Pith. sign in

REVIEW 3 major objections 6 minor 88 references

Above-ground Biomass Estimation with Geospatial Foundation Models

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that for global above-ground biomass regression, geospatial foundation models outperform a fully supervised state-of-the-art baseline only when delivered as pre-computed embedding products, not as frozen encoders; a…

desk verdict A genuinely new delivery-mode result for GFM benchmarking, but the AEF-specific advantage is entangled with GEDI-aligned pre-training; the missing TESSERA full-regime control is the decisive experiment. read the letter →

arxiv 2608.04792 v1 pith:6XPSTSO4 submitted 2026-08-05 cs.LG

Ghjulia Sialellia , Linus Scheibenreif , Jan Dirk Wegner , Konrad Schindler This is my paper
classification cs.LG
keywords above-groundbiomassgeospatialfoundationmodelsembeddingproductsfrozenencodersremotesensingregressionsatelliteimageryglobalmappinggeneralization
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

The paper tries to establish that the delivery format of a geospatial foundation model—weights the user runs, or embeddings the user consumes—decides whether the model helps at all for a hard quantitative regression task: mapping above-ground biomass from satellite data. It benchmarks eleven weight-distributed models as frozen encoders and two pre-computed embedding products (AlphaEarth Foundations and TESSERA) against a fully supervised convolutional baseline on the global AGBD dataset. The result is one-sided: relative to the baseline trained on the full raw feature stack, every frozen encoder underperforms, while an MLP trained on AEF embeddings reaches 52.22 Mg/ha RMSE versus 53.73 Mg/ha for the supervised baseline, and the same supervised network on AEF embeddings plus a few raw covariates reaches 50.79 Mg/ha, the best result. The authors attribute the gap to the input richness and multi-temporal context distilled into the embedding products, with AEF's reconstruction targets—including GEDI height metrics and L-band radar—a secondary refinement. If the claim holds, foundation-model builders should treat analysis-ready embedding layers as a first-class product, because a user who must run a frozen encoder is currently worse off than one who uses no foundation model at all.

What carries the argument

The load-bearing object is the pre-computed embedding layer: a per-pixel vector field (64 dimensions for AEF, 128 for TESSERA) distilled from a full year of multi-modal satellite observations, delivered as data rather than as model weights. The benchmark machinery is a frozen-encoder protocol in which eleven weight-distributed GFMs are paired with a standard dense-prediction decoder (UPerNet) and trained only on that head, while the embedding products are paired with a linear probe, an MLP, and the fully convolutional fcn_film network (a feature-wise linear-modulation architecture used both as supervised baseline and as high-capacity head). The mechanism proposed for the results is that pre-computation carries temporal depth into a single-date benchmark for free, and that AEF's dense per-pixel autoencoding objective—reconstructing targets that include GEDI relative-height metrics and L-band radar—preserves continuous structural gradients that patch-level masked-image-modeling or contrastive objectives do not. AGBD Lite, a distribution-matched 5% subset of the 16-million-sample dataset, is the screening device that makes benchmarking tractable while preserving model rankings.

What would settle it

Train the same linear-probe, MLP, and fcn_film heads on embeddings from a model with comparable multi-modal, multi-temporal pre-training but no GEDI-family or vegetation-structure reconstruction targets, and compare against the supervised baseline on the same AGBD test split. If the margin shrinks to near zero, the representation-quality reading fails; if the margin persists, the paper's delivery-mode conclusion survives. A complementary check is to fine-tune the best weight-distributed GFM with a dense per-pixel reconstruction objective over the same modalities and see whether it reaches AEF-level accuracy.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is a sharp dichotomy in how geospatial foundation models behave on a continuous regression task. Run as frozen encoders, none of the eleven weight-distributed GFMs matched the supervised baseline: the best, SSL4EO-MoCo, reaches 60.56 Mg/ha RMSE on the full test set versus 53.73 Mg/ha for the fully convolutional fcn_film baseline trained on AGBD features, and this gap is mainly blamed on restricted input support—most models accept only a subset of Sentinel-2 bands and cannot ingest the radar and ancillary variables in AGBD. Consumed as pre-computed embeddings, the picture inverts: a single-hidden-layer MLP on the 64-dimensional AEF embeddings reaches 52.22 Mg/ha, outperforming the supervised baseline, and fcn_film on AEF embeddings augmented with land cover, topography, and geographic coordinates reaches 50.79 Mg/ha, the best overall result, with lower RMSE in every region under within-region, cross-region, and general protocols and with much smaller degradation when a one-year gap separates training and evaluation data. The paper further argues that the bulk of this advantage comes from rich multi-modal, multi-temporal input context and analysis-ready delivery shared by AEF and TESSERA, while AEF's edge over TESSERA (2.7 Mg/ha with matched heads) is best explained by its dense per-pixel reconstruction targets, which include GEDI LiDAR height metrics and L-band radar. A side result is that a model trained on AEF embeddings over only eleven regions achieves near-parity with the globally calibrated ESA CCI biomass product when both are scored against independent AGBref plots.

Load-bearing premise

The results assume that AEF's advantage is a fair measure of representation quality rather than an artifact of its pre-training having already been pointed at the same kind of signal being predicted: its training targets include GEDI height measurements, and the labels are GEDI biomass estimates, so the representation has effectively practiced encoding the very structure it is then asked to predict.

