pith. machine review for the scientific record. sign in

arxiv: 2605.10958 · v1 · submitted 2026-05-04 · ⚛️ physics.ao-ph · cs.AI

Recognition: no theorem link

Multi-Fidelity Emulation of Atmospheric Correction Coefficients with Physics-Guided Kolmogorov-Arnold Networks

Md Abdullah Al Mazid , Naphtali Rishe
Authors on Pith no claims yet

Pith reviewed 2026-05-13 07:51 UTC · model grok-4.3

classification ⚛️ physics.ao-ph cs.AI
keywords atmospheric correctionmulti-fidelity emulationKolmogorov-Arnold networksradiative transfer modelingphysics-guided machine learningremote sensingSentinel-2
0
0 comments X

The pith

A physics-guided Kolmogorov-Arnold network predicts high-fidelity atmospheric correction coefficients by learning residuals from low-fidelity simulations.

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

The paper introduces a multi-fidelity surrogate that pairs 6S and libRadtran radiative transfer runs to train a model called pKANrtm. Atmospheric states and low-fidelity outputs feed into an Efficient-KAN architecture that predicts the difference to high-fidelity targets while a penalty term keeps the reconstructed coefficients physically consistent. The approach targets repeated generation of look-up tables for Sentinel-2 preprocessing, sensitivity studies, and operational correction. A reader would care because full libRadtran runs stay too slow for dense sampling or real-time use, yet the surrogate delivers stronger accuracy than standard regression models on both routine and out-of-distribution cases. Runtime tests show GPU inference reaches tens of thousands of samples per second.

Core claim

pKANrtm receives atmospheric and geometric inputs plus 6S-derived path reflectance, transmittance, and albedo, predicts the residual relative to libRadtran under matched conditions, and reconstructs the high-fidelity coefficients; when trained with an additional physics-consistency penalty in coefficient space, it records the best overall predictive performance among compared regression-based RTM surrogates across both standard and out-of-distribution test sets.

What carries the argument

pKANrtm, a physics-guided Efficient-KAN that ingests atmospheric state and low-fidelity 6S coefficients to output residuals, then applies a physics-consistency penalty in the original coefficient space before reconstruction.

If this is right

  • Dense look-up tables for Sentinel-2 atmospheric correction can be generated orders of magnitude faster than repeated libRadtran calls.
  • Operational preprocessing pipelines gain both speed and accuracy without sacrificing physical structure in the coefficients.
  • Sensitivity analysis and retrieval support become feasible at higher spatial or temporal density than full-physics runs allow.
  • The same multi-fidelity residual strategy applies to any paired low- and high-fidelity radiative transfer models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The framework could be retrained on other sensors by swapping the spectral-response functions and repeating the paired 6S-libRadtran sampling.
  • If the residual-prediction structure generalizes, it might reduce the cost of ensemble atmospheric correction in climate-model downscaling.
  • Integration into onboard satellite processors becomes plausible once the model is quantized and the inference speed reaches real-time rates.
  • The approach invites direct comparison with physics-informed neural operators that operate on the full radiative transfer equation rather than coefficient residuals.

Load-bearing premise

The physics-consistency penalty applied in coefficient space will enforce physical consistency on unseen atmospheric states without introducing systematic biases in the residual predictions.

What would settle it

Compute energy-balance or reciprocity violations for pKANrtm outputs versus direct libRadtran outputs on a held-out set of 10,000 atmospheric profiles drawn from regions outside the Latin Hypercube training distribution; systematic growth in violations would falsify the claim that the penalty preserves consistency.

Figures

Figures reproduced from arXiv: 2605.10958 by Md Abdullah Al Mazid, Naphtali Rishe.

