Pith. sign in

REVIEW 3 major objections 6 minor 48 references

A calibrated tree–neural fusion model hits 80% accuracy on 29 fine vegetation classes while cutting calibration error by 83%, without changing its class calls.

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 ↗

T0 review · grok-4.5

2026-07-31 22:19 UTC pith:BCNHN3SP

load-bearing objection Careful calibrated stacking on a hard 29-class East River table; the real product is honest protocol and calibration, not beating ExtraTrees. the 3 major comments →

arxiv 2607.24160 v1 pith:BCNHN3SP submitted 2026-07-27 cs.LG

Calibrated Tree-Neural Fusion for Fine-Grained Vegetation Community Classification

Dristi Datta , Md Khalid Hasan Sakib , Manoranjan Paul This is my paper
classification cs.LG
keywords vegetation classificationLiDARhyperspectral indicestree–neural fusionprobability calibrationecological remote sensingstackingsmall-sample multiclass
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.

Fine vegetation communities in mountain landscapes often look alike in spectral, terrain, and canopy measurements, and labelled samples are scarce and uneven. This paper builds a hybrid classifier, Calibrated EcoTreeFuseNet-Plus, that stacks out-of-fold probabilities from six strong tree ensembles with a compact modality-aware neural branch, picks the fusion source on validation data, then temperature-scales the probabilities. On a held-out East River test set of 370 points across 29 classes, it reaches accuracy 0.8000 and macro F1 0.7768, while calibration drops expected calibration error from 0.3866 to 0.0651 with no change in predicted labels. Five independent seed/split runs keep macro F1 near 0.77 with small spread. A sympathetic reader cares because ecological maps need both class calls and trustworthy confidence for field checks and management, and the work shows how to get that trade-off on a small multisource table without claiming to beat trees by a large margin.

Core claim

On a quality-controlled 1,833-observation, 8-predictor, 29-class East River table, Calibrated EcoTreeFuseNet-Plus delivers competitive discrimination (accuracy 0.8000, macro F1 0.7768, balanced accuracy and MCC 0.7903) together with substantially better probability reliability after temperature scaling (ECE 0.3866→0.0651), and that pattern is stable across five repeated spatially informed splits (macro F1 0.7717±0.0112). The gain over the best standalone tree is modest and not statistically significant; the main advance is the discrimination–calibration trade-off under a leakage-aware fusion protocol.

What carries the argument

Calibrated EcoTreeFuseNet-Plus: leakage-aware out-of-fold stacking of six tree probability branches plus EcoFuseNet-V2 (terrain/canopy/spectral encoders, gated fusion, embedding, prototypes), validation choice between meta-classifier and soft voting, then post-hoc temperature scaling on the selected raw probabilities.

Load-bearing premise

The class-preserving spatial grid split, with typical test points tens of metres from training points, is treated as enough protection against spatial dependence inflating held-out scores, even though it is not strict geographic blocking.

What would settle it

Repeat the full pipeline on an independent watershed or acquisition year with true spatial blocking (or fully nested OOF) and check whether calibrated ECE stays near 0.05–0.07 while macro F1 remains competitive with ExtraTrees; a large drop in either metric under true geographic holdout would falsify the claimed trade-off.

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

If this is right

  • Calibrated hybrid probabilities can flag low-confidence vegetation calls for targeted field verification without retraining the classifier.
  • For hard labels alone, a strong tree ensemble remains a simpler competitive baseline; fusion is most justified when reliable probabilities matter.
  • Multisource LiDAR terrain/canopy plus NDVI/NDWI tables are complementary: either domain alone collapses macro F1 on this 29-class task.
  • Sparse meta-learning will often keep tree probabilities and discard latent neural features on small ecological tables, clarifying what the fusion actually uses.

Where Pith is reading between the lines

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

  • If residual spatial autocorrelation is driving part of the score, operational maps at new sites may need spatial block CV before deployment budgets are set.
  • Extending MC-dropout or ensemble uncertainty to the full calibrated fusion output (not only EcoFuseNet-V2) would make confidence usable end-to-end for triage.
  • Adding seasonal spectra or denser canopy structure is a higher-leverage next step than deeper neural fusion, given that confusions cluster among ecologically similar classes.