Editorial extensions

If this is right

  • A 10 m biomass map can now be produced by a lightweight MLP over AEF embeddings at a fraction of the compute of running a frozen encoder, and it will beat a supervised model trained on the full raw feature stack.
  • Weight-distributed GFMs need to become input-flexible—accepting L-band SAR, ancillary covariates, variable patch sizes, and documented preprocessing—before they can serve quantitative regression tasks.
  • Embedding-based models degrade far less when training and evaluation years differ, so they can be updated near-real-time without waiting for new labels.
  • Zero-shot geographic transfer works for embedding-based models in most carbon-dense regions (17–25% RMSE reduction in Africa and South America), but fails in ecologically distinct regions such as South Asia unless raw features are retained alongside the embeddings.
  • A model trained on a handful of regions with AEF embeddings matches a globally calibrated operational product on independent reference data, suggesting that representation quality can substitute for geographic coverage in training data.

Reading between the lines

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

  • Editorial extension: if the delivery-mode explanation is right, benchmarks that evaluate GFMs only as frozen encoders will systematically undervalue the models for regression; future evaluation suites should include an embedding-product track, and model releases should ship both weights and embeddings.
  • Editorial extension: because AEF's reconstruction targets include GEDI height metrics and the benchmark labels are GEDI L4A biomass, one testable prediction follows—a multi-modal embedding product pre-trained with no GEDI-family or vegetation-structure targets would show a smaller margin over the supervised baseline; the current design cannot fully separate alignment from representation quality.
  • Editorial extension: the South Asia zero-shot failure suggests an operational checklist: before deploying an embedding-based model in a new region, compare the distribution of the target region in embedding space to the training coverage, and keep raw features when the target appears out-of-distribution.
  • Editorial extension: the paper's own temporal test only spans 2019–2020; a natural next experiment is to test embeddings on a year after a disturbance event such as fire, harvest, or storm, to see whether the multi-temporal context helps or hurts when land cover has changed.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper benchmarks geospatial foundation models (GFMs) for global above-ground biomass (AGB) regression on the AGBD dataset. It distinguishes weight-distributed GFMs, evaluated as frozen encoders within PANGAEA (11 models), from pre-computed embedding products (AEF and TESSERA), and compares both against a fully supervised fcn_film baseline. The central claims are: (i) frozen GFMs substantially underperform the supervised baseline; (ii) pre-computed embeddings, especially AEF, are highly effective, with an MLP on AEF embeddings beating the supervised baseline and fcn_film on AEF+ achieving the best overall RMSE; (iii) AEF embeddings generalize better across space and time and reach near-parity with the ESA CCI biomass product on the independent AGBref reference data. The paper concludes that the delivery mode of GFM features, not just model quality, determines their utility for quantitative regression.

Significance. If fully substantiated, the paper would make a useful contribution: it provides a broad, global-scale comparison that most existing GFM benchmarks avoid for continuous regression, ships public code and data, reports three-seed means with small standard deviations for the main AEF and baseline comparisons, excludes AEF pre-training locations from evaluation, and includes an independent validation against AGBref. However, the central delivery-mode claim is weakened by a known confound: AEF's pre-training explicitly reconstructs GEDI L2A structural metrics, the signal family from which the GEDI L4A regression labels are derived. The paper is candid about this alignment in Section 6.2, but the abstract and conclusion nevertheless assert the stronger claim that pre-computed embedding products as a class are highly effective. The contribution is therefore best viewed as a careful demonstration that AEF embeddings are informative for GEDI-derived AGB prediction, with the broader conclusion about delivery format still requiring a non-GEDI-aligned embedding control at full scale.

major comments (3)
  1. [Section 6.2, Table C.8, Appendix A.6] The central attribution of AEF's success to delivery mode is not yet supported. AEF's pre-training reconstruction targets include GEDI L2A relative-height metrics and ALOS PALSAR-2 L-band (Table A.6), while the AGBD regression labels are GEDI L4A biomass, a product derived from the same GEDI L2A waveform family (Section 2.1). Excluding AEF training locations from the test sets prevents spatial leakage, but the representation has still been optimized to encode exactly the structural signal that generates the labels. The only non-GEDI-aligned embedding product, TESSERA, was evaluated only in the Lite regime; there AEF's advantage over TESSERA with the fcn_film head is 2.7 Mg/ha (53.70 vs. 56.43, Table C.8) and with the MLP head 5.9 Mg/ha (53.91 vs. 59.79). The paper's statement that 'the bulk of the gap is attributable to the factors AEF and TESSERA share' therefore extrapolates from a Lite-only control. A full-regime TESSERA fcn_film/MLP result, or an equivalent non-GEDI-aligned embedding product, is needed before the headline 'pre-computed embedding products prove highly effective' can be attributed to delivery format rather than to AEF's task-aligned pre-training.
  2. [Section 5.1, Tables 1 and 2] The frozen-encoder GFM results are reported as single runs without standard deviations, while all other full-regime results are three-seed means. The full-regime comparison that supports the 'substantially underperform' claim is close: SSL4EO-MoCo (60.56) is only 2.0 Mg/ha worse than fcn_film on Sentinel-2 only (58.57), and in the Lite regime the ordering is reversed (64.34 vs. 66.51, Table C.8). With a single run, the full-regime reversal could be within run-to-run variability. Please provide multiple seeds for at least the best-performing GFMs, such as SSL4EO-MoCo and Prithvi-2, or otherwise quantify the uncertainty of the frozen-encoder results.
  3. [Sections 3.1 and 6.2, Abstract] The headline that weight-distributed GFMs 'substantially underperform' conflates representation quality with input support. Because none of the benchmarked GFMs accepts L-band SAR or the AGBD ancillary features, the frozen-encoder evaluation is an endpoint comparison under restricted inputs. The paper acknowledges this in Section 6.1, but the abstract and conclusion state the result without the caveat. In the matched-input Lite setting, the best GFM actually outperforms the supervised Sentinel-2-only baseline (64.34 vs. 66.51, Table C.8), and the full-regime reversal is based on single runs. I recommend stating in the abstract and conclusion that the underperformance holds in the realistic input-restricted deployment setting, and that representation quality alone is not the tested quantity.
minor comments (6)
  1. [Section 2.1] AGBD Lite is described as containing about 600,000 samples and as a subsample of about 5% of the approximately 16 million patches; 5% of 16 million is 800,000, so the numbers should be reconciled, for example by clarifying whether the 5% is applied to GEDI footprints before patch construction.
  2. [Section 3.2] The statement that TESSERA training locations are not publicly available and may overlap with AGBD test locations should be revisited in the discussion; at minimum, report whether the temporal overlap (same-year Sentinel-1/2 imagery) could introduce coincident data in the embedding generation.
  3. [Table 2] The caption says the fcn_film baseline rows are shaded, but the shading is not visible in the typeset version; please ensure the final PDF displays the intended shading.
  4. [Figure 7 caption] The caption contains a typo ('Here„'); it should be 'Here,'.
  5. [Section 4.2] Please clarify whether the all-years temporal baseline is trained on the full joint 2019+2020 training set or on the size-matched subsample; the sentence 'for those experiments, we subsample the 2020 training set to match the 2019 training set size' is ambiguous.
  6. [Section 5.5 and Appendix E] The 1500 km buffer radius for the AGBref comparison is selected by visual inspection; this is a reasonable screening device, but the sensitivity of the near-parity result to the buffer radius should be reported or at least acknowledged in the main text.

