Pith. sign in

REVIEW 4 major objections 5 minor 38 references

Modelling benthic animals in space and time using Bayesian Point Process with cross validation: the case of Holoturians

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read The paper argues that heterogeneous presence-only sea-cucumber surveys can be fused into one spatio-temporal log-Gaussian Cox process, and that a k-fold cross-validation scheme for point processes, extended to bounded spatial subregions…

desk verdict A solid spatio-temporal extension of point-process CV with CRPS, honestly applied to a sea cucumber dataset, but the habitat effect is confounded with survey method and the top models are nearly tied. read the letter →

arxiv 2506.01763 v2 pith:OWO64I7O submitted 2025-06-02 stat.AP stat.OT

classification stat.APstat.OT MSC 62M3060G5562F1562P12
keywords seacucumbersHolothuroidealog-GaussianCoxprocessspatio-temporalpointcross-validationCRPSINLAhabitatmodelling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that sea-cucumber positions collected by very different survey protocols across nine campaigns can be modelled together as a spatio-temporal log-Gaussian Cox process, and that the resulting models can be compared by out-of-sample predictive skill rather than by in-sample fit. To do this it extends k-fold cross-validation for point processes to a spatio-temporal setting, splitting each campaign's domain into bounded subregions and scoring raw residuals with the Continuous Ranked Probability Score. Applied to 240 model specifications from Giglio Island data, the procedure selects a model in which transplanted Posidonia meadow raises expected sea-cucumber intensity (posterior mean 0.308, 95% CI 0.113 to 0.504) while P. oceanica lowers it (posterior mean -0.388, 95% CI -0.528 to -0.248), with residual spatial correlation decaying around 56 metres. If the central claim is right, predictive cross-validation is a practical tool for comparing spatio-temporal point-process models in ecology, and sea-cucumber habitat associations can be estimated from heterogeneous presence-only surveys.

What carries the argument

The argument turns on the Log-Gaussian Cox Process (LGCP), a Poisson point process whose log intensity is a Gaussian random field, estimated with INLA through the SPDE approximation. The intensity is decomposed as campaign-specific random effects $\mu_t$, a habitat coefficient $\gamma$ for the indicator $z(s)$ of P. oceanica, covariate effects $x(s)^\top\beta$, and a shared spatial GP $w(s)$. The cross-validation step marks each point with a $K$-fold multinomial label; the training and validation processes then have intensities scaled by $(K-1)/K$ and $1/(K-1)$, so raw residuals $R^g_t = \|u^{g}_t\| - \int_{B^g_t}\lambda_t(s)\,ds$ can be computed on subregions and summarized by CRPS. This machinery is what turns a collection of heterogeneous presence-only surveys into comparable predictive models.

What would settle it

Re-estimate the selected model with a survey-method indicator (photogrammetry versus diver visual census) included as a covariate. If the posterior of $\gamma$ for P. oceanica shifts substantially or its credible interval crosses zero, or if out-of-sample CRPS improves materially, then the reported habitat effect is contaminated by method-specific detectability rather than being a pure habitat effect.

Watch

Extended reading notes

Core claim

The paper claims to be the first practical implementation of k-fold cross-validation for point processes in a spatio-temporal setting. The idea is to assign each observed location a random fold mark, treat the held-out fraction as an independent thinning of the process, and evaluate raw residuals on bounded spatial subsets; averaging the CRPS over folds gives a predictive score for each candidate model. When applied to the Holothurian data, the best-scoring model has a negative coefficient for P. oceanica (posterior mean -0.388, 95% CI -0.528 to -0.248), a positive coefficient for transplanted P. oceanica (0.308, 95% CI 0.113 to 0.504), and a spatial Gaussian process with range about 56.7 m. The paper also reports that ranking by CRPS differs from ranking by DIC, which it presents as evidence that predictive model comparison is preferable in this setting.

Load-bearing premise

The habitat comparison stands or falls on the assumption that the detection gap between photogrammetry and diver-based censuses is the same in every year and habitat; if that ratio drifts, the habitat coefficient absorbs the difference in survey methods, not the ecology.

