2 science products

    Earth Resources Observation and Science (EROS) Center, 2020. Landsat 8-9 oli/tirs collection 2 level-2 science products. URL:https://doi.org/10.5066/P9OGBGM6, doi:10.5066/P9OGBGM6

    2020 doi

  8. [16]

    Sentinel-1 radar data assessment to estimate crop water stress

    El-Shirbeny, M.A., Abutaleb, K., 2017. Sentinel-1 radar data assessment to estimate crop water stress. World Journal of Engineering and Technology 5, –. URL:https://doi.org/10.4236/wjet.2017. 52B006, doi:10.4236/wjet.2017.52B006

    2017 doi

  9. [17]

    How Hillshade works—ArcMap Documenta- tion

    Environmental Systems Research Institute (Esri), 2021. How Hillshade works—ArcMap Documenta- tion. URL:https://desktop.arcgis.com/en/arcmap/latest/tools/spatial-analyst-toolbox/ how-hillshade-works.htm. accessed: September 11, 2025

    2021

  10. [18]

    Normalised difference vegetation index 2014-2020 (raster 300 m), global, 10-daily – version 1

    European Environment Agency (EEA), 2016. Normalised difference vegetation index 2014-2020 (raster 300 m), global, 10-daily – version 1. URL:https://land.copernicus.eu/en/products/vegetation/ normalized-difference-vegetation-index-300m-v1.0

    2016

  11. [19]

    Normalised difference vegetation index 2020-2025 (raster 300 m), global, 10-daily – version 2

    European Environment Agency (EEA), 2022. Normalised difference vegetation index 2020-2025 (raster 300 m), global, 10-daily – version 2. URL:https://land.copernicus.eu/en/products/vegetation/ normalised-difference-vegetation-index-v2-0-300m. 28

    2022

  12. [20]

    Sentinel-1 class c, level-1 ground range detected (grd)

    European Space Agency (ESA), 2026. Sentinel-1 class c, level-1 ground range detected (grd). URL: https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS_S1_GRD. accessed via Google Earth Engine

    2026

  13. [21]

    Forster, P.M., Walsh, T., Smith, C., Lamb, W.F., Lamboll, R., Cassou, C., Hauser, M., Hausfather, Z., Lee, J.Y., Palmer, M.D., von Schuckmann, K., Slangen, A.B.A., Szopa, S., Trewin, B., Yun, J., Gillett, N.P., Jenkins, S., Matthews, H.D., Raghavan, K., Ribes, A., Rogelj, J., ...

  14. [22]

    Global land sur- face temperature from the along-track scanning radiometers

    Ghent, D.J., Corlett, G.K., Göttsche, F.M., Remedios, J.J., 2017. Global land sur- face temperature from the along-track scanning radiometers. Journal of Geophysical Research: Atmospheres 122, 12,167–12,193. URL:https://agupubs.onlinelibrary. wiley.com/doi/abs/10.1002/2017JD02...

    2017 doi

  15. [23]

    The impact of canyon geometry on intra urban and urban-suburban night temperature differences under warm weather conditions

    Giannopoulou, K., Santamouris, M., Livada, I., Georgakis, C., Caouris, Y., 2010. The impact of canyon geometry on intra urban and urban-suburban night temperature differences under warm weather conditions. Pure and Applied Geophysics 167, 1433–1449. doi:10.1007/s00024-010-0099-8

    2010 doi

  16. [24]

    Google earth engine: Planetary-scale geospatial analysis for everyone

    Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017. Google earth engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment 202, 18–27. doi:10.1016/j.rse.2017.06.031

    2017 doi

  17. [25]

    Mapping environmen- tal suitability changes for arbovirus mosquitoes in southeast asia: 1960–2020

    Hou, W., Zhou, Y., Luo, W., Wang, L., Kwan, M.P., Cook, A.R., 2024. Mapping environmen- tal suitability changes for arbovirus mosquitoes in southeast asia: 1960–2020. iScience 27, 110498. URL:https://www.sciencedirect.com/science/article/pii/S2589004224017231, doi:https:// doi...

