Pith. sign in

REVIEW 4 major objections 6 minor 71 references

Scalable deep fusion of spaceborne lidar and synthetic aperture radar for global forest structural complexity mapping

T0 review · 4 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A compact neural net trained on GEDI lidar and SAR backscatter can map forest structural complexity continuously at 25 m resolution, reproducing 82% of held-out variability and yielding a global quarterly time series from 2015 to 2022.

desk verdict Solid, well-validated GEDI–SAR fusion for WSCI; the holdout R²=0.82 is real, but the 'global' 2015–2022 claim runs ahead of validation in boreal/tundra and pre-2019 years. read the letter →

arxiv 2510.06299 v1 pith:WNUWONDD submitted 2025-10-07 cs.CV cs.LGstat.AP

Tiago de Conto , John Armston , Ralph Dubayah This is my paper
classification cs.CVcs.LGstat.AP
keywords foreststructuralcomplexityGEDIsyntheticapertureradardatafusiondeeplearninguncertaintyquantificationglobalmappingWaveformIndex
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 sets out to show that sparse spaceborne lidar measurements can be expanded into continuous, wall-to-wall maps of forest structural complexity by training a small convolutional network to predict the Waveform Structural Complexity Index (WSCI) from radar backscatter plus location. It claims the model explains 82% of variability in independent GEDI observations while using fewer than 400,000 parameters, and that its uncertainty estimates are well calibrated. If correct, this gives ecologists a 25 m, quarterly, global record of forest structural complexity covering years and latitudes that GEDI itself never sampled, useful for tracking disturbance, recovery, and habitat quality.

What carries the argument

The central object is the Waveform Structural Complexity Index (WSCI), a single number derived from GEDI lidar waveforms that integrates multiple canopy structural attributes. The carrying mechanism is a compact adapted EfficientNetV2 with FusedMBConv and MBConv blocks that preserves pixel-level spatial context, a masked Gaussian negative-log-likelihood loss for aleatoric uncertainty, and Monte Carlo dropout plus five overlapping tile passes for epistemic uncertainty. The input stack combines L-band and C-band SAR backscatter and incidence angles, topography, and cyclic longitude/latitude, which together let the network learn region-specific baselines and local gradients from sparse lidar ta

What would settle it

Take the model's quarterly 2015–2018 predictions at NEON and EBA sites that have ALS-derived structural complexity measurements from the same years, and regress predictions against those observations; if stationarity holds, R² and bias should match the 2019–2022 validation values, while a clear decline in R² or systematic seasonal bias would falsify the extrapolation claim.

Watch

Extended reading notes

Core claim

The authors claim that a carefully scaled EfficientNetV2-style CNN, trained on about 133 million GEDI footprints gridded to 25 m in three-month intervals, can predict WSCI from L-band PALSAR, C-band Sentinel-1, a digital elevation model, and cyclic-encoded geographic coordinates. On 26.9 million held-out footprints the model achieves R² = 0.82, RMSE = 0.50, and near-zero bias, with 71% of observations falling within one predicted standard deviation. The same architecture, with the feature-extraction layers frozen, transfers to canopy height (R² = 0.69) and canopy cover (R² = 0.66). The trained model is then applied wall-to-wall to every quarter from 2015 to 2022, producing the global, multi-

Load-bearing premise

The relationship learned from 2019–2022 GEDI footprints between 51.6°N and 51.6°S is assumed to remain the same in 2015–2018 and at latitudes outside that range, so all pre-GEDI and high-latitude predictions rest on spatial and temporal stationarity of the SAR–WSCI relationship.

Editorial extensions

If this is right

  • Quarterly 25 m wall-to-wall WSCI maps for 2015–2022 become possible, extending forest structural records back before GEDI's 2019 launch and beyond its 51.6° orbital limits.
  • The compact architecture (365,682 trainable parameters) means global inference can run on ordinary CPUs, making high-resolution structural mapping accessible without specialized GPU infrastructure.
  • Calibrated uncertainty estimates allow users to distinguish low-confidence extrapolations in boreal regions and pre-2017 periods from high-confidence tropical predictions.
  • Transfer learning with frozen feature-extraction weights predicts canopy height and cover at competitive accuracy (R² ≈ 0.66–0.69) while training four times faster than updating the whole network.
  • Feature-importance analysis indicates L-band penetrates to woody structure, C-band adds temporal matching, and geographic coordinates act as a spatial prior—guidance for designing future fusion products.
  • ALS-based validation shows the model captures forest-to-non-forest transitions and spatial gradients in heterogeneous landscapes, supporting use for disturbance and habitat mapping.

Reading between the lines

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

  • The 2015–2018 portions of the time series are extrapolations under a stationarity assumption; users should weight those years by the elevated epistemic uncertainty the paper itself reports, and validate them against independent ALS or ICESat-2 records where available.
  • Some of the observed seasonal cycles above 23°N may be SAR artifacts (the paper acknowledges this possibility); separating true phenology from signal artifacts likely requires adding optical data or InSAR coherence.
  • Because WSCI compresses many structural attributes into one index, a single drop in predicted complexity does not reveal which canopy layer changed; a multi-head extension predicting height, cover, and foliage metrics separately could decompose the signal.
  • The remaining spatial structure in residuals at homogeneous Amazon sites suggests the global model misses fine-scale variation there; a testable extension is to fine-tune on regional ALS and see whether disturbance detection in those forests improves enough to justify the added data.
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

4 major / 6 minor

Summary. The manuscript presents a deep-learning framework that fuses GEDI L4C footprint WSCI observations with multi-source SAR data (ALOS-PALSAR-2, Sentinel-1, Copernicus DEM) and geographic coordinates to produce 25 m wall-to-wall estimates of forest structural complexity. The model is an adapted EfficientNetV2 with 365,682 trainable parameters, trained on ~133 million GEDI footprints gridded into 1 km chips over Apr 2019–Dec 2022 within roughly 51.6°N/S. Holdout validation uses a spatial-block split and reports R²=0.82, RMSE=0.50, bias=-0.02 over n=26.9M samples, with uncertainty coverage near the nominal level. The authors then apply the model to generate quarterly global WSCI maps for 2015–2022, including areas and periods outside the training distribution. They also validate against ALS-based CE_XYZ from NEON/EBA networks, analyze feature importance with SHAP, and demonstrate transfer learning to GEDI RH98 and canopy cover.