Figure 1
Figure 1. Figure 1: Architecture and physics-guided training workflow of pKANrtm 3.7. Physics-guided training objective Targets are standardized using a StandardScaler fitted on the training set. The supervised term is computed in standardized target space using MSE. For the physics penalty, predictions are inverse transformed back to the original coefficient scale, and the penalty is computed on the unscaled predicted coeffi… view at source ↗
Figure 2
Figure 2. Figure 2: summarizes the sampled state-space coverage and accepted band distribution. The continuous variables show broad coverage over the prescribed sampling ranges, consistent with the use of LHS. The band histogram also reveals that most Sentinel-2 bands contribute similar sample counts after quality control, while B10 has substantially fewer valid rows. This reduction is expected because B10 lies in a cirrus an… view at source ↗
Figure 3
Figure 3. Figure 3: Band-wise discrepancy between libRadtran and 6S coefficients. Lines show mean and median absolute difference, while the shaded region indicates the interquartile range The conditional error heatmaps in [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Median absolute 6S-libRadtran discrepancy conditioned on atmospheric and geometric variables. Rows correspond to binned values of AOD550, CWV, solar zenith angle, and elevation; columns correspond to Sentinel-2 bands; and color denotes median absolute error for each coefficient 5.2. Overall accuracy Tables 3 and 4 report overall accuracy on the standard and OOD split regimes. On the standard split, the bas… view at source ↗
Figure 5
Figure 5. Figure 5: compares mean RMSE across coefficients for the standard and OOD splits. The OOD split increases error for Ttotal and ρpath, whereas spherical albedo shows slightly lower aggregate RMSE in the OOD summary. This pattern suggests that the impact of OOD sampling is coefficient-specific rather than uniform. In particular, transmittance is more sensitive to changes in aerosol and water-vapour state, while spheri… view at source ↗
Figure 6
Figure 6. Figure 6: Band- and coefficient-wise SMAPE under standard and OOD splits. B10 is the dominant high-error band, with additional sensitivity in B9, B11, and B12 5.4. Prediction-versus-truth behavior Figures 7 and 8 show predicted versus true libRadtran coefficients for the standard and OOD splits. For ρpath and Ttotal, predictions align closely with the one-to-one line in both regimes, indicating that the residual mod… view at source ↗
Figure 7
Figure 7. Figure 7: Predicted versus true libRadtran coefficients on the standard split. The dashed red line denotes the one-to-one relationship [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Predicted versus true libRadtran coefficients on the OOD AOD-CWV split Spherical albedo shows a different pattern. Most predictions follow the one-to-one line in the dense low-to-moderate coefficient region, but sparse high-value cases appear as outliers. These outliers are important because they may correspond to absorption-sensitive or numerically fragile regimes. The model therefore performs well for th… view at source ↗
Figure 9
Figure 9. Figure 9: Representative standard-split coefficient curves across Sentinel-2 bands. pKANrtm closely follows libRadtran while correcting the low-fidelity 6S discrepancy [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Representative OOD-split coefficient curves across Sentinel-2 bands. The surrogate remains close to libRadtran even under shifted AOD-CWV conditions The broader representative-case panel in [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Representative atmospheric cases comparing libRadtran, 6S, and model-predicted coeffi￾cient curves. Rows correspond to selected atmospheric regimes and columns correspond to the three target coefficients 5.6. Runtime and acceleration [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
read the original abstract

Atmospheric correction is a critical preprocessing step in optical remote sensing, but repeated high-fidelity radiative transfer simulations remain computationally expensive for dense look-up-table generation, sensitivity analysis, retrieval support, and operational preprocessing. This study presents a physics-aware multi-fidelity surrogate framework for emulating atmospheric correction coefficients using paired 6S and libRadtran simulations. Atmospheric and geometric states are sampled using Latin Hypercube Sampling, and both radiative transfer models are evaluated under matched conditions for Sentinel-2 bands using spectral-response-function-aware coefficient generation. The high-fidelity targets are path reflectance, total transmittance, and spherical albedo. A physics-guided Kolmogorov-Arnold Network, termed pKANrtm, receives the atmospheric state and low-fidelity 6S coefficients, predicts the residual relative to libRadtran, and reconstructs the high-fidelity coefficients. The pKANrtm model uses an Efficient-KAN architecture and is trained with a physics-consistency penalty applied in the original coefficient space. The proposed model is evaluated against state-of-the-art regression-based RTM surrogates. Across both standard and out-of-distribution evaluation settings, pKANrtm achieves the strongest overall predictive performance among the compared models. Runtime benchmarking demonstrates substantial acceleration relative to libRadtran, with GPU inference providing approximately four orders of magnitude single-sample speedup and batched inference reaching tens of thousands of samples per second. These results indicate that physics-aware multi-fidelity pKANrtm emulation provides an accurate, physically structured, and computationally efficient strategy for atmospheric correction coefficient generation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces pKANrtm, a physics-guided multi-fidelity Kolmogorov-Arnold Network that emulates atmospheric correction coefficients (path reflectance, total transmittance, spherical albedo) for Sentinel-2 by predicting residuals between 6S (low-fidelity) and libRadtran (high-fidelity) simulations. Atmospheric states are sampled via Latin Hypercube Sampling, and the model is trained with a physics-consistency penalty in coefficient space. It reports outperforming state-of-the-art regression surrogates in both in-distribution and out-of-distribution tests, with substantial runtime speedups (four orders of magnitude on GPU).

