Bayesian Inference: Kernel-Based Model for Surface Temperature Reconstruction in Ice Borehole Thermometry

Kshema Shaju; Peter Zaspel; Thomas Laepple

REVIEW 3 major objections 5 minor 40 references

Bayesian Inference: Kernel-Based Model for Surface Temperature Reconstruction in Ice Borehole Thermometry

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Kernel-based Bayesian inversion reconstructs past surface temperature from shallow ice boreholes.

desk verdict Genuine efficiency advance for borehole inversion, but the 'reliable uncertainties' claim needs a coverage check because the approximation-error model ignores vertical correlations. read the letter →

arxiv 2607.20322 v1 pith:4GXD5TLK submitted 2026-07-22 physics.comp-ph

Kshema Shaju , Thomas Laepple , Peter Zaspel This is my paper · ORCID

classification physics.comp-ph

keywords Bayesianinferenceiceboreholethermometrysurfacetemperaturereconstructionkernel-basedmodelMarkovchainMonteCarloensemblesampleruncertaintyquantificationapproximation

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 introduces a kernel-based representation of surface temperature history for Bayesian inversion of ice borehole temperature measurements. The representation uses a fixed set of Gaussian kernels, which turns the inverse problem into a fixed-dimensional parameter estimation problem that can be sampled with a parallel ensemble Markov chain Monte Carlo sampler. The authors find that reconstruction quality is insensitive to the number of kernels once the basis is dense, and that reducing measurement uncertainty improves reconstructions more than adding measurement points. Because the kernel model cannot represent short-term climate variability, they augment the likelihood with an approximation-uncertainty term derived from surrogate climate simulations. With this correction, the method recovers past surface temperature trends with reliable posterior credibility intervals at much lower computational cost than the standard reversible-jump piecewise-linear approach.

What carries the argument

The key object is the kernel-based surface temperature model: surface temperature is expressed as the pre-observational mean plus a linear combination of squared-exponential (Gaussian) kernels with fixed centers and length-scale. This gives a fixed-dimensional parameter vector (kernel weights plus mean), which makes the posterior amenable to the affine-invariant ensemble MCMC sampler. The second load-bearing mechanism is the approximation-uncertainty correction: the likelihood is modified by adding a depth-wise variance computed as the root-mean-square error between forward simulations of realistic and kernel-smooth baseline surface histories, so that unresolved climate variability is explic

What would settle it

Run a large ensemble of surrogate experiments in which the unresolved climate variability is generated with strong temporal autocorrelation (e.g., power-law exponent beta = 1) and realistic vertical correlation, then count how often the 95% pointwise credible intervals cover the true baseline signal; if coverage falls well below 95%, the independent-Gaussian approximation-error correction is misspecified.

Watch

Extended reading notes

Core claim

The paper's central claim is that a kernel-based surface temperature model, combined with a parallel affine-invariant ensemble MCMC sampler, provides an efficient and reliable Bayesian framework for reconstructing past surface temperature from shallow ice borehole measurements. The kernel basis is fixed in number and length-scale, so for realistic surrogate histories it cannot explain short-term stochastic variability; the authors treat this as model approximation uncertainty and incorporate it into the likelihood as a depth-dependent Gaussian variance. They demonstrate on synthetic and realistic surrogate experiments that this corrected likelihood yields posterior means that track the under

Load-bearing premise

The load-bearing premise is that the forward heat-diffusion-advection equation is an exact description of the data-generating process and that the surface-temperature-model approximation error is well captured by a depth-wise independent Gaussian with variance equal to the RMSE between realistic and baseline simulations.

Editorial extensions

If this is right

Once the kernel basis is sufficiently dense, reconstruction quality does not improve with more kernels, so a modest fixed number of kernels can be chosen without harming the result.
Reducing the instrumental measurement uncertainty has a substantially larger effect on reconstruction accuracy than increasing the number of borehole measurement points.
Unresolved short-term climate variability, not the inversion method itself, is the dominant source of posterior uncertainty in realistic scenarios; when it is perfectly known, reconstructions match the idealized case.
The corrected likelihood produces posterior credible intervals that are more reliable than those from the reference reversible-jump MCMC under the tested settings.
The framework handles very small measurement uncertainties (e.g., 1 mK) without additional tuning, whereas the reference implementation becomes computationally expensive in that regime.