Significance. The core in-domain result—a compact EfficientNetV2 trained on a very large GEDI sample with a spatially blocked holdout (R²=0.82, n≈26.9M)—is a solid and useful contribution to GEDI/SAR fusion. The explicit uncertainty calibration assessment, the computational efficiency claim, the SHAP-based feature attribution, and the transfer-learning experiments are all valuable. However, the manuscript's headline deliverable, a global multi-temporal 2015–2022 WSCI dataset, rests on extrapolation beyond the GEDI spatial domain (51.6°N/S) and before the GEDI era, with no independent validation for those extrapolations. The ALS validation, which is the only wall-to-wall check, shows substantially lower performance (R²=0.47) and is partially entangled with the WSCI definition because WSCI is derived from CE_XYZ. The paper is transparent about several of these limitations, but the abstract and conclusions overstate the strength of the evidence.

major comments (4)
  1. [§3.3, §2.1, Fig. 7, Sup. Table 1] The global 2015–2022 product is an extrapolation. Training data are only Apr 2019–Dec 2022 within 51.6°N/S, yet inference is run for 2015–2018 and for all landmasses, including boreal and tundra regions. The paper's own Sup. Table 1 shows boreal forests/taiga R²=0.59 and tundra R²=0.21, and Fig. 7 shows elevated epistemic uncertainty beyond GEDI's latitudinal limit and before 2017. No independent reference data are provided for either the pre-2019 period or the >51.6°N region; the Discussion explicitly calls for further boreal ALS validation. This makes the abstract claim of 'accurate predictions... across biomes and time periods' unsupported. The authors should either provide independent validation for the extrapolation domain or explicitly scope the product and claims to the training domain.
  2. [§3.5, Fig. 10, Fig. 11] The ALS validation is weak and partially circular. CE_XYZ is the source quantity from which WSCI was derived (§2.1, ref. 13), so comparing model predictions to CE_XYZ is not fully independent of the target definition. More importantly, the global model explains only 47% of ALS CE_XYZ variability at NEON sites, and Fig. 11b shows spatially structured residuals, indicating that fine-scale patterns are not fully captured. This weakens the abstract's claim of 'preserving fine-scale spatial patterns' and suggests that the high GEDI holdout R² may not transfer to continuous wall-to-wall mapping. The authors should analyze the GEDI-vs-ALS performance gap and validate against a more independent reference (e.g., ICESat-2 or national ALS data) before claiming wall-to-wall accuracy.
  3. [§3.1] The model cannot predict WSCI below 7.4, i.e., unforested or sparsely vegetated pixels. Because the generated maps are global and wall-to-wall, this saturation must affect non-forest and transition areas. The reported GEDI R²=0.82 is dominated by forested pixels and may mask the severity of this issue. The authors should quantify the area affected by saturation, report errors in non-forest/transition classes, or restrict the global mapping claims to the forested domain where the model is valid.
  4. [§3.3, §4] The seasonal cycles at temperate and boreal latitudes are presented as a finding, yet the paper concedes in the Discussion that input layers have mismatched temporal resolutions (PALSAR yearly vs Sentinel-1 quarterly) and that 'it is unclear whether observed seasonal variations truly reflect forest phenology or are artifacts of signal saturation.' Without separating sensor artifacts from ecological signal, the multi-temporal dataset cannot support the stated monitoring applications. A concrete test—e.g., comparing WSCI seasonal anomalies against independent optical phenology or snow/freeze-thaw data—is needed before the seasonal patterns can be interpreted as forest structural dynamics.
minor comments (6)
  1. [Eq. 1, Eq. 2] The variable definitions are inconsistent: Eq. (1) uses y as the observation and μ as the predicted mean in the standard NLL form, but the text below Eq. (2) states 'y = model prediction; μ = target value (GEDI WSCI)'. This should be corrected. Also, Eq. (2) appears to contain a typo ('NlLL' versus 'NLL').
  2. [Sup. Fig. 1] The caption says the temporal stability is shown for '2019–2023', but the training period is described elsewhere as Apr 2019–Dec 2022. Please clarify whether the 2023 points are inference-only or a typo.
  3. [§3, §2.1] The text inconsistently refers to the GEDI latitudinal limit as '52° North to 52° South' in §3 and '51.6°N/S' elsewhere. Use one value consistently.
  4. [Table 1] The table header says 'PALSAR 1 & 2', but the text and methods describe the Global PALSAR-2 Yearly Mosaic only. Clarify whether PALSAR-1 data are used or remove the reference to PALSAR-1.
  5. [§3.4, references] In the text, 'Reichstein et al . 2019' and 'Tang et al. 2015' are cited as narrative text instead of the numbered references [53] and [54]. Also, reference 9 (Liu et al.) appears to be missing volume/page details; please standardize the reference list.
  6. [Fig. 11] The caption says 'Moran's I spatial cross-correlation' but the figures report cross-correlation between predicted and observed WSCI (panel a) and spatial autocorrelation of residuals (panel b). Please clarify the terminology so readers do not conflate the two statistics.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor self-referential ALS validation; the SAR-to-WSCI mapping itself is not circular.

