Pith. sign in

REVIEW 3 major objections 4 minor 56 references

Tabular foundation model TabPFN beats classical soil-spectroscopy regressors on every scale, and PLS latent variables make it even better.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

TabPFN, a tabular foundation model, outperformed PLSR, Cubist, Random Forest, and CNN on 85 soil spectroscopy regression tasks, with PLS-compressed features giving the best results.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A solid, reproducible IID benchmark showing TabPFN + PLS wins on soil-spectroscopy tasks; the abstract overstates generality before OOD is tested. the 3 major comments →

arxiv 2608.00608 v1 pith:7JZOY645 submitted 2026-08-01 cs.LG

From field-scale to large-scale spectral libraries: Tabular foundation models in soil spectroscopy

Viacheslav Barkov , Jonas Schmidinger , Robin Gebbers , Martin Atzmueller This is my paper
classification cs.LG
keywords soil spectroscopytabular foundation modelsTabPFNpartial least squaresdigital soil mappingdimensionality reductionchemometricsvis-NIR and MIR spectroscopy
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper sets out to show that a tabular foundation model, TabPFN, is the best default regressor for soil spectroscopy at every operational scale, and that its performance can be pushed further by feeding it partial least squares (PLS) latent variables instead of raw spectra. It runs a uniform benchmark across 85 regression tasks—39 field-scale tasks from LimeSoDa and 46 large-scale tasks from the Open Soil Spectral Library—comparing TabPFN to a CNN, Cubist, Random Forest, and PLSR, each on full spectra and on PCA and PLS features. The consistent ordering is the same at both scales: TabPFN first, TabPFN on PLS features even better, and Cubist the strongest classical baseline. If true, practitioners get a clear, evidence-based rule: reach for a tabular foundation model, pair it with supervised PLS features when accuracy matters, and skip explicit dimensionality reduction when a simple workflow is preferred.

Core claim

The central claim is that in-context learning tabular foundation models, specifically TabPFN v2.5, deliver the best predictive accuracy for soil property estimation from vis-NIR and MIR spectra across scales, including global spectral libraries with tens of thousands of samples. Directly applying TabPFN to full SNV-preprocessed spectra already outperformed all classical baselines—PLSR, Cubist, Random Forest, and CNN—demonstrating that explicit dimensionality reduction is not required for strong performance. Adding PLS latent variables as inputs improved every model family, and PLS + TabPFN achieved the best predictions overall. PLS was consistently more effective than PCA as a feature-extrac

What carries the argument

TabPFN (Tabular Prior-data Fitted Network), an in-context learning transformer pre-trained on synthetic tabular tasks that conditions on a labeled context set at inference time and needs no per-dataset training or hyperparameter search; and PLS latent variables, target-aware components computed by maximizing covariance between spectra and the soil property, used as supervised dimensionality-reduction input features for any regressor. The paper's mechanism is the pairing of these two: PLS compresses the high-dimensional, collinear spectra into a small set of target-aligned features, and TabPFN learns the residual mapping from those features to the property. A second mechanism is the runtime s

Load-bearing premise

The load-bearing premise is that randomly splitting samples into training and test folds estimates real-world accuracy fairly; because soil samples close together are similar, random splits can leak spatial structure into training and may inflate accuracy or change which model ranks best.

What would settle it

A spatially blocked cross-validation that holds out entire fields or regions on the same 85 tasks could settle it: if under such splits PLSR or Cubist matches or beats TabPFN in RMSE ranking, the claimed consistent TabPFN advantage would not generalize to realistic deployment.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • TabPFN becomes a strong default for soil spectroscopy across operational scales: on full spectra it beats classical baselines without dataset-specific tuning.
  • When accuracy is the priority, PLS latent variables plus TabPFN is the best configuration; PLS improves Cubist, Random Forest, and TabPFN relative to full spectra and PCA.
  • PLS is the more effective dimensionality-reduction strategy for spectroscopy than PCA, and PLSR remains the strongest linear model, confirming the lasting value of supervised compression.
  • Computational burden shifts to prediction: TabPFN inference grows quadratically with test-set size and can take roughly 24 minutes on the largest libraries; PLS cuts that by about 61%, while training time stays minimal.
  • Whether the advantage holds under out-of-distribution conditions such as new sites, instruments, or laboratories remains open; the paper identifies a systematic out-of-distribution evaluation as the most immediate next step.

Where Pith is reading between the lines

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

  • Beyond the paper: because folds were random splits and soil samples are spatially autocorrelated, the IID benchmark may overstate absolute accuracy; a spatially blocked cross-validation could produce different RMSEs and possibly compress the gap between TabPFN and the best classical models.
  • Beyond the paper: if PLS latent variables capture the target-relevant linear structure, then even larger gains might come from target-aware feature selection or from fine-tuning a spectroscopy-specific foundation model on real spectra, an avenue the paper flags as likely.
  • Beyond the paper: the quadratic inference cost could be mitigated by distillation into a lightweight student model, which would make TabPFN-level accuracy practical for real-time field sensing.
  • Beyond the paper: TabPFN's built-in predictive distributions could turn this accuracy advantage into calibrated uncertainty estimates for spectroscopy, an extension the paper mentions but does not test.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This paper reports a large benchmark comparing TabPFN v2.5, CNN, Cubist, Random Forest, PLSR, and linear regression, each with full spectra, PCA, or PLS latent variables, on 85 soil-spectroscopy regression tasks (39 field-scale LimeSoDa, 46 global OSSL). Using nested 5-fold CV with random splits, shared folds, and Bayesian hyperparameter optimization, it finds that TabPFN variants dominate, PLS+TabPFN is best, PLS is the most effective dimensionality reduction, and Cubist is the strongest classical baseline. It also reports runtime scaling showing that TabPFN shifts computational cost to the prediction stage.