Significance. If the performance and generalization claims hold with quantitative support, this provides an efficient, physically structured surrogate for radiative transfer simulations that could accelerate look-up table generation, sensitivity analysis, and operational atmospheric correction in remote sensing while reducing reliance on repeated high-fidelity runs.

major comments (2)
  1. [Abstract] Abstract: the central claim that pKANrtm 'achieves the strongest overall predictive performance among the compared models' across standard and OOD settings is stated without any numerical metrics, error bars, sample counts, or details on validation splits and data exclusion, leaving the empirical superiority only partially supported.
  2. [Evaluation] Evaluation section: no post-hoc diagnostics (e.g., sign or magnitude of mean residuals, transmittance bound violations, or coefficient-space consistency checks) are reported on the OOD split, so it is not verified whether the physics-consistency penalty (applied only during training on paired 6S-libRadtran residuals) prevents systematic extrapolation biases on unseen atmospheric states.
minor comments (2)
  1. [Abstract] The acronym pKANrtm is introduced without expansion on first use.
  2. [Methods] The exact mathematical form of the physics-consistency penalty (including which coefficients receive the penalty and the value of the weighting coefficient) should be stated explicitly for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed review. We address each major comment below and indicate the specific revisions that will be incorporated into the next version of the manuscript to strengthen the empirical support and validation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that pKANrtm 'achieves the strongest overall predictive performance among the compared models' across standard and OOD settings is stated without any numerical metrics, error bars, sample counts, or details on validation splits and data exclusion, leaving the empirical superiority only partially supported.

    Authors: We agree that the abstract would be strengthened by including quantitative support for the performance claims. In the revised manuscript we will add representative numerical metrics (RMSE for each coefficient on both standard and OOD splits), sample counts, and a brief statement on the validation and OOD exclusion protocol so that the superiority statement is directly substantiated by numbers. revision: yes

  2. Referee: [Evaluation] Evaluation section: no post-hoc diagnostics (e.g., sign or magnitude of mean residuals, transmittance bound violations, or coefficient-space consistency checks) are reported on the OOD split, so it is not verified whether the physics-consistency penalty (applied only during training on paired 6S-libRadtran residuals) prevents systematic extrapolation biases on unseen atmospheric states.

    Authors: We acknowledge that explicit post-hoc diagnostics on the OOD split would provide additional verification of the physics-consistency penalty. We will add a dedicated subsection in the revised Evaluation section that reports (i) mean residual sign and magnitude per coefficient on the OOD set, (ii) fraction of transmittance values violating physical bounds, and (iii) coefficient-space consistency metrics, thereby demonstrating that the penalty continues to limit systematic extrapolation biases on unseen atmospheric states. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the multi-fidelity pKANrtm emulation framework

full rationale

The paper constructs a surrogate by training an Efficient-KAN on paired 6S and libRadtran radiative-transfer simulations generated via Latin Hypercube Sampling; the network predicts residuals to reconstruct high-fidelity coefficients while a physics-consistency penalty is added to the training loss in coefficient space. All performance claims rest on direct empirical comparison against regression baselines in both standard and out-of-distribution splits, with no derivation step that reduces by construction to a fitted parameter, self-definition, or self-citation chain. The external RTM runs supply independent grounding, and the penalty term does not redefine the target quantities.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

The framework depends on the assumption that low-fidelity 6S outputs form a reliable base for residual correction to libRadtran, on Latin Hypercube Sampling covering relevant atmospheric states, and on the KAN weights being fitted to paired simulation data.

free parameters (2)
  • KAN network weights and biases
    Learned parameters of the Efficient-KAN that map low-fidelity inputs to high-fidelity residuals.
  • Physics-consistency penalty coefficient
    Hyperparameter controlling the strength of the physics penalty term during training.
axioms (1)
  • domain assumption Low-fidelity 6S simulations provide a useful approximation whose residuals to high-fidelity libRadtran results can be learned from atmospheric state inputs.
    Invoked in the multi-fidelity residual prediction setup described in the abstract.
invented entities (1)
  • pKANrtm no independent evidence
    purpose: Physics-guided multi-fidelity emulator for path reflectance, transmittance, and spherical albedo coefficients.
    New model introduced to combine KAN architecture with physics penalty and multi-fidelity inputs.