    2024

  18. [26]

    Stgan: Swin transformer-based 29 gan to achieve remote sensing image super-resolution reconstruction

    Huo, W., Zhang, X., You, S., Zhang, Y., Zhang, Q., Hu, N., 2025. Stgan: Swin transformer-based 29 gan to achieve remote sensing image super-resolution reconstruction. Applied Sciences 15. URL: https://www.mdpi.com/2076-3417/15/1/305, doi:10.3390/app15010305

    2025 doi

  19. [27]

    Image-to-imagetranslation with conditionaladversarial networks, in: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Isola, P., Zhu, J.Y., Zhou, T., Efros, A.A., 2017. Image-to-imagetranslation with conditionaladversarial networks, in: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5967–

    2017

  20. [28]

    Assessment of forest fire severity and land surface temperature using google earth engine: a case study of gujarat state, india

    Jodhani, K.H., Patel, H., Soni, U., Patel, R., Valodara, B., Gupta, N., Patel, A., jee Omar, P., 2024. Assessment of forest fire severity and land surface temperature using google earth engine: a case study of gujarat state, india. Fire Ecology 20. doi:https://doi.org/10.1186/...

    2024 doi

  21. [29]

    Global land use/land cover with sentinel-2 and deep learning, in: 2021 IEEE International Geoscience and Re- mote Sensing Symposium IGARSS, IEEE

    Karra, K., Kontgis, C., Statman-Weil, Z., Mazzariello, J.C., Mathis, M., Brumby, S.P., 2021. Global land use/land cover with sentinel-2 and deep learning, in: 2021 IEEE International Geoscience and Re- mote Sensing Symposium IGARSS, IEEE. pp. 4704–4707. doi:10.1109/IGARSS47720...

    2021

  22. [30]

    Urban street canyons and heat islands: A system- atic review on morphological solutions

    Khalvandi, R., Karimimoshaver, M., 2025. Urban street canyons and heat islands: A system- atic review on morphological solutions. Results in Engineering 27, 106542. URL:https://www. sciencedirect.com/science/article/pii/S2590123025026118, doi:https://doi.org/10.1016/j. rineng....

    2025

  23. [31]

    Khedher, I., Favreau, J.M., Miguet, S., Gesquière, G., 2024. Rgb2lst: Enhancing deep learning-based land surface temperature estimation with multi-modality and artifacts removal, in: 2024 32nd Euro- pean Signal Processing Conference (EUSIPCO), pp. 2002–2006. doi:10.23919/EUSIP...

    2024

  24. [32]

    Spatial and temporal distribution of clouds observed by modis onboard the terra and aqua satellites

    King, M.D., Platnick, S., Menzel, W.P., Ackerman, S.A., Hubanks, P.A., 2013. Spatial and temporal distribution of clouds observed by modis onboard the terra and aqua satellites. IEEE Transactions on Geoscience and Remote Sensing 51, 3826–3852. doi:10.1109/TGRS.2012.2227333

    2013

  25. [33]

    Global revisit interval analysis of landsat-8/-9 and sentinel-2a/-2b data for terrestrial monitoring

    Li, J., Chen, B., 2020. Global revisit interval analysis of landsat-8/-9 and sentinel-2a/-2b data for terrestrial monitoring. Sensors 20, 6631. doi:10.3390/s20226631

    2020 doi

  26. [34]

    Generating spatiotemporal seamless data of clear-sky land surface temperature using synthetic aperture radar, digital elevation mode, and machine learning over vegetation areas

    Li, J., Yang, H., Chen, W., Li, C., Yang, G., 2024. Generating spatiotemporal seamless data of clear-sky land surface temperature using synthetic aperture radar, digital elevation mode, and machine learning over vegetation areas. Journal of Remote Sensing 4, 0071. URL:https: /...

    2024 doi

  27. [35]

    Evaluation of landsat 8-like land surface temperature by fusing landsat 8 and modis land surface temperature product

    Li, S., Wang, J., Li, D., Ran, Z., Yang, B., 2021. Evaluation of landsat 8-like land surface temperature by fusing landsat 8 and modis land surface temperature product. Processes 9. URL:https://www. mdpi.com/2227-9717/9/12/2262, doi:10.3390/pr9122262. 30

    2021 doi

  28. [36]

    Multi-feature combined cloud and cloud shadow detection in gaofen-1 wide field of view imagery

    Li, Z., Shen, H., Li, H., Xia, G., Gamba, P., Zhang, L., 2017. Multi-feature combined cloud and cloud shadow detection in gaofen-1 wide field of view imagery. Remote Sensing of Environment 191, 342–358. URL:https://www.sciencedirect.com/science/article/pii/S003442571730038X, d...

