Pith. sign in

REVIEW 4 major objections 6 minor 72 references

A self-supervised deep-learning framework, SIT-FUSE, maps harmful algal bloom severity and species by fusing reflectance from several satellite ocean-color sensors with solar-induced fluorescence, requiring no per-instrument labeled imagery

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

SIT-FUSE maps harmful algal bloom concentration and species from fused VIIRS/MODIS/Sentinel-3/PACE/TROPOMI data using self-supervised hierarchical clustering, with limited in-situ validation.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection A credible, well-documented extension of SIT-FUSE to HAB mapping with a genuinely new hierarchical IIC and a first PACE test, but the 'strong agreement' claim is not yet quantified and the matchup radii are too loose to support it. the 4 major comments →

arxiv 2510.02763 v3 pith:KQPVYWV3 submitted 2025-10-03 cs.LG cs.AI

Fusing Multi- and Hyperspectral Satellite Data for Harmful Algal Bloom Monitoring with Self-Supervised and Hierarchical Deep Learning

Nicholas LaHaye , Kelly M. Luis , Michelle M. Gierach This is my paper
classification cs.LG cs.AI
keywords Harmful algal bloomsSelf-supervised learningDeep clusteringInvariant Information ClusteringSatellite data fusionOcean colorSolar-induced fluorescencePhytoplankton speciation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper claims that a self-supervised machine-learning pipeline, called SIT-FUSE, can produce daily maps of harmful algal bloom severity and species from satellite data without needing labeled images for each instrument. It does this by learning an embedding from fused reflectance bands of several ocean-color sensors plus TROPOMI solar-induced fluorescence, then partitioning the embedding into a two-level hierarchy of clusters that are assigned meaning by matching them to binned in-situ concentration measurements. The central result is that one framework handles VIIRS, MODIS, Sentinel-3, and PACE OCI data—separately or fused—and the resulting maps agree with withheld in-situ counts of total phytoplankton, Karenia brevis, Pseudo-nitzschia, and Alexandrium in the Gulf of Mexico and Southern California. If true, operational HAB monitoring would no longer require a purpose-built retrieval algorithm and labeled dataset for every new sensor, and the method could extend to label-scarce regions.

Core claim

The paper's discovery is that coarse, human-readable labels are not needed during training: SIT-FUSE first learns, in a self-supervised way, to represent each 7-km pixel's neighborhood as an embedding vector from whatever spectral channels are available, then performs hierarchical deep clustering to split the data into 800 coarse classes and, below them, 100 subclasses per parent. After clustering, a simple histogram overlap between the unlabeled classes and binned in-situ phytoplankton concentrations assigns each class a concentration level, separately for total phytoplankton and for each HAB species. On held-out test periods, the confusion matrices between predicted bins and in-situ bins s

What carries the argument

The load-bearing mechanism is the pair of a Deep Belief Network encoder trained by contrastive divergence and a hierarchical deep-clustering head based on Invariant Information Clustering (IIC). IIC maximizes the mutual information between the cluster assignment of a pixel and the assignment of a perturbed version of the same pixel, so clusters emerge from the data rather than from labels. The hierarchy—a root level of 800 clusters and child levels of 100 subclusters per parent—gives the system two scales of specificity: the coarse level maps total bloom extent, the finer level separates species or concentration grades. A histogram-based context-assignment step then translates the unlabeled

Load-bearing premise

The validation treats a water sample taken within 2.5 km (Florida) or 10 km (Southern California) of a pixel center, at depth at or above 1 m, on the same day, as ground truth for a 7-km satellite pixel; if algal concentrations vary sharply within those radii, the agreement scores partly measure matchup mismatch rather than mapping skill.

What would settle it

Find pixels on the same day that contain two or more independent in-situ samples. If the within-pixel spread of measured concentrations routinely spans more than one concentration bin, then the confusion-matrix agreement cannot be attributed to the algorithm alone. A second falsifier: apply the trained system to a held-out region or season and compare bin agreement there, since context assignment currently uses the same regional networks that are later used for validation.

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

If this is right

  • One trained system can generate HAB products from any of VIIRS, MODIS, Sentinel-3 OLCI, or PACE OCI, either alone or fused with TROPOMI red SIF, so gaps or failures in a single sensor do not interrupt monitoring.
  • Because labels are attached after clustering, adding a new sensor requires only reflectance data and a handful of in-situ bins, not a new supervised training set; the PACE test case is an early demonstration of this claim.
  • Fusing solar-induced fluorescence with reflectance extends daily coverage relative to single-sensor ocean color products, since the SIF stream provides data where standard fluorescence line height does not.
  • The two-level hierarchy enables both total-phytoplankton mapping and species-level mapping from the same learned representation, and could support tracking individual bloom events across scenes and instruments.
  • The monthly-averaged and data-quality-tagged products provide a template for operational concentration maps that carry information about which data stream produced each pixel.

Where Pith is reading between the lines

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

  • If the matchup assumption holds, the same encoder/clustering recipe should transfer to other label-scarce coastal variables—such as colored dissolved organic matter or suspended sediment—because it never needs per-variable labeled imagery, only the final histogram assignment.
  • A natural extension would be to combine the per-instrument concentration maps with uncertainty weighting rather than the current stream-priority data-quality tag, and then test whether a globally fused product beats the best single sensor; the paper does not yet make that comparison.
  • The strongest test of generality would be to assign context in one region or season and validate in another without any in-situ data from the validation region; the current design assigns and validates within the same regional networks, so transferability is not yet demonstrated.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper presents SIT-FUSE, a framework that combines self-supervised representation learning (DBNs) with hierarchical invariant-information clustering (IIC) to segment and map harmful algal bloom (HAB) severity and speciation from multi-sensor satellite data. Inputs are daily Level-3 surface reflectance from VIIRS, MODIS, OLCI/OCI, optionally fused with TROPOMI red SIF, all gridded to 7 km. Cluster labels are converted to binned concentration classes via histogram overlap with in-situ observations (Section 2.3.3), and validation is performed on temporally held-out test periods (2018–2019 and 2024–2025) against in-situ matchups using confusion matrices (Section 2.3.5). The central claim is that SIT-FUSE shows 'strong agreement' with total phytoplankton, Karenia brevis, and Pseudo-nitzschia measurements and can generate HAB products without per-instrument labeled datasets.