Editorial extensions

If this is right

  • Spatio-temporal LGCP models can be compared by out-of-sample predictive score instead of DIC, and the ranking may change materially.
  • The selected habitat effects give concrete, testable predictions: restored P. transplanted meadow should be a hotspot of sea-cucumber intensity, while natural P. oceanica meadow should show lower counts after adjusting for campaign and spatial structure.
  • Residual spatial correlation decays around 56 metres, meaning local habitat features matter at that scale and higher-resolution covariates may improve predictions.
  • The same cross-validation scheme with bounded subsets and quadrature can be applied to any spatio-temporal point process with a computable posterior intensity.

Reading between the lines

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

  • Because in 2022 all retained points lie outside P. oceanica (photogrammetry) and in 2023 all lie inside it (diver census), the habitat coefficient is identifiable as a pure habitat effect only under a constant photogrammetry-to-diver detection ratio; adding an explicit survey-method covariate would test this.
  • The 2024 campaigns, which use both methods in both habitats, could serve as a natural holdout: fitting the habitat coefficient on 2024 alone and comparing it with the pooled estimate would reveal method contamination.
  • The bounded-subset CRPS residuals could be mapped to identify where a fitted LGCP over- or under-predicts, giving a spatial diagnostic for survey design.
  • A modest extension would use the cross-validation score to choose the number and placement of survey campaigns, optimizing future sampling effort rather than only ranking fixed models.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper develops a spatio-temporal Log-Gaussian Cox Process model for presence-only sea cucumber data collected over nine campaigns at Giglio Island, combining Structure-from-Motion photogrammetry and diver visual census. The methodological contribution is a k-fold cross-validation procedure for point processes, extended from Cronie et al. to time-varying bounded spatial subsets, evaluated through the Continuous Ranked Probability Score and implemented with INLA/inlabru. After testing 240 model specifications, the authors select a model with habitat covariates and campaign-specific random effects plus a shared spatial Gaussian process, and report habitat coefficients including a negative P. oceanica effect and a positive transplanted-meadow effect, with a GP range of about 56 meters.

Significance. If valid, the paper offers a useful, computationally efficient practical implementation of point-process k-fold cross-validation with CRPS for spatio-temporal LGCP models, which would be a genuine step beyond existing ecological SDM practice. The authors are transparent about the small CRPS differences, the instability of DIC rankings, and the computational cost, and the thinning/quadrature derivation in Section 2.5 follows standard theory. However, the ecological conclusion about P. oceanica habitat preference is not identified from the sampling design, and the cross-validation description leaves unclear whether models are refitted per fold. No code or data are released, so the claimed 'first practical implementation' cannot be independently verified. The methodological core is plausible, but the applied findings need reframing or additional identification.

major comments (4)
  1. [Section 2.3, Eq. (1); Section 3, Table 3] The identification of gamma as a habitat effect is not supported by the sampling design. In 2022 all campaigns are SfM surveys on D2 (z=0), in 2023 all are UVC surveys on D1 (z=1), and only 2024 uses both methods in both habitat types. gamma is therefore collinear with campaign effects in 2022-2023 and can absorb any method-specific detection ratio; Section 2.3 itself states that exp(gamma z(s)) 'corrects the intensity for differing sampling intensities within the P.oceanica region' and 'can be interpreted as a measure of the sampling effort.' Given this, the posterior mean -0.388 (95% CI -0.528 to -0.248) in Table 3 cannot be presented as a pure ecological habitat preference in Section 4. Please either include a survey-method indicator, restrict the habitat interpretation to the 2024 contrast, or explicitly reinterpret gamma as a combined habitat/detectability effect.
  2. [Section 2.5] The cross-validation description does not explicitly state that the model is refitted on each training fold. The text says that posterior samples of lambda_t are drawn 'from the posterior marginal distributions of the fitted model', which could mean the full-data posterior is used for all folds. If so, the CRPS values are not out-of-sample with respect to parameter learning, and the method no longer implements the Cronie et al. thinning scheme. Please specify the fitting protocol (e.g., 240 models times 5 folds, each refitted on u_train) and confirm that each fold's residual samples are conditional only on that fold's training data.
  3. [Section 3, Table 2] The CRPS ranking is not shown to be statistically meaningful: the top 10 models differ only in the fourth or fifth decimal (0.454320 to 0.454451), and no Monte Carlo standard errors or repeated-fold variability are reported for the CRPS estimates. With A=1000 posterior/quadrature draws, these differences may be pure Monte Carlo noise. The selection criterion then becomes 'covariates with 95% CIs excluding zero', which is a within-sample significance rule rather than a cross-validated predictive choice. Report uncertainty for CRPS differences, or restrict the claims to 'no strong predictive distinction among top models'.
  4. [Section 2.2] The assertion that excluding single-habitat years 'ensures that habitat comparisons are not confounded by sampling methodological inconsistencies across years' is contradicted by the resulting design: 2022 contributes only non-P. oceanica/SfM data, 2023 only P. oceanica/UVC data. The filtering removes within-method habitat overlap instead of creating it, and the only cross-method, cross-habitat comparison comes from 2024. The claim should be removed or reformulated as a limitation.
minor comments (5)
  1. [Section 2.3] The sentence 'C2,1 and C2.2 those from 2022' should read 'those from 2023', consistent with Table 1 and Table 3.
  2. [Section 3] The reference to 'Table ??' should be 'Table 2'.
  3. [Section 3 / Table 3] The text 'Campaigns in 2021 generally show positive values... campaigns from 2022 have strongly negative effects... Campaigns from 2023 fall in between' does not match the labels in Table 3: C1.* are 2022, C2.* are 2023, C3.* are 2024. Please correct the year labels in the discussion of campaign effects.
  4. [Figure 2 caption] The caption refers to 'slope (e) and depth (d)', but panels (d), (e), and (f) are the correlation plot and two conditional distributions; the panel labels should be checked and corrected.
  5. [Data availability] The data availability statement says data are available upon request, but no code repository is mentioned; providing the R-INLA/inlabru implementation and fold assignments would improve reproducibility of the cross-validation comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the cross-validation extension is genuinely out-of-sample, and the habitat-effect confounding is an identifiability limitation rather than a circular derivation.