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 / 6 minor

Summary. The manuscript introduces Calibrated EcoTreeFuseNet-Plus, a tree–neural probability-fusion classifier for 29-class vegetation-community mapping from eight LiDAR/hyperspectral tabular predictors at 1,833 East River reference points. The pipeline combines six tree ensembles, a compact modality-aware neural branch (EcoFuseNet-V2), out-of-fold meta-features, validation-selected elastic-net meta-learning or soft voting, and temperature scaling. On the seed-42 held-out test set it reports accuracy 0.8000, macro F1 0.7768, balanced accuracy/MCC 0.7903, and ECE reduction from 0.3866 to 0.0651 with unchanged labels; five-seed macro F1 is 0.7717±0.0112. The evaluation protocol is unusually careful for applied remote-sensing ML: training-only preprocessing, validation-only selection/calibration, untouched test labels, bootstrap CIs, paired bootstrap/McNemar comparisons, class-wise and ablation tables, and frank acknowledgment that the gain over ExtraTrees is small and non-significant. The central result is therefore best read as a strong within-dataset discrimination–calibration result rather than evidence of broad superiority over tree baselines.

Significance. If the numbers hold, the paper is a useful small-sample ecological tabular benchmark and a good example of leakage-aware stacking practice: OOF branch probabilities, explicit separation of training/validation/test roles, calibration treated as a first-class output, uncertainty diagnostics, ablations, and repeated-seed reruns are all valuable. I give explicit credit for the reproducibility-oriented saved-artifact design and for reporting non-significant differences against ExtraTrees rather than overstating the fusion gain. The practical significance is not a large accuracy advance; it is the demonstration that calibrated probabilities can be obtained without sacrificing discrimination, together with a transparent account of which feature families actually drive the meta-classifier. The main limitation on significance is external validity: the design is spatially informed but not spatially blocked, so the reported metrics are interpolation estimates for this watershed/survey, not estimates of geographic transfer.

major comments (3)
  1. [§4.4–§4.6, §6.7, Table 14] The headline test metrics are interpolation estimates under a split that deliberately interleaves train/validation/test points within each of 25 quantile blocks. The only independence diagnostic is median test–train NN distance 36.86 m with a 1 m near-duplicate flag (Eq. 14–15); this tests duplicate leakage, not spatial autocorrelation in smooth terrain/spectral fields or clustered ecological classes. Because all five seeds reuse the same 5×5 within-block scheme, Table 14 demonstrates precision of the estimate under one partitioning philosophy, not robustness to spatial transfer. Please add a spatially blocked sensitivity analysis (e.g., leave-blocks-out or buffered/block CV using the same 25 blocks, or a coarser external block scheme), report the resulting macro F1/ECE alongside Table 14, and rescope claims accordingly if blocked performance drops materially.
  2. [§5.6.1, §6.3, Table 10, Fig. 3] The selected temperature is T*=0.50, exactly the lower edge of the prespecified grid G_T⊆[0.5,5.0] (Eq. 102–103). The authors disclose this boundary condition, but for the paper’s largest quantitative claim (ECE 0.3866→0.0651) the calibration optimum being clipped by the search interval is load-bearing. Please extend the grid below 0.5 on a log-spaced scale, show the validation-NLL curve around the minimum, and report whether test ECE/Brier/NLL are stable to that extension. If the optimum remains at a boundary, state explicitly that the reported calibration is the best within the implemented family rather than an unconstrained optimum, and consider a validation-only isotonic/Platt comparator to show the ECE gain is not an artifact of one-parameter sharpening.
  3. [§5.4.2, Algorithm 1, §8] The stacking procedure is described as leakage-aware rather than leakage-free: the outer-training scaler is estimated before inner folds are formed, and the held-out neural fold also supports checkpoint selection. Given that the fusion improvement over ExtraTrees is non-significant, even small optimism in the 432-dim meta-feature matrix could affect the central comparison. Please quantify the exact leakage paths and provide a primary-seed sensitivity run with fully nested fold-specific scaling and fold-held-out checkpoint selection, or bound the expected bias by re-estimating the outer scaler and neural checkpoints without using held-out-fold information. If the nested variant changes macro F1/ECE by less than a stated tolerance, say so explicitly.
minor comments (6)
  1. [§4.10.2, Eq. (29)] The ECE bins are defined as left-closed/right-open over [0,1), so a confidence exactly equal to 1.0 is not assigned to any bin. This is unlikely to matter often, but after sharpening with T<1 it should either be handled explicitly (e.g., include 1.0 in the final bin) or justified as measure-zero under the implementation.
  2. [§2.4–§3.5, Table 4] NDVI and NDWI contain values outside the conventional [-1,1] normalized-difference range (e.g., NDWI up to 4.7143) yet are retained without an independent quality threshold. Please add a short note on the provenance/plausibility of these out-of-range values and whether a sensitivity run with clipping or robust scaling changes conclusions.
  3. [§3.5, §5.2.1, §6.6, Fig. 5] CHM is 95.14% zero, which severely limits how much can be claimed about canopy-structure learning. The manuscript is mostly cautious, but the gate/CHM diagnostics and zero elastic-net coefficients should be framed even more explicitly as evidence that the canopy branch contributed little identifiable signal in this dataset.
  4. [§6.5, Fig. 4] Class-wise F1 for BareSoil and Frasera rests on four test observations each. Please include per-class test support next to class-wise F1 in the main table or supplement so readers can distinguish stable high scores from tiny-sample artifacts.
  5. [General] Please clarify the provenance of “Legacy EcoTreeFuseNet”: is it a prior published model, an ablated predecessor implemented for this paper, or an internal baseline? A one-sentence definition would make Table 9 and Table 13 easier to interpret.
  6. [References/reproducibility] The reference list relies heavily on arXiv preprints for some tabular/remote-sensing claims; where peer-reviewed versions exist, please cite them. A code/data-availability statement with the executable configuration and saved predictions would substantially strengthen the reproducibility claims in §4.8 and §5.9.