Reading between the lines

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

The fixed-dimensional kernel approach could be transplanted to other diffusion-based paleoclimate archives, such as land boreholes or permafrost temperature profiles, where the same trade-off between model flexibility and sampler efficiency applies.
Because the approximation-uncertainty correction is fit per depth as an independent Gaussian, strongly correlated or non-Gaussian model errors would likely produce overconfident intervals; a hierarchical or full model-error covariance would be a natural robustness check.
The method's efficiency opens the possibility of routine ensemble reconstructions across many borehole sites, which could support spatial field reconstructions of past surface temperature rather than single-site curves.
The paper's surrogate-based prescription for quantifying approximation uncertainty suggests a generic protocol: compare a realistic high-variability forward simulation against the smooth model's best fit, then fold the residuals into the likelihood.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Dear colleague,

Key takeaway: this paper deserves a serious referee. It replaces the serial RJ-MCMC inversion of borehole temperature profiles with a squared-exponential kernel representation of surface temperature that runs on emcee, and that is a genuine improvement in efficiency for this inverse problem. The evaluation is careful and mostly convincing. The one claim I'd push back on is 'reliable posterior uncertainties': the approximation-uncertainty term in Sec 5.2 is modeled as independent Gaussian noise at each depth, but diffusion–advection smears surface errors into strongly correlated depth patterns. A diagonal covariance treats these correlated errors as independent information, which risks understating posterior spread. No empirical coverage check is provided, so the reliability claim remains unvalidated.

What's good: the kernel basis is a sensible alternative to reversible jump; keeping the basis fixed lets them use a parallel ensemble sampler and the runtime comparison is striking. The synthetic experiments are extensive: they test kernel count, measurement density, measurement uncertainty, and temporal smearing, and the conclusion that measurement quality matters more than measurement density is well supported. The surrogate experiments with power-law variability and the known-variability upper-bound diagnostic are a nice way to separate approximation error from inversion error. They are also honest about the RJ-MCMC comparison: their implementation follows the literature defaults and they note that extensive tuning could change those results. Code and data are public.

Soft spots: the kernel setup (N=40, γ=20) is selected by RMSE fits to the same signals used for reconstruction; they do test 40 vs 60 in the inverse setting, but γ sensitivity is not examined and σ_α is hand-tuned per configuration to match a target prior variance. Those are tuning choices, not fatal flaws. The bigger issue is the diagonal approximation-error model. The forward operator is a diffusion equation, so errors in the surface forcing propagate as smooth anomalies in depth; neighboring depths are not independent. Using the depth-wise RMSE as a diagonal covariance is a reasonable first-order correction, but it should be tested. A simple calibration check — do the 95% intervals actually cover the true baseline signal in the surrogate ensemble? — would go a long way. Without that, the 'reliable posterior uncertainties' phrase in the abstract oversells the current evidence.

Verdict: this is a solid methods paper with a real contribution. The flaws are fixable and do not undermine the main efficiency argument. I'd send it to peer review, with a request for coverage checks and a discussion of correlated model error. If the authors add those, I'd cite it.

Referee Report

3 major / 5 minor

Summary. The paper proposes replacing the adaptive piecewise-linear surface temperature model (sampled with reversible-jump MCMC) by a fixed kernel-basis expansion with squared-exponential kernels, sampled with an affine-invariant ensemble MCMC sampler (emcee) for Bayesian inversion of shallow ice-borehole temperature measurements. Synthetic experiments investigate sensitivity to kernel count, measurement uncertainty, measurement density, borehole depth, and comparison with RJ-MCMC. A case study with realistic surrogate climate histories adds a depth-wise approximation-uncertainty term to the likelihood to account for unresolved short-term climate variability. The paper concludes that kernel-based Bayesian inversion is an efficient framework yielding robust reconstructions and reliable posterior uncertainties for shallow borehole climate reconstructions.