full rationale

I find no circular derivation in this paper. The cross-validation procedure is explicitly attributed to Cronie et al. (2024), and the paper's extension consists of partitioning the spatial domain into bounded subsets and computing CRPS from raw residuals. The thinning intensities in Eq. (2), lambda_train = ((K-1)/K) lambda and lambda_val = (1/(K-1)) lambda_train, are the standard mark-thinning identities used to construct validation sets; they do not require refitting the target quantity and are not equivalent to the model comparison result. Model selection is performed on held-out validation folds, so the CRPS comparison is genuinely out-of-sample. The posterior estimates of habitat coefficients, including gamma for P. oceanica, are fitted parameters, not predictions derived from themselves. The authors are transparent that gamma 'corrects the intensity for differing sampling intensities within the P.oceanica region' and 'can be interpreted as a measure of the sampling effort,' which means the reported negative coefficient likely mixes detectability and habitat preference given the single-habitat campaigns in 2022 and 2023. However, that is an identifiability and confounding limitation, not a circular step: the model does not define gamma in terms of the CRPS outcome, nor does it fit a quantity and then rename it as a prediction. Self-citations appear for data sources and prior ecological surveys, but the load-bearing statistical derivation does not reduce to those citations. The paper also acknowledges the limited discriminative power of the covariates, which further supports that no forced equivalence between inputs and conclusions is present.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The model is a standard LGCP with a shared GP and campaign effects; the main data-dependent parameters are the habitat coefficients and GP hyperparameters listed above. Key axioms: the LGCP assumption, the thinning property, the shared GP across disjoint domains, the unbiasedness of the post-hoc exclusions, and the separability of habitat from survey method. No new physical or biological entities are introduced.