Circularity Check

0 steps flagged

No significant circularity: empirical held-out evaluation with validation-only selection/calibration; claims do not reduce to inputs by construction.

full rationale

This is a standard supervised tabular ML paper. Discrimination metrics (accuracy, macro F1, balanced accuracy, MCC) and reliability metrics (ECE, Brier, NLL) are computed on an untouched test partition after train-only fitting and validation-only meta-selection and temperature scaling. Temperature scaling is post-hoc on fixed raw probabilities and explicitly does not change predicted labels, so the calibration claim is not a relabeling of the fit. Out-of-fold stacking is used for meta-features; the paper discloses residual leakage-awareness (outer scaler before inner folds; neural fold checkpointing) rather than hiding it as a uniqueness theorem. Author self-citations (prior Datta/Paul remote-sensing work) appear only as background on spectral indices and related methods, not as load-bearing uniqueness or uniqueness-forced architecture. There is no self-definitional identity, no fitted parameter renamed as an independent prediction of the same quantity, and no ansatz smuggled in via self-citation. Spatial-split optimism is a generalization/correctness concern, not circularity of the derivation chain. Score 0 with empty steps is the honest finding.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central empirical claim rests on standard supervised-learning and remote-sensing assumptions plus many hand-fixed architectural and selection hyperparameters. No new physical entities; invented items are model modules. Load-bearing domain assumptions are that eight point-sampled rasters suffice to separate 29 communities and that the spatially informed split yields meaningful generalization estimates.

free parameters (6)
  • Temperature T* = 0.50
    Scalar chosen by minimizing validation NLL on grid [0.5,5.0]; selected value is the lower boundary and fully determines calibrated probabilities.
  • Elastic-net meta-classifier C = 0.003
    Validation-selected regularization strength for the winning meta-model that defines raw Plus probabilities.
  • EcoFuseNet-V2 loss weights (λp, λc, λs) and OOF variants = full (0.03,0.01,0.001); OOF (0.01,0.005,0.0005)
    Hand-set auxiliary prototype/coarse/separation weights differ for full vs OOF training; they shape exported neural features.
  • Initial α_CHM and gate/embedding widths = α_CHM=0.05; df=64; de=128
    Architecture constants (α_CHM=0.05, df=64, de=128, label smoothing 0.03, prototype margin 1.0) chosen by design, not derived.
  • Soft-voting Dirichlet/search weights and class-bias bounds = ExtraTrees ~85.88% in best soft vote; hierarchy α*=0
    Validation-optimized branch weights and optional bias vector bounds [-1.25,1.25] are free selection degrees of freedom (bias/hierarchy secondary).
  • Partition fractions and 5×5 quantile grid = realized ~69.45/10.37/20.19%; seed 42 primary
    Nominal 70/10/20 split and spatial block construction are design choices that define all reported metrics.
axioms (5)
  • domain assumption Out-of-fold stacked generalization yields meta-features valid for unseen points when base models are refit on full outer train after OOF generation.
    Invoked throughout §5.4–5.5; paper itself notes the neural OOF path is leakage-aware, not fully nested.
  • domain assumption Temperature scaling on log-probabilities preserves argmax labels while improving confidence-reliability on this multiclass problem.
    §5.6.1 and Table 11; holds here empirically but is an assumption of the calibration method family.
  • domain assumption Eight point-sampled terrain/canopy/spectral variables adequately represent the ecological distinctions among 29 fine classes.
    §2.4–2.5 predictor design; residual confusions among similar substrates/communities show the limit of this axiom.
  • ad hoc to paper Spatially informed stratified blocks with median NN distance ≫1 m sufficiently control spatial leakage for reported test metrics.
    §4.4–4.6 diagnostic rule (flag if median <1 m); authors correctly refuse to call this strict geographic isolation.
  • standard math Standard multiclass metrics (macro F1, BA, MCC, ECE with 10 bins, Brier, NLL) are appropriate decision criteria under class imbalance.
    §4.10 definitions follow common evaluation practice.
invented entities (2)
  • EcoFuseNet-V2 no independent evidence
    purpose: Compact modality-aware neural branch exporting probabilities, embedding, gates, and prototype distances for fusion.
    New architecture name combining gated terrain/canopy/spectral encoders with prototype auxiliaries; independent evidence is only ablation performance inside this paper.
  • Calibrated EcoTreeFuseNet-Plus no independent evidence
    purpose: End-to-end leakage-aware tree–neural probability fusion plus temperature scaling as the principal proposed model.
    Pipeline label for stacking six tree branches with EcoFuseNet-V2 and validation-selected meta/soft-vote source; no external falsifiable handle beyond this dataset’s metrics.

