Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Contrasting local and global modeling with machine learning and satellite data: A case study estimating tree canopy height in African savannas

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read In the Karingani Game Reserve, a small model trained only on local LiDAR labels and Sentinel-2 imagery predicts tree canopy height more accurately than four published global maps and more accurately than globally pretrained models…

desk verdict Solid case study showing local FCN beats global TCH maps in Mozambican savanna; the fine-tuning comparison is under-supported because only the last 3 layers of XceptionS2 were tuned. read the letter →

arxiv 2411.14354 v1 pith:DUEXERRM submitted 2024-11-21 cs.LG cs.AIcs.CV

Esther Rolf , Lucia Gordon , Milind Tambe , Andrew Davies This is my paper · ORCID
classification cs.LGcs.AIcs.CV
keywords treecanopyheightsatellitemachinelearninglocalversusglobalmodelsfine-tuningSentinel-2UAV-LiDARAfricansavannaspatialgeneralization
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 asks whether global satellite-machine-learning models help build accurate local maps, using tree canopy height in Karingani Game Reserve, Mozambique. It establishes that a small five-layer fully convolutional network trained only on locally collected UAV-LiDAR labels and Sentinel-2 imagery outperforms four published global canopy height maps, cutting average RMSE from 2.43 to 4.51 m down to 1.64 m. It further establishes that fine-tuning globally pretrained models on local data does not close the gap: the small local-only model beats the best fine-tuned XceptionS2 model's 1.98 m RMSE. The paper argues that local high-fidelity labels are the decisive resource for local mapping, and that some design choices made for global performance, such as location embeddings, do not help and may even hurt local extrapolation.

What carries the argument

The load-bearing object is a five-layer fully convolutional network with 128 filters per layer, roughly 604,417 parameters, and an 11-by-11-pixel receptive field, trained from scratch on 12-band Sentinel-2 imagery with UAV-LiDAR-derived tree canopy height labels as supervision. The experimental design also depends on site-stratified train, validation, and test splits across 24 surveyed sites, which force every reported error metric to reflect prediction at sites never seen during training or validation.

What would settle it

Collect new UAV-LiDAR data over sites outside the existing 24 in Karingani, or in a comparable savanna reserve, and compare the locally trained fully convolutional network's predictions against the Pauls et al. map on those sites; if the gap in RMSE largely disappears or reverses, the claim that local training is decisive for local accuracy would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that, for mapping tree canopy height in this savanna region, local modeling and global modeling are currently distinct goals: a model trained from scratch on local data outperforms both published global maps and globally pretrained models fine-tuned on the same local data. The paper reaches this conclusion through site-disjoint train, validation, and test splits over 24 UAV-LiDAR survey sites, so every reported result is out-of-sample with respect to whole sites. It also finds that the best global map in Karingani is not the newest one, that the choice of model architecture, input spectral bands, and amount of training data each affect performance by similar magnitudes, and that while local data are essential for evaluation, straightforward fine-tuning of global models does not substitute for training a small local model.

Load-bearing premise

The 24 LiDAR-surveyed sites are representative enough of the whole Karingani Game Reserve that the held-out test sites measure how the models will behave in the roughly two-thirds of the reserve without flight data.

Editorial extensions

If this is right

  • Locally collected high-fidelity labels, not global pretraining, are the decisive resource for accurate canopy height maps in this savanna; improving global models will not automatically improve local mapping.
  • Fine-tuning a globally pretrained model on local data provides little or no benefit over training a small local model from scratch, so transfer-learning pipelines should be benchmarked against local-only training before being adopted.
  • Global map quality rankings do not transfer to local quality rankings, since the older 30 m GLAD map can beat newer global maps in a specific region; local validation data remain necessary even when global maps look good globally.
  • Design choices that help global performance, such as absolute location embeddings, can be neutral or harmful for local spatial extrapolation, so local models need their own design validation with site-disjoint splits.