Circularity Check

1 steps flagged · score 6.0 of 10

AEF headline results are partially confounded by construction: AEF's pretraining reconstructs GEDI L2A height metrics, the parent signal of the GEDI L4A labels used for evaluation, so the benchmark partly measures target-aligned representational content rather than generic GFM transfer.

  1. fitted input called prediction [Section 6.2 (Discussion); Section 3.2; Table A.6; Section 2.1]
    "AEF is trained as a self-supervised autoencoder whose reconstruction targets include GEDI relative height metrics. Importantly, AEF is trained as a self-supervised autoencoder that reconstructs a broad set of target modalities (including GEDI LiDAR and L-band ALOS PALSAR-2 radar) from a smaller set of input modalities. Additional pre-training targets: ALOS PALSAR-2 ScanSAR; Copernicus DEM GLO-30; GEDI L2A; ERA5-Land; GRACE; NLCD Land Cover; Wikipedia; GBIF. Each sample pairs AGB values from NASA's GEDI L4A mission"

    Labels are GEDI L4A (Sec. 2.1), a product fitted to GEDI L2A waveform/height metrics. AEF's pretraining reconstructs those GEDI L2A relative-height metrics as dense targets (Sec. 3.2, Table A.6), so the embedding is optimized, by construction, to encode the structural signal whose close transform is the regression label. Excluding AEF training locations removes spatial leakage, not this representational alignment; the hard inversion from optical/SAR imagery to canopy structure is pre-solved in the embedding. Headline margins (MLP 52.22 vs 53.73; fcn_film AEF+ 50.79) thus partly measure re-readout of a target-aligned representation rather than generic GFM transfer.

full rationale

The paper is unusually transparent: Section 6.2 acknowledges the GEDI alignment, and Table C.8 provides a TESSERA control (no GEDI targets) that beats the supervised baseline in the Lite regime, so the 'embedding product' idea has independent support. However, the full-regime headline numbers are AEF-only (Section 5.1 restricts full-scale experiments to AEF for cost), and the decisive TESSERA full-regime control is absent. Because AEF's pretraining reconstructs GEDI L2A relative height, the parent measurement of the GEDI L4A labels, the AEF-specific superiority is partially forced by pretraining-target overlap rather than by general representation quality. The paper's own limitation statement ('our training labels come from GEDI L4A, itself a model-derived product') reinforces but does not resolve this confound. I do not see a formal equation-level circularity or a load-bearing self-citation chain; the issue is a target-aligned pretraining input presented as an independent benchmark result. Score 6 reflects that one central predictive comparison reduces, in large part, to re-reading a representation fitted to a close proxy of the labels.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central empirical claims rest on four external inputs: GEDI labels as ground truth, AGBref as an independent reference, the completeness of the public AEF training location list, and the faithfulness of the Lite subsample. None of these is established by the paper itself; the first two are acknowledged limitations, and the third and fourth are reasonable but unverified assumptions. The 1500 km buffer is a free parameter tuned to the evaluation geometry.

free parameters (3)
  • AGBD Lite subsample fraction = 5%
    Selected as a compute and accuracy tradeoff; the paper claims the Wasserstein-minimizing subsampling preserves model rankings (Section 2.1, Table C.8).
  • Subset buffer radius for AGBref comparison = 1500 km
    Chosen by visual inspection of the geographical distribution (Appendix E); affects the reported near-parity with ESA CCI.
  • MLP hidden width = 256 units
    Fixed at 256 for both embedding products to equalize head capacity; not fitted to the data, so low concern.
assumptions (4)
  • domain assumption GEDI L4A footprint biomass values are treated as ground-truth labels for training and evaluation.
    AGBD pairs each patch with GEDI L4A AGB; the benchmark therefore measures the ability to predict a model-derived product with documented biases (Section 7).
  • domain assumption AGBref 10 km aggregates provide an independent biomass reference.
    Used in Section 5.5 to compare against ESA CCI; the paper relies on AGBref's harmonized inventory and LiDAR based values.
  • domain assumption Google DeepMind's released AEF pre-training locations are complete and were fully excluded from AGBD validation and test.
    Section 3.2; if the release is incomplete, some AEF training patches could contaminate the test set.
  • domain assumption The 5% AGBD Lite subset preserves rankings across all configurations.
    Claimed in Section 2.1; Table C.8 supports it only for a subset of models and configurations, yet Lite is used to screen all 11 GFMs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Above-ground Biomass Estimation with Geospatial Foundation Models." pith.science (2026). https://pith.science/paper/6XPSTSO4