  1. self definitional [§2.1 (ALS Data); §2.3 (ALS independent validation)]
    "The GEDI WSCI product was derived from ALS measurements of 3D canopy entropy (CE XYZ) [9] matched to GEDI footprints [13]. This makes the fusion estimates directly comparable to the ALS measurements at 25 m resolution."

    The paper validates its WSCI predictions against ALS CE_XYZ, which is the source quantity from which WSCI itself was constructed in the authors' prior work (refs 9, 13). Thus the ALS comparison is not a fully independent external benchmark of the target: it re-uses the same definitional relationship between WSCI and CE_XYZ. This does not make the supervised SAR-to-WSCI mapping circular — the held-out GEDI test set provides genuinely independent validation of the learned relationship — but it weakens the claim that the ALS data serve as an independent ground-truth check on the WSCI product.

full rationale

The core derivation chain is a supervised regression of GEDI L4C WSCI on multi-source SAR inputs, trained on ~133 million GEDI footprints and tested on spatially blocked 20% holdout GEDI data. That central mapping is not circular: the target values come from GEDI waveforms and are not constructed from the SAR predictors. The only notable self-referential element is that WSCI is an author-defined index, and the 'independent' ALS validation uses CE_XYZ, the quantity from which WSCI itself was derived in prior work; this is a validation-entanglement limitation rather than a derivation that reduces to its inputs. No equation in the paper equates a prediction to a fitted parameter by construction. The temporal and latitudinal extrapolation (2015–2018, beyond 51.6°N/S) is a correctness and generalization risk, not a circularity, because no fitted value is renamed as a prediction. Score 2 reflects the minor self-referential validation; it is not higher because the main GEDI-holdout evaluation is genuinely independent.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The model is an empirical regression; its free parameters are the network weights and hand-chosen training/config choices. It assumes GEDI WSCI ground truth, input-feature sufficiency, and spatiotemporal stationarity. The target WSCI is an author-defined composite index rather than an independently measured physical property.

free parameters (4)
  • Neural network weights (365,682 trainable parameters) = 365,682 real-valued weights
    Learned by optimizing Gaussian NLL on ~133M GEDI footprints; these weights define the SAR-to-WSCI mapping and are the core fitted parameters of the model.
  • Training hyperparameters = lr=0.001, 50 epochs, batch=96, dropout=0.2
    Chosen by the authors; influence accuracy and uncertainty calibration and are not derived from theory.
  • Chip retention threshold (≥1% non-empty pixels) = 1% (≥16 of 1600 pixels)
    Ad hoc cutoff balancing data quality and tropical coverage; shapes training domain and causes inability to predict low WSCI.
  • Edge buffer (10% / 4 pixels) = 4 pixels
    Ad hoc design choice to mitigate edge artifacts in inference and loss computation.
assumptions (5)
  • domain assumption GEDI L4C WSCI v2 footprint values are treated as true targets with negligible error.
    The paper states in §2.3 that it 'considered the GEDI WSCI estimates as true targets when training'; any systematic error in WSCI is inherited by the model.
  • domain assumption The chosen SAR/DEM features (PALSAR HH/HV, Sentinel-1 VV/VH, incidence angles, DEM, cyclic coords) are sufficient to predict WSCI at 25m.
    Central data-fusion premise; the paper tests ablations (§3.4) but cannot prove sufficiency.
  • domain assumption The SAR–WSCI relationship is stationary across time and space, enabling extrapolation to 2015–2018 and latitudes >51.6°.
    Inference in §3.3 extends beyond training domain; the paper reports higher epistemic uncertainty there, and §4 flags seasonal-signal ambiguity.
  • domain assumption ALS-derived 3D canopy entropy (CE_XYZ) is directly comparable to WSCI and valid as ground truth.
    WSCI was originally derived from CE_XYZ (§2.1), making this comparison partially circular but still an independent sensor source.
  • standard math Gaussian NLL loss and MC-dropout yield well-calibrated predictive uncertainty.
    Standard ML assumptions used in §2.2; validity is empirically checked but not guaranteed.
invented entities (1)
  • WSCI (Waveform Structural Complexity Index)
    purpose: Composite target variable used to define forest structural complexity from GEDI waveforms; the model is trained to predict it.
    WSCI is an author-defined index (from prior work, ref 13), not a directly measured physical quantity. Its ecological meaning rests on correlations with other structural metrics; no external falsifiable handle is provided for the index itself in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scalable deep fusion of spaceborne lidar and synthetic aperture radar for global forest structural complexity mapping." pith.science (2026). https://pith.science/paper/WNUWONDD

@misc{pith2026251006299,
  author       = {Pith},
  title        = {Pith review of: Scalable deep fusion of spaceborne lidar and synthetic aperture radar for global forest structural complexity mapping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WNUWONDD}},
  note         = {Machine review of arXiv:2510.06299}
}
read the original abstract

Forest structural complexity metrics integrate multiple canopy attributes into a single value that reflects habitat quality and ecosystem function. Spaceborne lidar from the Global Ecosystem Dynamics Investigation (GEDI) has enabled mapping of structural complexity in temperate and tropical forests, but its sparse sampling limits continuous high-resolution mapping. We present a scalable, deep learning framework fusing GEDI observations with multimodal Synthetic Aperture Radar (SAR) datasets to produce global, high-resolution (25 m) wall-to-wall maps of forest structural complexity. Our adapted EfficientNetV2 architecture, trained on over 130 million GEDI footprints, achieves high performance (global R2 = 0.82) with fewer than 400,000 parameters, making it an accessible tool that enables researchers to process datasets at any scale without requiring specialized computing infrastructure. The model produces accurate predictions with calibrated uncertainty estimates across biomes and time periods, preserving fine-scale spatial patterns. It has been used to generate a global, multi-temporal dataset of forest structural complexity from 2015 to 2022. Through transfer learning, this framework can be extended to predict additional forest structural variables with minimal computational cost. This approach supports continuous, multi-temporal monitoring of global forest structural dynamics and provides tools for biodiversity conservation and ecosystem management efforts in a changing climate.