Significance. If the validation evidence were fully reported and robust, this would be a useful contribution to label-scarce HAB monitoring: the approach is instrument-agnostic, fuses multiple operational sensors, and makes code, data, and model weights publicly available. The temporally held-out evaluation is a strength, as is the honest acknowledgment of low matchup counts in some test cases. The hierarchical IIC structure is a novel extension of deep clustering and offers a potentially useful exploratory tool. However, the manuscript currently does not contain the numerical content needed to substantiate the 'strong agreement' claim, and the matchup protocol raises a load-bearing validity concern. The contribution is therefore promising but not yet established.

major comments (4)
  1. [§3.1.1–3.2.2, Tables 1–9] The central claim of 'strong agreement' is not supported by any numerical summary in the manuscript. The tables are described as confusion matrices with raw counts and percentages, but the actual entries are not reported or interpreted. There are no accuracy, kappa, precision/recall, or confidence-interval statistics. Section 2.3.5 admits the number of matchups is small, and Section 3.2.1 states the PACE Gulf counts are 'far too low to get a good evaluation.' Without per-table sample sizes, per-class metrics, and uncertainty bounds, the reader cannot assess whether agreement exceeds chance, especially for rare taxa such as Alexandrium spp. Please report and interpret the actual matrix values.
  2. [§2.3.3/§2.3.5] The matchup protocol treats an in-situ surface sample (depth ≤1 m) within 2.5 km (Florida) or 10 km (Southern California) and on the same calendar day as the mean of a 7-km satellite pixel. The text says the radii were chosen to obtain a 'large enough set' of matchups, not from any spatial decorrelation analysis. In coastal waters with fronts, plumes, and patchy HABs, sub-kilometer gradients could make the confusion matrices partly measure matchup mismatch rather than algorithm skill. The same protocol is used both to construct the cluster-to-concentration calibration (Section 2.3.3) and to validate the products (Section 2.3.5), so the mismatch noise enters twice. Please add a representativeness check (e.g., sensitivity to smaller radii, comparison with higher-resolution retrievals, or a spatial decorrelation analysis) or otherwise bound the uncertainty this introduces.
  3. [§2.3.3] The description as 'self-supervised' should be qualified. Cluster labels are converted to concentration bins by a supervised histogram-overlap step using in-situ data; without this step the product has no concentration units. The abstract's phrase 'without requiring per-instrument labeled datasets' is technically true but could be read as label-free. The framework is self-supervised for representation and clustering, followed by supervised calibration on a shared in-situ matchup set. Please state this explicitly and clarify whether the same matchup data can be used for both calibration and validation without circular reasoning.
  4. [§2.3.2–2.3.3, Fig. 4] The framework has many free hyperparameters: 800 root clusters, 100 child clusters, hierarchy depth, 50 k-means stratification classes, Gaussian noise level, concentration-bin definitions, and the two matchup radii. No sensitivity analysis is reported. The hierarchy size and binning directly define the output products, so the robustness of the 'strong agreement' claim to these choices is unknown. Please provide an ablation or sensitivity study for the most consequential settings (number of clusters, bin boundaries, matchup radii) or justify them using held-out criteria.
minor comments (6)
  1. [Abstract] Typo: 'Karena brevis' should be 'Karenia brevis'.
  2. [§2.4] Hardware is listed as 'NVIDIA GeForce Titan V100 GPU'; Titan V and V100 are distinct products. Please clarify the exact GPU used.
  3. [§2.3.2] The Gaussian noise perturbation for IIC is mentioned but its magnitude or relative scale is never specified. A value (or a range) is needed for reproducibility.
  4. [§2.3.4] The statement that 'OC +TROPOSIF, TROPOSIF only, and OC only outputs will not overlap, by definition' is unclear. Please explain what definition guarantees disjointness, especially since OC and TROPOSIF grids are colocated.
  5. [§3.3] The comparisons to C-HARM and Chl-a are described as 'significant agreement' based on visual overlay. Since the section is explicitly qualitative, please avoid language that could be read as a quantitative validation; report overlap statistics or clearly label these as qualitative demonstrations.
  6. [Fig. 9 caption] The note that the Clam Pass event was undetectable due to 'lack of matchups' is an important limitation. Consider moving this caveat into the main text discussion of limits, rather than only the figure caption.

Circularity Check

0 steps flagged

No significant circularity: held-out temporal validation and independent method description keep the central claim from reducing to its inputs.

full rationale

The derivation chain is: (1) self-supervised encoders produce context-free per-pixel embeddings; (2) IIC deep clustering partitions those embeddings into hierarchical labels without using in-situ targets; (3) Section 2.3.3 maps each cluster label to the most frequently co-occurring binned in-situ concentration, using only training-period scenes ('Over the entire training set... done collectively over the set of scenes in the training set'); (4) Section 2.3.5 validates on held-out time periods ('using the time periods held out for testing... matching up the daily binned concentration products to the in situ sites, binned in the same way'). Because the cluster-to-concentration mapping is fixed during training and then applied to temporally disjoint test scenes, the reported confusion matrices are not a restatement of the training fit. The matchup-radius issue raised by the reader is a validation-design / representativeness concern, not circularity: using the same 2.5 km / 10 km radii for both context assignment and validation could bias both phases if coastal gradients decorrelate at sub-pixel scales, but it does not make the test prediction equal to its input by construction. The paper's self-citations ([33], [50], [68]) are methodological background and software provenance, not load-bearing evidence for the empirical HAB claim; no uniqueness theorem or ansatz is imported from author-only prior work to force the result. The only notable overstatement is the phrase 'fully self-supervised paradigm' despite the supervised context-assignment step, but that is a framing issue, not circular reasoning.