    2017 doi

  29. [37]

    Satellite- derived land surface temperature: Current status and perspectives

    Li, Z.L., Tang, B.H., Wu, H., Ren, H., Yan, G., Wan, Z., Trigo, I.F., Sobrino, J.A., 2013. Satellite- derived land surface temperature: Current status and perspectives. Remote Sensing of Environment 131, 14–37. URL:https://www.sciencedirect.com/science/article/pii/S00344257120...

    2013 doi

  30. [38]

    Swint-srgan: Swintransformerenhancedgenerativeadversarial network for image super-resolution

    Liu, Q., Chen, L., Sun, Y., Liu, L., 2025. Swint-srgan: Swintransformerenhancedgenerativeadversarial network for image super-resolution. Electronics 14. URL:https://www.mdpi.com/2079-9292/14/17/ 3511, doi:10.3390/electronics14173511

    2025 doi

  31. [39]

    A review of reconstructing remotely sensed land surface temperature under cloudy conditions

    Mo, Y., Xu, Y., Chen, H., Zhu, S., 2021. A review of reconstructing remotely sensed land surface temperature under cloudy conditions. Remote Sensing 13. URL:https://www.mdpi.com/2072-4292/ 13/14/2838, doi:10.3390/rs13142838

    2021 doi

  32. [40]

    Leveraging geographic information system for dengue surveillance: A scoping review

    Nayak, P.P., Pai B., J., Govindan, S., 2025. Leveraging geographic information system for dengue surveillance: A scoping review. Tropical Medicine and Health 53, 783. URL:https://doi.org/10. 1186/s41182-025-00783-9, doi:10.1186/s41182-025-00783-9

    2025 doi

  33. [41]

    Analysis of land surface temperature (lst) in bur ni geureudong geothermal field, aceh, indonesia using landsat 8 oli / tirs images

    Putri, D.R., Ismail, N., Idroes, R., Rizal, S., Nur, S., Nanda, M., 2021. Analysis of land surface temperature (lst) in bur ni geureudong geothermal field, aceh, indonesia using landsat 8 oli / tirs images. Chiang Mai University Journal of Natural Sciences URL:https://api.sema...

    2021

  34. [42]

    Characterizing wildfire behav- ior with ecostress land surface temperature across four california case studies

    Rivera, M., Lowman, L.E., Mrad, A., Bar, S., Banerjee, T., 2026. Characterizing wildfire behav- ior with ecostress land surface temperature across four california case studies. Ecological Infor- matics , 103777URL:https://www.sciencedirect.com/science/article/pii/S157495412600...

    2026

  35. [43]

    Simonyan, K., Zisserman, A., 2015. Very deep convolutional networks for large-scale image recognition, in: 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings. URL:https://arxiv.org/abs/1409.1556

    2015 arXiv

  36. [44]

    Impact of spatial reso- lution and satellite overpass time on evaluation of the surface urban heat island effects

    Sobrino, J., Oltra-Carrió, R., Sòria, G., Bianchi, R., Paganini, M., 2012. Impact of spatial reso- lution and satellite overpass time on evaluation of the surface urban heat island effects. Remote Sensing of Environment 117, 50–56. URL:https://www.sciencedirect.com/science/art...

    2012 doi

  37. [45]

    Resolution-robust large mask inpainting with fourier convolutions

    Suvorov, R., Logacheva, E., Mashikhin, A., Remizova, A., Ashukha, A., Silvestrov, A., Kong, N., Goka, H., Park, K., Lempitsky, V., 2021. Resolution-robust large mask inpainting with fourier convolutions. arXiv preprint arXiv:2109.07161

    2021 arXiv

  38. [46]

    Downwelling longwave radiation and sensible heat flux observations are critical for surface temperature and emissivity estimation from flux tower data

    Thakur, G., Schymanski, S.J., Mallick, K., Trebs, I., Sulis, M., 2022. Downwelling longwave radiation and sensible heat flux observations are critical for surface temperature and emissivity estimation from flux tower data. Scientific Reports 12, 8592. URL:https://doi.org/10.10...