pith-pipeline@v0.9.0 · 5584 in / 1450 out tokens · 44364 ms · 2026-05-13T07:51:57.163365+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    Second simulation of the satellite signal in the solar spectrum, 6S: An overview.IEEE Transactions on Geoscience and Remote Sensing 1997,35, 675–686

    Vermote, E.; Tanre, D.; Deuze, J.; Herman, M.; Morcette, J.J. Second simulation of the satellite signal in the solar spectrum, 6S: An overview.IEEE Transactions on Geoscience and Remote Sensing 1997,35, 675–686. https://doi.org/10.1109/36.581987

    work page doi:10.1109/36.581987 1997

  2. [2]

    Validation of a vector version of the 6S radiative transfer code for atmospheric correction of satellite data

    Kotchenova, S.Y.; Vermote, E.F.; Matarrese, R.; Klemm, F.J.J. Validation of a vector version of the 6S radiative transfer code for atmospheric correction of satellite data. Part I: Path radiance. Applied Optics2006,45, 6762–6774. https://doi.org/10.1364/AO.45.006762. https://doi.org/ Version May 13, 2026 submitted toRemote Sens. 20 of 22

    work page doi:10.1364/ao.45.006762 2026

  3. [3]

    MODTRAN6: A major upgrade of the MODTRAN radiative transfer code

    Berk, A.; Conforti, P .; Kennett, R.; Perkins, T.; Hawes, F.; van den Bosch, J. MODTRAN6: A major upgrade of the MODTRAN radiative transfer code. In Proceedings of the Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XX. SPIE, 2014, Vol. 9088, p. 90880H. https://doi.org/10.1117/12.2050433

    work page doi:10.1117/12.2050433 2014

  4. [4]

    Technical note: The libRadtran software package for radiative transfer calculations—description and examples of use.Atmospheric Chemistry and Physics2005,5, 1855–

    Mayer, B.; Kylling, A. Technical note: The libRadtran software package for radiative transfer calculations—description and examples of use.Atmospheric Chemistry and Physics2005,5, 1855–

    work page

  5. [5]

    https://doi.org/10.5194/acp-5-1855-2005

    work page doi:10.5194/acp-5-1855-2005 2005

  6. [6]

    Numerically stable algorithm for discrete- ordinate-method radiative transfer in multiple scattering and emitting layered media.Applied Optics1988,27, 2502–2509

    Stamnes, K.; Tsay, S.C.; Wiscombe, W.; Jayaweera, K. Numerically stable algorithm for discrete- ordinate-method radiative transfer in multiple scattering and emitting layered media.Applied Optics1988,27, 2502–2509. https://doi.org/10.1364/AO.27.002502

    work page doi:10.1364/ao.27.002502

  7. [7]

    SBDART: A research and teaching software tool for plane-parallel radiative transfer in the Earth’s atmosphere.Bulletin of the American Meteorological Society1998,79, 2101–2114

    Ricchiazzi, P .; Yang, S.; Gautier, C.; Sowle, D. SBDART: A research and teaching software tool for plane-parallel radiative transfer in the Earth’s atmosphere.Bulletin of the American Meteorological Society1998,79, 2101–2114. https://doi.org/10.1175/1520-0477(1998)079<2101: SARATS>2.0.CO;2

    work page doi:10.1175/1520-0477(1998)079 1998

  8. [8]

    A Landsat surface reflectance dataset for North America, 1990–2000.IEEE Geoscience and Remote Sensing Letters2006,3, 68–72

    Masek, J.G.; Vermote, E.F.; Saleous, N.E.; Wolfe, R.; Hall, F.G.; Huemmrich, K.F.; Gao, F.; Kutler, J.; Lim, T.K. A Landsat surface reflectance dataset for North America, 1990–2000.IEEE Geoscience and Remote Sensing Letters2006,3, 68–72. https://doi.org/10.1109/LGRS.2005.857030

    work page doi:10.1109/lgrs.2005.857030 1990

  9. [9]

    Preliminary analysis of the performance of the Landsat 8/OLI land surface reflectance product.Remote Sensing of Environment2016,185, 46–56