Axiom & Free-Parameter Ledger

10 free parameters · 5 axioms · 0 invented entities

The paper contributes a pipeline, not a physical model; its scientific payload rests on assumptions about unsupervised cluster semantics and matchup data. No new physical entities are introduced; the DQI is a data-provenance flag, not a scientific entity.

free parameters (10)
  • Number of root clusters in hierarchical IIC = 800
    Manual choice; sets coarsest segmentation granularity (Section 2.3.2).
  • Number of child clusters per root = 100
    Manual choice; sets fine-level classes (Section 2.3.2).
  • Hierarchy depth = 2 levels
    Manually specified; authors note future automated node splitting (Section 2.3.2).
  • k-means stratification class count = 50
    Chosen via elbow method for subsampling ~3M training samples (Section 2.2.2).
  • Training sample count per encoder = ~3,000,000
    Subsampled using k-means stratification (Section 2.2.2).
  • Gaussian noise perturbation for IIC = not specified
    Perturbation magnitude for mutual-information clustering is not quantified (Section 2.3.2).
  • Florida matchup radius = 0.0225 deg (~2.5 km)
    Chosen because the Florida in-situ network is denser (Section 2.3.3).
  • Southern California matchup radius = 0.09 deg (~10 km)
    Chosen because coastal sites are sparse and pier-based (Section 2.3.3).
  • Depth cutoff for in-situ samples = 1 m
    Samples deeper than 1 m were removed (Section 2.3.3).
  • Concentration bin definitions = see Figure 4
    Bins used for both context assignment and validation; exact edges are not defined in text (Sections 2.3.3, 2.3.5).
axioms (5)
  • domain assumption Optical reflectance and TROPOMI red SIF, after reprojection and standardization, contain sufficient information to separate phytoplankton abundance and species via unsupervised embeddings.
    The entire method is built on this; no direct ablation or spectral feature analysis is provided.
  • domain assumption In-situ samples at <=1 m depth within 2.5 km (Florida) or 10 km (Southern California) represent the 7-km satellite pixel for the same day.
    Needed for context assignment and validation; radius chosen by data availability rather than a physical decorrelation scale (Section 2.3.3).
  • domain assumption IIC's mutual-information objective produces clusters that are stable and transferable across scenes and time periods.
    Hierarchical clusters are trained on the training period and applied to the test period without clustering retraining (Sections 2.3.2, 2.3.5).
  • ad hoc to paper Modal-overlap histogram assignment (cluster label -> concentration bin) is a valid calibration rule.
    Authors state 'we found this simple approach to be suitable'; no comparison to assignment alternatives is given (Section 2.3.3).
  • domain assumption DBN encoders trained via contrastive divergence provide adequate representations for this task.
    Authors validate DBNs against larger architectures only qualitatively, with details promised in future work (Section 2.3.1).

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Fusing Multi- and Hyperspectral Satellite Data for Harmful Algal Bloom Monitoring with Self-Supervised and Hierarchical Deep Learning." pith.science (2026). https://pith.science/paper/KQPVYWV3

@misc{pith2026251002763,
  author       = {Pith},
  title        = {Pith review of: Fusing Multi- and Hyperspectral Satellite Data for Harmful Algal Bloom Monitoring with Self-Supervised and Hierarchical Deep Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KQPVYWV3}},
  note         = {Machine review of arXiv:2510.02763}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We present a self-supervised machine learning framework for detecting and mapping the severity and speciation of harmful algal blooms (HABs) using multi-sensor satellite data. By fusing reflectance data from operational polar-orbiting satellite-based instruments (VIIRS, MODIS, OLCI, and OCI) with TROPOMI solar-induced fluorescence (SIF), our framework, called SIT-FUSE, generates HAB severity and speciation products without requiring per-instrument labeled datasets. The framework employs self-supervised representation learning and hierarchical deep clustering to segment phytoplankton cell abundance and species into interpretable classes, validated against in-situ data from the Gulf of Mexico and Southern California (2018-2025). Results show strong agreement with total phytoplankton, Karena brevis, and Pseudo-nitzschia spp. measurements. This work advances scalable HAB monitoring in environments where ground truth observations are limited, while enabling exploratory analysis via hierarchical embeddings - a critical step toward operationalizing self-supervised learning for global aquatic biogeochemistry.

Figures

Figures reproduced from arXiv: 2510.02763 by Kelly M. Luis, Michelle M. Gierach, Nicholas LaHaye.

Figure 1
Figure 1. Figure 1: A flow diagram for the processing of one input type (single instrument or fusion [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A 2-layer example of the setup for hierarchical deep clustering. Each box labeled [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: 10 [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: A depiction of all of the locations where in situ data was collected and used for [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: A depiction of the multi-tiered context assignment process based on the hierar [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: A depiction of the combination of the various data streams. First OC only, [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Daily products from each of the instrument/data streams for September 13, [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Daily products for total phytoplankton concentration and each of the potential [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Daily products for total phytoplankton concentration and each of the potential [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Daily context free segmentation products (top) and K. brevis concentration [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Daily context free segmentation products (top), total phytoplankton concen [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Overlays of PACE-based SIT-FUSE maps of total phytoplankton, with (column [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Overlays of AQUA MODIS-based SIT-FUSE maps of total phytoplankton, [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Preliminary per-pixel certainty (left) from the lowest layer (layer 2) of [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Preliminary SIT-FUSE-based context-free segmentation products from EMIT [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

72 extracted references · 27 canonical work pages · 1 internal anchor

  1. [1]

    Some methods for classification and analysis of multi- variate observations

    J. MacQueen. “Some methods for classification and analysis of multi- variate observations”. In: 1967.url:https://api.semanticscholar. org/CorpusID:6278891

    1967

  2. [2]

    Dynamics and physiology of saxitoxin produc- tion by the dinoflagellatesAlexandrium spp

    D. M. Anderson et al. “Dynamics and physiology of saxitoxin produc- tion by the dinoflagellatesAlexandrium spp.” In:Marine Biology104.3 (Oct. 1990), pp. 511–524.issn: 1432-1793.doi:10.1007/bf01314358. url:http://dx.doi.org/10.1007/BF01314358

    work page doi:10.1007/bf01314358 1990

  3. [3]

    Harm- ful algal blooms and red tide problems on the U.S. west coast

    Rita A. Horner, David L. Garrison, and F. Gerald Plumley. “Harm- ful algal blooms and red tide problems on the U.S. west coast”. In: Limnology and Oceanography42.5part2 (July 1997), pp. 1076–1088. issn: 1939-5590.doi:10.4319/lo.1997.42.5_part_2.1076.url: http://dx.doi.org/10.4319/lo.1997.42.5_part_2.1076

    work page doi:10.4319/lo.1997.42.5_part_2.1076.url: 1997

  4. [4]