free parameters (7)
  • P. oceanica habitat coefficient gamma = -0.388 (posterior mean; 95% CI -0.528 to -0.248)
    Central inferential claim about habitat effect; estimated from data; identification confounded by survey method.
  • P. transplanted coefficient = 0.308 (posterior mean; 95% CI 0.113 to 0.504)
    Supports the finding that restored Posidonia transplants are positively associated with sea cucumber intensity.
  • Dead matte coefficient = -0.134 (posterior mean; 95% CI -0.239 to -0.028)
    Habitat effect entered in the selected model.
  • Intercept mu0 = -4.718 (posterior mean)
    Baseline log-intensity.
  • GP range rho = 56.72 m (95% CI 43.4 to 73.6)
    Spatial correlation decay; used to interpret residual structure.
  • GP standard deviation sigma = 2.05 (95% CI 1.598 to 2.61)
    Amplitude of shared spatial effect.
  • Precision of campaign random effects = 1.33 (posterior mean)
    Variance of campaign effects.
assumptions (6)
  • domain assumption Observed Holothurian positions are a realization of a log-Gaussian Cox process with log-intensity given by the linear predictor.
    Section 2.3; the entire inference depends on this.
  • standard math The thinning property in Equation (2) holds: assigning iid uniform marks yields training and validation sets that are still LGCPs with scaled intensities.
    Section 2.5, Eq. (2); from Poisson thinning, correct under the LGCP model.
  • domain assumption The shared spatial GP w(s) has the same realization across all campaigns, allowing pooling of spatial information across disjoint domains D1 and D2.
    Section 2.3; load-bearing for borrowing strength across years.
  • ad hoc to paper Post-hoc exclusion of 2022 P. oceanica observations and 2023 non-P. oceanica observations does not bias the model.
    Section 2.2; exclusions are justified by detectability but could induce selection bias.
  • ad hoc to paper The binary habitat indicator gamma captures sampling-intensity differences without a separate survey-method effect; habitat and method are confounded in 2022 and 2023 campaigns.
    Section 2.3, Equation (1); 2022 data are exclusively D2 (SfM) and 2023 data exclusively D1 (UVC), so gamma may absorb method effects.
  • domain assumption INLA/SPDE provides a sufficiently accurate approximation to the LGCP posterior.
    Section 2.4; computational approximation is treated as exact for inference.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Modelling benthic animals in space and time using Bayesian Point Process with cross validation: the case of Holoturians." pith.science (2026). https://pith.science/paper/OWO64I7O

@misc{pith2026250601763,
  author       = {Pith},
  title        = {Pith review of: Modelling benthic animals in space and time using Bayesian Point Process with cross validation: the case of Holoturians},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OWO64I7O}},
  note         = {Machine review of arXiv:2506.01763}
}
read the original abstract

Understanding the spatial distribution of Holothurians is an essential task for ecosystem monitoring and sustainable management, particularly in the Mediterranean habitats. However, species distribution modeling is often complicated by the presence-only nature of the data and heterogeneous sampling designs. This study develops a spatio-temporal framework based on Log-Gaussian Cox Processes to analyze Holothurians' positions collected across nine survey campaigns conducted from 2022 to 2024 near Giglio Island, Italy. The surveys combined high-resolution photogrammetry with diver-based visual censuses, leading to varying detection probabilities across habitats, especially within Posidonia oceanica meadows. We adopt a model with a shared spatial Gaussian process component to accommodate this complexity, accounting for habitat structure, environmental covariates, and temporal variability. Model estimation is performed using Integrated Nested Laplace Approximation. We evaluate the predictive performances of alternative model specifications through a novel k-fold cross-validation strategy for point processes, using the Continuous Ranked Probability Score. Our approach provides a flexible and computationally efficient framework for integrating heterogeneous presence-only data in marine ecology and comparing the predictive ability of alternative models.

Figures

Figures reproduced from arXiv: 2506.01763 by the authors.

Figure 1
Figure 1. Map of the study area placed at the north-east coast of the Giglio Island. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Spatial distribution of the variables (a) Slope, (b) Depth and (c) Habitat. The correlation [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Spatial distribution of specimens (a) and relative frequencies of observed sea cucumbers [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Average value of the residuals for the chosen model across the different sampling times [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 34 canonical work pages

  1. [1]

    Bachl, Finn Lindgren, David L

    Fabian E. Bachl, Finn Lindgren, David L. Borchers, and Janine B. Illian. inlabru: an R package for Bayesian spatial modelling from ecological survey data. Methods in Ecology and Evolution, 10:760–766, 2019. doi: 10.1111/2041-210X.13168

  2. [2]

    Residual analysis for spatial point processes (with discussion)

    Adrian Baddeley, Rolf Turner, Jesper Møller, and Martin Hazelton. Residual analysis for spatial point processes (with discussion). Journal of the Royal Statistical Society Series B: Statistical Methodology, 67(5):617–666, 2005

  3. [3]

    Approximating point process likelihoods with glim

    Mark Berman and T Rolf Turner. Approximating point process likelihoods with glim. Journal of the Royal Statistical Society: Series C (Applied Statistics) , 41(1):31–38, 1992

  4. [4]

    High-performance computing with terastat

    Edoardo Bompiani, Umberto Ferraro Petrillo, Giovanna Jona Lasinio, and Francesco Palini. High-performance computing with terastat. In 2020 IEEE Intl Conf on Dependable, Au- tonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiC...