pith-pipeline@v1.2.0-grok45-kimik3 · 40924 in / 4023 out tokens · 92373 ms · 2026-07-31T22:19:37.120771+00:00 · methodology

0 comments
read the original abstract

Accurate vegetation-community classification is essential for ecological monitoring, habitat assessment, and evidence-based environmental management in heterogeneous landscapes. Existing studies often rely on standalone tree ensembles or generic neural networks, although fine-grained ecological classes frequently exhibit overlapping spectral, topographic, and structural characteristics. Many frameworks also provide limited protection against stacking leakage, insufficient probability calibration, weak minority-class evaluation, and little evidence of stability across repeated data splits. To address these limitations, this study proposes Calibrated EcoTreeFuseNet-Plus, a tree-neural probability-fusion framework that combines out-of-fold tree probabilities, EcoFuseNet-V2 outputs, validation-selected meta-learning, and post-hoc temperature scaling. Raster values from six LiDAR-derived terrain and canopy variables and two hyperspectral vegetation indices were extracted at coordinate-based reference locations. Quality control removed 26 samples with missing elevation and one sample with non-finite NDWI, producing 1,833 complete records across 29 vegetation and non-vegetation classes. On the held-out test set, the proposed model achieved an accuracy of 0.8000, a macro F1-score of 0.7768, a balanced accuracy of 0.7903, and an MCC of 0.7903. Calibration reduced the expected calibration error from 0.3866 to 0.0651 without changing class predictions. Five-seed evaluation yielded a macro F1-score of 0.7717 +/- 0.0112, indicating stable performance across repeated splits. The results demonstrate a reliable discrimination-calibration trade-off for small-sample, fine-grained ecological classification.

Figures

Figures reproduced from arXiv: 2607.24160 by Dristi Datta, Manoranjan Paul, Md Khalid Hasan Sakib.

Figure 1
Figure 1. Figure 1: Spatial distribution of the training, validation, and test observations for the primary seed-42 partition in the East River study area. The partition contained 1,273 training, 190 validation, and 370 test observations. The three subsets retained broad coverage of the study domain under the spatially informed, class-preserving partitioning strategy. The visualization documents the partition geometry and doe… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed Calibrated EcoTreeFuseNet-Plus framework. Coordinate-based raster extraction and quality control produce eight predictors from LiDAR-derived terrain and canopy products and hyperspectral-derived vegetation indices. Following spatially informed partitioning and training-set preprocessing, parallel tree-probability and EcoFuseNet-V2 branches generate the inputs to the 432-dimensional… view at source ↗
Figure 3
Figure 3. Figure 3: Reliability diagrams of EcoTreeFuseNet-Plus before and after temperature scaling. Calibration reduced the expected calibration error from 0.3866 to 0.0651 without altering the predicted class labels. First Author et al.: Preprint submitted to Elsevier Page 35 of 48 [PITH_FULL_IMAGE:figures/full_fig_p035_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Row-normalized confusion matrix of Calibrated EcoTreeFuseNet-Plus on the seed-42 test partition (𝑁 = 370). The model correctly classified 296 observations and misclassified 74 observations. First Author et al.: Preprint submitted to Elsevier Page 38 of 48 [PITH_FULL_IMAGE:figures/full_fig_p038_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Normalized absolute coefficient contributions of the meta-feature families retained by the selected elastic-net meta-classifier. Tree-probability features accounted for 93.70% of the total contribution, followed by EcoFuseNet-V2 class probabilities at 6.19% and the original predictors at 0.11%. The embedding, gate, and prototype-distance feature families received zero coefficient contribution. First Author… view at source ↗
Figure 6
Figure 6. Figure 6: Change in macro F1 relative to Calibrated EcoTreeFuseNet-Plus for the principal architectural and modality ablations. All evaluated variants produced lower macro F1, with the largest reductions observed for the LiDAR-only and spectral-only configurations. First Author et al.: Preprint submitted to Elsevier Page 44 of 48 [PITH_FULL_IMAGE:figures/full_fig_p044_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of MC-dropout predictive entropy for correct and incorrect EcoFuseNet-V2 classifications. Incorrect predictions exhibited higher mean entropy than correct predictions (1.7940 versus 1.5115; one-sided Mann–Whitney 𝑝 = 1.18 × 10−10; Cohen’s 𝑑 = 0.716). This diagnostic applies to EcoFuseNet-V2 and not to the calibrated final probability￾fusion model. First Author et al.: Preprint submitted to Els… 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