Figures

Figures reproduced from arXiv: 2510.06299 by the authors.

Figure 1
Figure 1. Methodological workflow for the development of a fusion model to map GEDI WSCI at high spatial resolution. The process integrates three primary data sources: GEDI L4C footprint WSCI measurements (top), multi-source SAR datasets (middle), and airborne laser scanning [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Detailed architecture of our adapted EfficientNetV2 neural network for WSCI estimation. The network (left) processes 40×40×10 input images (7 SAR layers + 3 channels with encoded geographical coordinates) and outputs 32×32×2 predictions (mean and variance) after removing a 4-pixel border to avoid edge artifacts. The network maintains spatial dimensions [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Spatial and channel-wise feature importance analysis using Deep SHAP explainers for a single predicted pixel. This visualization illustrates the various factors influencing the model's prediction at the individual pixel level. (a) Wall-to-wall WSCI fusion values for a 1×1 km image chip at 25m resolution, with the red star marking the target pixel being analyzed. (b) Corresponding standard deviation map showing predi… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Performance evaluation of the GEDI-SAR WSCI fusion model using independent test data. (a) Density scatterplot showing the relationship between predicted fusion WSCI and observed GEDI WSCI values (R² = 0.82, RMSE = 0.50, bias = -0.02, n = 26,932,199). (b) Relationship b…
Figure 7
Figure 7. Figure 7: Latitudinal patterns in WSCI and estimated uncertainty components from 2015 to 2022. Data are averaged in 5-degree latitude bands and 3-month intervals. Top panel: Predicted WSCI values showing seasonal cycles at temperate and boreal latitudes. Middle panel: Data uncer…
Figure 7
Figure 7. Figure 7: Transfer learning performance for GEDI canopy height (RH98) [PITH_FULL_IMAGE:figures/full_fig_p049_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 1 canonical work pages

  1. [1]

    Ecosystem Structure throughout the Brazilian Amazon from Landsat Observations and Automated Spectral Unmixing

    Asner GP, Knapp DE, Cooper AN, Bustamante MMC, Olander LP. Ecosystem Structure throughout the Brazilian Amazon from Landsat Observations and Automated Spectral Unmixing. Earth Interactions. 2005 Jun 1;9(7):1–31

    2005

  2. [2]

    Global patterns and climatic controls of forest structural complexity

    Ehbrecht M, Seidel D, Annighöfer P, Kreft H, Köhler M, Zemp DC, et al. Global patterns and climatic controls of forest structural complexity. Nature Communications. 2021 Dec 1;12(1):519

    2021

  3. [3]

    Remotely sensed forest structural complexity predicts multi species occurrence at the landscape scale

    Zellweger F, Braunisch V, Baltensweiler A, Bollmann K. Remotely sensed forest structural complexity predicts multi species occurrence at the landscape scale. Forest Ecology and Management. 2013 Nov 1;307:303–12

    2013

  4. [4]

    Integrating forest structural diversity measurement into ecological research

    Atkins JW, Bhatt P, Carrasco L, Francis E, Garabedian JE, Hakkenberg CR, et al. Integrating forest structural diversity measurement into ecological research. Ecosphere. 2023 Sep;14(9):e4633

    2023

  5. [5]

    Unravelling the relationship between plant diversity and vegetation structural complexity: A review and theoretical framework

    Coverdale TC, Davies AB. Unravelling the relationship between plant diversity and vegetation structural complexity: A review and theoretical framework. Journal of Ecology. 2023 Jul;111(7):1378–95

    2023

  6. [6]

    Forest and woodland stand structural complexity: Its definition and measurement

    McElhinny C, Gibbons P, Brack C, Bauhus J. Forest and woodland stand structural complexity: Its definition and measurement. Forest Ecology and Management. 2005 Oct 24;218(1–3):1–24

    2005

  7. [7]

    High rates of primary production in structurally complex forests

    Gough CM, Atkins JW, Fahey RT, Hardiman BS. High rates of primary production in structurally complex forests. Ecology. 2019 Oct 1;100(10):e02864

    2019

  8. [8]

    Quantifying stand structural complexity and its relationship with forest management, tree species diver sity and microclimate

    Ehbrecht M, Schall P, Ammer C, Seidel D. Quantifying stand structural complexity and its relationship with forest management, tree species diver sity and microclimate. Agricultural and Forest Meteorology. 2017 Aug 15;242:1–9

    2017

Show all 71 references
  1. [9]

    A novel entropy -based method to quantify forest canopy structural complexity from multiplatform lidar point clouds

    Liu X, Ma Q, Wu X, Hu T, Liu Z, Liu L, et al. A novel entropy -based method to quantify forest canopy structural complexity from multiplatform lidar point clouds. Remote S ensing of Environment. 2022 Dec;282:113280. 35

    2022

  2. [10]

    A new index of forest structural heterogeneity using tree architectural attributes measured by terrestrial laser scanning

    Reich KF, Kunz M, von Oheimb G. A new index of forest structural heterogeneity using tree architectural attributes measured by terrestrial laser scanning. Ecological Indicators. 2021 Dec 1;133

    2021

  3. [11]

    Measuring habitat complexity and spatial heterogeneity in ecology

    Loke L HL, Chisholm RA. Measuring habitat complexity and spatial heterogeneity in ecology. Ecology Letters [Internet]. 2022 Aug 17; Available from: https://onlinelibrary.wiley.com/doi/10.1111/ele.14084

    2022 doi

  4. [12]

    The Global Ecosystem Dynamics Investigation: High -resolution laser ranging of the Earth’s forests and topography