  5. [5]

    Spatiotemporal prediction for log-gaussian cox processes

    Anders Brix and Peter J Diggle. Spatiotemporal prediction for log-gaussian cox processes. 19 Depth Slope Xcoord Ycoord P . T rans- planted Dead ’Matte’ Hard Bot- tom Sandy Bot- tom CRPS ✖ ✖ ✖ ✖ ✔ ✔ ✖ ✔ 0.454320 ✖ ✖ ✖ ✖ ✔ ✔ ✖ ✖ 0.454329 ✖ ✖ ✔ ✖ ✔ ✔ ✖ ✖ 0.454373 ✖ ✖ ✔ ✔ ✖ ✔ ✖ ✔ 0.454388 ✖ ✖ ✔ ✔ ✔ ✖ ✖ ✖ 0.454389 ✖ ✖ ✖ ✔ ✔ ✔ ✖ ✔ 0.454400 ✖ ✖ ✖ ✖ ✖ ✔ ✖ ✔ 0.454...

  6. [6]

    Assessment of the impact of salvaging the costa concordia wreck on the deep coralligenous habitats

    E Casoli, D Ventura, L Cutroneo, M Capello, G Jona-Lasinio, R Rinaldi, A Criscoli, A Bel- luscio, and GD Ardizzone. Assessment of the impact of salvaging the costa concordia wreck on the deep coralligenous habitats. Ecological Indicators, 80:124–134, 2017

  7. [7]

    A cross-validation-based statistical theory for point processes

    Ottmar Cronie, Mehdi Moradi, and Christophe AN Biscio. A cross-validation-based statistical theory for point processes. Biometrika, 111(2):625–641, 2024

  8. [8]

    Combining rov-based acoustic data and underwater photogrammetry to characterize hakon mosby mud volcano (barents sea) cold seep systems

    Luca Fallati, Giuliana Panieri, Claudio Argentino, Andrea Giulia Varzi, Stefan B¨ unz, and Alessandra Savini. Combining rov-based acoustic data and underwater photogrammetry to characterize hakon mosby mud volcano (barents sea) cold seep systems. In EGU General Assembly Conference Abstracts, page 10620, 2024

Show all 38 references
  1. [9]

    Bayesian inference and computing for spatial point patterns

    Alan E Gelfand and Erin M Schliep. Bayesian inference and computing for spatial point patterns. In NSF-CBMS regional conference series in probability and statistics , volume 10, pages i–125. JSTOR, 2018. 20 Parameter Mean SD 2.5% Median 97.5% Intercept -4.718 0.075 -4.864 -4.7...

  2. [10]

    Strictly proper scoring rules, prediction, and esti- mation

    Tilmann Gneiting and Adrian E Raftery. Strictly proper scoring rules, prediction, and esti- mation. Journal of the American statistical Association , 102(477):359–378, 2007

  3. [11]

    Assessment of sea cucumber populations from the aegean sea (turkey): First insights to sustainable management of new fisheries

    Mercedes Gonz´ alez-Wang¨ uemert, Mehmet Aydin, and Chantal Conand. Assessment of sea cucumber populations from the aegean sea (turkey): First insights to sustainable management of new fisheries. Ocean & Coastal Management , 92:87–94, 2014

  4. [12]

    Global knowledge on the commercial sea cucumber holothuria scabra.Advances in marine biology , 91:1–286, 2022

    Jean-Fran¸ cois Hamel, Igor Eeckhaut, Chantal Conand, Jiamin Sun, Guillaume Caulier, and Annie Mercier. Global knowledge on the commercial sea cucumber holothuria scabra.Advances in marine biology , 91:1–286, 2022

  5. [13]

    Evaluating probabilistic forecasts with scoringrules

    Alexander Jordan, Fabian Kr¨ uger, and Sebastian Lerch. Evaluating probabilistic forecasts with scoringrules. Journal of Statistical Software , 90:1–37, 2019. 21

  6. [14]