    Vermote, E.; Justice, C.; Claverie, M.; Franch, B. Preliminary analysis of the performance of the Landsat 8/OLI land surface reflectance product.Remote Sensing of Environment2016,185, 46–56. https://doi.org/10.1016/j.rse.2016.04.008

    work page doi:10.1016/j.rse.2016.04.008 2016

  10. [10]

    Geo-atmospheric processing of airborne imaging spectrometry data

    Richter, R.; Schläpfer, D. Geo-atmospheric processing of airborne imaging spectrometry data. Part 2: Atmospheric/topographic correction.International Journal of Remote Sensing2002, 23, 2631–2649. https://doi.org/10.1080/01431160110115834

    work page doi:10.1080/01431160110115834

  11. [11]

    Hagolle, O.; Huc, M.; Villa Pascual, D.; Dedieu, G. A multi-temporal and multi-spectral method to estimate aerosol optical thickness over land, for the atmospheric correction of FormoSat-2, LandSat, VEN µS and Sentinel-2 images.Remote Sensing2015,7, 2668–2691. https://doi.org/10.3390/rs70302668

    work page doi:10.3390/rs70302668

  12. [12]

    The harmonized Landsat and Sentinel-2 surface reflectance data set.Remote Sensing of Environment2018,219, 145–161

    Claverie, M.; Ju, J.; Masek, J.G.; Dungan, J.L.; Vermote, E.F.; Roger, J.C.; Skakun, S.V .; Justice, C. The harmonized Landsat and Sentinel-2 surface reflectance data set.Remote Sensing of Environment2018,219, 145–161. https://doi.org/10.1016/j.rse.2018.09.002

    work page doi:10.1016/j.rse.2018.09.002 2018

  13. [13]

    Atmospheric correction algorithms for hyperspectral remote sensing data of land and ocean.Remote Sensing of Environment2009, 113, S17–S24

    Gao, B.C.; Montes, M.J.; Davis, C.O.; Goetz, A.F.H. Atmospheric correction algorithms for hyperspectral remote sensing data of land and ocean.Remote Sensing of Environment2009, 113, S17–S24. https://doi.org/10.1016/j.rse.2007.12.015

    work page doi:10.1016/j.rse.2007.12.015 2007

  14. [14]

    Atmospheric correction inter-comparison exercise.Remote Sensing 2018,10, 352

    Doxani, G.; Vermote, E.; Roger, J.C.; Gascon, F.; Adriaensen, S.; Frantz, D.; Hagolle, O.; Hollstein, A.; Kirches, G.; Li, F.; et al. Atmospheric correction inter-comparison exercise.Remote Sensing 2018,10, 352. https://doi.org/10.3390/rs10020352

    work page doi:10.3390/rs10020352 2018

  15. [15]

    Doxani, G.; Vermote, E.F.; Roger, J.C.; Skakun, S.; Gascon, F.; Collison, A.; De Keukelaere, L.; Desjardins, C.; Frantz, D.; Hagolle, O.; et al. Atmospheric correction inter-comparison exercise, ACIX-II land: An assessment of atmospheric correction processors for Landsat 8 and Sentinel-2 over land.Remote Sensing of Environment2023,285, 113412. https://doi...

    work page doi:10.1016/j.rse.2022 2022

  16. [16]

    Atmospheric correction inter-comparison exercise, ACIX-III land: An assessment of atmospheric correction processors for EnMAP and PRISMA over land.Remote Sensing2025,17, 3790

    Cremer, N.; Alonso, K.; Doxani, G.; Chlus, A.; Thompson, D.R.; Brodrick, P .; Townsend, P .A.; Palombo, A.; Santini, F.; Gao, B.C.; et al. Atmospheric correction inter-comparison exercise, ACIX-III land: An assessment of atmospheric correction processors for EnMAP and PRISMA over land.Remote Sensing2025,17, 3790. https://doi.org/10.3390/rs17233790

    work page doi:10.3390/rs17233790

  17. [17]

    Optimal estimation for imaging spectrometer atmospheric correction.Remote Sensing of Environment 2018,216, 355–373

    Thompson, D.R.; Natraj, V .; Green, R.O.; Helmlinger, M.C.; Gao, B.C.; Eastwood, M.L. Optimal estimation for imaging spectrometer atmospheric correction.Remote Sensing of Environment 2018,216, 355–373. https://doi.org/10.1016/j.rse.2018.07.003

    work page doi:10.1016/j.rse.2018.07.003 2018

  18. [18]