    Phytoplankton blooms

    “Phytoplankton blooms”. In: (2001)

    2001

  5. [5]

    A Loop Current Intrusion Case Study on the West Florida Shelf

    Ruoying He and Robert H. Weisberg. “A Loop Current Intrusion Case Study on the West Florida Shelf”. In:Journal of Physical Oceanography 33.2 (Feb. 2003), pp. 465–477.issn: 1520-0485.doi:10.1175/1520- 0485(2003)033<0465:alcics>2.0.co;2.url:http://dx.doi.org/ 10.1175/1520-0485(2003)033%3C0465:ALCICS%3E2.0.CO;2

    doi:10.1175/1520- 2003

  6. [6]

    Phytoplankton response to intrusions of slope water on the West Florida Shelf: Models and observations

    John J. Walsh et al. “Phytoplankton response to intrusions of slope water on the West Florida Shelf: Models and observations”. In:Journal of Geophysical Research: Oceans108.C6 (June 2003).issn: 0148-0227. doi:10.1029/2002jc001406.url:http://dx.doi.org/10.1029/ 2002JC001406. 28

    work page doi:10.1029/2002jc001406.url:http://dx.doi.org/10.1029/ 2003

  7. [7]

    Red tides in the Gulf of Mexico: Where, when, and why?

    J. J. Walsh et al. “Red tides in the Gulf of Mexico: Where, when, and why?” In:Journal of Geophysical Research: Oceans111.C11 (Nov. 2006).issn: 0148-0227.doi:10 . 1029 / 2004jc002813.url:http : //dx.doi.org/10.1029/2004JC002813

    work page doi:10.1029/2004jc002813 2006

  8. [9]

    Why is the Northern End of the Cal- ifornia Current System So Productive?

    Barbara Hickey and Neil Banas. “Why is the Northern End of the Cal- ifornia Current System So Productive?” In:Oceanography21.4 (Dec. 2008), pp. 90–107.issn: 1042-8275.doi:10.5670/oceanog.2008.07. url:http://dx.doi.org/10.5670/oceanog.2008.07

    work page doi:10.5670/oceanog.2008.07 2008

  9. [10]

    2009.url:http://qgis.org

    QGISDevelopmentTeam.QGIS Geographic Information System.Open Source Geospatial Foundation. 2009.url:http://qgis.org

    2009

  10. [11]

    Historical perspective on Karenia brevis red tide research in the Gulf of Mexico

    Karen A. Steidinger. “Historical perspective on Karenia brevis red tide research in the Gulf of Mexico”. In:Harmful Algae8.4 (Mar. 2009), pp. 549–561.issn: 1568-9883.doi:10 . 1016 / j . hal . 2008 . 11 . 009. url:http://dx.doi.org/10.1016/j.hal.2008.11.009

    work page doi:10.1016/j.hal.2008.11.009 2009

  11. [12]

    The physical oceanography of upwelling systems and the development of harmful algal blooms

    G.C. Pitcher et al. “The physical oceanography of upwelling systems and the development of harmful algal blooms”. In:Progress in Oceanog- raphy85.1–2 (Apr. 2010), pp. 5–32.issn: 0079-6611.doi:10.1016/ j. pocean. 2010. 02.002.url:http: // dx. doi. org/ 10. 1016 /j . pocean.2010.02.002

    2010

  12. [13]

    Examination of the Seasonal Dynamics of the Toxic Dinoflagellate Alexandrium catenella at Redondo Beach, Cal- ifornia, by Quantitative PCR

    Marie-Ève Garneau et al. “Examination of the Seasonal Dynamics of the Toxic Dinoflagellate Alexandrium catenella at Redondo Beach, Cal- ifornia, by Quantitative PCR”. In:Applied and Environmental Micro- biology77.21 (Nov. 2011), pp. 7669–7680.issn: 1098-5336.doi:10. 1128 / aem . 06174 - 11.url:http : / / dx . doi . org / 10 . 1128 / AEM . 06174-11

    2011

  13. [14]

    Scikit-learn: Machine Learning in Python

    Fabian Pedregosa et al. “Scikit-learn: Machine Learning in Python”. In: J. Mach. Learn. Res.12.null (Nov. 2011), pp. 2825–2830.issn: 1532- 4435. 29

    2011

  14. [15]

    Retrievalandglobalassessmentofterrestrialchloro- phyllfluorescencefromGOSATspacemeasurements

    LuisGuanteretal.“Retrievalandglobalassessmentofterrestrialchloro- phyllfluorescencefromGOSATspacemeasurements”.In:Remote Sens- ing of Environment121(June2012),pp.236–251.issn:0034-4257.doi: 10.1016/j.rse.2012.02.006.url:http://dx.doi.org/10.1016/ j.rse.2012.02.006

    work page doi:10.1016/j.rse.2012.02.006.url:http://dx.doi.org/10.1016/ 2012

  15. [16]

    Harmful algal blooms along the North American west coast region: History, trends, causes, and impacts

    Alan J. Lewitus et al. “Harmful algal blooms along the North American west coast region: History, trends, causes, and impacts”. In:Harmful Algae19 (Sept. 2012), pp. 133–159.issn: 1568-9883.doi:10.1016/j. hal.2012.06.009.url:http://dx.doi.org/10.1016/j.hal.2012. 06.009

    doi:10.1016/j 2012

  16. [17]

    Joiner et al

    J. Joiner et al. “Global monitoring of terrestrial chlorophyll fluores- cence from moderate spectral resolution near-infrared satellite mea- surements: methodology, simulations, and application to GOME-2”. In: (Apr. 2013).doi:10 . 5194 / amtd - 6 - 3883 - 2013.url:http : //dx.doi.org/10.5194/amtd-6-3883-2013

    work page doi:10.5194/amtd-6-3883-2013 2013

  17. [18]

    Far-red sun-induced chlorophyll fluorescence shows ecosystem-specific relationships to gross primary production: An as- sessment based on observational and modeling approaches