    Dubayah R, Blair JB, Goetz S, Fatoyinbo L, Hansen M, Hea ley S, et al. The Global Ecosystem Dynamics Investigation: High -resolution laser ranging of the Earth’s forests and topography. Science of Remote Sensing. 2020 Jun;1:100002

    2020

  5. [13]

    Characterizing the structural complexity of the Earth’s forests with spaceborne lidar

    De Conto T, Armston J, Dubayah R. Characterizing the structural complexity of the Earth’s forests with spaceborne lidar. Nat Commun. 2024 Sep 16;15(1):8116

    2024

  6. [14]

    A high- resolution canopy height model of the Earth

    Lang N, Jetz W, Schindler K, Wegner JD. A high- resolution canopy height model of the Earth. Nat Ecol Evol. 2023 Nov;7(11):1778–89

    2023

  7. [15]

    Un ified Deep Learning Model for Global Prediction of Aboveground Biomass, Canopy Height and Cover from High -Resolution, Multi-Sensor Satellite Imagery [Internet]

    Weber M, Beneke C, Wheeler C. Un ified Deep Learning Model for Global Prediction of Aboveground Biomass, Canopy Height and Cover from High -Resolution, Multi-Sensor Satellite Imagery [Internet]. arXiv; 2024 [cited 2024 Aug 23]. Available from: http://arxiv.org/abs/2408.11234

    2024 arXiv

  8. [16]

    Monitoring of Forest Structure Dynamics by Means of L -Band SAR Tomography

    Cazcarra-Bes V, Tello-Alonso M, Fischer R, Heym M, Papathanassiou K. Monitoring of Forest Structure Dynamics by Means of L -Band SAR Tomography. Remote Sensing. 2017 Nov 28;9(12):1229

    2017

  9. [17]

    Sensitivity of Multi -Source SAR Backscatter to Changes in Forest Aboveground Biomass

    Huang W, Sun G, Ni W, Zhang Z, Dubayah R. Sensitivity of Multi -Source SAR Backscatter to Changes in Forest Aboveground Biomass. Remote Sensing. 2015 Jul 28;7(8):9587– 609

    2015

  10. [18]

    Remote sensing approaches to monitor tropical forest restoration: Current methods and future possibilities

    De Almeida DRA, Vedovato LB, Fuza M, Molin P, Cassol H, Resende AF, et al. Remote sensing approaches to monitor tropical forest restoration: Current methods and future possibilities. Journal of Applied Ecology. 2025 Feb;62(2):188–206. 36

    2025

  11. [19]

    Multisource and multitemporal data fusion in remote sensing: A comprehensive review of the state of the art

    Ghamisi P, Rasti B, Yokoya N, Wang Q, Hofle B, Bruzzone L, et al. Multisource and multitemporal data fusion in remote sensing: A comprehensive review of the state of the art. IEEE Geoscience and Remote Sensing Magazine. 2019 Mar 1;7(1):6–39

    2019

  12. [20]

    Spatiotemporal Fusion of Multisource Remote Sensing Data: Literature Survey, Taxonomy, Principles, Applications, and Future Directions

    Zhu X, Cai F, Tian J, Williams T. Spatiotemporal Fusion of Multisource Remote Sensing Data: Literature Survey, Taxonomy, Principles, Applications, and Future Directions. Remote Sensing. 2018 Mar 29;10(4):527

    2018

  13. [21]

    A Deep Learning Approach to Estimate Canopy Height and Uncertainty by Integrating Seasonal Optical, SAR and Limited GEDI LiDAR Data over Northern Forests [Internet]

    Castro JB, Rogers C, Sothe C, Cyr D, Gonsamo A. A Deep Learning Approach to Estimate Canopy Height and Uncertainty by Integrating Seasonal Optical, SAR and Limited GEDI LiDAR Data over Northern Forests [Internet]. arXi v; 2024 [cited 2025 Feb 9]. Available from: http://arxiv.o...

    2024 arXiv

  14. [22]

    Fusing Sentinel -1 and -2 to Model GEDI -Derived Vegetation Structure Characteristics in GEE for the Paraguayan Chaco

    Kacic P, Hirner A, Da Ponte E. Fusing Sentinel -1 and -2 to Model GEDI -Derived Vegetation Structure Characteristics in GEE for the Paraguayan Chaco. Remote Sensing. 2021 Jan;13(24):5105

    2021

  15. [23]

    Mapping large -scale pantropical forest canopy height by integrating GEDI lidar and TanDEM -X InSAR data

    Qi W, Armston J, Choi C, Stovall A, Saarela S, Pardini M, et al. Mapping large -scale pantropical forest canopy height by integrating GEDI lidar and TanDEM -X InSAR data. Remote Sensing of Environment. 2025 Mar;318:114534

    2025

  16. [24]

    Improved forest height estimation by fusion of simulated GEDI Lidar data and TanDEM-X InSAR data

    Qi W, Lee SK, Hancock S, Luthcke S, Tang H, Armston J, et al. Improved forest height estimation by fusion of simulated GEDI Lidar data and TanDEM-X InSAR data. Remote Sensing of Environment. 2019 Feb 1;221:621–34

    2019

  17. [25]

    Forest biomass estimation over three distinct forest types using TanDEM-X InSAR data and simulated GEDI lidar data

    Qi W, Saarela S, Armston J, Ståhl G, Dubayah R. Forest biomass estimation over three distinct forest types using TanDEM-X InSAR data and simulated GEDI lidar data. Remote Sensing of Environment. 2019 Oct;232:111283

    2019

  18. [26]

    Fusing simulated GEDI, ICESat-2 and NISAR data for regional aboveground biomass mapping