Significance. If the conclusions hold, the method offers a substantial reduction in computational cost for shallow-borehole surface-temperature reconstruction while providing posterior uncertainties that account for model approximation error. The extensive synthetic experiments, explicit autocorrelation-time convergence diagnostics, large surrogate ensembles (1000 realizations), and open code are clear strengths. However, the reliability claim depends on an unvalidated diagonal approximation-error covariance and on kernel hyperparameters chosen by fitting to the same synthetic targets; these issues must be addressed before the headline conclusion is supportable.

major comments (3)

[§5.2, Eq. (3.2)] The approximation-uncertainty term in the likelihood is a diagonal covariance whose entries are the depth-wise RMSE between forward simulations of realistic and baseline signals. The forward operator (Eq. 2.2) is diffusive, so approximation errors at neighboring depths are expected to be strongly positively correlated; treating them as independent generally overstates information and narrows credible intervals. The paper claims 'reliable posterior uncertainties' (Abstract and Conclusions) but gives no calibration check, such as the empirical coverage of the 95% credible intervals on held-out surrogate realizations. Please add such a check; if coverage is below nominal, use a correlated approximation-uncertainty model or otherwise inflate the uncertainty.
[§4.1, §4.2] The kernel configuration (N_KB=40, gamma=20 years) is selected by direct RMSE fits of the kernel expansion to the very same four synthetic signals used in the subsequent inversion benchmarks. This can bias the synthetic evaluation. The demonstrated insensitivity to 40 vs. 60 kernels does not cover gamma or the prior scale sigma_alpha (Section 3.2), which is hand-tuned per configuration. To support the introduction's goal of 'minimal dependency on the particular choice of parameters', please show posterior stability with respect to gamma and sigma_alpha, or validate the configuration on independent signals.
[§4.5, Fig. 7] The conclusion that the kernel-based approach provides 'more reliable posterior uncertainty estimates' than RJ-MCMC rests on a single comparison with an RJ-MCMC implementation whose hyperparameters are taken from [20,28] and not tuned for these signals; the paper itself states that 'extensive, signal specific hyperparameter tuning' could improve the RJ-MCMC results. This asymmetry weakens the comparative reliability claim. Please provide an RJ-MCMC hyperparameter sensitivity analysis, or rephrase the conclusion to refer only to the tested RJ-MCMC setup rather than the RJ-MCMC method as a whole.

minor comments (5)

[Eq. (2.7)] The notation ||t-t_i||^2 is confusing because t is a scalar time; it should read (t-t_i)^2.
[Appendix B.2] The expressions 'u_1 =∼ N(0,12)' and similar appear to intend N(0,1). Please correct the notation.
[§5.2] The text says the approximation uncertainty is added 'in addition to the measurement uncertainty', but later states these reconstructions are 'performed with only approximation uncertainty, assuming no measurement uncertainty'. This is contradictory; please clarify whether the 1 mK measurement uncertainty was included in Fig. 11(b-e).
[Fig. 11 caption] The caption describes (b)-(e) as 'reconstructions of the realistic signals', whereas the text states the objective is to reconstruct the baseline signals; please align the caption with the text.
[§3.2] The hand-tuned values sigma_alpha = 0.6, 0.49, and 1.2 are reported for different (N,gamma) configurations without explaining how they are calibrated; a formula relating sigma_alpha to the target prior standard deviation sigma_theta would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the synthetic benchmarks and approximation-uncertainty model are self-contained rather than reductions to inputs.

full rationale

I find no circular step. The evaluation is a genuine synthetic benchmark: borehole observations are generated by forward-solving Eq. (2.2) from known surface-temperature signals and then inverted under the likelihood (3.2), so the reconstruction target is not identical to the model input by construction. The kernel hyperparameters in Sec. 4.1 are chosen by direct L2 fits, but this only sets the basis size and length-scale; the posterior kernel weights are inferred by MCMC and are not the same fitted weights. Similarly, the prior calibration of σ_α in Sec. 3.2 ensures a target prior variance, but it does not determine the posterior or the reported credible intervals. In Sec. 5.2, the approximation-error term is an explicit, independently computed likelihood covariance (depth-wise RMSE between forward simulations of realistic and baseline surrogates), and the posterior is not algebraically equal to that RMSE. The use of [35] for β=0.6 and m=11 is a minor self-citation, but it supplies only surrogate/forward-model constants and is not load-bearing for the central inversion claim. The skeptic's concern about vertical correlations in the diagonal approximation-error covariance is a model-misspecification or correctness risk, not circularity.