    A. Damm et al. “Far-red sun-induced chlorophyll fluorescence shows ecosystem-specific relationships to gross primary production: An as- sessment based on observational and modeling approaches”. In:Remote Sensing of Environment166 (Sept. 2015), pp. 91–105.issn: 0034-4257. doi:10.1016/j.rse.2015.06.004.url:http://dx.doi.org/10. 1016/j.rse.2015.06.004

    work page doi:10.1016/j.rse.2015.06.004.url:http://dx.doi.org/10 2015

  18. [19]

    Simplified physically based retrieval of sun-induced chlorophyll fluorescence from GOSAT data

    Philipp Kohler, Luis Guanter, and Christian Frankenberg. “Simplified physically based retrieval of sun-induced chlorophyll fluorescence from GOSAT data”. In:IEEE Geoscience and Remote Sensing Letters12.7 (July 2015), pp. 1446–1450.issn: 1558-0571.doi:10 . 1109 / lgrs . 2015 . 2407051.url:http : / / dx . doi . org / 10 . 1109 / lgrs . 2015 . 2407051

    2015

  19. [20]

    Numba:aLLVM- based Python JIT compiler

    SiuKwanLam,AntoinePitrou,andStanleySeibert.“Numba:aLLVM- based Python JIT compiler”. In:Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC. SC15. ACM, Nov. 2015, pp. 1–6.doi:10.1145/2833157.2833162.url:http://dx.doi.org/ 10.1145/2833157.2833162

    arXiv 2015

  20. [21]

    Vertical distribution of chlorophyll-a concentration and phytoplankton community composition from in situ fluorescence profiles: a first database for the global ocean

    R. Sauzède et al. “Vertical distribution of chlorophyll-a concentration and phytoplankton community composition from in situ fluorescence profiles: a first database for the global ocean”. In:Earth System Science Data7.2 (Oct. 2015), pp. 261–273.issn: 1866-3516.doi:10 . 5194 / 30 essd-7-261-2015.url:http://dx.doi.org/10.5194/essd-7-261- 2015

    work page doi:10.5194/essd-7-261- 2015

  21. [22]

    Initial skill assessment of the California Harmful Algae Risk Mapping (C-HARM) system

    Clarissa R. Anderson et al. “Initial skill assessment of the California Harmful Algae Risk Mapping (C-HARM) system”. In:Harmful Algae 59 (Nov. 2016), pp. 1–18.issn: 1568-9883.doi:10.1016/j.hal.2016. 08.006.url:http://dx.doi.org/10.1016/j.hal.2016.08.006

    work page doi:10.1016/j.hal.2016 2016

  22. [23]

    An unprecedented coastwide toxic algal bloom linkedtoanomalousoceanconditions

    Ryan M. McCabe et al. “An unprecedented coastwide toxic algal bloom linkedtoanomalousoceanconditions”.In:Geophysical Research Letters 43.19 (Oct. 2016).issn: 1944-8007.doi:10.1002/2016gl070023.url: http://dx.doi.org/10.1002/2016GL070023

    work page doi:10.1002/2016gl070023.url: 2016

  23. [24]

    On calibration of modern neural networks

    Chuan Guo et al. “On calibration of modern neural networks”. In: Proceedings of the 34th International Conference on Machine Learn- ing - Volume 70. ICML’17. Sydney, NSW, Australia: JMLR.org, 2017, pp. 1321–1330

    2017

  24. [25]

    xarray: N-D labeled Arrays and Datasets in Python

    Stephan Hoyer and Joe Hamman. “xarray: N-D labeled Arrays and Datasets in Python”. In:Journal of Open Research Software5.1 (Apr. 2017), p. 10.issn: 2049-9647.doi:10.5334/jors.148.url:http: //dx.doi.org/10.5334/jors.148

    work page doi:10.5334/jors.148.url:http: 2017

  25. [26]

    CuPy:ANumPy-CompatibleLibraryforNVIDIA GPU Calculations

    RyosukeOkutaetal.“CuPy:ANumPy-CompatibleLibraryforNVIDIA GPU Calculations”. In:Proceedings of Workshop on Machine Learn- ing Systems (LearningSys) in The Thirty-first Annual Conference on Neural Information Processing Systems (NIPS). 2017.url:http:// learningsys.org/nips17/assets/papers/paper_16.pdf

    2017

  26. [27]

    On the Loop Current Pene- tration into the Gulf of Mexico

    Robert H. Weisberg and Yonggang Liu. “On the Loop Current Pene- tration into the Gulf of Mexico”. In:Journal of Geophysical Research: Oceans122.12 (Dec. 2017), pp. 9679–9694.issn: 2169-9291.doi:10. 1002/2017jc013330.url:http://dx.doi.org/10.1002/2017JC013330

    work page doi:10.1002/2017jc013330 2017

  27. [28]

    Pseudo-nitzschia, Nitzschia, and domoic acid: New research since 2011

    Stephen S. Bates et al. “Pseudo-nitzschia, Nitzschia, and domoic acid: New research since 2011”. In:Harmful Algae79 (Nov. 2018), pp. 3–43. issn: 1568-9883.doi:10 . 1016 / j . hal . 2018 . 06 . 001.url:http : //dx.doi.org/10.1016/j.hal.2018.06.001

    work page doi:10.1016/j.hal.2018.06.001 2011

  28. [29]

    Feature expansion by a continuous re- stricted Boltzmann machine for near-infrared spectrometric calibra- tion

    Peter de Boves Harrington. “Feature expansion by a continuous re- stricted Boltzmann machine for near-infrared spectrometric calibra- tion”. In:Analytica Chimica Acta1010 (June 2018), pp. 20–28.issn: 31 0003-2670.doi:10.1016/j.aca.2018.01.026.url:http://dx.doi. org/10.1016/j.aca.2018.01.026

    work page doi:10.1016/j.aca.2018.01.026.url:http://dx.doi 2018

  29. [30]

    Henriques, and Andrea Vedaldi.Invariant Informa- tion Clustering for Unsupervised Image Classification and Segmenta- tion