    Predictive inference based on markov chain monte carlo output

    Fabian Kr¨ uger, Sebastian Lerch, Thordis Thorarinsdottir, and Tilmann Gneiting. Predictive inference based on markov chain monte carlo output. International Statistical Review , 89(2): 274–301, 2021

  7. [15]

    Bayesian inference and model assessment for spatial point patterns using posterior predictive samples

    Thomas J Leininger and Alan E Gelfand. Bayesian inference and model assessment for spatial point patterns using posterior predictive samples. Bayesian Analysis, 12(1):1–30, 2017

  8. [16]

    Finn Lindgren, H ˚ avard Rue, and Johan Lindstr¨ om. An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach.Jour- nal of the Royal Statistical Society: Series B (Statistical Methodology) , 73(4):423–498...

  9. [17]

    Ecology of deposit-feeding animals in marine sediments

    Glenn R Lopez and Jeffrey S Levinton. Ecology of deposit-feeding animals in marine sediments. The quarterly review of biology , 62(3):235–260, 1987

  10. [18]

    Impact of the costa concordia shipwreck on a posidonia oceanica meadow: A multi-scale assessment from a population to a landscape level

    G Mancini, E Casoli, D Ventura, G Jona-Lasinio, A Criscoli, A Belluscio, and GD Ardizzone. Impact of the costa concordia shipwreck on a posidonia oceanica meadow: A multi-scale assessment from a population to a landscape level. Marine Pollution Bulletin , 148:168–181, 2019

  11. [19]

    Transplantation on a posidonia oceanica meadow to facilitate its recovery after the concordia shipwrecking

    G Mancini, D Ventura, E Casoli, A Belluscio, and GD Ardizzone. Transplantation on a posidonia oceanica meadow to facilitate its recovery after the concordia shipwrecking. Marine Pollution Bulletin , 179:113683, 2022

  12. [20]

    Integration of presence-only data from several sources: a case study on dolphins’ spatial distribution

    Sara Martino, Daniela Silvia Pace, Stefano Moro, Edoardo Casoli, Daniele Ventura, Alessan- dro Frachea, Margherita Silvestri, Antonella Arcangeli, Giancarlo Giacomini, Giandomenico Ardizzone, et al. Integration of presence-only data from several sources: a case study on dolphi...

  13. [21]

    Species distribution models with masking: The case of holothurians in a posidonia rich area

    Gianluca Mastrantonio, Daniele Ventura, Edoardo Casoli, Arnold Rakaj, Giovanna Jona Lasinio, Daniele Poggio, Cecilia Vitiello, and Crescenza Calculli. Species distribution models with masking: The case of holothurians in a posidonia rich area. In Scientific Meeting of the Ital...

  14. [22]

    Scoring rules for continuous probability distribu- tions

    James E Matheson and Robert L Winkler. Scoring rules for continuous probability distribu- tions. Management science, 22(10):1087–1096, 1976

  15. [23]

    Statistical rethinking: A Bayesian course with examples in R and Stan

    Richard McElreath. Statistical rethinking: A Bayesian course with examples in R and Stan . Chapman and Hall/CRC, 2018

  16. [24]

    Log gaussian cox processes

    Jesper Møller, Anne Randi Syversveen, and Rasmus Plenge Waagepetersen. Log gaussian cox processes. Scandinavian journal of statistics , 25(3):451–482, 1998

  17. [25]

    Viviana Pasquini, Cristina Porcu, Martina Francesca Marongiu, Maria Cristina Follesa, Am- bra Angelica Giglioli, and Pierantonio Addis. New insights upon the reproductive biology of the sea cucumber holothuria tubulosa (echinodermata, holothuroidea) in the mediterranean: Impli...

  18. [26]

    Ecological roles of exploited sea cucumbers

    Steven W Purcell, Chantal Conand, Sven Uthicke, and Maria Byrne. Ecological roles of exploited sea cucumbers. In Oceanography and marine biology , pages 375–394. CRC press, 2016

  19. [27]

    Mediterranean sea cucumbers—biology, ecology, and exploitation

    Arnold Rakaj and Alessandra Fianchini. Mediterranean sea cucumbers—biology, ecology, and exploitation. In The world of sea cucumbers , pages 753–773. Elsevier, 2024

  20. [28]

    Approximate bayesian inference for latent gaussian models by using integrated nested laplace approximations

    H ˚ avard Rue, Sara Martino, and Nicolas Chopin. Approximate bayesian inference for latent gaussian models by using integrated nested laplace approximations. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 71(2):319–392, 2009. doi: https://doi. o...