    2022 doi

  39. [47]

    Sar data applications in earth observation: An overview

    Tsokas, A., Rysz, M., Pardalos, P.M., Dipple, K., 2022. Sar data applications in earth observation: An overview. Expert Systems with Applications 205, 117342. URL:https://www.sciencedirect.com/ science/article/pii/S0957417422006960, doi:https://doi.org/10.1016/j.eswa.2022.117342

    2022

  40. [48]

    Geological Survey, n.d

    U.S. Geological Survey, n.d. Landsat collection 2 surface temperature.https://www.usgs.gov/ landsat-missions/landsat-collection-2-surface-temperature. Accessed: 2025-01-05

    2025

  41. [49]

    Atmospheric correction algorithm: spectral reflectances (mod09)

    Vermote, E., Vermeulen, A., 1999. Atmospheric correction algorithm: spectral reflectances (mod09). ATBD version 4, 1–107

    1999

  42. [50]

    Urban heat island: Causes, consequences, and mitigation measures with emphasis on reflective and permeable pavements

    Vujovic, S., Haddad, B., Karaky, H., Sebaibi, N., Boutouil, M., 2021. Urban heat island: Causes, consequences, and mitigation measures with emphasis on reflective and permeable pavements. CivilEng 2, 459–484. URL:https://www.mdpi.com/2673-4109/2/2/26, doi:10.3390/civileng2020026

    2021 doi

  43. [51]

    Generating daily land surface temperature at landsat resolution by fusing landsat and modis data

    Weng, Q., Fu, P., Gao, F., 2014. Generating daily land surface temperature at landsat resolution by fusing landsat and modis data. Remote Sensing of Environment 145, 55–67. URL:https://www. sciencedirect.com/science/article/pii/S0034425714000479, doi:https://doi.org/10.1016/j....

    2014 doi

  44. [52]

    Spatially continuous and high-resolution land surface temperature product generation: A review of reconstruction and spatiotemporal fusion techniques

    Wu, P., Yin, Z., Zeng, C., Duan, S.B., Göttsche, F.M., Ma, X., Li, X., Yang, H., Shen, H., 2021. Spatially continuous and high-resolution land surface temperature product generation: A review of reconstruction and spatiotemporal fusion techniques. IEEE Geoscience and Remote Se...

    2021

  45. [53]

    Shared insights for heat health risk adaptation in metropolitan areas of developing countries

    Yu, W., Yang, J., Sun, D., Xue, B., Sun, W., Ren, J., Yu, H., Xiao, X., Xia, J., Li, X., 2024. Shared insights for heat health risk adaptation in metropolitan areas of developing countries. iScience 27, 109728. URL:https://www.sciencedirect.com/science/article/pii/S25890042240...

    2024

  46. [54]

    Stgan-cr: A swin transformer-enhanced gan framework for effective cloud removal in satellite imagery, pp

    Zhu, H., Wang, Z., Xu, M., Zhang, J., Liu, Q., Du, B., 2024. Stgan-cr: A swin transformer-enhanced gan framework for effective cloud removal in satellite imagery, pp. 51–56. doi:10.18293/SEKE2024-059. 32

    2024 doi

  47. [55]

    An enhanced spatial and temporal adaptive reflectance fusion model for complex heterogeneous regions

    Zhu, X., Chen, J., Gao, F., Chen, X., Masek, J.G., 2010. An enhanced spatial and temporal adaptive reflectance fusion model for complex heterogeneous regions. Remote Sensing of Environment 114, 2610–

    2010

  48. [56]

    Reconstruction of land surface temperatureundercloudyconditionsfromlandsat8datausingannualtemperaturecyclemodel

    Zhu, X., Duan, S.B., Li, Z.L., Wu, P., Wu, H., Zhao, W., Qian, Y., 2022. Reconstruction of land surface temperatureundercloudyconditionsfromlandsat8datausingannualtemperaturecyclemodel. Remote Sensing of Environment 281, 113261. URL:https://www.sciencedirect.com/science/articl...

    2022

  49. [2026]

    Earth System Science Data 18, 3889–3933

    Indicators of global climate change 2025: annual update of key indicators of the state of the climate system and human influence. Earth System Science Data 18, 3889–3933. URL:https://essd. copernicus.org/articles/18/3889/2026/, doi:10.5194/essd-18-3889-2026

    2025 doi

  50. [2623]

    doi:10.1016/j.rse.2010.05.032

    2010 doi

  51. [5976]

    doi:10.1109/CVPR.2017.632

    2017 doi

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.