48 extracted references · 7 canonical work pages

  1. [9]

    doi:10.1109/ICMEW63481.2024.10645460

    Unveiling soil-vegetation interactions: Reflection relationships and an attention-based deep learning approach for carbon estimation, in: 2024 IEEE International Conference on Multimedia and Expo Workshops (ICMEW), IEEE. doi:10.1109/ICMEW63481.2024.10645460. Datta, D., Paul, M., Murshed, M., Teng, S.W., Schmidtke, L.M.,

    arXiv 2024

  2. [10]

    Computers and Electronics in Agriculture 237, 110406

    Hybrid bayesian attention model for estimating soil organic carbon from Landsat 8 satellite data. Computers and Electronics in Agriculture 237, 110406. doi:10.1016/j.compag.2025.110406. First Author et al.:Preprint submitted to ElsevierPage 46 of 48 Short Title of the Article Datta, D., Paul, M., Murshed, M., Teng, S.W., Schmidtke, L.M.,

    arXiv 2025

  3. [14]

    arXiv preprint arXiv:2104.02301

    Hyperspectral and LiDAR data classification based on linear self-attention. arXiv preprint arXiv:2104.02301 . Foody, G.M.,

    Pith/arXiv arXiv

  4. [15]

    The Annals of Statistics 29, 1189–1232

    Greedy function approximation: A gradient boosting machine. The Annals of Statistics 29, 1189–1232. doi:10.1214/aos/ 1013203451. Gal,Y.,Ghahramani,Z.,2016. Dropoutasabayesianapproximation:Representingmodeluncertaintyindeeplearning,in:Proceedingsofthe33rd International Conference on Machine Learning, pp. 1050–1059. Gao, B.C.,

    doi:10.1214/aos/ 2016

  5. [19]

    arXiv preprint arXiv:2106.11959

    Revisiting deep learning models for tabular data. arXiv preprint arXiv:2106.11959 . Gorodkin, J.,

    Pith/arXiv arXiv

  6. [24]

    Vadose Zone Journal 17, 180061

    The east river, colorado, watershed: A mountainous community testbed for improving predictive understanding of multiscale hydrological–biogeochemical dynamics. Vadose Zone Journal 17, 180061. doi:10.2136/vzj2018.03.0061. IOGPGeomaticsCommittee,2022.WGS84/UTMZone13N:EPSG32613.EPSGGeodeticParameterDataset.URL:https://epsg.io/32613. coordinate reference syst...

    work page doi:10.2136/vzj2018.03.0061 2022

  7. [25]

    arXiv preprint arXiv:2112.01800

    A survey: Deep learning for hyperspectral image classification with few labeled samples. arXiv preprint arXiv:2112.01800 . Kampe, T.U., Johnson, B.R., Kuester, M.A., Keller, M.,

    Pith/arXiv arXiv

  8. [27]

    Korkeala,J.,Muhojoki,J.,Taher,J.,Salolahti,K.,Hyyppä,M.,Kukko,A.,Hyyppä,J.,2025

    LightGBM: A highly efficient gradient boosting decision tree, in: Advances in Neural Information Processing Systems. Korkeala,J.,Muhojoki,J.,Taher,J.,Salolahti,K.,Hyyppä,M.,Kukko,A.,Hyyppä,J.,2025. Normalview:sensor-agnostictreespeciesclassification from backpack and aerial lidar data using geometric projections. arXiv preprint arXiv:2512.05610 . van der ...