    Generalized radiative transfer emulation for imaging spectroscopy reflectance retrievals.Remote Sensing of Environment2021,261, 112476

    Brodrick, P .G.; Thompson, D.R.; Fahlen, J.E.; Eastwood, M.L.; Sarture, C.M.; Lundeen, S.R.; Olson-Duvall, W.; Carmon, N.; Green, R.O. Generalized radiative transfer emulation for imaging spectroscopy reflectance retrievals.Remote Sensing of Environment2021,261, 112476. https://doi.org/10.1016/j.rse.2021.112476

    work page doi:10.1016/j.rse.2021.112476 2021

  19. [19]

    Efficient emulation of radiative transfer codes using Gaussian processes and application to land surface parameter inferences.Remote Sensing2016, 8, 119

    Gómez-Dans, J.L.; Lewis, P .E.; Disney, M. Efficient emulation of radiative transfer codes using Gaussian processes and application to land surface parameter inferences.Remote Sensing2016, 8, 119. https://doi.org/10.3390/rs8020119. https://doi.org/ Version May 13, 2026 submitted toRemote Sens. 21 of 22

    work page doi:10.3390/rs8020119 2026

  20. [20]

    Emulation of leaf, canopy and atmosphere radiative transfer models for fast global sensitivity analysis.Remote Sensing2016,8, 673

    Verrelst, J.; Rivera, J.P .; Moreno, J.; Camps-Valls, G. Emulation of leaf, canopy and atmosphere radiative transfer models for fast global sensitivity analysis.Remote Sensing2016,8, 673. https://doi.org/10.3390/rs8080673

    work page doi:10.3390/rs8080673

  21. [21]

    Systematic assessment of MODTRAN emulators for atmospheric correction.IEEE Transactions on Geoscience and Remote Sensing2022,60, 1–17

    Vicent Servera, J.; Rivera-Caicedo, J.P .; Verrelst, J.; Muñoz-Marí, J.; Sabater, N.; Berthelot, B.; Camps-Valls, G.; Moreno, J. Systematic assessment of MODTRAN emulators for atmospheric correction.IEEE Transactions on Geoscience and Remote Sensing2022,60, 1–17. https://doi.org/ 10.1109/TGRS.2021.3071376

    work page doi:10.1109/tgrs.2021.3071376 2021

  22. [22]

    Neural network radiative transfer for imaging spectroscopy.Atmospheric Measurement Techniques2019,12, 2567–2578

    Bue, B.D.; Thompson, D.R.; Deshpande, S.; Eastwood, M.; Green, R.O.; Natraj, V .; Mullen, T.; Parente, M. Neural network radiative transfer for imaging spectroscopy.Atmospheric Measurement Techniques2019,12, 2567–2578. https://doi.org/10.5194/amt-12-2567-2019

    work page doi:10.5194/amt-12-2567-2019 2019

  23. [23]

    Surrogate models of radiative transfer codes for atmospheric trace gas retrievals from satellite observations.Machine Learning2023, 112, 1337–1363

    Brence, J.; Tanevski, J.; Adams, J.; Malina, E.; Dzeroski, S. Surrogate models of radiative transfer codes for atmospheric trace gas retrievals from satellite observations.Machine Learning2023, 112, 1337–1363. https://doi.org/10.1007/s10994-022-06155-2

    work page doi:10.1007/s10994-022-06155-2

  24. [24]

    Using deep neural networks as cost-effective surrogate models for super-parameterized E3SM radiative transfer.Geophysical Research Letters2019, 46, 6069–6079

    Pal, A.; Mahajan, S.; Norman, M.R. Using deep neural networks as cost-effective surrogate models for super-parameterized E3SM radiative transfer.Geophysical Research Letters2019, 46, 6069–6079. https://doi.org/10.1029/2018GL081646

    work page doi:10.1029/2018gl081646

  25. [25]

    RadNet 1.0: Exploring deep learning architectures for longwave radiative transfer.Geoscientific Model Development2020,13, 4399–4412

    Liu, Y.; Caballero, R.; Monteiro, J.M. RadNet 1.0: Exploring deep learning architectures for longwave radiative transfer.Geoscientific Model Development2020,13, 4399–4412. https: //doi.org/10.5194/gmd-13-4399-2020

    work page doi:10.5194/gmd-13-4399-2020 2020

  26. [26]