Assumptions & free parameters 3 free parameters · 7 assumptions · 1 invented entities

The paper's central claims depend on hand-tuned prior scales, a kernel configuration chosen by fitting to the same synthetic signals, an assumed Gaussian/depth-independent approximation error, and the surrogate model of climate variability. These are not derived from external data, so they must be counted as free parameters or domain assumptions.

free parameters (3)

kernel weight prior std σ_α = 0.6 (N=40, γ=20), 0.49 (N=60, γ=20), 1.2 (for σ_θ=2 K)
Hand-set in Section 3.2 to make kernel weights produce a desired pointwise surface-temperature std; the value controls prior width and hence posterior spread.
kernel count N_KB and length scale γ = N_KB=40, γ=20 yr
Selected in Section 4.1 by RMSE of direct kernel fits to the four synthetic signals (Fig. 2). Not selected from borehole data but from the same target histories used later for reconstruction.
approximation-uncertainty covariance = depth-wise RMSE from 1000 surrogate realizations per PA category (Fig. 11a)
Estimated from the same surrogate class and added to the likelihood in Section 5.2; its magnitude controls posterior width under realistic data.

assumptions (7)

domain assumption Heat diffusion–advection Eq. (2.2) with Dirichlet boundary conditions is an adequate model of borehole temperatures.
The entire inversion uses this forward operator as truth; no forward-model uncertainty is propagated (Section 3, before Eq. 3.1).
domain assumption Measurement errors are i.i.d. Gaussian with known σ_m.
Used in the likelihood, Section 3.1, Eq. (3.2).
ad hoc to paper Squared-exponential kernels with γ=20 and 40 kernels span the reconstruction target (baseline signals).
Chosen by RMSE fits in Section 4.1; no inverse-problem sensitivity analysis for γ is provided.
ad hoc to paper Surface-temperature model approximation error is Gaussian, independent across depths, with variance equal to ensemble RMSE.
Assumed in Section 5.2; correlations or non-Gaussian tails in model error are not considered.
domain assumption Surrogate climate variability with power-law exponent β=0.6 and polar amplification factors PA=0..3 is representative of real short-term variability.
Used to construct the realistic surrogate histories in Section 5, following [35].
domain assumption Forward Euler time discretization (Δt=2^-4 yr) and ~4 m spatial resolution are sufficient.
Stated in Section 2.1; numerical error is not quantified or included in uncertainty.
domain assumption MCMC convergence is diagnosed by the autocorrelation-time heuristic (sample size ~50× maximum autocorrelation time).
Used in Section 3.4 following [12]; this is a heuristic, not a formal convergence guarantee.

invented entities (1)

baseline signal (best-possible kernel approximation of a surrogate temperature history)
purpose: Defines the reconstruction target when the kernel model cannot represent short-term variability (Section 5).
It is not observed; it is defined from the same model class being inverted, so its use as ground truth rests on accepting that the kernel basis is the right target resolution.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Bayesian Inference: Kernel-Based Model for Surface Temperature Reconstruction in Ice Borehole Thermometry." pith.science (2026). https://pith.science/paper/4GXD5TLK

@misc{pith2026260720322,
  author       = {Pith},
  title        = {Pith review of: Bayesian Inference: Kernel-Based Model for Surface Temperature Reconstruction in Ice Borehole Thermometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4GXD5TLK}},
  note         = {Machine review of arXiv:2607.20322}
}

read the original abstract

Reconstructing past surface temperature from shallow ice borehole temperature profiles requires solving an ill-posed inverse problem while quantifying uncertainties arising from measurements and prior assumptions. Bayesian formulations enable probabilistic reconstruction of surface temperature histories and uncertainty quantification. Existing reversible jump-Markov chain Monte Carlo approach based on adaptive piecewise-linear surface temperature models can, however, be computationally demanding. Here, we introduce a kernel-based surface temperature model that enables the use of a parallel ensemble Markov chain Monte Carlo sampler for efficient exploration of the solution space and quantification of the posterior. Using synthetic experiments, we investigate the effects of kernel configuration, measurement uncertainty, measurement density, and temporal smearing on reconstruction performance. We find that reconstruction quality is largely insensitive to the number of kernels once the kernel basis is sufficiently dense. Reducing measurement uncertainty substantially improves reconstructions, whereas increasing the number of borehole temperature measurements provides only marginal benefit. Finally, we evaluate the method using realistic surrogate climate histories that combine long-term temperature changes with stochastic climate variability. The kernel-based surface temperature model cannot represent short-term variability and therefore cannot fully explain the realistic measurements, highlighting the need to account for this approximation uncertainty. The likelihood is adapted to include the approximation uncertainty of the surface temperature model, yielding robust reconstructions with reliable posterior uncertainties. Overall, our results demonstrate that kernel-based Bayesian inversion provides an efficient framework for shallow ice borehole based climate reconstructions.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 10 canonical work pages