Reading between the lines

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

  • Inference: if the local-data advantage holds beyond this reserve, the marginal value of a global model for local applications may be small, and collection of local LiDAR or equivalent ground-referenced data should be prioritized over further global-pretraining investment.
  • Inference: the results suggest a testable scaling law—as the local training footprint grows past some radius, the local-only advantage over global maps should shrink, and mapping that crossover would tell practitioners when global models become cost-effective.
  • Inference: the failure of absolute location embeddings hints that spatial extrapolation may require relative or learned positional encodings rather than absolute coordinates, a hypothesis the paper does not directly test.
  • Inference: replicating the study in a different biome, such as a dense tropical forest or a boreal landscape, would separate savanna-specific effects from a general local-versus-global discrepancy in geospatial machine learning.
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 / 5 minor

Summary. This paper uses a case study of tree canopy height (TCH) mapping in Karingani Game Reserve, Mozambique, to contrast local and global training paradigms in satellite machine learning. The authors compare (i) four published global TCH maps evaluated out of the box, (ii) locally trained models (a small FCN, a U-Net, an XceptionS2, and a random forest) trained from scratch on local UAV-LiDAR labels, and (iii) globally pretrained models (XceptionS2 with supervised global TCH weights and U-Net with SSL4EO self-supervised weights) fine-tuned on local data. Using spatially disjoint site-stratified splits, they find that a small locally trained FCN achieves the best RMSE (1.64 m), outperforming the best global map (Pauls et al., 2.43 m) and the best fine-tuned global model (XceptionS2 with global initialization, 1.98 m). They also analyze the relative influence of training data quantity, spectral bands, and model architecture, and discuss points of conflict and synergy between local and global modeling.

Significance. The evaluation design is a strength: spatially disjoint site-stratified splits (four splits), ten random seeds per model, comparison against four published global maps, and error stratification by height, geology, and distance to rivers provide a rigorous benchmark. The downstream aboveground biomass calibration analysis (Section 5.1.3) is a useful additional check of practical utility. If the main finding survives, it provides concrete evidence that local high-fidelity labels can outweigh global pretraining for this task, with implications for data collection and model design in geospatial ML. The paper also delivers falsifiable predictions and transparent metrics.

major comments (3)
  1. [§4.2.2, Table 1, Table 2] The claim in the abstract and Section 5.2 that locally trained models 'even outperform globally pretrained models that we fine-tune using local data' is not fully supported because the only supervised global TCH model with available weights (XceptionS2) is fine-tuned only in its last 1–3 layers (Section 4.2.2; Table 2), rather than fully fine-tuned. The RMSE gap between the best fine-tuned XceptionS2 (1.98 m) and the local FCN (1.64 m) is 0.34 m, which could close if earlier layers were allowed to adapt, especially since the global model's early layers and positional embeddings were optimized on global data. Please add a full fine-tuning condition (or at least a condition tuning a substantially larger portion of the network) or restrict the claim to the specific fine-tuning protocols used.
  2. [§4.2.2, Abstract, §5.2] The fine-tuning comparison is also limited by model availability: among the four recent global maps, only the ETH/XceptionS2 model weights are publicly available; the Pauls et al. weights are not released, and the Meta model would require proprietary imagery for fine-tuning. Thus the conclusion that 'globally pretrained models are not necessarily the right starting point' (Section 6.3) is based on a single supervised global TCH model and one self-supervised U-Net. The paper should state this limitation more prominently and frame the conclusion as specific to the models and protocols tested.
  3. [§3.3, §6.1] The evaluation is confined to the 24 UAV-LiDAR sites; no independent data are available for the roughly two-thirds of Karingani not covered by flight data. The site-stratified splits simulate deployment to new sites, but they cannot validate the model on truly unseen portions of the reserve if those areas differ systematically in vegetation or geology from the 24 sites. The paper acknowledges coverage gaps but the abstract's unqualified statement that local models 'outperform published global TCH maps' in the study region should be scoped to the evaluated sites or supplemented with an explicit acknowledgment of this limitation.
minor comments (5)
  1. [§5.1.1] 'RSME' should be 'RMSE'; also the reported local FCN RMSE is given as 1.65 in the bullet list but 1.64 in Table 1 – please align.
  2. [§6.1] 'Karangani' is spelled 'Karingani' elsewhere; fix the typo.
  3. [Table 2 caption, Table 3] 'XcpetionS2' should be 'XceptionS2'.
  4. [§4.2.2] The choice to stop at 'last 3 layers' after observing small differences between 2 and 3 layers is not a substitute for a full fine-tuning sweep; please add a sentence justifying this choice (e.g., computational budget) or report the full sweep.
  5. [Table 1] The paper does not report statistical significance tests comparing the local FCN RMSE to the best fine-tuned model; given the small standard deviations across seeds (0.01 vs 0.03 m), a paired test across splits would strengthen the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the local-vs-global TCH comparison rests on external global maps and independent local LiDAR labels.