    Silva CA, Duncanson L, Hancock S, Neuenschwander A, Thomas N, Hofton M, et al. Fusing simulated GEDI, ICESat-2 and NISAR data for regional aboveground biomass mapping. Remote Sensing of Environment. 2021 Feb 1;253. 37

    2021

  19. [27]

    Estimation of Aboveground Biomass for Different Forest Types Using Data from Sentinel- 1, Sentinel-2, ALOS PALSAR-2, and GEDI

    Wang C, Zhang W, Ji Y, Marino A, Li C, Wang L, et al. Estimation of Aboveground Biomass for Different Forest Types Using Data from Sentinel- 1, Sentinel-2, ALOS PALSAR-2, and GEDI. Forests. 2024 Jan;15(1):215

    2024

  20. [28]

    Towards the next generation of Geospatial Artificial Intelligence

    Mai G, Xie Y, Jia X, Lao N, Rao J, Zhu Q, et al. Towards the next generation of Geospatial Artificial Intelligence. International Journal of Applied Earth Observation and Geoinformation. 2025 Feb 1;136:104368

    2025

  21. [29]

    Global Ecosystem Dynamics Investigation (GEDI)GEDI L4C Footprint Level Waveform Structural Complexity Index, Version 2 [Internet]

    De Conto T, Armston J, Dubayah RO. Global Ecosystem Dynamics Investigation (GEDI)GEDI L4C Footprint Level Waveform Structural Complexity Index, Version 2 [Internet]. ORNL Distributed Active Archive Center; 2024 [cited 2024 Aug 20]. p. 0 MB. Available from: https://daac.ornl.go...

    2024

  22. [30]

    Global Ecosystem Dynamics Investigation (GEDI)GEDI L3 Gridded Land Surface Metrics, Version 2 [Internet]

    Dubayah RO, Luthcke SB, Sabaka TJ, Nicholas JB, Preaux S, Hofton MA. Global Ecosystem Dynamics Investigation (GEDI)GEDI L3 Gridded Land Surface Metrics, Version 2 [Internet]. ORNL Distributed Active Archive Center; 2021 [cited 2025 Apr 7]. p. 0 MB. Available from: https://daac...

    2021

  23. [31]

    New global forest/non-forest maps from ALOS PALSAR data (2007–2010)

    Shimada M, Itoh T, Motooka T, Watanabe M, Shiraishi T, Thapa R, et al. New global forest/non-forest maps from ALOS PALSAR data (2007–2010). Remote Sensing of Environment. 2014 Dec;155:13–31

    2007

  24. [32]

    Sentinel -1 SAR Backscatter Analysis Ready Data Preparation in Google Earth Engine

    Mullissa A, Vollrath A, Odongo -Braun C, Slagter B, Balling J, Gou Y, et al. Sentinel -1 SAR Backscatter Analysis Ready Data Preparation in Google Earth Engine. Remote Sensing. 2021 May 17;13(10):1954

    2021

  25. [33]

    Copernicus DEM [Internet]

    European Space Agency, Airbus. Copernicus DEM [Internet]. 2022 [cited 2025 Mar 11]. Available from: https://dataspace.copernicus.eu/explore -data/data-collections/copernicus- contributing-missions/collections-description/COP-DEM

    2022

  26. [34]

    TanDE M-X - Digital Elevation Model (DEM) - Global, 90m [Internet]

    German Aerospace Center. TanDE M-X - Digital Elevation Model (DEM) - Global, 90m [Internet]. German Aerospace Center (DLR); 2018 [cited 2025 Mar 12]. Available from: https://geoservice.dlr.de/data-assets/ju28hc7pui09.html 38

    2018

  27. [35]

    The Shuttle Radar Topography Mission

    Farr TG, Rosen PA, Caro E, Crippen R, Duren R, Hensley S, et al. The Shuttle Radar Topography Mission. Reviews of Geophysics. 2007 Jun;45(2):2005RG000183

    2007

  28. [36]

    ASTER Global Digital Elevation Model V003 [Internet]

    NASA, METI, AIST. ASTER Global Digital Elevation Model V003 [Internet]. NASA EOSDIS Land Processes Distributed Active Archive Center; 2019 [cited 2025 Mar 12]. Available from: https://lpdaac.usgs.gov/products/astgtmv003/

    2019

  29. [37]

    Google Earth Engine: Planetary-scale geospatial analysis for everyone

    Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment. 2017 Dec;202:18–27

    2017

  30. [38]

    Discrete return LiDAR point cloud (DP1.30003.001) [Internet]

    National Ecological Observatory Network (NEON). Discrete return LiDAR point cloud (DP1.30003.001) [Internet]. National Ecological Observatory Network (NEON); 2024. Available from: https://data.neonscience.org/data-products/DP1.30003.001/RELEASE-2024

    2024

  31. [39]

    L1A - Discrete airborne LiDAR transects collected by EBA in the Brazilian Amazon (Roraima e Amapá) [Internet]

    Ometto J, Gorgens EB, Pereira FR de S, Sato L, Assis MLR, Cantinho R, et al. L1A - Discrete airborne LiDAR transects collected by EBA in the Brazilian Amazon (Roraima e Amapá) [Internet]. Zenodo; 2023 [cited 2024 Feb 13]. Available from: https://zenodo.org/records/7689693

    2023

  32. [40]

    A biomass map of the Brazilian Amazon from multisource remote sensing

    Ometto JP, Gorgens EB, de Souza Pereira FR, Sato L, de Assis MLR, Cantinho R, et al. A biomass map of the Brazilian Amazon from multisource remote sensing. Sci Data. 2023 Sep 30;10(1):668

    2023

  33. [41]

    EfficientNetV2: Sm aller Models and Faster Training

    Tan M, Le QV. EfficientNetV2: Sm aller Models and Faster Training. 2021 [cited 2025 Mar 11]; Available from: https://arxiv.org/abs/2104.00298