[1]

and Beck, A

Shen, P.-Y. and Beck, A. E. , title =. Journal of Geophysical Research , year =
[2]

, title =

Kakuta, S. , title =. Palaeogeography, Palaeoclimatology, Palaeoecology , year =
[3]

and Beltrami, H

Mareschal, J.-C. and Beltrami, H. , title =. Climate Dynamics , year =
[4]

1992 , issn =

A comparative study of inverse methods for estimating climatic history from borehole temperature data , journal =. 1992 , issn =. doi:10.1016/0031-0182(92)90192-8 , author =

work page doi:10.1016/0031-0182(92)90192-8 1992
[5]

and Mosegaard, K

Dahl-Jensen, D. and Mosegaard, K. and Gundestrup, N. and Clow, G. D. and Johnsen, S. J. and Hansen, A. W. and Balling, N. , title =. Science , year =
[6]

Woodbury, A. D. and Ferguson, G. , title =. Geophysical Research Letters , year =
[7]

and Langer, M

Groenke, B. and Langer, M. and Nitzbon, J. and Westermann, S. and Gallego, G. and Boike, J. , title =. The Cryosphere , volume =. 2023 , number =

2023
[8]

and Pereyra, Marcelo , journal =

Cai, Xiaohao and McEwen, Jason D. and Pereyra, Marcelo , journal =. 2022 , volume =

2022

Show all 40 references

[9]

and Chiozzi, P

Verdoya, M. and Chiozzi, P. and Pasquale, V. , title =. Climate of the Past , volume =. 2007 , number =

2007
[10]

Scheichl, Robert and Cullen, Mike and Freitag, Melina A and Kindermann, Stefan , year=
[11]

and Laepple, Thomas

Shaju, Kshema. and Laepple, Thomas. and Hirsch, Nora. and Zaspel, Peter. , title =. Geoscientific Instrumentation, Methods and Data Systems , volume =. 2025 , number =

2025
[12]

and Steffen, Konrad and Slater, Andrew G

Muto, Atsuhiro and Scambos, Ted A. and Steffen, Konrad and Slater, Andrew G. and Clow, Gary D. , title =. Geophysical Research Letters , volume =. doi:10.1029/2011GL048086 , year =

doi
[13]

2011 , url =

Muto, Akira , title =. 2011 , url =

2011
[14]

, title =

Beltrami, Hugo and Ferguson, Grant and Harris, Robert N. , title =. Geophysical Research Letters , volume =. doi:https://doi.org/10.1029/2005GL023714 , year =

doi
[15]

Lie, H. C. and Sullivan, T. J. and Teckentrup, A. L. , title =. SIAM/ASA Journal on Uncertainty Quantification , volume =. 2018 , doi =

2018
[16]

and Burstedde, Carsten and Ghattas, Omar , title =

Martin, James and Wilcox, Lucas C. and Burstedde, Carsten and Ghattas, Omar , title =. SIAM Journal on Scientific Computing , volume =. 2012 , doi =

2012
[17]

and Rosenbluth, Marshall N

Metropolis, Nicholas and Rosenbluth, Arianna W. and Rosenbluth, Marshall N. and Teller, Augusta H. and Teller, Edward , title =. The Journal of Chemical Physics , volume =. 1953 , month =

1953
[18]

Hastings, W. K. , title =. Biometrika , volume =. 1970 , month =

1970
[19]

Annual Review of Statistics and Its Application , volume =

Roy, Vivekananda , title =. Annual Review of Statistics and Its Application , volume =. 2020 , doi =

2020
[20]

, year =

Bishop, Christopher M. , year =
[21]

1994 , edition =

Physics of Glaciers , author=. 1994 , edition =

1994
[22]

2015 , address =

Reich, Sebastian and Cotter, Colin , title =. 2015 , address =

2015
[23]

Communications in Applied Mathematics and Computational Science , volume =

Goodman, Jonathan and Weare, Jonathan , title =. Communications in Applied Mathematics and Computational Science , volume =. 2010 , doi =

2010
[24]

, title =

Green, Peter J. , title =. Biometrika , volume =. 1995 , doi =

1995
[25]

