See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/363039315

Contrasting occupancy models with presence-only models: Does accounting

for detection lead to better predictions?

Article in Ecological Modelling · August 2022

DOI: 10.1016/j.ecolmodel.2022.110105

CITATIONS READS

10 431

3 authors:

Ashish Jha Nameer, Prof. PO

Wildlife Institute of India Kerala Agricultural University
27 PUBLICATIONS 58 CITATIONS 262 PUBLICATIONS 1,941 CITATIONS

SEE PROFILE SEE PROFILE

Praveen Jayadevan
Nature Conservation Foundation
153 PUBLICATIONS 893 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ashish Jha on 28 August 2022.

The user has requested enhancement of the downloaded file.

 Ecological Modelling 472 (2022) 110105

Contents lists available at ScienceDirect

Ecological Modelling
journal homepage: www.elsevier.com/locate/ecolmodel

Short communication

Contrasting occupancy models with presence-only models: Does accounting

for detection lead to better predictions?
Ashish Jha a, *, Praveen J b, P.O. Nameer a
a
College of Climate Change and Environmental Science, Kerala Agricultural University, Thrissur, Kerala, 680656 India
b
KeralaBirder, Ambadi, Kunnathurmedu, Palakkad, Kerala, 678013 India

A R T I C L E I N F O A B S T R A C T

Keywords: Species distribution models are popular statistical tools for inferring potential distribution range of species across
Kerala bird atlas space and time and are extensively used in conservation planning. Models based on presence-only data (e.g.,
MaxEnt MaxEnt) are widely used; however, these models assume perfect species detectability. Occupancy modelling is
SDM
considered a better modelling technique since it accounts for species detectability. Presence-only models are
Site occupancy
relatively simpler, requiring only presence locations, while occupancy models are data hungry models requiring
Species detection
Western Ghats detection/non-detection data from multiple visits to the survey sites. We utilized data from the Kerala Bird Atlas
(India) and modelled current distribution for 109 species using MaxEnt and occupancy approaches. MaxEnt
performed well even with less occurrences, while occupancy model failed for species with fewer than 40 records.
In terms of evaluation metrics, AUC and Root Mean Square Error, both models performed relatively better for
species with low occurrences than those with high occurrences (generalist species). The comparison metrics
(Relative-rank scores, Root Mean Square Error, Hellinger distance and Expectation of Shared Presences) were
significantly correlated with the number of occurrences; MaxEnt and occupancy based SDMs for widespread
species had more concordance than SDMs of narrowly distributed species. There was some discordance between
algorithms with regards to diversity hotspot. Selection of best combination of variables and correction for
overprediction can help improve performances of both models and improve consistency between them. Given the
data hungry nature of occupancy models and marginal difference with the MaxEnt models; it appears that latter
is better suited for predicting the distribution of rare species and studies dealing with cumulative data from
multiple species.

1. Introduction can be extrapolated in space and time (Elith and Leathwick, 2009). A
variety of algorithms are available to predict the spatial distribution of
Accurate information on species distribution and ecological con­ species but the ‘presence-only’ SDM approaches (e.g., MaxEnt, Random
straints are prerequisite for effective conservation measures (Margules Forest, Generalized Additive Models) are particularly popular since
and Pressey, 2000). Extensive on-ground surveys across a species’ dis­ ascertaining absences with confidence is tedious (Elith et al., 2006).
tribution range to obtain this information are logistically infeasible. Ecologists have used SDMs for a variety of applications such as conser­
Statistical models called species distribution models or SDMs provide an vation planning and surveys, invasive species management, and pre­
alternative; SDMs describes habitat suitability in ecological space and dicting the impact of future climate change (Engler et al., 2017; Rahman
yield a potential distribution map when projected into geographic space et al., 2019; Jha and Vasudevan, 2020; Sreekumar and Nameer, 2021).
(Phillips et al., 2006). SDMs achieve this by correlating field observa­ An effective SDM can accurately capture occurrence-environment
tions (presences only or presences/absences or relationship and can reliably predict the species occurrence under
presences/pseudo-absences or presences/background points) to envi­ given environmental conditions. However, most SDM algorithms as­
ronmental predictor variables, based on statistically or theoretically sume a constant detection probability across sites; failure to account for
derived response surfaces (Guisan and Zimmermann, 2000). The imperfect detection can compromise estimated prediction maps (Tyre
correlative SDMs predict distributions across geographical space and et al., 2003; Gu and Swihart 2004). Occupancy models offer a substitute