    Evaluation of neural network emulations for radiation parameterization in cloud resolving model.Geophysical Research Letters2020,47, e2020GL089444

    Roh, S.; Song, H.J. Evaluation of neural network emulations for radiation parameterization in cloud resolving model.Geophysical Research Letters2020,47, e2020GL089444. https://doi.org/ 10.1029/2020GL089444

    work page doi:10.1029/2020gl089444

  27. [27]

    Surro- gate modeling of MODTRAN physical radiative transfer code using deep-learning regression

    Aghdami-Nia, M.; Shah-Hosseini, R.; Homayouni, S.; Rostami, A.; Ahmadian, N. Surro- gate modeling of MODTRAN physical radiative transfer code using deep-learning regression. Environmental Sciences Proceedings2024,29, 16. https://doi.org/10.3390/ECRS2023-16294

    work page doi:10.3390/ecrs2023-16294

  28. [28]

    Gaussian process and deep learning atmospheric correction.Remote Sensing2023,15, 649

    Basener, B.; Basener, A. Gaussian process and deep learning atmospheric correction.Remote Sensing2023,15, 649. https://doi.org/10.3390/rs15030649

    work page doi:10.3390/rs15030649

  29. [29]

    RTM surrogate modeling in optical remote sensing: A review of emulation for vegetation and atmosphere applications.Remote Sensing2025,17, 3618

    Verrelst, J.; Morata, M.; Garcia-Soria, J.L.; Sun, Y.; Qi, J.; Rivera-Caicedo, J.P . RTM surrogate modeling in optical remote sensing: A review of emulation for vegetation and atmosphere applications.Remote Sensing2025,17, 3618. https://doi.org/10.3390/rs17213618

    work page doi:10.3390/rs17213618

  30. [30]

    Predicting the output from a complex computer code when fast approximations are available.Biometrika2000,87, 1–13

    Kennedy, M.C.; O’Hagan, A. Predicting the output from a complex computer code when fast approximations are available.Biometrika2000,87, 1–13. https://doi.org/10.1093/biomet/87.1.1

    work page doi:10.1093/biomet/87.1.1

  31. [31]

    Multi-fidelity optimization via surrogate modelling

    Forrester, A.I.J.; Sóbester, A.; Keane, A.J. Multi-fidelity optimization via surrogate modelling. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences2007,463, 3251–

    work page

  32. [32]

    https://doi.org/10.1098/rspa.2007.1900

    work page doi:10.1098/rspa.2007.1900 2007

  33. [33]

    Perdikaris, P .; Raissi, M.; Damianou, A.; Lawrence, N.D.; Karniadakis, G.E. Nonlinear in- formation fusion algorithms for data-efficient multi-fidelity modelling.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences2017,473, 20160751. https: //doi.org/10.1098/rspa.2016.0751

    work page doi:10.1098/rspa.2016.0751 2016

  34. [34]

    Survey of multifidelity methods in uncertainty propagation, inference, and optimization.SIAM Review2018,60, 550–591

    Peherstorfer, B.; Willcox, K.; Gunzburger, M. Survey of multifidelity methods in uncertainty propagation, inference, and optimization.SIAM Review2018,60, 550–591. https://doi.org/10.1 137/16M1082469

    work page

  35. [35]

    A deep learning approach to fast radiative transfer.Journal of Quantitative Spectroscopy and Radiative Transfer2022,280, 108088

    Stegmann, P .G.; Johnson, B.; Moradi, I.; Karpowicz, B.; McCarty, W. A deep learning approach to fast radiative transfer.Journal of Quantitative Spectroscopy and Radiative Transfer2022,280, 108088. https://doi.org/10.1016/j.jqsrt.2022.108088

    work page doi:10.1016/j.jqsrt.2022.108088 2022

  36. [36]

    Using deep learning to emulate and accelerate a radiative transfer model.Journal of Atmospheric and Oceanic Technology 2021,38, 1673–1696

    Lagerquist, R.; Turner, D.; Ebert-Uphoff, I.; Stewart, J.; Hagerty, V . Using deep learning to emulate and accelerate a radiative transfer model.Journal of Atmospheric and Oceanic Technology 2021,38, 1673–1696. https://doi.org/10.1175/JTECH-D-21-0007.1

    work page doi:10.1175/jtech-d-21-0007.1 2021

  37. [37]

    A physics-incorporated deep learning framework for parameterization of atmospheric radiative transfer.Journal of Advances in Modeling Earth Systems 2023,15, e2022MS003445

