Pith. sign in

REVIEW 3 major objections 4 minor 64 references

VibrantVS: A high-resolution multi-task transformer for forest canopy height estimation

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Using 4-band NAIP imagery, VibrantVS estimates canopy height at 0.5 m resolution with a median MAE of 2.71 m, beating Meta, LANDFIRE, and ETH across western US ecoregions.

desk verdict A credible new benchmark and likely a real accuracy gain, but the random tile split with no spatial buffer makes the headline gap to baselines soft. read the letter →

arxiv 2412.10351 v3 pith:RHXFQMRT submitted 2024-12-13 cs.CV

Tony Chang , Kiarie Ndegwa , Andreas Gros , Vincent A. Landau , Luke J. Zachmann , Bogdan State , Mitchell A. Gritts , Colton W. Miller
show 3 more authors
Nathan E. Rutenbeck Scott Conway Guy Bayes
This is my paper · ORCID
classification cs.CV
keywords canopyheightmodelvisiontransformerNAIPimagerylidarremotesensingforeststructuremulti-tasklearningwildfirerisk
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a multi-task vision transformer called VibrantVS, trained on 4-band NAIP aerial imagery and USGS 3DEP lidar, can produce 0.5-meter canopy height models for the western United States that are more accurate and more precise than three published alternatives: Meta's global map, LANDFIRE's national layer, and ETH's Sentinel-2/GEDI global model. Across 67,227 held-out test tiles, VibrantVS has a median MAE of 2.71 meters versus 4.83, 5.96, and 7.05 meters for the baselines, and it keeps this advantage across most EPA Level 3 ecoregions and most height bins. If true, the practical payoff is that forest managers could get wall-to-wall, sub-meter canopy height updates every three years or less from freely available imagery, without commissioning new lidar flights.

What carries the argument

The carrying object is a multi-task Vision Transformer with a Swin-v2 encoder, a dense prediction transformer (DPT) decoder, and two prediction heads: a metric-bin module head that estimates each pixel's distribution over 64 height bins and linearly combines them into a 0.5-meter canopy height, and a lightweight convolutional head that predicts canopy cover at 10-meter resolution. The encoder uses grouped-query attention, Flash Attention 2, SWIGLU activation, and RMSNorm, and the inference context window is extended to 1,536x1,536 pixels to reduce artifacts on NAIP mosaics. The model is trained end to end with L1 loss on 4-band NAIP input and lidar CHM labels from USGS 3DEP, with an in-memory buffer that stitches large-area inferences without checkerboard artifacts.

What would settle it

Retrain or evaluate with strict spatial separation: hold out entire HUC12 watersheds, or enforce a buffer of several kilometers between every training and test tile, then recompute the median MAE. If the VibrantVS advantage over Meta, LANDFIRE, and ETH shrinks toward zero under spatial holdout, the claimed generalizability is largely an artifact of spatial autocorrelation; if the 2.71-meter gap persists, the claim is confirmed.

Watch

Extended reading notes

Core claim

The central claim is that a single multi-task vision transformer, trained on roughly 195,000 half-kilometer tiles of four-band NAIP imagery paired with lidar-derived canopy height models, generalizes across 24 EPA Level 3 ecoregions in the western United States better than three peer-reviewed benchmark CHMs. On the held-out test set, VibrantVS achieves lower median MAE (2.71 m), better Block-R2 (0.69 versus negative values for the baselines), and lower edge error (0.08) than Meta, LANDFIRE, and ETH; it also has less bias than Meta and ETH in the 2-25 meter height range that dominates the lidar pixels. The paper attributes the improvement mainly to the larger and more ecologically diverse training set and to architectural choices in the transformer, while acknowledging that trees above 50 meters remain underestimated.

Load-bearing premise

The load-bearing premise is that the 67,227 test tiles are statistically independent of the 195,416 training tiles even though the random split is by 1-kilometer tile with no stated minimum distance or watershed holdout; if neighboring tiles share lidar flights, NAIP mosaics, or stand conditions, the reported errors are optimistic.