Significance. If the IID ranking holds, this is a valuable empirical contribution: it would provide evidence that tabular foundation models are a strong default for soil spectroscopy across scales and that supervised PLS feature extraction remains useful even for modern models. The strengths are the breadth of the benchmark (85 tasks across two operational scales), the careful protocol (same folds across configurations, nested CV, inner validation for hyperparameters), and the open code and data. I find no circularity in the PLS-as-features design: PLS is fitted only on training folds within the nested CV. However, the external-validity gap described below is substantial, so the overarching claims currently outrun the evidence.

major comments (3)
  1. [§2.2.4 and §4.4] The abstract's central claim ('TabPFN consistently delivered the best overall performance across scales') and the conclusion's recommendation of TabPFN 'as default ... across operational scales' outrun the evidence. §2.2.4 states outer folds were 'created by random splitting'; §4.4 concedes this is an IID evaluation and that OOD behavior 'requires further evaluation.' Because soil spectra are spatially autocorrelated and the LimeSoDa field datasets are small and site-clustered (32–460 samples, §2.1.1), random splits can place near-duplicate spectra from the same field in both training and test folds. This is not neutral: TabPFN and Cubist are memory/instance-based (context set; nearest-neighbor correction) and can exploit spectral similarity, while PLSR and CNN cannot. The ranking may therefore be inflated in favor of these models for deployment to new sites. Please add site/field-groupe
  2. [§3 (Figures 5–6 and A1)] No uncertainty quantification is provided for the ranks that carry the conclusions. Mean/median ranks across 39 and 46 tasks are shown, but there are no confidence intervals, paired tests, or repeated resampling. The text itself notes that on OSSL 'performance across top-ranking configurations was more comparable,' and Figure A1 shows substantial overlap between PLS TabPFN and PCA TabPFN. A Friedman-style or Wilcoxon signed-rank test would substantiate the 'consistently best' claim; otherwise the top-1 vs top-2 separation may be within noise.
  3. [§2.2.4 and §3] The rank computation is under-specified. Eq. (3) defines RMSE with n as the number of samples in a held-out test fold, which suggests a per-fold RMSE; the text then reports 'mean and median ranks' across tasks. It is unclear whether ranks were computed per fold and then averaged, or computed once from task-level RMSE (pooled or averaged across folds). This ambiguity affects the headline ordering. Please state the aggregation explicitly and, if per-fold ranks were used, also report task-level ranks from pooled test predictions.
minor comments (4)
  1. [Appendix A] On OSSL, TabPFN on full spectra has a mean relative gap of 10.63%, while PLS Cubist has 8.74%. The phrase 'surpassed all classical baselines' should be stated as 'ranked higher' or reconciled with the gap metric, since the two criteria do not always coincide.
  2. [§2.2.3] A standard gradient-boosted tree baseline (XGBoost/LightGBM) is absent. Given its prevalence in tabular ML, a sentence acknowledging this exclusion would improve completeness.
  3. [Figure 7] The prediction-time comparison mixes GPU (CNN, TabPFN) and CPU (all others) runtimes. The caption notes this, but the text should explicitly caution that absolute times are not directly comparable across those platforms.
  4. [Table B1] For CNN, batch_size is a range on LimeSoDa but a single value (300) on OSSL. Please confirm this is intentional and state the reason.

Circularity Check

0 steps flagged

No significant circularity: the central rankings come from a nested-CV benchmark on open data, and the paper's self-citations are background, not load-bearing.

full rationale

The paper's load-bearing claims—TabPFN's ranking and the benefit of PLS latent variables—are empirical outcomes of a held-out evaluation, not consequences of definitions or fitted inputs. Section 2.2.4 states that folds were 'created by random splitting' and that 'Hyperparameters of each model were optimized within an inner validation loop'; PLS components, though supervised, are fitted only on outer training folds, so the test-fold predictions do not use test labels. Section 2.2.2 explicitly describes PLS as a decomposition 'involves Y,' but this does not make the later evaluation circular because the decomposition is refit within each training fold. The paper's citations to the authors' own prior work (Barkov et al., 2026; Huang et al., 2026) are contextual and supportive, not the source of the new ranking; the ranking itself is generated from LimeSoDa and OSSL benchmark tasks with shared fold partitions across all models. Section 4.4 concedes that the evaluation is IID and that OOD behavior 'requires further evaluation'—this is an external-validity caveat, not a circularity. No equation or fitted parameter is renamed as a prediction, and no uniqueness claim is imported from the authors' prior work. Therefore, the derivation chain is self-contained with respect to the benchmark results, and the appropriate circularity score is 0.