@misc{pith2026260804792,
  author       = {Pith},
  title        = {Pith review of: Above-ground Biomass Estimation with Geospatial Foundation Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6XPSTSO4}},
  note         = {Machine review of arXiv:2608.04792}
}
read the original abstract

Accurate estimation of Above-Ground Biomass (AGB) from satellite imagery is essential for the large-scale monitoring of carbon stocks, yet it remains a challenging regression task at global scale. Geospatial Foundation Models (GFMs) have recently emerged as a promising machine learning paradigm to derive general-purpose representations from Earth observation data, but their utility for quantitative regression tasks like biomass estimation remains largely unexplored, as most benchmarks emphasize classification and segmentation. Here, we present a comprehensive benchmark of GFMs for global-scale AGB estimation using the AGBD dataset, a machine learning-ready benchmark spanning diverse biomes and geographies. We distinguish two ways in which GFMs reach practitioners: (i) models distributed as weights to be run by the user, which we evaluate as frozen encoders within the PANGAEA benchmarking framework; and (ii) models distributed as ready-to-use, pre-computed embedding products, for which we evaluate AlphaEarth Foundations (AEF) and TESSERA. We compare 11 GFMs available on PANGAEA and both embedding products against a fully supervised state-of-the-art (SOTA) model, assess their geographical and temporal generalization abilities, as well as agreement with the ESA CCI biomass product on independent reference data. Our results show that GFMs run as frozen encoders substantially underperform with respect to the supervised SOTA model, whereas pre-computed embedding products prove highly effective. An MLP trained on AEF embeddings outperforms the supervised SOTA model trained on AGBD features, and the same SOTA model trained on AEF embeddings (optionally augmented with selected raw features) achieves the best overall result, while also generalizing better across space and time.

Figures

Figures reproduced from arXiv: 2608.04792 by the authors.

Figure 1
Figure 1. Spatial distribution of the AGBref plots (Robinson projection, H3 hexagonal [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. GFM-based biomass estimation framework. At each GEDI footprint, the AGBD [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Residuals (AGBpred− AGBref in Mg/ha) stratified by AGB labels bins for fcn_film trained on AGBD features (blue) and AEF embeddings (orange), together with the best-performing GFM, SSL4EO-MoCo (green), all evaluated on the full AGBD test set [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Density scatter plots of predicted vs. GEDI AGB (Mg/ha) for [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: AGBD Lite test RMSE (Mg/ha) (↓) vs. inference throughput (samples/s) for all benchmarked GFMs, with color indicating total GPU-hours required for fine-tuning on the AGBD Lite dataset until convergence (values >20 are marked as outliers) on a single NVIDIA GeForce RTX 4…
Figure 6
Figure 6. Figure 6: Qualitative comparison of predicted AGB maps across four sources on three [PITH_FULL_IMAGE:figures/full_fig_p023_6.png]
Figure 7
Figure 7. Figure 7: Learned representations for the same three tiles, over [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Comparison of predicted AGB against reference values (AGBref) for our model [PITH_FULL_IMAGE:figures/full_fig_p027_8.png]
Figure 9
Figure 9. Figure 9: Per-plot map comparison for a selection of AGBref plots spanning the biomass [PITH_FULL_IMAGE:figures/full_fig_p028_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

88 extracted references · 65 canonical work pages

  1. [1]

    Compressing Earth Embeddings , year =

  2. [2]

    FiLM-Ensemble: Probabilistic Deep Learning via Feature-wise Linear Modulation , url =

    Turkoglu, Mehmet Ozgur and Becker, Alexander and G\". FiLM-Ensemble: Probabilistic Deep Learning via Feature-wise Linear Modulation , url =. Advances in Neural Information Processing Systems , editor =

  3. [3]

    2024 , eprint=

    Bridging Remote Sensors with Multisensor Geospatial Foundation Models , author=. 2024 , eprint=

    2024

  4. [4]

    and Schindler, Konrad , title =

    Sialelli, Ghjulia and Peters, Torben and Wegner, Jan D. and Schindler, Konrad , title =. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences , volume =

  5. [5]

    2024 , eprint=

    PANGAEA: A Global and Inclusive Benchmark for Geospatial Foundation Models , author=. 2024 , eprint=

    2024

  6. [6]

    and Adeli, Ehsan and others , title =

    Bommasani, Rishi and Hudson, Drew A. and Adeli, Ehsan and others , title =. arXiv preprint arXiv:2108.07258 , year =

    arXiv

  7. [7]

    Jakubik, Johannes and Roy, Sujit and Phillips, C. E. and others , title =. arXiv preprint arXiv:2310.18660 , year =

    arXiv

  8. [8]

    and Gupta, Ritwik and Li, Shufan and others , title =

    Reed, Colorado J. and Gupta, Ritwik and Li, Shufan and others , title =

Show all 88 references
  1. [9]