, title =

Murphy, Kevin P. , title =. 2012 , isbn =

2012
[26]

and Lang, Dustin and Goodman, Jonathan , title =

Foreman-Mackey, Daniel and Hogg, David W. and Lang, Dustin and Goodman, Jonathan , title =. doi:10.1086/670067 , year =

doi
[27]

, title =

Bardsley, Johnathan M. , title =. SIAM Journal on Scientific Computing , volume =. 2012 , doi =

2012
[28]

2014 , doi =

Duvenaud, David Kristjanson , title =. 2014 , doi =

2014
[29]

and Gallagher, Kerry and Pain, Chris C

Hopcroft, Peter O. and Gallagher, Kerry and Pain, Chris C. , title =. Geophysical Journal International , volume =. 2007 , month =

2007
[30]

and Gallagher, Kerry , title =

Hopcroft, Peter O. and Gallagher, Kerry , title =. Geophysical Research Letters , volume =. doi:10.1029/2023GL104631 , year =

doi
[31]

and Cornuelle, Bruce D

Orsi, Anais J. and Cornuelle, Bruce D. and Severinghaus, Jeffrey P. , title =. Geophysical Research Letters , volume =. doi:10.1029/2012GL051260 , year =

doi
[32]

Journal of Geophysical Research: Atmospheres , author =

Modeling the densification of polar firn including heat diffusion:. Journal of Geophysical Research: Atmospheres , author =. doi:10.1029/2002JD003319 , language =

doi
[33]

, year =

Lliboutry, L. , year =. A
[34]

Towards the Application of Skewed Detailed Balance in Lattice Gauge Theories

Barros, Joao Pinto and Krstic Marinkovic, Marina. Towards the Application of Skewed Detailed Balance in Lattice Gauge Theories. Proceedings of The 39th International Symposium on Lattice Field Theory PoS(LATTICE2022). doi:10.22323/1.430.0028

doi
[35]

Dan Foreman-Mackey & contributors , title =
[36]

Journal of Glaciology , author =

Firn. Journal of Glaciology , author =. 1980 , pages =. doi:10.3189/S0022143000015239 , language =

1980 doi
[37]

Journal of Geophysical Research: Earth Surface , author =

In situ measurements of. Journal of Geophysical Research: Earth Surface , author =. 2010 , keywords =. doi:10.1029/2009JF001306 , language =

2010 doi
[38]

and Paterson, W

Cuffey, Kurt M. and Paterson, W. S. B. , month = jun, year =. The
[39]

Global and Planetary Change , author =

The extent of temporal smearing in surface-temperature histories derived from borehole temperature measurements , volume =. Global and Planetary Change , author =. 1992 , pages =. doi:10.1016/0921-8181(92)90027-8 , number =

1992 doi
[40]

SIAM/ASA Journal on Uncertainty Quantification , volume =

Cvetkovi\'. SIAM/ASA Journal on Uncertainty Quantification , volume =. 2024 , doi =

2024

Pith tools

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

[1] [1]

and Beck, A

Shen, P.-Y. and Beck, A. E. , title =. Journal of Geophysical Research , year =

[2] [2]

, title =

Kakuta, S. , title =. Palaeogeography, Palaeoclimatology, Palaeoecology , year =

[3] [3]

and Beltrami, H

Mareschal, J.-C. and Beltrami, H. , title =. Climate Dynamics , year =

[4] [4]

1992 , issn =

A comparative study of inverse methods for estimating climatic history from borehole temperature data , journal =. 1992 , issn =. doi:10.1016/0031-0182(92)90192-8 , author =

work page doi:10.1016/0031-0182(92)90192-8 1992

[5] [5]

and Mosegaard, K

Dahl-Jensen, D. and Mosegaard, K. and Gundestrup, N. and Clow, G. D. and Johnsen, S. J. and Hansen, A. W. and Balling, N. , title =. Science , year =

[6] [6]

Woodbury, A. D. and Ferguson, G. , title =. Geophysical Research Letters , year =

[7] [7]

and Langer, M

Groenke, B. and Langer, M. and Nitzbon, J. and Westermann, S. and Gallego, G. and Boike, J. , title =. The Cryosphere , volume =. 2023 , number =

2023

[8] [8]

and Pereyra, Marcelo , journal =

Cai, Xiaohao and McEwen, Jason D. and Pereyra, Marcelo , journal =. 2022 , volume =

2022