Axiom & Free-Parameter Ledger

1 free parameters · 4 axioms · 0 invented entities

The central claim is an empirical ranking, so the ledger contains no derivation constants. The main free parameter is the number of PLS/PCA components, tuned per task. The axioms are the experimental assumptions: metric choice, random-split validity, dataset representativeness, and uniform preprocessing. No new entities are postulated.

free parameters (1)
  • n_components for PCA/PLS/PLSR = selected per task by inner CV; search range 2-64 (LimeSoDa), 2-200 (OSSL)
    This controls the dimensionality reduction step that is central to the PLS TabPFN result. It is tuned on validation folds, not fixed by theory.
axioms (4)
  • domain assumption RMSE is the primary performance metric and rank-based aggregation is a valid summary
    Defined in Section 2.2.4; the paper does not weight tasks or account for metric uncertainty.
  • domain assumption Random cross-validation splits yield valid generalization estimates
    Section 2.2.4 uses random splits; Section 4.4 acknowledges only IID evaluation. Spatial autocorrelation in soil samples may make these splits optimistic.
  • domain assumption LimeSoDa and OSSL are representative benchmark collections for the two operational scales
    Sections 2.1.1 and 2.1.2; OSSL tasks require at least 10,000 observations, so very small or specialized libraries are not represented.
  • domain assumption SNV preprocessing is uniformly appropriate
    Section 2.2.1 applies SNV to all spectra; supported by cited literature, but not tested against other preprocessings for each model.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of From field-scale to large-scale spectral libraries: Tabular foundation models in soil spectroscopy." pith.science (2026). https://pith.science/paper/7JZOY645

@misc{pith2026260800608,
  author       = {Pith},
  title        = {Pith review of: From field-scale to large-scale spectral libraries: Tabular foundation models in soil spectroscopy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7JZOY645}},
  note         = {Machine review of arXiv:2608.00608}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Visible and near-infrared (vis-NIR) and mid-infrared (MIR) spectroscopy enable rapid, cost-effective prediction of soil properties. Yet, translating high-dimensional, highly collinear spectra into accurate soil property predictions remains challenging, particularly when employing machine learning. We systematically investigated regression models and dimensionality reduction approaches for spectroscopic modeling across 85 regression tasks from open benchmark datasets in pedometrics spanning field-scale digital soil mapping and a global soil spectral library. We compared an in-context learning tabular foundation model (TabPFN), a convolutional neural network (CNN), rule-based regression (Cubist), Random Forest, and partial least squares regression (PLSR) using full spectra as well as features derived from principal component analysis (PCA) and partial least squares (PLS) latent variables. TabPFN consistently delivered the best overall performance across scales, including large spectral library tasks with tens of thousands of soil samples. Notably, TabPFN applied directly to full spectra already surpassed all classical baselines, showing that explicit dimensionality reduction is not strictly required for strong performance. Further improvements were achieved through PLS, which proved to be an effective dimensionality reduction strategy for all models. Combining PLS latent variables with TabPFN yielded the best predictions overall. Our findings provide evidence-based guidance for spectroscopic calibration model selection across operational scales, demonstrating that the long-standing advantages of PLSR and modern tabular foundation models complement each other in chemometrics.

Figures

Figures reproduced from arXiv: 2608.00608 by Jonas Schmidinger, Martin Atzmueller, Robin Gebbers, Viacheslav Barkov.

Figure 1
Figure 1. Figure 1: Spatial distribution of the 13 field-scale soil spectroscopy datasets from the LimeSoDa benchmark collection employed in this study. The datasets span seven countries across three continents, encompassing diverse geological settings, spectrometer instruments, study area sizes, and measurement protocols. Base map source: Natural Earth. 2. Materials and methods 2.1. Datasets 2.1.1. Field-scale spectroscopy d… view at source ↗
Figure 2
Figure 2. Figure 2: Mean spectra of the individual spectroscopy datasets from LimeSoDa: (a) eight vis-NIR datasets, and (b) five MIR datasets. Each line is the per-band mean spectrum across all soil samples of one dataset. Colors differ to visually distinguish the individual datasets. Vis-NIR spectra are shown as reflectance against wavelength, and MIR spectra as absorbance against wavenumber. fluvioglacial deposits, Cretaceo… view at source ↗
Figure 3
Figure 3. Figure 3: Spatial distribution of soil samples in the Open Soil Spectral Library (OSSL) employed in this study. The OSSL aggregates harmonized spectral data from multiple national and international soil spectral collections, providing near-global spatial coverage. Base map source: Natural Earth. of spectral data additionally introduces substantial computational overhead at this scale (Viscarra Rossel et al., 2016; G… view at source ↗
Figure 4
Figure 4. Figure 4: Mean spectra of the individual spectroscopy regression tasks from OSSL: (a) 16 vis-NIR regression tasks, and (b) 30 MIR regression tasks. Each line is the per-band mean spectrum across all soil samples for each target soil property. Colors differ to visually distinguish the individual regression tasks. Vis-NIR spectra are shown as reflectance against wavelength, and MIR spectra as absorbance against wavenu… view at source ↗
Figure 5
Figure 5. Figure 5: Ranking of 14 model and dimensionality reduction configurations evaluated across 39 field-scale regression tasks from the LimeSoDa dataset collection, showing: (a) mean ranks, and (b) distribution of ranks across tasks, with the large dot marking the median rank. Ranking is based on RMSE. Lower ranks indicate better predictive performance. Results are reported separately for the 39 field-scale LimeSoDa tas… view at source ↗
Figure 6
Figure 6. Figure 6: Ranking of 14 model and dimensionality reduction configurations evaluated across 46 regression tasks from the Open Soil Spectral Library (OSSL), showing: (a) mean ranks, and (b) distribution of ranks across tasks, with the large dot marking the median rank. Ranking is based on RMSE. Lower ranks indicate better predictive performance. Cubist was the strongest classical baseline. PLS Cubist ranked directly b… view at source ↗
Figure 7
Figure 7. Figure 7: Runtime scaling of the benchmarked models. Left: full hyperparameter tuning plus final fit, as a function of training-set size. Right: prediction time as a function of test-set size. Axes are log–log. Each point corresponds to one regression task, reported as the median wall-clock time across its cross-validation folds. CNN and TabPFN were run on a GPU, while the remaining models were run on a CPU. The wal… view at source ↗