    Bastani, Favyen and Wolters, Piper and Gupta, Ritwik and Ferdinando, Joe and Kembhavi, Aniruddha , title =

  2. [10]

    arXiv preprint arXiv:2403.15356 , year =

    Xiong, Zhitong and Wei, Yi and others , title =. arXiv preprint arXiv:2403.15356 , year =

  3. [11]

    , title =

    Fuller, Anthony and Millard, Koreen and Green, James R. , title =

  4. [12]

    IEEE Trans

    Liu, Fan and Chen, Delong and Guan, Zhangqingyun and others , title =. IEEE Trans. Geosci. Remote Sens. , year =

  5. [13]

    Hong, Danfeng and Zhang, Bing and others , title =

  6. [14]

    and Kazmierski, Michal R

    Brown, Christopher F. and Kazmierski, Michal R. and Pasquarella, Valerie J. and Rucklidge, William J. and Samsikova, Masha and Zhang, Chenhui and Shelhamer, Evan and Lahera, Estefania and Wiles, Olivia and Ilyushchenko, Simon and Gorelick, Noel and Zhang, Lihui Lydia and Alj, ...

    2025 doi arXiv

  7. [15]

    , title =

    Lang, Nico and Jetz, Walter and Schindler, Konrad and Wegner, Jan D. , title =. Nature Ecology & Evolution , volume =

  8. [16]

    Towards Geospatial Foundation Models via Continual Pretraining , booktitle = ICCV, pages =

    Mendieta, Mat. Towards Geospatial Foundation Models via Continual Pretraining , booktitle = ICCV, pages =

  9. [17]

    IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    TerraMind: Large-Scale Generative Multimodality for Earth Observation , author=. IEEE/CVF International Conference on Computer Vision (ICCV) , year=

  10. [18]

    2024 , eprint =

    Szwarcman, Daniela and Roy, Sujit and Fraccaro, Paolo and others , title =. 2024 , eprint =

    2024

  11. [19]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Tessera: Temporal embeddings of surface spectra for earth representation and analysis , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  12. [20]

    Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

    BioMassters: A Benchmark Dataset for Forest Biomass Estimation using Multi-modal Satellite Time-series , author=. Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

  13. [21]

    2026 , eprint=

    MMEarth-Bench: Global Model Adaptation via Multimodal Test-Time Training , author=. 2026 , eprint=

    2026

  14. [22]

    2025 , eprint=

    Towards a Unified Copernicus Foundation Model for Earth Vision , author=. 2025 , eprint=

    2025

  15. [23]

    2024 , eprint=

    Fibaek, Casper and Camilleri, Luke and Luyts, Andreas and Dionelis, Nikolaos and. 2024 , eprint=

    2024

  16. [24]

    arXiv preprint arXiv:2312.10114 , year=

    FoMo-Bench: a multi-modal, multi-scale and multi-task Forest Monitoring Benchmark for remote sensing foundation models , author=. arXiv preprint arXiv:2312.10114 , year=

    arXiv

  17. [25]

    NeurIPS , year =

    Alexandre Lacoste and Nils Lehmann and Pau Rodriguez and Evan David Sherwin and Hannah Kerner and Bj. NeurIPS , year =

  18. [26]

    Thirty-fifth Conference on Neural Information Processing Systems, Datasets and Benchmarks Track (Round 2) , year =

    SustainBench: Benchmarks for Monitoring the Sustainable Development Goals with Machine Learning , author =. Thirty-fifth Conference on Neural Information Processing Systems, Datasets and Benchmarks Track (Round 2) , year =

  19. [27]

    Santoro, Maurizio and Cartus, Oliver , keywords =. ESA Biomass Climate Change Initiative (Biomass\_cci): Global datasets of forest above-ground biomass for the years 2007, 2010, 2015, 2016, 2017, 2018, 2019, 2020, 2021 and 2022, v6.0 , publisher =. 2025 , copyright =. doi:10.5...

    2007 doi

  20. [28]

    2025 , copyright =

    Araza, Arnan , title =. 2025 , copyright =. doi:10.5281/ZENODO.15495068 , url =

    2025 doi

  21. [29]

    Saatchi and L

    S. Saatchi and L. Xu and Y. Yang , year = 2021, howpublished =

    2021

  22. [30]

    and Armston, J

    Dubayah, R.O. and Armston, J. and Healey, S.P. and Yang, Z. and Patterson, P.L. and Saarela, S. and Stahl, G. and Duncanson, L. and Kellner, J.R. , keywords =. doi:10.3334/ORNLDAAC/2017 , url =

    2017 doi

  23. [31]

    doi:10.5067/D2E2L5GK7ER3 , url =

    Montesano, Paul and Neuenschwander, Amy and Guenther, Eric and Duncanson, Laura , title =. doi:10.5067/D2E2L5GK7ER3 , url =

    doi

  24. [32]

    and Armston, J

    Dubayah, R.O. and Armston, J. and Kellner, J.R. and Duncanson, L. and Healey, S.P. and Patterson, P.L. and Hancock, S. and Tang, H. and Bruening, J.M. and Hofton, M.A. and Blair, J.B. and Luthcke, S.B. , keywords =. doi:10.3334/ORNLDAAC/2056 , url =

    doi

  25. [33]

    and Smith, Benjamin and Yang, Yuekui and Zwally, Jay , year =

    Markus, Thorsten and Neumann, Tom and Martino, Anthony and Abdalati, Waleed and Brunt, Kelly and Csatho, Beata and Farrell, Sinead and Fricker, Helen and Gardner, Alex and Harding, David and Jasinski, Michael and Kwok, Ron and Magruder, Lori and Lubin, Dan and Luthcke, Scott a...

    doi

  26. [34]

    and Nieusma, Jordan M

    Lu, Siqi and Guo, Junlin and Zimmer-Dauphinee, James R. and Nieusma, Jordan M. and Wang, Xiao and VanValkenburgh, Parker and Wernke, Steven A. and Huo, Yuankai , year =. Vision Foundation Models in Remote Sensing: A survey , volume =. IEEE Geoscience and Remote Sensing Magazin...