Editorial extensions

If this is right

  • VibrantVS can be rerun on new NAIP acquisitions to refresh canopy height maps on a three-year-or-less cycle across the western US, matching post-disturbance monitoring needs.
  • Forest managers can use the 0.5-meter CHM as input to individual-tree segmentation, yielding trees per acre, basal area, and canopy base height without new lidar.
  • The high-resolution height surfaces can feed wildfire spread models with fuel discontinuities at scales coarse products miss.
  • The advantage is concentrated in low and mid-height vegetation; all models underestimate very tall trees, so tall-structure applications still need dedicated retraining.

Reading between the lines

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

  • If the random tile split has leaked spatial autocorrelation, real-world accuracy on unvisited landscapes will be lower than 2.71 m; a watershed-level holdout would quantify the gap.
  • Because the training window is 2014-2021, the three-year update claim presumes NAIP-to-height relationships stay stable; the paper does not test forward transfer to post-2022 imagery.
  • The canopy cover head is trained but not evaluated here; if it performs comparably, the same 4-band input could produce a multi-layer fuels product for fire modeling in one pass.
  • The method could extend to the central and eastern US and to shrublands, but the current training sample is western forests, so claims about those regions are untested.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper introduces VibrantVS, a multi-task vision transformer (SWINv2 encoder with DPT decoder, a metric-bin height head, and a convolutional canopy-cover head) that estimates canopy height at 0.5 m resolution from 4-band NAIP imagery over the western United States. The authors compare VibrantVS against three existing products (Meta, LANDFIRE, ETH) using withheld 3DEP lidar tiles and report a median tile-level MAE of 2.71 m versus 4.83 m, 5.96 m, and 7.05 m, respectively. Evaluations are aggregated by EPA Level 3 ecoregion and by lidar height bin, with additional metrics including Block-R2, mean error, RMSE, MAPE, and an edge-error metric. The paper also claims a three-year-or-less update cadence, enabled by NAIP's revisit cycle.

Significance. If the reported accuracy advantage holds under a statistically valid evaluation, the result is practically significant: a 0.5 m CHM with broad western-US coverage at a three-year cadence would be valuable for wildfire risk assessment, forest management, and downstream structural products. The manuscript has real strengths: a large training/evaluation sample (262,643 tiles from 24 ecoregions), evaluation on held-out tiles, multiple metrics, aggregated ecoregion and height-bin analyses, a temporal-mismatch sensitivity check in Fig. A3, and honest discussion of known failures (tall-tree underestimation, low-bin overestimation, class imbalance). These elements make the empirical claim testable. However, the central comparison is only as convincing as the independence of the test set, and the current split protocol does not establish that independence. The paper is therefore not yet ready for acceptance, but the main defects are fixable through a spatial hold-out analysis and targeted sensitivity checks.

major comments (3)
  1. [Section 2, Figure 2, and Table 2] The train/test split is described as an approximately 85/15 'randomly sampled' tile split by ecoregion, with no reported minimum distance, buffer, or spatial clustering. Because 3DEP lidar is delivered in contiguous work units and NAIP is acquired as state-level mosaics, adjacent tiles share flight lines, illumination, and atmospheric conditions, and forest structure is strongly spatially autocorrelated. The withheld test tiles are therefore not conditionally independent of the training tiles. This matters because only VibrantVS is fit to the training data; leakage would selectively lower its test error relative to the fixed baseline products, inflating the 2.71 m vs. 4.83-7.05 m MAE gap that is the paper's central claim. Please add a spatially blocked evaluation: for example, exclude all test tiles within a buffer of several kilometers of any training tile, or hold out entire HUC12 watersheds, and report the metrics for the spatially separated subset.
  2. [Section 3.2, Table 2, and Section 5 (height-bin discussion)] The decision to mask out all lidar pixels below 2 m before computing every error metric is not a neutral preprocessing choice. The paper itself notes in the Discussion that VibrantVS overestimates values where lidar heights are close to zero (the lowest height bin), and the mask removes exactly those pixels. This can preferentially reduce VibrantVS's errors relative to models with different low-height behavior, and the magnitude of this effect is not quantified. Please report sensitivity analyses with no height mask or with alternative thresholds, and include both masked and unmasked versions of the headline metrics in Table 2.
  3. [Section 2.3 and Fig. A3] The baseline products do not share a single acquisition epoch with the lidar labels: Meta and ETH represent 2020, LANDFIRE includes 2016 layers, and the lidar labels span 2015-2021. Forest disturbances and regrowth over five to six years can be substantial, so analyzing all years together may conflate model error with temporal mismatch. The Appendix Fig. A3 subset for 2019-2021 lidar is reassuring, but it is not in the main text and its construction is not described in detail. Please report the main error metrics (or at minimum the median MAE of Table 2) stratified by lidar year or restricted to 2019-2021 lidar in the main results.
minor comments (4)
  1. [Section 2 vs. Table 1] The tile footprint is stated as '0.5 x 0.5 km2' in Section 2 but Table 1 says 'Sample tiles are 1 x 1 km2'; please reconcile this discrepancy because it affects the reported sampling area and the interpretation of spatial autocorrelation.
  2. [Section 3.1] The training recipe is described only qualitatively (e.g., '8-bit Adam', 'L1 loss', '2,688 hours on an A100'); please provide concrete hyperparameters (learning rate, batch size, number of epochs, context-window overlap, loss weights) or a public code/model release to make the method reproducible.
  3. [Figure 6b] The caption refers to 'all validation and test tiles', but no separate validation set is defined anywhere in the text; please clarify what is plotted.
  4. [Section 3.2] The sentence beginning 'Weapplieda numberoferrormetricstoallbaselinemodels...' appears to have lost its spaces; please fix the typo.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central result is an empirical held-out comparison against lidar, not a derivation from the fitted parameters.