    Xu Ji, João F. Henriques, and Andrea Vedaldi.Invariant Informa- tion Clustering for Unsupervised Image Classification and Segmenta- tion. 2018.doi:10.48550/ARXIV.1807.06653.url:https://arxiv. org/abs/1807.06653

    Pith/arXiv arXiv doi:10.48550/arxiv.1807.06653 2018

  30. [31]

    Global Retrievals of Solar-Induced Chlorophyll Fluorescence With TROPOMI: First Results and Intersensor Compar- ison to OCO-2

    Philipp Köhler et al. “Global Retrievals of Solar-Induced Chlorophyll Fluorescence With TROPOMI: First Results and Intersensor Compar- ison to OCO-2”. In:Geophysical Research Letters45.19 (Oct. 2018). issn: 1944-8007.doi:10.1029/2018gl079031.url:http://dx.doi. org/10.1029/2018GL079031

    work page doi:10.1029/2018gl079031.url:http://dx.doi 2018

  31. [32]

    A decade and a half of Pseudo-nitzschia spp. and domoic acid along the coast of southern California

    Jayme Smith et al. “A decade and a half of Pseudo-nitzschia spp. and domoic acid along the coast of southern California”. In:Harmful Algae 79 (Nov. 2018), pp. 87–104.issn: 1568-9883.doi:10.1016/j.hal. 2018.07.007.url:http://dx.doi.org/10.1016/j.hal.2018.07. 007

    work page doi:10.1016/j.hal 2018

  32. [33]

    Multi-Modal Object Tracking and Image Fu- sion With Unsupervised Deep Learning

    Nicholas LaHaye et al. “Multi-Modal Object Tracking and Image Fu- sion With Unsupervised Deep Learning”. In:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing12.8 (Aug. 2019), pp. 3056–3066.issn: 2151-1535.doi:10.1109/jstars.2019. 2920234.url:http://dx.doi.org/10.1109/JSTARS.2019.2920234

    arXiv 2019

  33. [34]

    2019.doi:10.48550/ARXIV.1912.01703.url: https://arxiv.org/abs/1912.01703

    Adam Paszke et al.PyTorch: An Imperative Style, High-Performance Deep Learning Library. 2019.doi:10.48550/ARXIV.1912.01703.url: https://arxiv.org/abs/1912.01703

    Pith/arXiv arXiv doi:10.48550/arxiv.1912.01703 2019

  34. [35]

    Restricted Boltzmann Machine as Image Pre-processing Method for Deep Neural Classifier

    Szymon Sobczak and Rafal Kapela. “Restricted Boltzmann Machine as Image Pre-processing Method for Deep Neural Classifier”. In:2019 First International Conference on Societal Automation (SA). IEEE, Sept. 2019, pp. 1–5.doi:10 . 1109 / sa47457 . 2019 . 8938039.url: http://dx.doi.org/10.1109/SA47457.2019.8938039

    arXiv 2019

  35. [36]

    The Coastal Ocean Circulation Influence on the 2018 West Florida Shelf <scp>K. brevis</scp> Red Tide Bloom

    Robert H. Weisberg et al. “The Coastal Ocean Circulation Influence on the 2018 West Florida Shelf <scp>K. brevis</scp> Red Tide Bloom”. In:Journal of Geophysical Research: Oceans124.4(Apr.2019),pp.2501– 2512.issn: 2169-9291.doi:10.1029/2018jc014887.url:http://dx. doi.org/10.1029/2018JC014887. 32

    work page doi:10.1029/2018jc014887.url:http://dx 2018

  36. [37]

    2019.doi:10

    Fan Yang, Mengnan Du, and Xia Hu.Evaluating Explanation With- out Ground Truth in Interpretable Machine Learning. 2019.doi:10. 48550 / ARXIV . 1907 . 06831.url:https : / / arxiv . org / abs / 1907 . 06831

    2019

  37. [38]

    The Systematic Review of K-Means Clustering Algorithm

    ArdavanAshabi,ShamsulBinSahibuddin,andMehdiSalkhordehHaghighi. “The Systematic Review of K-Means Clustering Algorithm”. In:2020 The 9th International Conference on Networks, Communication and Computing. ICNCC 2020. ACM, Dec. 2020, pp. 13–18.doi:10.1145/ 3447654 . 3447657.url:http : / / dx . doi . org / 10 . 1145 / 3447654 . 3447657

    2020

  38. [39]

    Climate Change and Harmful Algal Blooms: Insights and perspective

    Christopher J. Gobler. “Climate Change and Harmful Algal Blooms: Insights and perspective”. In:Harmful Algae91 (Jan. 2020), p. 101731. issn: 1568-9883.doi:10 . 1016 / j . hal . 2019 . 101731.url:http : //dx.doi.org/10.1016/j.hal.2019.101731

    arXiv 2020

  39. [40]

    2020.doi:10.48550/ARXIV.2006.07733

    Jean-Bastien Grill et al.Bootstrap your own latent: A new approach to self-supervised Learning. 2020.doi:10.48550/ARXIV.2006.07733. url:https://arxiv.org/abs/2006.07733

    Pith/arXiv arXiv doi:10.48550/arxiv.2006.07733 2020

  40. [41]

    Enhanced zippy restricted Boltzmann machine for feature expansion and improved classification of analytical data

    Peter B. Harrington. “Enhanced zippy restricted Boltzmann machine for feature expansion and improved classification of analytical data”. In:Journal of Chemometrics34.3 (Feb. 2020).issn: 1099-128X.doi: 10.1002/cem.3228.url:http://dx.doi.org/10.1002/cem.3228

    work page doi:10.1002/cem.3228.url:http://dx.doi.org/10.1002/cem.3228 2020

  41. [42]

    Array programming with NumPy

    Charles R. Harris et al. “Array programming with NumPy”. In:Nature 585.7825 (Sept. 2020), pp. 357–362.issn: 1476-4687.doi:10.1038/ s41586-020-2649-2.url:http://dx.doi.org/10.1038/s41586- 020-2649-2

    doi:10.1038/s41586- 2020

  42. [43]

    KelseyJordahletal.geopandas/geopandas: v0.8.1.2020.doi:10.5281/ ZENODO.3946761.url:https://zenodo.org/record/3946761

    arXiv 2020

  43. [44]