    Yao, Y.; Zhong, X.; Zheng, Y.; Wang, Z. A physics-incorporated deep learning framework for parameterization of atmospheric radiative transfer.Journal of Advances in Modeling Earth Systems 2023,15, e2022MS003445. https://doi.org/10.1029/2022MS003445

    work page doi:10.1029/2022ms003445 2023

  38. [38]

    KAN: Kolmogorov–Arnold networks

    Liu, Z.; Wang, Y.; Vaidya, S.; Ruehle, F.; Halverson, J.; Soljacic, M.; Hou, T.Y.; Tegmark, M. KAN: Kolmogorov–Arnold networks. In Proceedings of the The Thirteenth International Conference on Learning Representations, 2025. https://doi.org/ Version May 13, 2026 submitted toRemote Sens. 22 of 22

    work page 2025

  39. [39]

    Raissi, M.; Perdikaris, P .; Karniadakis, G.E. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational Physics2019,378, 686–707. https://doi.org/10.1016/j.jcp.20 18.10.045

    work page doi:10.1016/j.jcp.20

  40. [40]

    Physics-informed machine learning.Nature Reviews Physics2021,3, 422–440

    Karniadakis, G.E.; Kevrekidis, I.G.; Lu, L.; Perdikaris, P .; Wang, S.; Yang, L. Physics-informed machine learning.Nature Reviews Physics2021,3, 422–440. https://doi.org/10.1038/s42254-021 -00314-5

    work page doi:10.1038/s42254-021

  41. [41]

    A comparison of three methods for selecting values of input variables in the analysis of output from a computer code.Technometrics1979, 21, 239–245

    McKay, M.D.; Beckman, R.J.; Conover, W.J. A comparison of three methods for selecting values of input variables in the analysis of output from a computer code.Technometrics1979, 21, 239–245. https://doi.org/10.1080/00401706.1979.10489755

    work page doi:10.1080/00401706.1979.10489755 1979

  42. [42]

    A deep learning perspective to atmospheric correction of satellite images

    Shah, M.; Raval, M.S.; Divakaran, S. A deep learning perspective to atmospheric correction of satellite images. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2022, pp. 3469–3472. https://doi.org/10.1109/IGARSS46834.2022.9884 900

    work page doi:10.1109/igarss46834.2022.9884 2022

  43. [43]

    A spatio-temporal deep learning model for enhanced atmospheric correction.Modeling Earth Systems and Environment2025,11, 3

    Shah, M.; Raval, M.S.; Divakaran, S.; Dhar, D.; Parmar, H. A spatio-temporal deep learning model for enhanced atmospheric correction.Modeling Earth Systems and Environment2025,11, 3. https://doi.org/10.1007/s40808-024-02175-0

    work page doi:10.1007/s40808-024-02175-0

  44. [44]

    Learning representations by back-propagating errors.Nature1986,323, 533–536

    Rumelhart, D.E.; Hinton, G.E.; Williams, R.J. Learning representations by back-propagating errors.Nature1986,323, 533–536. https://doi.org/10.1038/323533a0

    work page doi:10.1038/323533a0

  45. [45]

    Multilayer perceptron (MLP)

    Taud, H.; Mas, J.F. Multilayer perceptron (MLP). InGeomatic Approaches for Modeling Land Change Scenarios; Springer, 2018; pp. 451–455. https://doi.org/10.1007/978-3-319-60801-3_27

    work page doi:10.1007/978-3-319-60801-3_27 2018

  46. [46]

    efficient-kan: An efficient PyTorch implementation of Kolmogorov–Arnold networks

    Blealtan. efficient-kan: An efficient PyTorch implementation of Kolmogorov–Arnold networks. https://github.com/Blealtan/efficient-kan, 2024

    work page 2024

  47. [47]

    Backpropagation through time and the brain.Current Opinion in Neurobiology, 55:82–89, 2019

    Drusch, M.; Del Bello, U.; Carlier, S.; Colin, O.; Fernandez, V .; Gascon, F.; Hoersch, B.; Isola, C.; Laberinti, P .; Martimort, P .; et al. Sentinel-2: ESA’s optical high-resolution mission for GMES operational services.Remote Sensing of Environment2012,120, 25–36. https://doi.org/10.1016/j. rse.2011.11.026. Disclaimer/Publisher’s Note:The statements, o...

    work page doi:10.1016/j 2011