full rationale

VibrantVS is trained end-to-end on NAIP imagery paired with 3DEP lidar CHM labels, and the headline result in Section 4 / Table 2 is its error computed on 67,227 withheld test tiles (Section 2, Figure 2). No fitted constant is fed back into the evaluation to produce the 2.71 m MAE; the test tiles are described as withheld from model training, and the comparison to Meta, LANDFIRE, and ETH is made against the same held-out lidar labels. The model's loss (L1 on height and cover) is irrelevant to the test-time computation, so the 'prediction' is not forced by construction. The paper's acknowledged limitations—random 85/15 tile split without an explicit spatial buffer, temporal mismatch between baseline products and lidar acquisition years, and under-representation of very tall trees—are real threats to external validity and fairness of comparison, but they are statistical or scope concerns, not circularity: the claimed outcome is not an input to the derivation. There are no load-bearing self-citations or uniqueness theorems invoked to forbid alternatives; architectural choices cite external sources, and the benchmark models are independent published products. Accordingly, no circular step can be exhibited, and the appropriate score is 0.

Assumptions & free parameters 3 free parameters · 7 assumptions · 0 invented entities

The central claim is an empirical model-performance result, not a derivation, so the ledger contains no fitted physical constants or invented entities. It does depend on several domain assumptions: lidar labels are treated as ground truth, NAIP-to-lidar temporal matching is treated as sufficient to avoid disturbance-driven label error, the tile-level split is treated as statistically independent, and the fixed-2020 baseline products are treated as comparable to lidar from other years after resampling. The most consequential hand-chosen settings are the 2m height mask and the one-year temporal matching window.

free parameters (3)
  • Lidar height mask threshold = 2 m
    Section 3.2: all error metrics are computed only on pixels with lidar CHM at or above 2m. This hand-set threshold removes low vegetation and affects every reported MAE and RMSE value.
  • NAIP-to-lidar temporal matching window = 1 year
    Section 2.1.1: NAIP tiles more than one year from the coincident lidar acquisition are excluded. The choice controls label noise but also gives VibrantVS temporally matched inputs that fixed-year baselines do not receive.
  • Metric-bin count in regression head = 64 bins
    Section 3.1: the metric-bin module head discretizes each pixel's height distribution into 64 bins; bin count is a hand-chosen architectural hyperparameter that influences the predicted height distribution.
assumptions (7)
  • domain assumption 3DEP aerial lidar CHMs are an accurate ground truth for canopy height.
    Section 2.2.2; all accuracy and precision claims are computed against these CHMs, but lidar point density, DTM/DSM processing, and normalization noise are not independently validated.
  • domain assumption NAIP imagery acquired within one year of lidar represents the same canopy state.
    Section 2.1.1; assumes growth, disturbance, and phenology differences over less than one year are negligible at 0.5m resolution and do not bias labels.
  • domain assumption Tile-level random train/test split yields independent test samples.
    Section 2 and Figure 2; no spatial buffer or watershed-level withholding is described, so neighboring tiles can share lidar flights, NAIP mosaics, and stand conditions.
  • domain assumption Fixed-2020 baseline products can be fairly compared to lidar spanning 2014-2021.
    Section 2.3 and Section 5; Meta, LANDFIRE, and ETH represent nominal 2020 states while test lidar includes 2015-2021 acquisitions; the paper offers only a partial sensitivity check.
  • domain assumption Nearest-neighbor resampling of coarse baseline CHMs to 0.5m preserves comparability.
    Section 2.3 and Figure 12; block artifacts from upsampling 10m/30m products are acknowledged, and the paper checks that aggregation to 10m/30m preserves the MAE trend.
  • ad hoc to paper The described architecture and training recipe perform as reported.
    Section 3.1; no external verification of the SWINv2/DPT/metric-bin implementation, training stability, or inference pipeline is provided beyond the authors' own runs.
  • domain assumption Sobel edge error is a meaningful proxy for spatial fidelity of canopy height.
    Appendix A.2; the edge error metric is borrowed from Tolan et al. (2024) and assumes edge statistics of resampled CHMs capture fine-scale structural accuracy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VibrantVS: A high-resolution multi-task transformer for forest canopy height estimation." pith.science (2026). https://pith.science/paper/RHXFQMRT