  21. [29]

    Inorganic carbon turnover caused by digestion of carbonate sands and metabolic activity of holothurians

    Kenneth Schneider, Jacob Silverman, Ben Kravitz, Tanya Rivlin, Aya Schneider-Mor, Sergio Barbosa, Maria Byrne, and Ken Caldeira. Inorganic carbon turnover caused by digestion of carbonate sands and metabolic activity of holothurians. Estuarine, Coastal and Shelf Science , 133:...

  22. [30]

    Spatio-temporal log-gaussian cox processes for modelling wildfire occurrence: the case of catalonia, 1994–2008

    Laura Serra, Marc Saez, Jorge Mateu, Diego Varga, Pablo Juan, Carlos D ´ ıaz- ´Avalos, and 23 H ˚ avard Rue. Spatio-temporal log-gaussian cox processes for modelling wildfire occurrence: the case of catalonia, 1994–2008. Environmental and ecological statistics, 21:531–563, 2014

  23. [31]

    Accounting for spatial varying sampling effort due to accessibility in citizen science data: A case study of moose in norway

    Jorge Sicacha-Parada, Ingelin Steinsland, Benjamin Cretois, and Jan Borgelt. Accounting for spatial varying sampling effort due to accessibility in citizen science data: A case study of moose in norway. Spatial Statistics, 42:100446, 2021

  24. [32]

    Going off grid: Computationally efficient inference for log-gaussian cox processes

    Daniel Simpson, Janine Baerbel Illian, Finn Lindgren, Sigrunn H Sørbye, and Havard Rue. Going off grid: Computationally efficient inference for log-gaussian cox processes. Biometrika, 103(1):49–70, 2016

  25. [33]

    Penalising model component complexity: a principled, practical approach to constructing priors

    Daniel Simpson, H ˚ avard Rue, Andrea Riebler, Thiago G Martins, and Sigrunn H Sørbye. Penalising model component complexity: a principled, practical approach to constructing priors. Statistical Science, 32(1):1–28, 2017

  26. [34]

    Seagrass posidonia oceanica (l.) delile as a marine biomarker: a metabolomic and toxicological analysis

    Chiara Toniolo, Antonella Di Sotto, Silvia Di Giacomo, Daniele Ventura, Edoardo Casoli, An- drea Belluscio, Marcello Nicoletti, and Giandomenico Ardizzone. Seagrass posidonia oceanica (l.) delile as a marine biomarker: a metabolomic and toxicological analysis. Ecosphere, 9(3):...

  27. [35]

    High spatial resolution underwater data for mapping seagrass transplantation: A powerful tool for visualization and analysis

    Daniele Ventura, Luca Castoro, Gianluca Mancini, Edoardo Casoli, Daniela Silvia Pace, An- drea Belluscio, and Giandomenico Ardizzone. High spatial resolution underwater data for mapping seagrass transplantation: A powerful tool for visualization and analysis. Data in Brief, 40...

  28. [36]

    Detecting habitat preferences and monitoring population dynamics of sea cucumbers in coastal ecosystems through underwater photogrammetry

    Daniele Ventura, Arnold Rakaj, Giovanna Jona Lasinio, Gian Mario Sangiovanni, Daniele Pog- gio, Gianluca Mastrantonio, Alessio Pollice, Gaia Grasso, Edoardo Casoli, Gianluca Mancini, et al. Detecting habitat preferences and monitoring population dynamics of sea cucumbers in co...

  29. [37]

    Poisson point process models solve the” pseudo-absence problem” for presence-only data in ecology

    David I Warton and Leah C Shepherd. Poisson point process models solve the” pseudo-absence problem” for presence-only data in ecology. The Annals of Applied Statistics, pages 1383–1402, 2010. 24

  30. [38]

    Point process models for spatio-temporal distance sampling data from a large-scale survey of blue whales

    Yuan Yuan, Fabian E Bachl, Finn Lindgren, David L Borchers, Janine B Illian, Stephen T Buckland, H ˚ avard Rue, and Tim Gerrodette. Point process models for spatio-temporal distance sampling data from a large-scale survey of blue whales. Annals of Applied Statistics , 11(4): 2...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.