  27. [35]

    and Ganguly, Auroop R

    Benavides-Martinez, Ivan Felipe and Guthrie, Justin and Arias, Jhon Edwin and Garces-Gomez, Yeison Alberto and Guzman-Alvis, Angela Ines and Portilla-Cabrera, Cristiam Victoriano and Mondal, Somnath and Allyn, Andrew J. and Ganguly, Auroop R. , keywords =. What on Earth is Alp...

    2026 doi

  28. [36]

    How Much of a Model Do We Need? Redundancy and Slimmability in Remote Sensing Foundation Models , publisher =

    Hackel, Leonard and Burgert, Tom and Demir, Beg\". How Much of a Model Do We Need? Redundancy and Slimmability in Remote Sensing Foundation Models , publisher =. 2026 , copyright =. doi:10.48550/ARXIV.2601.22841 , url =

    2026 doi arXiv

  29. [37]

    Hide and Seek: Investigating Redundancy in Earth Observation Imagery , publisher =

    Papazafeiropoulos, Tasos and Bountos, Nikolaos Ioannis and Papadopoulos, Nikolas and Papoutsis, Ioannis , keywords =. Hide and Seek: Investigating Redundancy in Earth Observation Imagery , publisher =. 2026 , copyright =. doi:10.48550/ARXIV.2603.13524 , url =

    2026 doi

  30. [38]

    and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , keywords =

    Hu, Edward J. and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , keywords =. LoRA: Low-Rank Adaptation of Large Language Models , publisher =. 2021 , copyright =. doi:10.48550/ARXIV.2106.09685 , url =

    2021 doi arXiv

  31. [39]

    and Ishida, H

    Tadono, T. and Ishida, H. and Oda, F. and Naito, S. and Minakawa, K. and Iwamoto, H. , TITLE =. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences , VOLUME =. 2014 , PAGES =

    2014

  32. [40]

    Copernicus Global Land Service: Land Cover 100m: collection 3: epoch 2019: Globe , publisher =

    Buchhorn, Marcel and Smets, Bruno and Bertels, Luc and Roo, Bert De and Lesiv, Myroslava and Tsendbazar, Nandin-Erdene and Herold, Martin and Fritz, Steffen , keywords =. Copernicus Global Land Service: Land Cover 100m: collection 3: epoch 2019: Globe , publisher =. 2020 , cop...

    2019 doi

  33. [41]

    AAAI , year=

    FiLM: Visual Reasoning with a General Conditioning Layer , author=. AAAI , year=

  34. [42]

    Proceedings of the European Conference on Computer Vision (ECCV) , month =

    Xiao, Tete and Liu, Yingcheng and Zhou, Bolei and Jiang, Yuning and Sun, Jian , title =. Proceedings of the European Conference on Computer Vision (ECCV) , month =

  35. [43]

    and Kommareddy, Anil and Pickens, Amy and Turubanova, Svetlana and Tang, Hao and Silva, Carlos Edibaldo and Armston, John and Dubayah, Ralph and Blair, J

    Potapov, Peter and Li, Xinyuan and Hernandez-Serna, Andres and Tyukavina, Alexandra and Hansen, Matthew C. and Kommareddy, Anil and Pickens, Amy and Turubanova, Svetlana and Tang, Hao and Silva, Carlos Edibaldo and Armston, John and Dubayah, Ralph and Blair, J. Bryan and Hofto...

  36. [44]

    and Schwartz, Martin and Saatchi, Sassan and Ciais, Philippe and Pokutta, Sebastian and Brandt, Martin and Gieseke, Fabian , keywords =

    Pauls, Jan and Zimmer, Max and Kelly, Una M. and Schwartz, Martin and Saatchi, Sassan and Ciais, Philippe and Pokutta, Sebastian and Brandt, Martin and Gieseke, Fabian , keywords =. Estimating Canopy Height at Scale , publisher =. 2024 , copyright =. doi:10.48550/ARXIV.2406.01...

    2024 doi

  37. [45]

    2026 , eprint=

    THOR: A Versatile Foundation Model for Earth Observation Climate and Society Applications , author=. 2026 , eprint=

    2026

  38. [46]

    OlmoEarth: Stable Latent Image Modeling for Multimodal Earth Observation , publisher =

    Herzog, Henry and Bastani, Favyen and Zhang, Yawen and Tseng, Gabriel and Redmon, Joseph and Sablon, Hadrien and Park, Ryan and Morrison, Jacob and Buraczynski, Alexandra and Farley, Karen and Hansen, Joshua and Howe, Andrew and Johnson, Patrick Alan and Otterlee, Mark and Sch...

    2025 doi

  39. [47]

    Europe-wide maps of biomass density based on satellite remote sensing data for 2017, 2020, 2021 and 2023 , journal =

    Maurizio Santoro and Oliver Cartus and Arnan Araza and Martin Herold and Jukka Miettinen and Ake Rosenqvist and Kazufumi Kobayashi and Takeo Tadono and Frank Martin Seifert , keywords =. Europe-wide maps of biomass density based on satellite remote sensing data for 2017, 2020,...