@misc{pith2026241210351,
  author       = {Pith},
  title        = {Pith review of: VibrantVS: A high-resolution multi-task transformer for forest canopy height estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RHXFQMRT}},
  note         = {Machine review of arXiv:2412.10351}
}
read the original abstract

This paper explores the application of a novel multi-task vision transformer (ViT) model for the estimation of canopy height models (CHMs) using 4-band National Agriculture Imagery Program (NAIP) imagery across the western United States. We compare the effectiveness of this model in terms of accuracy and precision aggregated across ecoregions and class heights versus three other benchmark peer-reviewed models. Key findings suggest that, while other benchmark models can provide high precision in localized areas, the VibrantVS model has substantial advantages across a broad reach of ecoregions in the western United States with higher accuracy, higher precision, the ability to generate updated inference at a cadence of three years or less, and high spatial resolution. The VibrantVS model provides significant value for ecological monitoring and land management decisions, including for wildfire mitigation.

Figures

Figures reproduced from arXiv: 2412.10351 by the authors.

Figure 1
Figure 1. Sampling of tiles within Hydrologic Units 12 (HUC12) watersheds of the western United [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Sample tile counts within each of the randomly sampled train and test (approx. 85% to [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Map of EPA L3 Ecoregions into which our test tiles were aggregated to evaluate baseline [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: VibrantVS multi-task vision transformer architecture with 4 band NAIP input to predict [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Box and whisker plots of test tile level MAE by model and ecoregion sorted by lowest to [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: (a) Results of MAE by ecoregion within all test tiles. (b) Map of MAE by ecoregions [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: 2D histogram plot of VibrantVS canopy heights vs lidar-derived canopy heights across [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Box plots of mean error in different height bins and histogram of sample counts per height [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Comparison of model performance by ecoregion for every error metric. [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Qualitative evaluation of the various baseline models compared to the original lidar data. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Comparison of model performance at different height class bins. [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: Comparison of model performance at varying resolutions with the target lidar data [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 57 canonical work pages

  1. [1]

    J. T. Abatzoglou and A. P. Williams. Impact of anthropogenic climate change on wildfire across westernusforests. Proceedings of the National Academy of sciences, 113(42):11770–11775, 2016

    work page 2016

  2. [2]

    Ainslie, J

    J. Ainslie, J. Lee-Thorp, M. de Jong, Y. Zemlyanskiy, F. Lebrón, and S. Sanghai. Gqa: Train- ing generalized multi-query transformer models from multi-head checkpoints.arXiv preprint arXiv:2305.13245, 2023

    arXiv 2023

  3. [3]

    Andersen, R

    H.-E. Andersen, R. J. McGaughey, and S. E. Reutebuch. Estimating forest canopy fuel param- eters using lidar data.Remote sensing of Environment, 94(4):441–449, 2005

    work page 2005

  4. [4]

    Naip: National agriculture imagery program, 2024

    AWS Open Data Registry. Naip: National agriculture imagery program, 2024. URLhttps: //registry.opendata.aws/naip/. Accessed: 2024-09-26

    work page 2024

  5. [5]