full rationale

The paper's central claim—that a small local-only FCN (RMSE 1.64 m) outperforms published global TCH maps (best 2.43 m) and locally fine-tuned global models (best 1.98 m)—is established by comparing predictions against held-out UAV-LiDAR labels at spatially disjoint test sites (Section 3.3, Section 4.2.4, Table 1). The global maps are external public products (GLAD, ETH, Meta, Pauls et al.) and the local labels are independent of the authors' prior results; no parameter is fitted to the test labels, and no equation in the paper defines the predicted quantity in terms of the comparison metric. The XceptionS2 fine-tuning choice (tuning only the last 3 layers, Section 4.2.2) is a modeling decision that may understate the value of global pretraining, but that is a validity/robustness concern, not circularity: the comparison is not forced by construction. Self-citations (Rolf et al. 2022; Rolf 2023; Rolf et al. 2024; Aiken et al. 2023) appear in contextual framing and evaluation-practice discussion, but none is load-bearing for the empirical result, which is self-contained against external benchmarks. No circular step is exhibited.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claims depend on data quality (UAV-LiDAR reference labels, Sentinel-2 imagery), site representativeness, and hand-chosen preprocessing values such as a 30 m label cap, 90th percentile aggregation, and limited fine-tuning budgets for global models. No new physical or model entities are posited; the work is an empirical comparison of existing models.

free parameters (5)
  • LiDAR label cap at 30 m = 30 m
    Pixels above 30 m are set to NaN in preprocessing (Section 4.2.3), excluding 0.05% of data from training and evaluation. This hand-chosen threshold could, in principle, reduce errors for models that under-predict tall trees.
  • 90th percentile coarsening of labels = 90th percentile
    1 m LiDAR labels aggregated to 10 m using the 90th percentile (Section 4.2.3), matching earlier global mapping practice. The choice affects the label distribution and therefore the RMSE comparison.
  • Number of XceptionS2 layers fine-tuned = 3 layers
    In the transfer learning setup, only the last one to three layers are updated (Table 2), as determined by validation. This restriction limits the potential benefit of pretrained weights.
  • U-Net evaluation margin = innermost 4x4 pixels per 64x64 patch
    Edge instability led the authors to evaluate only the inner 4x4 pixels of each patch for U-Net (Section 5.2), which changes the evaluation grid relative to other models.
  • FCN architecture size = 5 layers, 128 filters
    The winning architecture is a hand-designed five-layer FCN with 604,417 parameters; comparisons were made against several alternatives, but the specific size is a design choice.
assumptions (4)
  • domain assumption UAV-LiDAR canopy height model (Boucher et al. 2023) provides accurate reference TCH.
    The paper treats the 1 m LiDAR-derived TCH maps as ground truth for training and evaluation (Section 3.1). Systematic bias in these reference data would bias the comparison.
  • domain assumption The 24 LiDAR sites represent the full Karingani reserve.
    Site-stratified splits simulate deployment to uncovered regions, so the test sites must reflect the ecological variation of the reserve (Section 3.3). The paper notes that roughly two thirds of the reserve were not flown.
  • domain assumption Sentinel-2 12-band imagery is sufficient for local TCH prediction.
    All local models use Sentinel-2 bands and no elevation data (Section 3.2); if these inputs are inadequate in this landscape, the local-global performance gap could be misattributed.
  • domain assumption Allometric equation F from Colgan et al. 2013 is valid for savanna AGB estimation.
    Used in Section 5.1.3 to convert TCH predictions to aboveground biomass; the AGB comparison is secondary to the main claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Contrasting local and global modeling with machine learning and satellite data: A case study estimating tree canopy height in African savannas." pith.science (2026). https://pith.science/paper/DUEXERRM

@misc{pith2026241114354,
  author       = {Pith},
  title        = {Pith review of: Contrasting local and global modeling with machine learning and satellite data: A case study estimating tree canopy height in African savannas},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DUEXERRM}},
  note         = {Machine review of arXiv:2411.14354}
}
read the original abstract