    2017

  40. [48]

    Remote Sensing , VOLUME =

    Li, Hantao and Kato, Tomomichi and Hayashi, Masato and Wu, Lan , TITLE =. Remote Sensing , VOLUME =. 2022 , NUMBER =

    2022

  41. [49]

    Assessing temporal trends of forest aboveground biomass density in Japan from 2009 to 2018 under disturbance regimes using multisource remote sensing data , volume =

    Li, Hantao and Hiroshima, Takuya and Li, Xiaoxuan and Kato, Tomomichi and Hayashi, Masato , year =. Assessing temporal trends of forest aboveground biomass density in Japan from 2009 to 2018 under disturbance regimes using multisource remote sensing data , volume =. Journal of...

    2009 doi

  42. [50]

    and Numata, Izaya and Elmore, Andrew J

    Jia, Duo and Wang, Cangjiao and Hakkenberg, Christopher R. and Numata, Izaya and Elmore, Andrew J. and Cochrane, Mark A. , year =. Accuracy evaluation and effect factor analysis of GEDI aboveground biomass product for temperate forests in the conterminous United States , volum...

  43. [51]

    and Duncanson, Laura I

    Pascual, Adrián and Guerra-Hernández, Juan and Armston, John and Minor, David M. and Duncanson, Laura I. and May, Paul B. and Kellner, James R. and Dubayah, Ralph , year =. Assessing the performance of NASA’s GEDI L4A footprint aboveground biomass density models using National...

  44. [52]

    Evaluation of GEDI footprint level biomass models in Southern African Savannas using airborne LiDAR and field measurements , volume =

    Li, Xiaoxuan and Wessels, Konrad and Armston, John and Duncanson, Laura and Urbazaev, Mikhail and Naidoo, Laven and Mathieu, Renaud and Main, Russell , year =. Evaluation of GEDI footprint level biomass models in Southern African Savannas using airborne LiDAR and field measure...

  45. [53]

    Savannah woody structure modelling and mapping using multi-frequency (X-, C- and L-band) Synthetic Aperture Radar data , volume =

    Naidoo, Laven and Mathieu, Renaud and Main, Russell and Kleynhans, Waldo and Wessels, Konrad and Asner, Gregory and Leblon, Brigitte , year =. Savannah woody structure modelling and mapping using multi-frequency (X-, C- and L-band) Synthetic Aperture Radar data , volume =. doi...

    doi

  46. [54]

    , year =

    Imhoff, Marc L. , year =. Radar backscatter and biomass saturation: ramifications for global biomass inventory , volume =. IEEE Transactions on Geoscience and Remote Sensing , publisher =. doi:10.1109/tgrs.1995.8746034 , number =

    1995

  47. [55]

    Nature , volume =

    The enduring world forest carbon sink , author =. Nature , volume =. 2024 , doi =

    2024

  48. [56]

    Forests, biodiversity and people , author =

    The State of the World's Forests 2020. Forests, biodiversity and people , author =. 2020 , doi =

    2020

  49. [57]

    Climate Change 2007: Mitigation

    Forestry , author =. Climate Change 2007: Mitigation. Contribution of Working Group III to the Fourth Assessment Report of the IPCC , publisher =

    2007

  50. [58]

    Manual for building tree volume and biomass allometric equations: from field measurement to prediction , author =

  51. [59]

    Global Change Biology , volume =

    Improved allometric models to estimate the aboveground biomass of tropical trees , author =. Global Change Biology , volume =. 2014 , doi =

    2014

  52. [60]

    iForest - Biogeosciences and Forestry , volume =

    GlobAllomeTree: international platform for tree allometric equations to support volume, biomass and carbon assessment , author =. iForest - Biogeosciences and Forestry , volume =. 2013 , doi =

    2013

  53. [61]

    Global Change Biology , volume =

    Allometric equations for integrating remote sensing imagery into forest monitoring programmes , author =. Global Change Biology , volume =. 2017 , doi =

    2017

  54. [62]

    Manual for measurement, monitoring and reporting of carbon stocks in forests and other land uses in Ethiopia , author =

  55. [63]

    Scandinavian Journal of Forest Research , volume =

    Remote sensing and forest inventories in Nordic countries -- roadmap for the future , author =. Scandinavian Journal of Forest Research , volume =. 2018 , doi =

    2018

  56. [64]

    Interface Focus , volume =

    Weighing trees with lasers: advances, challenges and opportunities , author =. Interface Focus , volume =. 2018 , doi =

    2018

  57. [65]

    Remote Sensing of Environment , volume =

    Terrestrial and mobile laser scanning for national forest inventories: From theory to implementation , author =. Remote Sensing of Environment , volume =. 2025 , doi =

    2025

  58. [66]

    Remote Sensing of Environment , volume =

    A nationwide forest attribute map of Sweden predicted using airborne laser scanning data and field data from the National Forest Inventory , author =. Remote Sensing of Environment , volume =. 2017 , doi =

    2017

  59. [67]

    The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences , volume =

    Wide-area mapping of forest with national airborne laser scanning and field inventory datasets , author =. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences , volume =. 2016 , doi =

    2016

  60. [68]

    ISPRS Journal of Photogrammetry and Remote Sensing , volume =

    Status and future of laser scanning, synthetic aperture radar and hyperspectral remote sensing data for forest biomass assessment , author =. ISPRS Journal of Photogrammetry and Remote Sensing , volume =. 2010 , doi =

    2010

  61. [69]