    Beland, G

    M. Beland, G. Parker, B. Sparrow, D. Harding, L. Chasmer, S. Phinn, A. Antonarakis, and A. Strahler. On promoting the use of lidar systems in forest ecosystem research.Forest Ecology and Management, 450:117484, 2019

    work page 2019

  6. [6]

    S. F. Bhat, R. Birkl, D. Wofk, P. Wonka, and M. Müller. Zoedepth: Zero-shot transfer by combining relative and metric depth.arXiv preprint arXiv:2302.12288, 2023

    arXiv 2023

  7. [7]

    T. Dao. Flashattention-2: Faster attention with better parallelism and work partitioning.arXiv preprint arXiv:2307.08691, 2023

    arXiv 2023

  8. [8]

    Darcet, M

    T. Darcet, M. Oquab, J. Mairal, and P. Bojanowski. Vision transformers need registers.arXiv preprint arXiv:2309.16588, 2023

    arXiv 2023

Show all 64 references
  1. [9]

    Defazio, H

    A. Defazio, H. Mehta, K. Mishchenko, A. Khaled, A. Cutkosky, et al. The road less scheduled. arXiv preprint arXiv:2405.15682, 2024

    2024 arXiv

  2. [10]

    Dettmers, M

    T. Dettmers, M. Lewis, S. Shleifer, and L. Zettlemoyer. 8-bit optimizers via block-wise quan- tization. CoRR, abs/2110.02861, 2021. URLhttps://arxiv.org/abs/2110.02861

    2021 arXiv

  3. [11]

    Drusch, U

    M. Drusch, U. Del Bello, S. Carlier, O. Colin, V. Fernandez, F. Gascon, B. Hoersch, C. Isola, P. Laberinti, P. Martimort, et al. Sentinel-2: Esa’s optical high-resolution mission for gmes operational services. Remote sensing of Environment, 120:25–36, 2012

    2012

  4. [12]

    Dubayah, J

    R. Dubayah, J. Armston, S. P. Healey, J. M. Bruening, P. L. Patterson, J. R. Kellner, L. Dun- canson, S. Saarela, G. Ståhl, Z. Yang, et al. Gedi launches a new era of biomass inference from space. Environmental Research Letters, 17(9):095001, 2022. 16

    2022

  5. [13]

    R. O. Dubayah and J. B. Drake. Lidar remote sensing for forestry.Journal of forestry, 98(6): 44–46, 2000

    2000

  6. [14]

    National Agriculture Im- agery Program (NAIP), 2017

    Earth Resources Observation And Science (EROS) Center. National Agriculture Im- agery Program (NAIP), 2017. URL https://www.usgs.gov/centers/eros/science/ usgs-eros-archive-aerial-photography-national-agriculture-imagery-program-naip? qt-science_center_objects=0#qt-science_cen...

    2017

  7. [15]

    Global canopy height map for the year 2020 derived from sentinel-2 and gedi (version 1), 2023

    ETH Zurich. Global canopy height map for the year 2020 derived from sentinel-2 and gedi (version 1), 2023. URLhttps://doi.org/10.3929/ethz-b-000609802. ETH Zurich Research Collection

    2020 doi

  8. [16]

    Ferraz, F

    A. Ferraz, F. Bretar, S. Jacquemoud, G. Gonçalves, L. Pereira, M. Tomé, and P. Soares. 3d mapping of a multi-layered mediterranean forest using als data.Remote Sensing of Environ- ment, 121:210–223, 2012. doi: 10.1016/j.rse.2012.01.020

    2012 doi

  9. [17]

    S. E. Franklin, M. B. Lavigne, M. A. Wulder, and G. B. Stenhouse. Change detection and landscape structure mapping using remote sensing. The Forestry Chronicle, 78(5):618–625, 2002

    2002

  10. [18]

    P. F. Hessburg, J. K. Agee, and J. F. Franklin. Dry forests and wildland fires of the inland northwest usa: contrasting the landscape ecology of the pre-settlement and modern eras.Forest Ecology and management, 211(1-2):117–139, 2005

    2005

  11. [19]

    K. M. Hoffman, E. L. Davis, S. B. Wickham, K. Schang, A. Johnson, T. Larking, P. N. Lauriault, N. Quynh Le, E. Swerdfager, and A. J. Trant. Conservation of earth’s biodiversity is embedded in indigenous fire stewardship.Proceedings of the National Academy of Sciences, 118(32):...