While advances in machine learning with satellite imagery (SatML) are facilitating environmental monitoring at a global scale, developing SatML models that are accurate and useful for local regions remains critical to understanding and acting on an ever-changing planet. As increasing attention and resources are being devoted to training SatML models with global data, it is important to understand when improvements in global models will make it easier to train or fine-tune models that are accurate in specific regions. To explore this question, we contrast local and global training paradigms for SatML through a case study of tree canopy height (TCH) mapping in the Karingani Game Reserve, Mozambique. We find that recent advances in global TCH mapping do not necessarily translate to better local modeling abilities in our study region. Specifically, small models trained only with locally-collected data outperform published global TCH maps, and even outperform globally pretrained models that we fine-tune using local data. Analyzing these results further, we identify specific points of conflict and synergy between local and global modeling paradigms that can inform future research toward aligning local and global performance objectives in geospatial machine learning.

Figures

Figures reproduced from arXiv: 2411.14354 by the authors.

Figure 1
Figure 1. Our case study simulates a common local mapping use case: we use locally￾collected tree canopy height (TCH) maps derived from data collected via UAV￾mounted LiDAR sensors at sites distributed across Karingani Game Reserve, Mozam￾bique to train a predictive model, then deploy it throughout the study region. (a) The local supervised learning problem consists of high-resolution TCH labels derived from LiDAR mea￾suremen… view at source ↗
Figure 2
Figure 2. A locally-trained fully convolutional network (FCN) outperforms the three existing global maps in quantitative performance. Average performance is shown across splits (gray dots), and averaged over splits (larger, colored dots). Models on the horizontal axis are ordered by date of publication [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. A locally-trained fully convolutional network (FCN) exhibits less prediction bias than the four existing global TCH maps, across ecologically-relevant strata: (a) binned tree canopy height, (b) geology type, and (c) binned distance to nearest river. performance for each split individually, showing that these differences are consistent across different data splits [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Existing TCH maps exhibit different types of visual error structures, which are largely alleviated by our local model. TCH labels (aggregated to 10m resolution) derived from locally collected LiDAR data, our model trained only with local labeled data (5-layer FCN with …
Figure 5
Figure 5. Figure 5: Different design decisions have similar magnitude of effect on local model performance. Performance varies across: (a) different machine learning architectures, (b) amount of training data (number of training sites) available, and (c) different sets of spectral bands u…
Figure 6
Figure 6. Figure 6: Data composition and model choice interact to affect model performance (train/val/test split 0 shown here; similar trends hold across splits — see [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Distribution of predictions and errors from each model (horizontal axis) vs. the LiDAR [PITH_FULL_IMAGE:figures/full_fig_p029_7.png]
Figure 8
Figure 8. Figure 8: Aboveground biomass (AGB) per site, as estimated using the 1m LiDAR-derived TCH [PITH_FULL_IMAGE:figures/full_fig_p029_8.png]
Figure 9
Figure 9. Figure 9: Results extending Figure 6 for all four data splits. [PITH_FULL_IMAGE:figures/full_fig_p031_9.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Localized, High-resolution Geographic Representations with Slepian Functions

    cs.LG 2026-01 conditional novelty 6.0 of 10

    Spherical Slepian functions, used as a region-concentrated geographic position encoder, outperform global spherical-harmonic and Fourier location encoders on localized geospatial prediction tasks.

  2. A Novel Large Vision Foundation Model (LVFM)-based Approach for Generating High-Resolution Canopy Height Maps in Plantations for Precision Forestry Management

    cs.CV 2025-06 conditional novelty 5.0 of 10

    A DINOv2-based CHM model with self-supervised feature upsampling reports 0.09 m MAE against lidar in a Beijing plantation test site, plus tree detection and AGB estimation applications.