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

56 extracted references · 47 canonical work pages · 1 internal anchor

  1. [7]

    Surveys in Geophysics 40, 361–399

    Imaging Spectroscopy for Soil Mapping and Monitoring. Surveys in Geophysics 40, 361–399. doi:10.1007/s10712-019-09524-0. Correa,J.,Tavakoli,H.,Vogel,S.,Gebbers,R.,2025. Overfittingduetodataleakageinsoilsensorcalibration:Examplesfromlab-basedandin-situ soil nir spectroscopy. Computers and Electronics in Agriculture 239, 110920. doi:10.1016/j.compag.2025.11...

    work page arXiv 2025

  2. [8]

    Padarian, J., Minasny, B., McBratney, A.B.,

    doi:10.3390/soilsystems8040128. Padarian, J., Minasny, B., McBratney, A.B.,

    work page doi:10.3390/soilsystems8040128

  3. [9]

    Soil Science Society of America Journal 86, 879–903

    Calibration set optimization and library transfer for soil carbon estimation using soil spectroscopy—A review. Soil Science Society of America Journal 86, 879–903. doi:10.1002/saj2.20435. dosSantos,E.P.,Moreira,M.C.,Fernandes-Filho,E.I.,Demattê,J.A.M.,dosSantos,U.J.,daSilva,D.D.,Cruz,R.R.P.,Moura-Bueno,J.M.,Santos, I.C.,deSáBarretoSampaio,E.V.,2023. Impro...

    work page arXiv 2023

  4. [12]

    CRC Press, Boca Raton

    Diffuse Reflectance Spectroscopy Environmental Problem Solving. CRC Press, Boca Raton. doi:10.1201/9781351071413. Gamagedara,Y.,Feng,G.,Waiser,T.,Tagert,M.L.,Martins,V.S.,Wijewardane,N.K.,2025. Applicationofcalibrationtransfertechniquesbetween different mid-infrared spectrometers/modules to improve accuracy in estimating soil properties. Soil Science Soci...

    work page doi:10.1201/9781351071413 2025

  5. [13]

    arXiv preprint arXiv:2605.13986

    Tabpfn-3: Technical report. arXiv preprint arXiv:2605.13986 . Grinsztajn, L., Flöge, K., Key, O., Hayler, A., Manium, M., Garg, A., Robertson, J., Hoo, S.B., Birkel, F., Jund, P., Jäger, B., Yu, R.T.Y., Schölkopf, B., Hollmann, N., Hutter, F.,

    Pith/arXiv arXiv

  6. [14]

    URL:https: //openreview.net/forum?id=6lv642eq4y

    TabPFN-2.5: a preview, in: EurIPS 2025 Workshop: AI for Tabular Data. URL:https: //openreview.net/forum?id=6lv642eq4y. Gyawali, A.J., Wiseman, M., Ackerson, J.P., Coffman, S., Meissner, K., Morgan, C.L.S.,

    work page 2025

  7. [15]

    Geoderma 453, 117152

    Measuringin situsoil carbon stocks: A study using a novel handheld VisNIR probe. Geoderma 453, 117152. doi:10.1016/j.geoderma.2024.117152. Hateffard, F., Steinbuch, L., Heuvelink, G.B.,

    work page arXiv 2024

  8. [18]

    Nature 637, 319–326

    Accurate predictions on small data with a tabular foundation model. Nature 637, 319–326. Holzmüller,D.,Grinsztajn,L.,Steinwart,I.,2024. Betterbydefault:Strongpre-tunedMLPsandboostedtreesontabulardata,in:TheThirty-eighth Annual Conference on Neural Information Processing Systems. URL:https://openreview.net/forum?id=3BNPUDvqMt. Huang,Y.C.,Padarian,J.,Minasn...

    doi:10.5194/egusphere- 2024

  9. [19]

    Geoderma 471, 117880

    Zero-shot inference with tabular prior-data fitted network (tabpfn) for soil mir spectral analysis. Geoderma 471, 117880. doi:10.1016/j.geoderma.2026.117880. Hutengs, C., Eisenhauer, N., Schädler, M., Cesarz, S., Lochner, A., Seidel, M., Vohland, M.,

    work page arXiv 2026

  10. [20]