    2021

  12. [20]

    S. M. A. Jeronimo, V. R. Kane, D. J. Churchill, R. J. McGaughey, and J. F. Franklin. Applying lidar individual tree detection to management of structurally diverse forest landscapes.Journal of Forestry, 116(4):336–346, June 2018. ISSN 1938-3746. doi: 10.1093/jofore/fvy023. URL...

    2018 doi

  13. [21]

    H. A. Kramer, B. M. Collins, M. Kelly, and S. L. Stephens. Quantifying ladder fuels: A new approach using lidar.Forests, 5(6):1432–1453, 2014

    2014

  14. [22]

    M. L. Lane, T. W. Swetnam, and T. Oki. Fuel models and landfire: Standardizing inputs for fire simulation. International Journal of Wildland Fire, 20(7):845–856, 2011. doi: 10.1071/ WF10158. URL https://doi.org/10.1071/WF10158

    2011 doi

  15. [23]

    N. Lang, W. Jetz, K. Schindler, and J. D. Wegner. A high-resolution canopy height model of the earth. Nature Ecology & Evolution, 7(11):1778–1789, 2023

    2023

  16. [24]

    M. A. Lefsky, W. B. Cohen, G. G. Parker, and D. J. Harding. Lidar remote sensing for ecosystem studies. BioScience, 52(1):19–30, 2002

    2002

  17. [25]

    S. Li, M. Brandt, R. Fensholt, A. Kariryaa, C. Igel, F. Gieseke, T. Nord-Larsen, S. Oehmcke, A. H. Carlsen, S. Junttila, et al. Deep learning enables image-based tree counting, crown segmentation, and height prediction at national scale.PNAS nexus, 2(4):pgad076, 2023. 17

    2023

  18. [26]

    Z. Liu, H. Hu, Y. Lin, Z. Yao, Z. Xie, Y. Wei, J. Ning, Y. Cao, Z. Zhang, L. Dong, et al. Swin transformer v2: Scaling up capacity and resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12009–12019, 2022

    2022

  19. [27]

    Maxar vivid2 mosaic imagery data

    Maxar Technologies. Maxar vivid2 mosaic imagery data. https://developers.maxar.com/ docs/streaming-basemap, 2024. Accessed August 8, 2024

    2024

  20. [28]

    J. E. Means, S. A. Acker, D. J. Harding, J. B. Blair, M. A. Lefsky, W. B. Cohen, M. E. Harmon, and W. A. McKee. Use of large-footprint scanning airborne lidar to estimate forest stand characteristics in the western cascades of oregon.Remote sensing of environment, 67(3): 298–308, 1999

    1999

  21. [29]

    Highresolutioncanopyheightmaps(chm)

    MetaandWorldResourcesInstitute(WRI). Highresolutioncanopyheightmaps(chm). https: //registry.opendata.aws/dataforgood-fb-forests, 2024

    2024

  22. [30]

    M. A. Moritz, E. Batllori, R. A. Bradstock, A. M. Gill, J. Handmer, P. F. Hessburg, J. Leonard, S. McCaffrey, D. C. Odion, T. Schoennagel, et al. Learning to coexist with wildfire.Nature, 515(7525):58–66, 2014

    2014

  23. [31]

    J. H. Olszewski and J. D. Bailey. Lidar as a tool for assessing change in vertical fuel continuity following restoration. Forests, 13(4):503, 2022

    2022

  24. [32]

    J. M. Omernik and G. E. Griffith. Ecoregions of the conterminous united states: evolution of a hierarchical spatial framework.Environmental management, 54:1249–1266, 2014

    2014

  25. [33]

    S. C. Popescu and R. H. Wynne. Seeing the trees in the forest.Photogrammetric Engineering & Remote Sensing, 70(5):589–604, 2004

    2004

  26. [34]

    Ranftl, K

    R. Ranftl, K. Lasinger, D. Hafner, K. Schindler, and V. Koltun. Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer.IEEE transactions on pattern analysis and machine intelligence, 44(3):1623–1637, 2020

    2020

  27. [35]

    Modelingairbornelaserscanning data for the spatial generation of critical forest parameters in fire behavior modeling.Remote Sensing of Environment, 86(2):177–186, 2003