    Remote Sensing of Environment , volume =

    Sentinel-2: ESA's optical high-resolution mission for GMES operational services , author =. Remote Sensing of Environment , volume =. 2012 , doi =

    2012

  62. [70]

    IEEE Transactions on Geoscience and Remote Sensing , volume =

    ALOS PALSAR: A pathfinder mission for global-scale monitoring of the environment , author =. IEEE Transactions on Geoscience and Remote Sensing , volume =. 2007 , doi =

    2007

  63. [71]

    IEEE Geoscience and Remote Sensing Magazine , volume =

    A tutorial on synthetic aperture radar , author =. IEEE Geoscience and Remote Sensing Magazine , volume =. 2013 , doi =

    2013

  64. [72]

    Science of Remote Sensing , volume =

    The Global Ecosystem Dynamics Investigation: High-resolution laser ranging of the Earth's forests and topography , author =. Science of Remote Sensing , volume =. 2020 , doi =

    2020

  65. [73]

    Science , volume =

    Spatial resolution for forest carbon maps , author =. Science , volume =. 2025 , doi =

    2025

  66. [74]

    Environmental Research Letters , volume =

    On the NASA GEDI and ESA CCI biomass maps: aligning for uptake in the UNFCCC global stocktake , author =. Environmental Research Letters , volume =. 2023 , doi =

    2023

  67. [75]

    Remote Sensing , volume =

    High-resolution aboveground biomass mapping: The benefits of biome-specific deep learning models , author =. Remote Sensing , volume =. 2025 , doi =

    2025

  68. [76]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    SatMAE: Pre-training transformers for temporal and multi-spectral satellite imagery , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  69. [77]

    arXiv preprint arXiv:1610.01644 , year =

    Understanding intermediate layers using linear classifier probes , author =. arXiv preprint arXiv:1610.01644 , year =

    arXiv

  70. [78]

    arXiv preprint arXiv:2211.07044 , year=

    SSL4EO-S12: A Large-Scale Multi-Modal, Multi-Temporal Dataset for Self-Supervised Learning in Earth Observation , author=. arXiv preprint arXiv:2211.07044 , year=

    arXiv

  71. [79]

    Jamie Tolan and Hung-I Yang and Benjamin Nosarzewski and Guillaume Couairon and Huy V. Vo and John Brandt and Justine Spore and Sayantan Majumdar and Daniel Haziza and Janaki Vamaraju and Theo Moutakanni and Piotr Bojanowski and Tracy Johns and Brian White and Tobias Tiecke an...

    2024

  72. [80]

    and Ciais, P

    Schwartz, M. and Ciais, P. and De Truchis, A. and Chave, J. and Ottl\'e, C. and Vega, C. and Wigneron, J.-P. and Nicolas, M. and Jouaber, S. and Liu, S. and Brandt, M. and Fayad, I. , TITLE =. Earth System Science Data , VOLUME =. 2023 , NUMBER =

    2023

  73. [81]

    Forests , volume=

    Review of remote sensing-based methods for forest aboveground biomass estimation: Progress, challenges, and prospects , author=. Forests , volume=. 2023 , publisher=

    2023

  74. [82]

    IEEE geoscience and remote sensing magazine , volume=

    Deep learning in remote sensing: A comprehensive review and list of resources , author=. IEEE geoscience and remote sensing magazine , volume=. 2017 , publisher=

    2017

  75. [83]

    Avitabile, Valerio and Herold, Martin and Heuvelink, Gerard B. M. and Lewis, Simon L. and Phillips, Oliver L. and Asner, Gregory P. and Armston, John and Ashton, Peter S. and Banin, Lindsay and Bayol, Nicolas and Berry, Nicholas J. and Boeckx, Pascal and de Jong, Bernardus H. ...

  76. [84]

    and Fransson, Johan E.S

    Rodríguez-Veiga, Pedro and Quegan, Shaun and Carreiras, Joao and Persson, Henrik J. and Fransson, Johan E.S. and Hoscilo, Agata and Ziółkowski, Dariusz and Stereńczak, Krzysztof and Lohberger, Sandra and St\". Forest biomass retrieval approaches from earth observation in diffe...

    2019 doi

  77. [85]

    and Armston, John and Dubayah, Ralph and Minor, David M

    Duncanson, Laura and Kellner, James R. and Armston, John and Dubayah, Ralph and Minor, David M. and Hancock, Steven and Healey, Sean P. and Patterson, Paul L. and Saarela, Svetlana and Marselis, Suzanne and Silva, Carlos E. and Bruening, Jamis and Goetz, Scott J. and Tang, Hao...

    2022

  78. [86]

    IEEE Geoscience and Remote Sensing Letters , year=

    HyBiomass: Global Hyperspectral Imagery Benchmark Dataset for Evaluating Geospatial Foundation Models in Forest Aboveground Biomass Estimation , author=. IEEE Geoscience and Remote Sensing Letters , year=

  79. [87]

    Cryo-Bench: Benchmarking Foundation Models for Cryosphere Applications , publisher =

    Kaushik, Saurabh and Maurya, Lalit and Tellman, Beth and Marsocci, Valerio , keywords =. Cryo-Bench: Benchmarking Foundation Models for Cryosphere Applications , publisher =. 2026 , copyright =. doi:10.48550/ARXIV.2603.01576 , url =

    2026 doi arXiv

  80. [88]

    arXiv preprint arXiv:2606.23503 , year =

    UniverSat: Resolution- and Modality-Agnostic Transformers for Earth Observation , author =. arXiv preprint arXiv:2606.23503 , year =

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.