Reference graph

Works this paper leans on

21 extracted references · 12 canonical work pages · cited by 2 Pith papers

  1. [2]

    doi: 10.24963/ijcai.2023/653

    ISBN 978-1-956792-03-4. doi: 10.24963/ijcai.2023/653. Cristina Aponte, Sabine Kasel, Craig R Nitschke, Mihai A Tanase, Helen Vickers, Linda Parker, Melissa Fedrigo, Michele Kohout, Paloma Ruiz-Benito, Miguel A Zavala, et al. Structural diver- sity underpins carbon storage in Australian temperate forests. Global Ecology and Biogeography, 29(5):789–802,

    doi:10.24963/ijcai.2023/653 2023

  2. [4]

    FoMo-Bench: a multi-modal, multi-scale and multi-task forest monitoring benchmark for remote sensing foundation models

    Nikolaos Ioannis Bountos, Arthur Ouaknine, and David Rolnick. FoMo-Bench: a multi-modal, multi-scale and multi-task forest monitoring benchmark for remote sensing foundation models. arXiv preprint arXiv:2312.10114 ,

    arXiv

  3. [6]

    doi: 10.1073/pnas.2113658119

    ISSN 1091-6490. doi: 10.1073/pnas.2113658119. URL http://dx.doi. org/10.1073/pnas.2113658119. Matthew S. Colgan, Gregory P. Asner, and Tony Swemmer. Harvesting tree biomass at the stand level to assess the accuracy of field and airborne biomass estimation in savannas. Ecological Applications, 23(5):1170–1184,

    doi:10.1073/pnas.2113658119

  4. [10]

    doi: https://doi.org/10.1016/j.rse.2022.113402

    ISSN 0034-4257. doi: https://doi.org/10.1016/j.rse.2022.113402. Adam Lilja, Junsheng Fu, Erik Stenborg, and Lars Hammarstrand. Localization is all you evaluate: Data leakage in online mapping datasets and how to fix it. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 22150–22159,

    work page arXiv 2022

  5. [12]

    OpenForest: A data catalogue for machine learning in forest monitoring

    Arthur Ouaknine, Teja Kattenborn, Etienne Lalibert´ e, and David Rolnick. OpenForest: A data catalogue for machine learning in forest monitoring. arXiv preprint arXiv:2311.00277 ,

    work page Pith review arXiv

  6. [13]

    doi: 10.1038/s41586-023-06825-8

    ISSN 1476-4687. doi: 10.1038/s41586-023-06825-8. Jan Pauls, Max Zimmer, Una M Kelly, Martin Schwartz, Sassan Saatchi, Philippe Ciais, Sebastian Pokutta, Martin Brandt, and Fabian Gieseke. Estimating canopy height at scale. International Conference of Machine Learning ,

    doi:10.1038/s41586-023-06825-8

  7. [14]

    doi: https://doi.org/10.1016/j.rse.2020.112165

    ISSN 0034-4257. doi: https://doi.org/10.1016/j.rse.2020.112165. Gyri Reiersen, David Dao, Bj¨ orn L¨ utjens, Konstantin Klemmer, Kenza Amara, Attila Steinegger, Ce Zhang, and Xiaoxiang Zhu. ReforesTree: A dataset for estimating tropical forest carbon stock with deep learning and aerial imagery. Proceedings of the 2022 Conference by the Association for the...

    arXiv 2020

  8. [17]

    Knowledge Transfer from Vision Foundation Models for Efficient Training of Small Task-specific Models

    ISSN 2072-4292. doi: 10.3390/rs15215162. Raviteja Vemulapalli, Hadi Pouransari, Fartash Faghri, Sachin Mehta, Mehrdad Farajtabar, Mo- hammad Rastegari, and Oncel Tuzel. Label-efficient training of small task-specific models by leveraging vision foundation models. arXiv preprint arXiv:2311.18237 ,

    work page Pith review arXiv

Show all 21 references
  1. [18]

    doi: 10.1109/MGRS.2023.3281651. Ben G. Weinstein, Sarah J. Graves, Sergio Marconi, Aditya Singh, Alina Zare, Dylan Stewart, Stephanie A. Bohlman, and Ethan P. White. A benchmark dataset for canopy crown detection 27 and delineation in co-registered airborne RGB, LiDAR and hype...

    2023

  2. [19]

    Phil Wilkes, Simon D

    doi: 10.1371/journal.pcbi.1009180. Phil Wilkes, Simon D. Jones, Lola Suarez, Andrew Mellor, William Woodgate, Mariela Soto- Berelov, Andrew Haywood, and Andrew K. Skidmore. Mapping forest canopy height across large areas by upscaling ALS estimates with freely available satelli...

    doi

  3. [21]

    X = CC ×H

    is calculated for each site using our high-fidelity LiDAR-derived local data at 1m resolution, and plotted on the vertical axis. On the horizontal axis of each panel, we plot the same “X = CC ×H”, this time estimated from each of the four different data products and our own pr...