    Geoderma 450, 117037

    Enhanced VNIR and MIR proximal sensing of soil organic matter and PLFA-derived soil microbial properties through machine learning ensembles and external parameter orthogonalization. Geoderma 450, 117037. doi:10.1016/j.geoderma.2024.117037. Janik, L.J., Merry, R.H., Skjemstad, J.O.,

    work page arXiv 2024

  11. [21]

    doi:10.3390/s21206745. Viscarra Rossel, R., Behrens, T., Ben-Dor, E., Brown, D., Demattê, J., Shepherd, K., Shi, Z., Stenberg, B., Stevens, A., Adamchuk, V., Aïchi, H., Barthès, B., Bartholomeus, H., Bayer, A., Bernoux, M., Böttcher, K., Brodský, L., Du, C., Chappell, A., Fouad, Y., Genot, V., Gomez, C., Grunwald,S.,Gubler,A.,Guerrero,C.,Hedley,C.,Knadel,...

    work page doi:10.3390/s21206745 2016

  12. [24]

    Rodríguez-Albarracín, H.S., Demattê, J.A.M., Rosin, N.A., Amorim, M.T.A., Contreras, A.E.D., Andreote, F.D., Rosas, J.T.F.,

    doi:10.3390/s24030849. Rodríguez-Albarracín, H.S., Demattê, J.A.M., Rosin, N.A., Amorim, M.T.A., Contreras, A.E.D., Andreote, F.D., Rosas, J.T.F.,

    work page doi:10.3390/s24030849

  13. [26]

    Patterns 4, 100791

    Paired evaluation of machine-learning models characterizes effects of confounders and outliers. Patterns 4, 100791. doi:10.1016/j.patter.2023.100791. Ng, W., Minasny, B., Jones, E., McBratney, A.,

    arXiv 2023

  14. [27]

    Geoderma 406, 115501

    To spike or to localize? Strategies to improve the prediction of local soil properties using regional spectral library. Geoderma 406, 115501. doi:10.1016/j.geoderma.2021.115501. Ng, W., Minasny, B., Montazerolghaem, M., Padarian, J., Ferguson, R., Bailey, S., McBratney, A.B.,

    work page arXiv 2021

  15. [28]

    Geoderma352,251–267

    Convolutional neural network for simultaneouspredictionofseveralsoilpropertiesusingvisible/near-infrared,mid-infrared,andtheircombinedspectra. Geoderma352,251–267. doi:10.1016/j.geoderma.2019.06.016. Nießl, C., Herrmann, M., Wiedemann, C., Casalicchio, G., Boulesteix, A.L.,

    work page doi:10.1016/j.geoderma.2019.06.016 2019

  16. [29]

    WileyInterdisciplinaryReviews:DataMiningandKnowledgeDiscovery12,e1441

    Over-optimism in benchmark studies and the multiplicity of designandanalysisoptionswheninterpretingtheirresults. WileyInterdisciplinaryReviews:DataMiningandKnowledgeDiscovery12,e1441. Nocita, M., Stevens, A., van Wesemael, B., Aitkenhead, M., Bachmann, M., Barthès, B., Ben Dor, E., Brown, D.J., Clairotte, M., Csorba, A., Dardenne, P., Demattê, J.A., Genot...

    work page 2015

  17. [31]

    Geoderma Regional 16, e00198

    Using deep learning to predict soil properties from regional spectral data. Geoderma Regional 16, e00198. doi:10.1016/j.geodrs.2018.e00198. Paltseva,A.A.,Deeb,M.,DiIorio,E.,Circelli,L.,Cheng,Z.,Colombo,C.,2022. Predictionofbioaccessibleleadinurbanandsuburbansoilswith Vis-NIR diffuse reflectance spectroscopy. Science of The Total Environment 809, 151107. d...

    work page arXiv 2018

  18. [32]

    Beyond IID: How General Are Tabular Foundation Models, Really?

    Beyond iid: How general are tabular foundation models, really? URL:https://arxiv.org/abs/2606.30410,arXiv:2606.30410. Qu, J., Holzmüller, D., Varoquaux, G., Morvan, M.L.,

    work page internal anchor Pith review Pith/arXiv arXiv

  19. [33]

    URL:https://openreview.net/forum?id=0VvD1PmNzM

    TabICL: A tabular foundation model for in-context learning on large data, in: Forty- second International Conference on Machine Learning. URL:https://openreview.net/forum?id=0VvD1PmNzM. Qu,J.,Holzmüller,D.,Varoquaux,G.,Morvan,M.L.,2026. Tabiclv2:Abetter,faster,scalable,andopentabularfoundationmodel. arXivpreprint arXiv:2602.11139 . Quinlan, J.R., et al.,

    arXiv 2026

  20. [36]

    Science of The Total Environment 947, 174652

    Soil organic carbon sequestration potential explained by mineralogical and microbiological activity using spectral transfer functions. Science of The Total Environment 947, 174652. doi:10.1016/j.scitotenv.2024.174652. Rossel, R.V., Behrens, T.,

    work page arXiv 2024

  21. [39]