    D.Riaño, E.Meier, B.Allgöwer, E.Chuvieco, andS.L.Ustin. Modelingairbornelaserscanning data for the spatial generation of critical forest parameters in fire behavior modeling.Remote Sensing of Environment, 86(2):177–186, 2003

    2003

  28. [36]

    T. Ritu, H. James, W. L. ad Reinke Karin, and J. Simon. Effect of fuel spatial resolution on predictive wildfire models.International Journal of Wildland Fire, 30:776–789, 2021. doi: 10.1071/WF20192. URL https://doi.org/10.1071/WF20192

    2021 doi

  29. [37]

    M. G. Rollins. Landfire: a nationally consistent vegetation, wildland fire, and fuel assessment. International Journal of Wildland Fire, 18(3):235–249, 2009

    2009

  30. [38]

    M. G. Rollins, R. E. Keane, Z. Zhu, and J. P. Menakis. An overview of the landfire prototype project. In: Rollins, Matthew G.; Frame, Christine K., tech. eds. 2006. The LANDFIRE Prototype Project: nationally consistent and locally relevant geospatial data for wildland fire man...

    2006

  31. [39]

    Roussel, D

    J.-R. Roussel, D. Auty, N. C. Coops, P. Tompalski, T. R. Goodbody, A. S. Meador, J.-F. Bourdon, F. de Boissieu, and A. Achim. lidr: An r package for analysis of airborne laser scanning (als) data. Remote Sensing of Environment, 251:112061, 2020. ISSN 0034-4257. doi: 10.1016/j....

    2020

  32. [40]

    N. Shazeer. Glu variants improve transformer.arXiv preprint arXiv:2002.05202, 2020

    2002 arXiv

  33. [41]

    C. A. Silva, A. T. Hudak, L. A. Vierling, E. L. Loudermilk, J. J. O’Brien, J. K. Hiers, S. B. Jack, C. Gonzalez-Benecke, H. Lee, M. J. Falkowski, et al. Imputation of individual longleaf pine (pinus palustris mill.) tree attributes from field and lidar data.Canadian journal of...

    2016

  34. [42]

    S. L. Steenburgh and K. R. Schmid. Landfire program: Fuel data for fire and resource management planning. Fire Ecology, 8(2):89–96, 2012. doi: 10.4996/fireecology08189. URL https://doi.org/10.4996/fireecology08189

    2012 doi

  35. [43]

    C. S. Stevens-Rumann, K. B. Kemp, P. E. Higuera, B. J. Harvey, M. T. Rother, D. C. Donato, P. Morgan, and T. T. Veblen. Evidence for declining forest resilience to wildfires under climate change. Ecology letters, 21(2):243–252, 2018

    2018

  36. [44]

    L. J. Sugarbaker, E. W. Constance, H. K. Heidemann, A. L. Jason, V. Lukas, D. L. Saghy, and J. M. Stoker. The 3d elevation program initiative: a call for action. Technical report, U.S. Geological Survey, 2014

    2014

  37. [45]

    R.-L. Team. Lidar data analysis with r: Canopy height models, 2024. URLhttps://r-lidar. github.io/lidRbook/dsm.html?utm_source=chatgpt.com. Accessed: 2024-12-03

    2024

  38. [46]

    K. M. Thibault, C. M. Laney, K. M. Yule, N. M. Franz, and P. M. Mabee. The us national ecological observatory network and the global biodiversity framework: national research infras- tructure with a global reach.Journal of Ecology and Environment, 47, 2023

    2023

  39. [47]

    Tolan, H.-I

    J. Tolan, H.-I. Yang, B. Nosarzewski, G. Couairon, H. V. Vo, J. Brandt, J. Spore, S. Ma- jumdar, D. Haziza, J. Vamaraju, et al. Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on aerial lidar. ...