    Pith/arXiv arXiv 2025

  9. [28]

    Lefsky, M.A., Cohen, W.B., Parker, G.G., Harding, D.J.,

    doi:10.2202/1544-6115.1309. Lefsky, M.A., Cohen, W.B., Parker, G.G., Harding, D.J.,

    arXiv

  10. [30]

    arXiv preprint arXiv:1910.12861

    Deep learning for hyperspectral image classification: An overview. arXiv preprint arXiv:1910.12861 . Lin, H., Nazari, M., Zheng, D.,

    Pith/arXiv arXiv 1910

  11. [31]

    arXiv preprint arXiv:2412.04714

    PCTreeS: 3D point cloud tree species classification using airborne LiDAR images. arXiv preprint arXiv:2412.04714 . Loshchilov, I., Hutter, F.,

    Pith/arXiv arXiv

  12. [32]

    International Conference on Learning Representations

    Decoupled weight decay regularization. International Conference on Learning Representations . Mazzia,V.,Khaliq,A.,Chiaberge,M.,2020. Improvementinlandcoverandcropclassificationbasedontemporalfeatureslearningfromsentinel-2 data using recurrent-convolutional neural network (R-CNN). arXiv preprint arXiv:2004.12880 . McNemar, Q.,

    Pith/arXiv arXiv 2020

  13. [34]

    MethodsinEcologyandEvolution13,1866–1884

    Spanning scales: The airborne spatial and temporal samplingdesignofthenationalecologicalobservatorynetwork. MethodsinEcologyandEvolution13,1866–1884. doi:10.1111/2041-210X. 13942. Niculescu-Mizil,A.,Caruana,R.,2005.Predictinggoodprobabilitieswithsupervisedlearning,in:Proceedingsofthe22ndInternationalConference on Machine Learning, pp. 625–632. doi:10.1145...

    arXiv 2041

  14. [37]

    Ecography 40, 913–929

    Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography 40, 913–929. doi:10.1111/ecog.02881. Rodriguez-Galiano,V.F.,Ghimire,B.,Rogan,J.,Chica-Olmo,M.,Rigol-Sanchez,J.P.,2012. Anassessmentoftheeffectivenessofarandomforest classifierforland-coverclassification. ISPRSJournalofPhotogrammetryandRemoteSens...

    doi:10.1111/ecog.02881 2012

  15. [38]

    Information Fusion 81, 84–90

    Tabular data: Deep learning is not all you need. Information Fusion 81, 84–90. doi:10.1016/j.inffus. 2021.11.011. Snell, J., Swersky, K., Zemel, R.S.,

    doi:10.1016/j.inffus 2021

  16. [39]

    Information Processing & Management 45, 427–437

    A systematic analysis of performance measures for classification tasks. Information Processing & Management 45, 427–437. doi:10.1016/j.ipm.2009.03.002. Stehman, S.V., Foody, G.M.,

    doi:10.1016/j.ipm.2009.03.002 2009

  17. [40]

    Remote Sensing of Environment 231, 111199

    Key issues in rigorous accuracy assessment of land cover products. Remote Sensing of Environment 231, 111199. doi:10.1016/j.rse.2019.05.018. Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., Wojna, Z.,

    work page doi:10.1016/j.rse.2019.05.018 2019

  18. [41]

    2818–2826

    Rethinking the inception architecture for computer vision, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2818–2826. doi:10.1109/CVPR.2016.308. Talukdar, S., Singha, P., Mahato, S., Shahfahad, Pal, S., Liou, Y.A., Rahman, A.,

    doi:10.1109/cvpr.2016.308 2016

  19. [43]

    Methods in Ecology and Evolution 10, 225–232

    blockCV: An R package for generating spatially or environmentally separated folds for k-fold cross-validation of species distribution models. Methods in Ecology and Evolution 10, 225–232. doi:10.1111/2041-210X. 13107. Venter, Z.S., Sydenham, M.A.K.,

    doi:10.1111/2041-210x 2041

  20. [44]

    arXiv preprint arXiv:2104.10922

    Continental-scale land cover mapping at 10 m resolution over europe (ELC10). arXiv preprint arXiv:2104.10922 . Vermeer, M., Hay, J.A., Völgyes, D., Koma, Z., Breidenbach, J., Fantin, D.S.M.,

    Pith/arXiv arXiv

  21. [47]

    arXiv preprint arXiv:2503.21072

    HSLiNets: Evaluating band ordering strategies in hyperspectral and lidar fusion. arXiv preprint arXiv:2503.21072 . Yang, J.X., Zhou, J., Wang, J., Tian, H., Liew, A.W.C., 2024a. LiDAR-guided cross-attention fusion for hyperspectral band selection and image classification. arXiv preprint arXiv:2404.03883 . Yang, J.X., Zhou, J., Wang, J., Tian, H., Liew, A....