    PLOS ONE 20, e0296545

    Open Soil Spectral Library (OSSL): Building reproducible soil calibration models through open development and community engagement. PLOS ONE 20, e0296545. doi:10.1371/journal.pone.0296545. Safanelli, J.L., Sanderman, J., Bloom, D., Todd-Brown, K., Parente, L.L., Hengl, T., Adam, S., Albinet, F., Ben-Dor, E., Boot, C.M., Bridson, J.H., Chabrillat, S., Deis...

    work page doi:10.1371/journal.pone.0296545

  22. [40]

    Geoderma 440, 116724

    An interlaboratory comparison of mid-infrared spectra acquisition: Instruments and procedures matter. Geoderma 440, 116724. doi:10.1016/j.geoderma.2023.116724. Sanderman, J., Partida, C., Safanelli, J.L., Shepherd, K., Ge, Y., Mitu, S.M., Ferguson, R.,

    work page arXiv 2023

  23. [41]

    European Journal of Soil Science 76, e70053

    Application of a Handheld Near Infrared Spectrophotometer to Farm-Scale Soil Carbon Monitoring. European Journal of Soil Science 76, e70053. doi:10.1111/ejss.70053. Schmidinger, J., Barkov, V., Vogel, S., Atzmueller, M., Heuvelink, G.B., 2026a. Kriging prior regression: A case for kriging-based spatial features with tabpfn in soil mapping. Computers and E...

    work page arXiv 2025

  24. [42]

    Precision Agriculture 25, 1529–1555

    Effect of training sample size, sampling design and prediction model on soil mapping with proximal sensing data for precision liming. Precision Agriculture 25, 1529–1555. doi:10.1007/s11119-024-10122-3. Schmidinger,J.,Vogel,S.,Barkov,V.,Pham,A.D.,Gebbers,R.,Tavakoli,H.,Correa,J.,Tavares,T.R.,Filippi,P.,Jones,E.J.,etal.,2025.LimeSoDa: A dataset collection ...

    work page doi:10.1007/s11119-024-10122-3 2025

  25. [47]

    Geoderma 468, 117764

    Self-supervised and multi-fidelity learning for extended predictive soil spectroscopy. Geoderma 468, 117764. doi:10.1016/j.geoderma.2026.117764. Tavakoli, H., Correa, J., Sabetizade, M., Vogel, S.,

    work page arXiv 2026

  26. [48]

    Soil and Tillage Research 229, 105684

    Predicting key soil properties from Vis-NIR spectra by applying dual-wavelength indices transformations and stacking machine learning approaches. Soil and Tillage Research 229, 105684. doi:10.1016/j.still.2023.105684. Thomas, V., Ma, J., Hosseinzadeh, R., Golestan, K., Yu, G., Volkovs, M., Caterini, A.,

    work page arXiv 2023

  27. [49]

    (Eds.), Advances in Neural Information Barkov et al

    Retrieval & fine-tuning for in-context tabular models, in: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C. (Eds.), Advances in Neural Information Barkov et al. Page 27 of 28 Tabular foundation models in soil spectroscopy Processing Systems, Curran Associates, Inc.. pp. 108439–108467. doi:10.52202/079017-3442. Tibsh...

    doi:10.52202/079017-3442

  28. [53]

    Machinelearningfordigitalsoilmapping:Applications,challengesandsuggestedsolutions

    Wadoux,A.M.C.,Minasny,B.,McBratney,A.B.,2020. Machinelearningfordigitalsoilmapping:Applications,challengesandsuggestedsolutions. Earth-Science Reviews 210, 103359. Wang,J.,Liu,T.,Zhang,J.,Yuan,H.,Acquah,G.E.,2022.Spectralvariableselectionforestimationofsoilorganiccarboncontentusingmid-infrared spectroscopy. European Journal of Soil Science 73, e13267. doi...

    work page doi:10.1111/ejss.13267 2020

  29. [56]

    volume 100 ofAdvances in Agronomy, pp

    Chapter 4 microbial distribution in soils: Physics and scaling, Academic Press. volume 100 ofAdvances in Agronomy, pp. 81–121. doi:10.1016/S0065-2113(08)00604-4. Zhao,S.,Ayoubi,S.,Mousavi,S.R.,Mireei,S.A.,Shahpouri,F.,Wu,S.x.,Chen,C.b.,Zhao,Z.y.,Tian,C.y.,2024. Integratingproximalsoilsensing dataandenvironmentalvariablestoenhancethepredictionaccuracyforso...

    work page arXiv 2024

  30. [91]

    EvaluationofOptimizedPreprocessingand Modeling Algorithms for Prediction of Soil Properties Using VIS-NIR Spectroscopy

    Vestergaard,R.J.,Vasava,H.B.,Aspinall,D.,Chen,S.,Gillespie,A.,Adamchuk,V.,Biswas,A.,2021. EvaluationofOptimizedPreprocessingand Modeling Algorithms for Prediction of Soil Properties Using VIS-NIR Spectroscopy. Sensors

    work page 2021

  31. [363]

    Digitalsoilmapping:Abriefhistoryandsomelessons

    Minasny,B.,McBratney,A.,2016. Digitalsoilmapping:Abriefhistoryandsomelessons. Geoderma264,301–311. doi:10.1016/j.geoderma. 2015.07.017. Minasny,B.,McBratney,A.B.,2008. Regressionrulesasatoolforpredictingsoilpropertiesfrominfraredreflectancespectroscopy. Chemometrics and Intelligent Laboratory Systems 94, 72–79. doi:10.1016/j.chemolab.2008.06.003. Mouazen,...