    2024

  40. [48]

    Forest Service

    U.S. Forest Service. Initial landscape investments to support the national wild- fire crisis strategy, 2022. URL https://www.fs.usda.gov/sites/default/files/ WCS-Initial-Landscape-Investments.pdf. U.S. Department of Agriculture

    2022

  41. [49]

    Geological Survey

    U.S. Geological Survey. Landfire fuels - forest canopy height, 2022. URL https://www. landfire.gov/fuel/ch. Accessed: 2024-04-27

    2022

  42. [50]

    Geological Survey

    U.S. Geological Survey. WESM Data Dictionary. https://www.usgs.gov/ ngp-standards-and-specifications/wesm-data-dictionary/ , 2024. [Online; accessed 7-August-2024]

    2024

  43. [51]

    Vihervaara, M.Maltamo, C.A

    R.Valbuena, B.O’Connor, F.Zellweger, W.Simonson, P. Vihervaara, M.Maltamo, C.A. Silva, D. R. A. d. Almeida, F. Danks, F. Morsdorf, et al. Standardizing ecosystem morphological traits from 3d information sources.Trends in Ecology & Evolution, 35(8):656–667, 2020. 19

    2020

  44. [52]

    Van Leeuwen and M

    M. Van Leeuwen and M. Nieuwenhuis. Retrieval of forest structural parameters using lidar remote sensing. European Journal of Forest Research, 129:749–770, 2010

    2010

  45. [53]

    K. T. Vierling, L. A. Vierling, W. A. Gould, S. Martinuzzi, and R. M. Clawges. Lidar: shedding new light on habitat characterization and modeling.Frontiers in Ecology and the Environment, 6(2):90–98, 2008

    2008

  46. [54]

    F. H. Wagner, S. Roberts, A. L. Ritz, G. Carter, R. Dalagnol, S. Favrichon, M. C. Hirye, M. Brandt, P. Ciais, and S. Saatchi. Sub-meter tree height mapping of california us- ing aerial images and lidar-informed u-net model. Remote Sensing of Environment, 305: 114099, 2024. ISS...

    2024

  47. [55]

    K. F. Watts, D. W. Koehler, D. Todd, and N. Gillett. Integration of landfire data into fire modeling: Enhancing accuracy and consistency. Journal of Applied Meteorology and Climatology, 56(9):2017JA024010, 2017. doi: 10.1175/JAMC-D-16-0253.1. URL https: //doi.org/10.1175/JAMC-...

    2017 doi

  48. [56]

    A. L. Westerling. Increasing western US forest wildfire activity: sensitivity to changes in the timing of spring. Philosophical Transactions of the Royal Society B: Biological Sciences, 371(1696):20150178, June 2016. doi: 10.1098/rstb.2015.0178. URL https:// royalsocietypublis...

    2016

  49. [57]

    J. C. White, N. C. Coops, M. A. Wulder, M. Vastaranta, T. Hilker, and P. Tompalski. Remote sensing technologies for enhancing forest inventories: A review.Canadian Journal of Remote Sensing, 42(5):619–641, 2016

    2016

  50. [58]

    M. A. Wulder, C. W. Bater, N. C. Coops, T. Hilker, and J. C. White. The role of lidar in sustainable forest management.The forestry chronicle, 84(6):807–826, 2008

    2008

  51. [59]

    Zhang and R

    B. Zhang and R. Sennrich. Root mean square layer normalization.Advances in Neural Infor- mation Processing Systems, 32, 2019. 20 A Appendix A.1 Additional Tables and Figures Figure A1: Lidar sample pixel-wise height distributions within each of the EPA Level 3 ecoregions. Red ...

    2019

  52. [60]

    Mean Absolute Error (MAE) MAE = 1 n nX i=1 |yi − ˆyi|

  53. [61]

    21 Model Spatial Reso- lution Spatial Ex- tent T empor

    Block-R2 R2 block = 1 − PB b=1 P i∈b(yi − ˆyi)2 PB b=1 P i∈b(yi − ¯yb)2 Where B is the number of blocks,yb is the ground truth value in blockb, ˆyb the model estimate for block b, and ¯yb is the mean of the ground-truth values in blockb. 21 Model Spatial Reso- lution Spatial E...

    2014

  54. [62]

    Root Mean Square Error (RMSE) RMSE = vuut 1 n nX i=1 (yi − ˆyi)2

  55. [63]

    Mean Error (ME) ME = 1 n nX i=1 (ˆyi − yi)

  56. [64]

    Edge Error Metric (EE) EE = 1 n nX i=1 |E(ˆyi) − E(yi)| Where E(·) represents Sobel edge detection operation on the data (also compare Tolan et al. (2024)). 23 Figure A3: Comparison of model performance at different year groups of lidar acquisition determine temporal mismatch ...

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.