    Pith/arXiv arXiv

  22. [77]

    Datta, D., Paul, M., Murshed, M., Teng, S.W., Schmidtke, L.M., 2023b

    doi:10.3390/environments10050077. Datta, D., Paul, M., Murshed, M., Teng, S.W., Schmidtke, L.M., 2023b. Novel dry soil and vegetation indices to predict soil contents from Landsat 8 satellite data, in: 2023 International Conference on Digital Image Computing: Techniques and Applications (DICTA), IEEE. pp. 152–159. doi:10.1109/DICTA60407.2023.00029. Datta,...

    arXiv 2023

  23. [1135]

    Valavi, R., Elith, J., Lahoz-Monfort, J.J., Guillera-Arroita, G.,

    doi:10.3390/rs12071135. Valavi, R., Elith, J., Lahoz-Monfort, J.J., Guillera-Arroita, G.,

    work page doi:10.3390/rs12071135

  24. [1947]

    Psychometrika 12, 153–157

    Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika 12, 153–157. doi:10.1007/BF02295996. Moore,I.D.,Grayson,R.B.,Ladson,A.R.,1991. Digitalterrainmodelling:Areviewofhydrological,geomorphological,andbiologicalapplications. Hydrological Processes 5, 3–30. doi:10.1002/hyp.3360050103. First Author et al.:Pr...

    Pith/arXiv arXiv 1991

  25. [1979]

    The Annals of Statistics 7, 1–26

    Bootstrap methods: Another look at the jackknife. The Annals of Statistics 7, 1–26. doi:10.1214/aos/1176344552. Falco, N., et al.,

    arXiv

  26. [1992]

    Neural Networks 5, 241–259

    Stacked generalization. Neural Networks 5, 241–259. doi:10.1016/S0893-6080(05)80023-1. Yang, J.X., Wang, J., Li, Z., Sui, C., Long, Z., Zhou, J.,

    doi:10.1016/s0893-6080(05)80023-1

  27. [1996]

    Remote Sensing of Environment 58, 257–266

    NDWI—a normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment 58, 257–266. doi:10.1016/S0034-4257(96)00067-3. GDAL/OGR Contributors,

    doi:10.1016/s0034-4257(96)00067-3

  28. [2001]

    Machine Learning 45, 5–32

    Random forests. Machine Learning 45, 5–32. doi:10.1023/A:1010933404324. Brier,G.W.,1950. Verificationofforecastsexpressedintermsofprobability. MonthlyWeatherReview78,1–3. doi:10.1175/1520-0493(1950) 078<0001:VOFEIT>2.0.CO;2. Brodersen,K.H.,Ong,C.S.,Stephan,K.E.,Buhmann,J.M.,2010. Thebalancedaccuracyanditsposteriordistribution,in:Proceedingsofthe20th Inter...

    doi:10.1023/a:1010933404324 1950

  29. [2002]

    BioScience 52, 19–30

    Lidar remote sensing for ecosystem studies. BioScience 52, 19–30. doi:10.1641/ 0006-3568(2002)052[0019:LRSFES]2.0.CO;2. Li, S., Song, W., Fang, L., Chen, Y., Ghamisi, P., Benediktsson, J.A.,

    2002

  30. [2004]

    Computational Biology and Chemistry 28, 367–374

    Comparing two𝑘-category assignments by a𝑘-category correlation coefficient. Computational Biology and Chemistry 28, 367–374. doi:10.1016/j.compbiolchem.2004.09.006. Grinsztajn, L., Oyallon, E., Varoquaux, G.,

    doi:10.1016/j.compbiolchem.2004.09.006 2004

  31. [2005]

    Journal of the Royal Statistical Society: Series B 67, 301–320

    Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society: Series B 67, 301–320. doi:10.1111/j.1467-9868.2005.00503.x. First Author et al.:Preprint submitted to ElsevierPage 48 of 48

    arXiv 2005

  32. [2006]

    Machine Learning 63, 3–42

    Extremely randomized trees. Machine Learning 63, 3–42. doi:10.1007/s10994-006-6226-1. Gillies, S., Contributors, R.,

    doi:10.1007/s10994-006-6226-1

  33. [2007]

    Journal of the American Statistical Association 102, 359–378

    Strictly proper scoring rules, prediction, and estimation. Journal of the American Statistical Association 102, 359–378. doi:10.1198/016214506000001437. Gorishniy, Y., Rubachev, I., Khrulkov, V., Babenko, A.,

    doi:10.1198/016214506000001437

  34. [2009]

    IEEE Transactions on Knowledge and Data Engineering 21, 1263–1284

    Learning from imbalanced data. IEEE Transactions on Knowledge and Data Engineering 21, 1263–1284. doi:10.1109/TKDE.2008.239. Huang, X., Khetan, A., Cvitkovic, M., Karnin, Z.,

    doi:10.1109/tkde.2008.239 2008

  35. [2010]

    Journal of Applied Remote Sensing 4, 043510

    NEON: The first continental-scale ecological observatory with airborne remote sensing of vegetation canopy biochemistry and structure. Journal of Applied Remote Sensing 4, 043510. doi:10.1117/1.3361375. Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., Liu, T.Y.,

    work page doi:10.1117/1.3361375

  36. [2016]