    2021 arXiv

  34. [42]

    EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks [Internet]

    Tan M, Le QV. EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks [Internet]. arXiv; 2020 [cited 2025 Feb 13]. Available from: http://arxiv.org/abs/1905.11946

    2020 arXiv

  35. [43]

    U-Net: Convolutional Networks for Biomedical Image Segmentation [Internet]

    Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation [Internet]. arXiv; 2015 [cited 2025 Mar 11]. Available from: https://arxiv.org/abs/1505.04597 39

    2015 arXiv

  36. [44]

    Dropout as a Bayesian approximation: representing model uncertainty in deep learning

    Gal Y, Ghahraman i Z. Dropout as a Bayesian approximation: representing model uncertainty in deep learning. Proceedings of the 33rd International Conference on International Conference on Machine Learning. 2016;48((ICML’16)):1050–9

    2016

  37. [45]

    Adam: A Method for Stochastic Optimization [Internet]

    Kingma DP, Ba J. Adam: A Method for Stochastic Optimization [Internet]. arXiv; 2014 [cited 2025 Jun 17]. Available from: https://arxiv.org/abs/1412.6980

    2014 arXiv

  38. [46]

    Terrestrial Ecoregions of the World: A New Ma p of Life on Earth

    Olson DM, Dinerstein E, Wikramanayake ED, Burgess ND, Powell GVN, Underwood EC, et al. Terrestrial Ecoregions of the World: A New Ma p of Life on Earth. BioScience. 2001;51(11):933

    2001

  39. [47]

    A Unified Approach to Interpreting Model Predictions

    Lundberg SM, Lee SI. A Unified Approach to Interpreting Model Predictions. In: Advances in Neural Information Processing Systems [Internet]. Curran Associates, Inc.; 2017 [cited 2024 Jan 23]. Available from: https://proceedings.neurips.cc/paper_files/paper/2017/hash/8a20a86219...

    2017

  40. [48]

    Burst Misalignment Evaluation for ALOS- 2 PALSAR-2 ScanSAR-ScanSAR Interferometry

    Natsuaki R, Motohka T, Shimada M, Suzuki S. Burst Misalignment Evaluation for ALOS- 2 PALSAR-2 ScanSAR-ScanSAR Interferometry. Remote Sensing. 2017 Feb 28;9(3):216

    2017

  41. [49]

    SAR interferometry using ALOS-2 PALSAR-2 data for the Mw 7.8 Gorkha, Nepal earthquake

    Natsuaki R, Nagai H, Motohka T, Ohki M, Watanabe M, Thapa RB, et al. SAR interferometry using ALOS-2 PALSAR-2 data for the Mw 7.8 Gorkha, Nepal earthquake. Earth Planets Space. 2016 Dec;68(1):15

    2016

  42. [50]

    Regenerated ALOS -2/PALSAR-2 global mosaics 2016 and 2014/2015 for forest observations

    Shimada M, Itoh T, Mot ooka T. Regenerated ALOS -2/PALSAR-2 global mosaics 2016 and 2014/2015 for forest observations. In: 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) [Internet]. Fort Worth, TX: IEEE; 2017 [cited 2025 Jun 10]. p. 2454–7. Available ...

    2016

  43. [51]

    Comparative Study on Remote Sensing Methods for Forest Height Mapping in Complex Mountainous Environments

    Huang X, Cheng F, Wang J, Yi B, Bao Y. Comparative Study on Remote Sensing Methods for Forest Height Mapping in Complex Mountainous Environments. Remote Sensing. 2023 Apr 25;15(9):2275. 40

    2023

  44. [52]

    Distribution Pattern of Woody Plants in a Mountain Forest Ecosystem Influenced by Topography and Monsoons

    Zhou X, Wang Z, Liu W, Fu Q, Shao Y, Liu F, et al. Distribution Pattern of Woody Plants in a Mountain Forest Ecosystem Influenced by Topography and Monsoons. Forests. 2022 Jun 19;13(6):957

    2022

  45. [53]

    Deep learning and process understanding for data -driven Earth system science

    Reichstein M, Camps -Valls G, Stevens B, Jung M, Denzler J, C arvalhais N, et al. Deep learning and process understanding for data -driven Earth system science. Nature. 2019 Feb;566(7743):195–204

    2019

  46. [54]

    Improving Image Classification with Location Context

    Tang K, Paluri M, Fei -Fei L, Fergus R, Bourdev L. Improving Image Classification with Location Context. In: 2015 IEEE International Conference on Computer Vision (ICCV) [Internet]. 2015 [cited 2025 Apr 8]. p. 1008–16. Available from: https://ieeexplore.ieee.org/document/7410478

    2015

  47. [55]

    Forest Biomass Estimation Using Deep Learning Data Fusion of Lidar, Multispectral, and Topographic Data Remote Sensing of Environment [Internet]

    Seely H, Coops NC, White JC, Montwé D, Ragab A. Forest Biomass Estimation Using Deep Learning Data Fusion of Lidar, Multispectral, and Topographic Data Remote Sensing of Environment [Internet]. 2024 [cited 2025 Mar 8]. Available from: https://www.ssrn.com/abstract=5006646

    2024

  48. [56]

    Generation of country-scale canopy height maps over Gabon using deep learning and TanDEM-X InSAR data

    Carcereri D, Rizzoli P, Dell’Amore L, Bueso -Bello JL, Ienco D, Bruzzone L. Generation of country-scale canopy height maps over Gabon using deep learning and TanDEM-X InSAR data. Remote Sensing of Environment. 2024 Sep;311:114270

    2024

  49. [57]

    Multimodal Deep Learning Enables Forest Height Mapping from Patchy Spaceborne Lidar Using Sar and Passive Optical Satellite Data [Internet]