    work page doi:10.1016/j.geoderma 2016

  32. [681]

    Jenny, H.,

    doi:10.1071/EA97144. Jenny, H.,

    work page doi:10.1071/ea97144

  33. [896]

    Fan, J., Lv, J.,

    doi:10.1007/s42452-025-07580-3. Fan, J., Lv, J.,

    work page doi:10.1007/s42452-025-07580-3

  34. [1165]

    Sun, L., Safanelli, J.L., Sanderman, J., Georgiou, K., Brungard, C., Grover, K., Hopkins, B.G., Liu, S., Bremer, P.T.,

    doi:10.2136/sssaj1981.03615995004500060031x. Sun, L., Safanelli, J.L., Sanderman, J., Georgiou, K., Brungard, C., Grover, K., Hopkins, B.G., Liu, S., Bremer, P.T.,

    work page arXiv

  35. [1299]

    McBratney, A., Mendonça Santos, M., Minasny, B.,

    doi:10.3390/rs9121299. McBratney, A., Mendonça Santos, M., Minasny, B.,

    work page doi:10.3390/rs9121299

  36. [1941]

    Agronomy Journal 33, 857–858

    Factors of soil formation, a system of quantitative pedology. Agronomy Journal 33, 857–858. doi:10.2134/agronj1941. 00021962003300090016x. Kühn,S.L.,Iden,S.C.,Kästner,F.,Sut-Lohmann,M.,2025. TrackingsoilmoisturedynamicswithVis-NIRspectroscopyduringlaboratorybare-soil evaporation. Geoderma 459, 117368. doi:10.1016/j.geoderma.2025.117368. LeCun, Y., Bottou,...

    work page arXiv 2025

  37. [1975]

    Journal of Applied Probability 12, 117–142

    Soft modelling by latent variables: The non-linear iterative partial least squares (nipals) approach. Journal of Applied Probability 12, 117–142. doi:10.1017/S0021900200047604. Ye, H.J., Liu, S.Y., Chao, W.L.,

    doi:10.1017/s0021900200047604

  38. [1989]

    Applied Spectroscopy 43, 772–777

    Standard Normal Variate Transformation and De-Trending of Near-Infrared Diffuse Reflectance Spectra. Applied Spectroscopy 43, 772–777. doi:10.1366/0003702894202201. Ben-Dor, E., Irons, J., Epema, G., et al.,

    doi:10.1366/0003702894202201

  39. [1992]

    Learning with continuous classes, in: 5th Australian joint conference on artificial intelligence, World Scientific. pp. 343–348. Ramirez-Lopez,L.,Behrens,T.,Schmidt,K.,Rossel,R.V.,Demattê,J.,Scholten,T.,2013. Distanceandsimilarity-searchmetricsforusewithsoil vis-nir spectra. Geoderma 199, 43–53. doi:10.1016/j.geoderma.2012.08.035. Reyes, J., Ließ, M.,

    work page doi:10.1016/j.geoderma.2012.08.035 2013

  40. [1994]

    Geoderma 62, 1–15

    The development of pedometrics. Geoderma 62, 1–15. doi:10.1016/0016-7061(94)90024-8. Wold, H.,

    work page doi:10.1016/0016-7061(94)90024-8

  41. [1996]

    Journal of the Royal Statistical Society: Series B (Methodological) 58, 267–288

    Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society: Series B (Methodological) 58, 267–288. doi:10.1111/j.2517-6161.1996.tb02080.x. Tong, W.L., Pehlevan, C.,

    work page arXiv 1996

  42. [1999]

    Remote sensing for the earth sciences: Manual of remote sensing 3, 111–188

    Soil reflectance. Remote sensing for the earth sciences: Manual of remote sensing 3, 111–188. Bergstra,J.,Bardenet,R.,Bengio,Y.,Kégl,B.,2011. Algorithmsforhyper-parameteroptimization,in:AdvancesinNeuralInformationProcessing Systems, Curran Associates, Inc. Boulesteix, A.L., Wilson, R., Hapfelmeier, A.,

    work page 2011

  43. [2000]

    Biometrika 87, 603–618

    Principal component analysis based on robust estimators of the covariance or correlation matrix: influence functions and efficiencies. Biometrika 87, 603–618. doi:10.1093/biomet/87.3.603. Cui,Z.,Hu,B.,Chen,S.,Hong,Y.,Wang,N.,Peng,J.,Mouazen,A.M.,Shi,Z.,2026. Improvingaccuracyandtransferabilityofsoiltotalnitrogen estimationmodelbyintegratingvisiblenear-inf...

    work page arXiv 2026

  44. [2003]

    Geoderma 117, 3–52

    On digital soil mapping. Geoderma 117, 3–52. doi:10.1016/S0016-7061(03) 00223-4. Mészáros, J., Kovács, Z., László, P., Vass-Meyndt, S., Koós, S., Pirkó, B., Szűcs-Vásárhelyi, N., Bakacsi, Z., Laborczi, A., Balog, K., et al.,

    work page doi:10.1016/s0016-7061(03

  45. [2005]