    XGBoost: A scalable tree boosting system, in: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 785–794. doi:10.1145/2939672.2939785. Cohen, J.,

    arXiv

  37. [2017]

    arXiv preprint arXiv:1709.00308

    A comprehensive survey of deep learning in remote sensing: Theories, tools, and challenges for the community. arXiv preprint arXiv:1709.00308 . Baltrušaitis, T., Ahuja, C., Morency, L.P.,

    Pith/arXiv arXiv

  38. [2018]

    URL:https://doi.org/10.15485/1602034, doi:10.15485/1602034

    ESS-DIVE Data Repository. URL:https://doi.org/10.15485/1602034, doi:10.15485/1602034. dataset. Feng, M., Gao, F., Fang, J., Dong, J.,

    work page doi:10.15485/1602034

  39. [2019]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 41, 423–443

    Multimodal machine learning: A survey and taxonomy. IEEE Transactions on Pattern Analysis and Machine Intelligence 41, 423–443. doi:10.1109/TPAMI.2018.2798607. Bannari, A., Morin, D., Bonn, F., Huete, A.R.,

    arXiv 2018

  40. [2020]

    arXiv preprint arXiv:2012.06678

    TabTransformer: Tabular data modeling using contextual embeddings. arXiv preprint arXiv:2012.06678 . Hubbard, S.S., Williams, K.H., Agarwal, D., Banfield, J., Beller, H., Bouskill, N., Brodie, E., Carroll, R., Dafflon, B., Dwivedi, D., Falco, N., Faybishenko, B., Maxwell, R., Nico, P., Steefel, C., Steltzer, H., Tokunaga, T., Tran, A., Wainwright, H., Var...

    Pith/arXiv arXiv 2012

  41. [2021]

    arXiv preprint arXiv:1908.07442

    TabNet: Attentive interpretable tabular learning. arXiv preprint arXiv:1908.07442 . Ball, J.E., Anderson, D.T., Chan, C.S.,

    Pith/arXiv arXiv 1908

  42. [2022]

    Guo,C.,Pleiss,G.,Sun,Y.,Weinberger,K.Q.,2017

    Why do tree-based models still outperform deep learning on typical tabular data? arXiv preprint arXiv:2207.08815 . Guo,C.,Pleiss,G.,Sun,Y.,Weinberger,K.Q.,2017. Oncalibrationofmodernneuralnetworks,in:Proceedingsofthe34thInternationalConference on Machine Learning, pp. 1321–1330. He, H., Garcia, E.A.,

    Pith/arXiv arXiv 2017

  43. [2023]

    arXiv preprint arXiv:2311.06066

    Lidar-based norwegian tree species detection using deep learning. arXiv preprint arXiv:2311.06066 . Wolpert, D.H.,

    Pith/arXiv arXiv

  44. [2024]

    arXiv preprint arXiv:2401.08837

    Image fusion in remote sensing: An overview and meta analysis. arXiv preprint arXiv:2401.08837 . Arik, S.O., Pfister, T.,

    Pith/arXiv arXiv

  45. [2025]

    Ecology 106, e70078

    Spectral biology across scales in changing environments. Ecology 106, e70078. doi:10.1002/ecy.70078. Cavender-Bares,J.,Schneider,F.D.,Santos,M.J.,Armstrong,A.,Carnaval,A.,Dahlin,K.M.,Figueiredo,F.O.G.,Loarie,S.,Purvis,A.,Schimel,D., Townsend,P.A.,Ustin,S.L.,Wang,Z.,Wilson,B.E.,etal.,2022. Integratingremotesensingwithecologyandevolutiontoadvancebiodiversit...

    work page doi:10.1002/ecy.70078 2022

  46. [2026]

    IEEE Transactions on Geoscience and Remote Sensing 64, 1–16

    ReflectGAN: Modeling vegetation effects for soil carbon estimation from satellite imagery. IEEE Transactions on Geoscience and Remote Sensing 64, 1–16. doi:10.1109/TGRS.2026.3676356. Efron, B.,

    arXiv 2026

  47. [3463]

    Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., et al.,

    doi:10.3390/rs17203463. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., et al.,

    work page doi:10.3390/rs17203463

  48. [4546]

    Neuralobliviousdecisionensemblesfordeeplearningontabulardata

    Popov,S.,Morozov,S.,Babenko,A.,2020. Neuralobliviousdecisionensemblesfordeeplearningontabulardata. arXivpreprintarXiv:1909.06312 . Prokhorenkova, L., Gusev, G., Vorobev, A., Dorogush, A.V., Gulin, A.,

    Pith/arXiv arXiv 2020