    Chen M, Dong W, Yu H, Woodhouse IH, Ryan CM, Liu H, et al. Multimodal Deep Learning Enables Forest Height Mapping from Patchy Spaceborne Lidar Using Sar and Passive Optical Satellite Data [Internet]. 2024 [cited 2025 Mar 8]. Available from: https://www.ssrn.com/abstract=4898106

    2024

  50. [58]

    Country -wide high- resolution vegetation height mapping with Sentinel-2

    Lang N, Schindler K, Wegner JD. Country -wide high- resolution vegetation height mapping with Sentinel-2. Remote Sensing of Environment. 2019 Nov 1;233:111347

    2019

  51. [59]

    Comparison of three global canopy height maps and their applicability to biodivers ity modeling: Accuracy issues revealed

    Moudrý V, Gábor L, Marselis S, Pracná P, Barták V, Prošek J, et al. Comparison of three global canopy height maps and their applicability to biodivers ity modeling: Accuracy issues revealed. Ecosphere. 2024 Oct;15(10):e70026. 41

    2024

  52. [60]

    Repeat GEDI footprints measure the effects of tropical forest disturbances

    Holcomb A, Burns P, Keshav S, Coomes DA. Repeat GEDI footprints measure the effects of tropical forest disturbances. Remote Sensing of Environment. 2024 Jul 1;308:114174

    2024

  53. [61]

    Airborne and Spaceborne Lidar Reveal Trends and Patterns of Functional Diversity in a Semi -Arid Ecosystem

    Ilangakoon N, Glenn NF, Schneider FD, Dashti H, Hancock S, Spaete L, et al. Airborne and Spaceborne Lidar Reveal Trends and Patterns of Functional Diversity in a Semi -Arid Ecosystem. Front Remote Sens. 2021 Nov 12;2:743320

    2021

  54. [62]

    High- Resolution Global Maps of 21st -Century Forest Cover Change

    Hansen MC, Potapov PV, Moore R, Hancher M, Turubanova SA, Tyukavina A, et al. High- Resolution Global Maps of 21st -Century Forest Cover Change. Science. 2013 Nov 15;342(6160):850–3

    2013

  55. [63]

    Forest disturbance alerts for the Congo Basin using Sentinel -1

    Reiche J, Mullissa A, Slagter B, Gou Y, Tsendbazar NE, Odongo -Braun C, et al. Forest disturbance alerts for the Congo Basin using Sentinel -1. Environ Res Lett. 2021 Feb 1;16(2):024005

    2021

  56. [64]

    Mapping global forest canopy height through integration of GEDI and Landsat data

    Potapov P, Li X, Hernandez -Serna A, Tyukavina A, Hansen MC, Kommareddy A, et al. Mapping global forest canopy height through integration of GEDI and Landsat data. Remote Sensing of Environment. 2021 Feb;253:112165

    2021

  57. [65]

    A New InSAR Temporal Decorrelation Model for Seasonal Vegetation Change With Dense Time- Series Data

    Bhogapurapu N, Siqueira P, Armston J. A New InSAR Temporal Decorrelation Model for Seasonal Vegetation Change With Dense Time- Series Data. IEEE Geosci Remote Sensing Lett. 2024;21:1–5

    2024

  58. [66]

    NASA -ISRO Synthetic Aperture Radar (NISAR) Mission

    Kellogg K, Hoffman P, Standley S, Shaffer S, Rosen P, Edelstein W, et al. NASA -ISRO Synthetic Aperture Radar (NISAR) Mission. In: 2020 IEEE Aerospace Conference [Internet]. Big Sky, MT, USA: IEEE; 2020 [cited 2025 Mar 11]. p. 1– 21. Available from: https://ieeexplore.ieee.org...

    2020

  59. [67]

    The European Space Agency BIOMASS mission: Measuring forest above-ground biomass from space

    Quegan S, Le Toan T, Chave J, Dall J, Exbrayat JF, Minh DHT, et al. The European Space Agency BIOMASS mission: Measuring forest above-ground biomass from space. Remote Sensing of Environment. 2019 Jun;227:44–60

    2019

  60. [68]

    Spatially Continuous Mapping of Forest Canopy Height in Canada by Combining GEDI and ICESat -2 with PALSAR and Sentinel

    Sothe C, Gonsamo A, Lourenço RB, Kurz WA, Snider J. Spatially Continuous Mapping of Forest Canopy Height in Canada by Combining GEDI and ICESat -2 with PALSAR and Sentinel. Remote Sensing. 2022 Oct 15;14(20):5158. 42

    2022

  61. [69]

    The Ice, Cloud, and land Elevation Satellite -2 (ICESat-2): Science requirements, concept, and implementation

    Markus T, Neumann T, Martino A, Abdalati W, Brunt K, Csatho B, et al. The Ice, Cloud, and land Elevation Satellite -2 (ICESat-2): Science requirements, concept, and implementation. Remote Sensing of Environment. 2017 Mar;190:260–73

    2017

  62. [70]

    Computational tools for assessing forest recovery with GEDI shots and forest change maps

    Holcomb A, Mathis SV, Coomes DA, Keshav S. Computational tools for assessing forest recovery with GEDI shots and forest change maps. Science of Remote Sensing. 2023 Dec;8:100106

    2023

  63. [71]

    Evidential Deep Learning: Enhancing Predictive Uncertai nty Estimation for Earth System Science Applications

    Schreck JS, Gagne DJ, Becker C, Chapman WE, Elmore K, Fan D, et al. Evidential Deep Learning: Enhancing Predictive Uncertai nty Estimation for Earth System Science Applications. Artificial Intelligence for the Earth Systems [Internet]. 2024 Dec 11 [cited 2025 Jan 27];3(4). Ava...

    2024

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.