    Global Retrievals of Solar-Induced Chlorophyll Fluorescence at Red Wavelengths With TROPOMI

    Philipp Köhler et al. “Global Retrievals of Solar-Induced Chlorophyll Fluorescence at Red Wavelengths With TROPOMI”. In:Geophysical Research Letters47.15 (July 2020).issn: 1944-8007.doi:10.1029/ 2020gl087541.url:http://dx.doi.org/10.1029/2020GL087541

    work page doi:10.1029/2020gl087541 2020

  44. [45]

    Exploring the potential value of satellite re- mote sensing to monitor chlorophyll-a for US lakes and reservoirs

    Michael Papenfus et al. “Exploring the potential value of satellite re- mote sensing to monitor chlorophyll-a for US lakes and reservoirs”. In: Environmental Monitoring and Assessment192.12 (Dec. 2020).issn: 33 1573-2959.doi:10.1007/s10661- 020- 08631- 5.url:http://dx. doi.org/10.1007/s10661-020-08631-5

    doi:10.1007/s10661- 2020

  45. [46]

    Learnergy: Energy-based Machine Learners

    MateusRoder,GustavoHenriquedeRosa,andJoãoPauloPapa.Learn- ergy: Energy-based Machine Learners. 2020.doi:10 . 48550 / ARXIV . 2003.07443.url:https://arxiv.org/abs/2003.07443

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2003.07443 2020

  46. [47]

    SciPy 1.0: fundamental algorithms for scientific computing in Python

    Pauli Virtanen et al. “SciPy 1.0: fundamental algorithms for scientific computing in Python”. In:Nature Methods17.3 (Feb. 2020), pp. 261– 272.issn: 1548-7105.doi:10.1038/s41592-019-0686-2.url:http: //dx.doi.org/10.1038/s41592-019-0686-2

    work page doi:10.1038/s41592-019-0686-2.url:http: 2020

  47. [48]

    2021.doi:10.48550/ARXIV.2111.06377.url:https://arxiv.org/ abs/2111.06377

    Kaiming He et al.Masked Autoencoders Are Scalable Vision Learners. 2021.doi:10.48550/ARXIV.2111.06377.url:https://arxiv.org/ abs/2111.06377

    Pith/arXiv arXiv doi:10.48550/arxiv.2111.06377 2021

  48. [49]

    Florida’sHarm- ful Algal Bloom (HAB) Problem: Escalating Risks to Human, Environ- mental and Economic Health With Climate Change

    CynthiaAnnHeilandAmandaLorraineMuni-Morgan.“Florida’sHarm- ful Algal Bloom (HAB) Problem: Escalating Risks to Human, Environ- mental and Economic Health With Climate Change”. In:Frontiers in Ecology and Evolution9 (June 2021).issn: 2296-701X.doi:10.3389/ fevo.2021.646080.url:http://dx.doi.org/10.3389/fevo.2021. 646080

    arXiv 2021

  49. [50]

    A Quantitative Validation of Multi-Modal Im- age Fusion and Segmentation for Object Detection and Tracking

    Nicholas LaHaye et al. “A Quantitative Validation of Multi-Modal Im- age Fusion and Segmentation for Object Detection and Tracking”. In: Remote Sensing13.12 (June 2021), p. 2364.issn: 2072-4292.doi:10. 3390/rs13122364.url:http://dx.doi.org/10.3390/rs13122364

    work page doi:10.3390/rs13122364 2021

  50. [51]

    On ‘ground’ truth and why we should abandon the term

    Iain H. Woodhouse. “On ‘ground’ truth and why we should abandon the term”. In:Journal of Applied Remote Sensing15.04 (Nov. 2021). issn: 1931-3195.doi:10.1117/1.jrs.15.041501.url:http://dx. doi.org/10.1117/1.JRS.15.041501

    work page doi:10.1117/1.jrs.15.041501.url:http://dx 2021

  51. [52]

    2022.doi: 10.48550/ARXIV.2210.10318.url:https://arxiv.org/abs/2210

    Renjie Liao et al.Gaussian-Bernoulli RBMs Without Tears. 2022.doi: 10.48550/ARXIV.2210.10318.url:https://arxiv.org/abs/2210. 10318

    Pith/arXiv arXiv doi:10.48550/arxiv.2210.10318 2022

  52. [53]

    Pool-Based Sequential Active Learning For Regres- sion Based on Incremental Cluster Center Selection

    Binwei Lin et al. “Pool-Based Sequential Active Learning For Regres- sion Based on Incremental Cluster Center Selection”. In:2021 Ninth International Conference on Advanced Cloud and Big Data (CBD). IEEE, Mar. 2022, pp. 176–182.doi:10.1109/cbd54617.2021.00038. url:http://dx.doi.org/10.1109/CBD54617.2021.00038. 34

    arXiv 2021

  53. [54]

    Development, calibration, and evaluation of a model of Pseudo-nitzschia and domoic acid production for regional ocean modeling studies

    Allison R. Moreno et al. “Development, calibration, and evaluation of a model of Pseudo-nitzschia and domoic acid production for regional ocean modeling studies”. In:Harmful Algae118 (Oct. 2022), p. 102296. issn: 1568-9883.doi:10 . 1016 / j . hal . 2022 . 102296.url:http : //dx.doi.org/10.1016/j.hal.2022.102296

    arXiv 2022

  54. [55]

    Local And Deep-Ocean Forc- ing Effects on the West Florida Continental Shelf Circulation and Ecol- ogy

    Robert H. Weisberg and Yonggang Liu. “Local And Deep-Ocean Forc- ing Effects on the West Florida Continental Shelf Circulation and Ecol- ogy”. In:Frontiers in Marine Science9 (June 2022).issn: 2296-7745. doi:10.3389/fmars.2022.863227.url:http://dx.doi.org/10. 3389/fmars.2022.863227

    arXiv 2022

  55. [56]

    2023.doi:10.48550/ARXIV

    Mahmoud Assran et al.Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture. 2023.doi:10.48550/ARXIV. 2301.08243.url:https://arxiv.org/abs/2301.08243

    Pith/arXiv arXiv doi:10.48550/arxiv.2301.08243 2023

  56. [57]