    2024

  4. [2003]

    doi: 10.1162/153244304322972667

    ISSN 1532-4435. doi: 10.1162/153244304322972667. Adam J Stewart, Caleb Robinson, Isaac A Corley, Anthony Ortiz, Juan M Lavista Ferres, and Arindam Banerjee. Torchgeo: deep learning with geospatial data. In Proceedings of the 30th International Conference on Advances in Geograp...

    doi

  5. [2013]

    Isaac Corley, Caleb Robinson, Rahul Dodhia, Juan M Lavista Ferres, and Peyman Najafirad

    doi: https://doi.org/10.1890/12-0922.1. Isaac Corley, Caleb Robinson, Rahul Dodhia, Juan M Lavista Ferres, and Peyman Najafirad. Revisiting pre-trained remote sensing model benchmarks: resizing and normalization matters. In Proceedings of the IEEE/CVF Conference on Computer Vi...

    doi

  6. [2015]

    doi: 10.3390/rs70912563

    ISSN 2072-4292. doi: 10.3390/rs70912563. Christopher Yeh, Anthony Perez, Anne Driscoll, George Azzari, Zhongyi Tang, David Lobell, Stefano Ermon, and Marshall Burke. Using publicly available satellite imagery and deep learning to understand economic well-being in Africa. Natur...

    doi

  7. [2016]

    Open- Canopy: A country-scale benchmark for canopy height estimation at very high resolution

    Fajwel Fogel, Yohann Perron, Nikola Besic, Laurent Saint-Andr´ e, Agn` es Pellissier-Tanon, Martin Schwartz, Thomas Boudras, Ibrahim Fayad, Alexandre d’Aspremont, Loic Landrieu, et al. Open- Canopy: A country-scale benchmark for canopy height estimation at very high resolution...

    arXiv

  8. [2020]

    doi: 10.3390/rs12172840

    ISSN 2072-4292. doi: 10.3390/rs12172840. Kaizhu Huang, Haiqin Yang, Irwin King, and Michael Lyu. Machine Learning: Modeling Data Locally and Globally . Zhejiang University Press and Springer Berlin Heidelberg, Hangzhou and Berlin,

    doi

  9. [2021]

    doi: 10.3390/rs13122392

    ISSN 2072-4292. doi: 10.3390/rs13122392. Peter B Boucher, Evan G Hockridge, Jenia Singh, and Andrew B Davies. Flying high: Sampling savanna vegetation with UA V-lidar. Methods in Ecology and Evolution , 14(7):1668–1686,

    doi

  10. [2022]

    doi: 10.1038/s41597-022-01307-4

    ISSN 2052-4463. doi: 10.1038/s41597-022-01307-4. Guanghua Chi, Han Fang, Sourav Chatterjee, and Joshua E. Blumenstock. Microestimates of wealth for all low- and middle-income countries.Proceedings of the National Academy of Sciences, 119(3), January

    doi

  11. [2023]

    Emily Aiken, Esther Rolf, and Joshua Blumenstock

    doi: 10.5194/essd-15-681-2023. Emily Aiken, Esther Rolf, and Joshua Blumenstock. Fairness and representation in satellite-based poverty maps: Evidence of urban-rural disparities and their impacts on downstream policy. In Proceedings of the Thirty-Second International Joint Con...

    2023 doi

  12. [2024]

    URL https://data.neonscience.org/data-products/DP3.30015.001/RELEASE-2024. Florent Noul` ekoun, Emiru Birhane, Sylvanus Mensah, Habtemariam Kassa, Alemayehu Berhe, Zefere Mulaw Gebremichael, Nuru Mohammed Adem, Yigremachew Seyoum, Tefera Mengistu, Bekele Lemma, et al. Structur...

    2024

  13. [4257]

    Angela Tsao, Ikenna Nzewi, Ayodeji Jayeoba, Uzoma Ayogu, and David B

    doi: https://doi.org/10.1016/j.rse.2023.113888. Angela Tsao, Ikenna Nzewi, Ayodeji Jayeoba, Uzoma Ayogu, and David B. Lobell. Canopy height mapping for plantations in Nigeria using GEDI, Landsat, and Sentinel-2. Remote Sensing , 15 (21),

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.