    Chemometrics and Intelligent Laboratory Systems 79, 55–64

    Partial robust m-regression. Chemometrics and Intelligent Laboratory Systems 79, 55–64. doi:10.1016/j.chemolab.2005.04.007. Söderström, M., Sohlenius, G., Rodhe, L., Piikki, K.,

    work page doi:10.1016/j.chemolab.2005.04.007 2005

  46. [2008]

    Journal of the Royal Statistical Society Series B: Statistical Methodology 70, 849–911

    Sure independence screening for ultrahigh dimensional feature space. Journal of the Royal Statistical Society Series B: Statistical Methodology 70, 849–911. doi:10.1111/j.1467-9868.2008.00674.x. Feuer,B.,Schirrmeister,R.T.,Cherepanova,V.,Hegde,C.,Hutter,F.,Goldblum,M.,Cohen,N.,White,C.,2024. Tunetables:Contextoptimization for scalable prior-data fitted ne...

    arXiv 2008

  47. [2010]

    Geoderma 158, 46–54

    Using data mining to model and interpret soil diffuse reflectance spectra. Geoderma 158, 46–54. Rosset,S.,Perlich,C.,Zadrozny,B.,2005. Ranking-basedevaluationofregressionmodels,in:FifthIEEEInternationalConferenceonDataMining (ICDM’05), pp. 8 pp.–. doi:10.1109/ICDM.2005.126. Barkov et al. Page 26 of 28 Tabular foundation models in soil spectroscopy Rutledg...

    work page doi:10.1109/icdm.2005.126 2005

  48. [2013]

    TrAC Trends in Analytical Chemistry 50, 22–32

    Independent components analysis with the jade algorithm. TrAC Trends in Analytical Chemistry 50, 22–32. doi:10.1016/j.trac.2013.03.013. Safanelli, J.L., Hengl, T., Parente, L.L., Minarik, R., Bloom, D.E., Todd-Brown, K., Gholizadeh, A., Mendes, W.D.S., Sanderman, J.,

    doi:10.1016/j.trac.2013.03.013 2013

  49. [2014]

    Applied Spectroscopy Reviews 49, 139–186

    The performance of visible, near-, and mid-infrared reflectance spectroscopy for prediction of soil physical, chemical, and biological properties. Applied Spectroscopy Reviews 49, 139–186. doi:10.1080/05704928.2013.811081. Stoner, E.R., Baumgardner, M.,

    work page arXiv 2013

  50. [2016]

    Precision Agriculture 17, 588–607

    Adaptation of regional digital soil mapping for precision agriculture. Precision Agriculture 17, 588–607. doi:10.1007/s11119-016-9439-8. Soriano-Disla, J.M., Janik, L.J., Rossel, R.A.V., Macdonald, L.M., McLaughlin, M.J.,

    work page doi:10.1007/s11119-016-9439-8

  51. [2019]

    Optuna: A next-generation hyperparameter optimization framework, in: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Association for Computing Machinery. pp. 2623–2631. AngelettiDaFonseca,A.,Pasquini,C.,CristinaCosta,D.,MercêsBarrosSoares,E.,2022. Effectofthesamplemeasurementrepresentativenesson soilcarbon...

    work page arXiv 2022

  52. [2022]

    Geoderma Regional 29, e00530

    Combining differentpre-processingandmultivariatemethodsforpredictionofsoilorganicmatterbynearinfraredspectroscopy(NIRS)inSouthernBrazil. Geoderma Regional 29, e00530. doi:10.1016/j.geodrs.2022.e00530. Chabrillat, S., Ben-Dor, E., Cierniewski, J., Gomez, C., Schmid, T., Van Wesemael, B.,

    work page doi:10.1016/j.geodrs.2022.e00530 2022

  53. [2023]

    Barkov et al

    Practical guidance for deciding whether to account for soil variability when managing for land health, agricultural production, and climate resilience. Barkov et al. Page 24 of 28 Tabular foundation models in soil spectroscopy Journal of Soil and Water Conservation 78, 125A–133A. doi:10.2489/jswc.2023.0706A. Hollmann, N., Müller, S., Eggensperger, K., Hutter, F.,

    work page arXiv 2023

  54. [2024]

    Geoderma 441, 116740

    Evaluating the extrapolation potential of random forest digital soil mapping. Geoderma 441, 116740. doi:10.1016/j.geoderma.2023.116740. Herrick, J.E., Maynard, J.M., Bestelmeyer, B.T., Carey, C.J., Salley, S.W., Shepherd, K., Stewart, Z.P., Wills, S.A., Ziadat, F.M.,

    work page arXiv 2023

  55. [2025]

    Pedosphere 35, 325–337

    Prediction of lead in agricultural soils: An integrated approach using machine learning, terrain attributes and reflectance spectra. Pedosphere 35, 325–337. doi:10.1016/j.pedsph. 2024.01.002. Akiba, T., Sano, S., Yanase, T., Ohta, T., Koyama, M.,

    work page doi:10.1016/j.pedsph 2024

  56. [2026]

    doi:10.1111/ejss.70299

    Modern neural networks for small tabular datasets: The new default for field-scale Digital Soil Mapping? European Journal of Soil Science 77, e70299. doi:10.1111/ejss.70299. Barnes, R.J., Dhanoa, M.S., Lister, S.J.,

    work page doi:10.1111/ejss.70299

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.