    Helm, and Carey E

    Brandon Duderstadt, Hayden S. Helm, and Carey E. Priebe.Com- paring Foundation Models using Data Kernels. 2023.doi:10.48550/ ARXIV.2305.05126.url:https://arxiv.org/abs/2305.05126

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.05126 2023

  57. [58]

    Alexander Kirillov et al.Segment Anything. 2023. eprint:arXiv:2304. 02643

    2023

  58. [59]

    First Light Demonstration of Red Solar Induced Fluorescence for Harmful Algal Bloom Monitoring

    Kelly Luis et al. “First Light Demonstration of Red Solar Induced Fluorescence for Harmful Algal Bloom Monitoring”. In:Geophysical Research Letters50.13 (July 2023).issn: 1944-8007.doi:10.1029/ 2022gl101715.url:http://dx.doi.org/10.1029/2022GL101715

    work page doi:10.1029/2022gl101715 2023

  59. [60]

    2023.doi:10

    Jenny Zhang et al.OMNI: Open-endedness via Models of human No- tions of Interestingness. 2023.doi:10 . 48550 / ARXIV . 2306 . 01711. url:https://arxiv.org/abs/2306.01711

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.01711 2023

  60. [61]

    RetrievalofoceanicchlorophyllconcentrationfromGOES-RAdvanced Baseline Imager using deep learning

    Guangming Zheng, Christopher W. Brown, and Paul M. DiGiacomo. “RetrievalofoceanicchlorophyllconcentrationfromGOES-RAdvanced Baseline Imager using deep learning”. In:Remote Sensing of Environ- ment295 (Sept. 2023), p. 113660.issn: 0034-4257.doi:10.1016/j. rse.2023.113660.url:http://dx.doi.org/10.1016/j.rse.2023. 113660

    arXiv 2023

  61. [62]

    J. Kravitz et al.Pushing the limits of aquatic remote sensing: Synthetic data and deep learning for fast inverse emulation of a coupled water- atmosphere radiative transfer model.2024.url:https://ntrs.nasa. gov/citations/20230002640. 35

    arXiv 2024

  62. [63]

    2024.doi:10.48550/ARXIV.2408.06292.url: https://arxiv.org/abs/2408.06292

    Chris Lu et al.The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery. 2024.doi:10.48550/ARXIV.2408.06292.url: https://arxiv.org/abs/2408.06292

    Pith/arXiv arXiv doi:10.48550/arxiv.2408.06292 2024

  63. [64]

    A Machine Learning Approach to Produce a Continuous Solar-Induced Chlorophyll Fluorescence Over the Arctic Ocean

    Nima Madani et al. “A Machine Learning Approach to Produce a Continuous Solar-Induced Chlorophyll Fluorescence Over the Arctic Ocean”. In:Journal of Geophysical Research: Machine Learning and Computation1.4(Dec.2024).issn:2993-5210.doi:10.1029/2024jh000215. url:http://dx.doi.org/10.1029/2024JH000215

    work page doi:10.1029/2024jh000215 2024

  64. [65]

    2024.doi:10.48550/ARXIV.2412

    Valerio Marsocci et al.PANGAEA: A Global and Inclusive Benchmark for Geospatial Foundation Models. 2024.doi:10.48550/ARXIV.2412. 04204.url:https://arxiv.org/abs/2412.04204

    Pith/arXiv arXiv doi:10.48550/arxiv.2412.04204 2024

  65. [66]

    2025.doi:10

    David Hoese et al.pytroll/pyresample: Version 1.34.2. 2025.doi:10. 5281/ZENODO.3372769.url:https://zenodo.org/doi/10.5281/ zenodo.3372769

    2025

  66. [67]

    2025.doi:10.5281/ ZENODO.15693706.url:%5Curl%7Bhttps://zenodo.org/doi/10

    Nicholas LaHaye, Kelly Luis, and Michelle Gierach.MultiSensor Harm- ful Algal Bloom Severity and Speciation Dataset. 2025.doi:10.5281/ ZENODO.15693706.url:%5Curl%7Bhttps://zenodo.org/doi/10. 5281/zenodo.15693706%7D

    2025

  67. [68]

    In:Remote Sensing17.7 (Apr

    NicholasLaHayeetal.“DevelopmentandApplicationofSelf-Supervised Machine Learning for Smoke Plume and Active Fire Identification from the Fire Influence on Regional to Global Environments and Air Quality Datasets”. In:Remote Sensing17.7 (Apr. 2025), p. 1267.issn: 2072- 4292.doi:10 . 3390 / rs17071267.url:http : / / dx . doi . org / 10 . 3390/rs17071267

    2025

  68. [69]

    2025.doi:10

    Nick LaHaye et al.SITFUSE V2.1.0. 2025.doi:10 . 5281 / ZENODO . 17117149.url:%5Curl % 7Bhttps : / / zenodo . org / doi / 10 . 5281 / zenodo.17117149%7D

    2025

  69. [70]

    2025.doi:10

    Alistair Miles et al.zarr-developers/zarr-python: v3.1.3. 2025.doi:10. 5281/ZENODO.3773449.url:https://zenodo.org/doi/10.5281/ zenodo.3773449

    2025

  70. [71]

    2025.doi:10

    Nick LaHaye.HAB Model Weights. 2025.doi:10 . 57967 / HF / 5837. url:%5Curl%7Bhttps://huggingface.co/njlahaye/OC_SIF_HABs_ 2025%7D

    2025

  71. [72]

    2025.doi:10

    Even Rouault et al.GDAL. 2025.doi:10 . 5281 / ZENODO . 5884351. url:https://zenodo.org/doi/10.5281/zenodo.5884351. 36

    doi:10.5281/zenodo.5884351 2025

  72. [73]

    https : / / ui

    Amanda Burke, Mark Carroll, and Caleb Spradlin.Finding the Trees in a (Random) Forest: How Do We Get a Representative Sample in a Training Dataset for a Global Land Cover Classification? — ui.adsabs.harvard.edu. https : / / ui . adsabs . harvard . edu / abs / 2023AGUFMIN51C0429B / abstract. [Accessed 09-10-2025]. [74]GitHub - gongzg/opencv-itseez: Open So...

    2025

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.