* Corresponding author.
E-mail address: [email protected] (A. Jha).

https://doi.org/10.1016/j.ecolmodel.2022.110105
Received 30 March 2022; Received in revised form 6 August 2022; Accepted 11 August 2022
0304-3800/© 2022 Elsevier B.V. All rights reserved.
A. Jha et al. Ecological Modelling 472 (2022) 110105

only a marginal improvement in the performance of occupancy interests or personal relationships that could have appeared to influence
approach over other SDM approaches (Rota et al., 2011; Perkins-Taylor the work reported in this paper.
and Frey, 2020).
Performances of both modelling approaches, as judged using AUC Data availability
and RMSE scores, were correlated with the number of species’ occur­
rences; abundant species consistently performed poorer than less Data is available from Dryad repository and supplementary files. The
abundant species. This feature of MaxEnt algorithm has been noted same has been mentioned in the manuscript.
across many studies (Rota et al., 2011; Connor et al., 2017). For over
3/4th of the species, the relative-rank score between occupancy and
MaxEnt models was above 0.8. Gormley et al. (2011) reported a strong Acknowledgements
positive correlation (rs =0.89) between the rankings of cells by the two
methods for Sambhar deer (Rusa unicolour). Relative-rank score was not Funding for the data analysis and publication was provided by the
related to number of occurrences, but Hellinger distance, ESP and RMSE Duleep Matthai Nature Conservation Trust, Gujarat, and is gratefully
scores were. It suggests that congruence between the two SDM ap­ acknowledged.
proaches is not a general rule and varies across species. Both SDMs
predicted high elevation areas of Kerala (the Western Ghats) as areas of Supplementary materials
high biodiversity, however, MaxEnt predicted a much larger area than
occupancy model. Supplementary material associated with this article can be found, in
SDM overprediction, depending on the extent of analysis, can lead to the online version, at doi:10.1016/j.ecolmodel.2022.110105.
misleading prioritization in conservation planning (Velazco et al.,
2020). Correction of SDM overprediction is crucial, particularly when References
extent of analysis is large. MaxEnt has inbuilt ‘clamping’ feature and
regularization parameter to constrain features to remain within the Altwegg, R.&., Nichols, J., 2019. Occupancy models for citizen-science data. Methods in
Ecol. Evolution 10, 8–21. https://doi.org/10.1111/2041-210X.13090.
range of values in the training data and restrict extrapolation (Elith BirdLife International., Country profile: india. 2022. Available from http://www.birdlife.
et al., 2011). Given the computational time constraint, we had set org/datazone/country/india. Checked: 2022-03-05.
‘clamping’ to true and selected best fitting regularization parameter for Broms, K.M., Hooten, M.B., Johnson, D.S., Altwegg, R., Conquest, L.L., 2016. Dynamic
occupancy models for explicit colonization processes. Ecology 97 (1), 194–204.
all 109 species from a set of 5 values (0.5, 1, 2, 3, 4). We did not include https://doi.org/10.1890/15-0416.1.
additional habitat features that can affect detections such as canopy Connor, T., Hull, V., Viña, A., Shortridge, A., Tang, Y., Zhang, J., Wang, F., Liu, J., 2017.
cover and undergrowth density; these were not available for the survey Effects of grain size and niche breadth on species distribution modelling. Ecography
(Cop.) 41 (2018), 1270–1282.
sites. Addition of these data could have altered the performance of both Chen, G., Li, X., Liu, X., 2022. Global land projection based on plant functional types with
algorithms and the final SDMs. Thus, there is potential to further fine a 1-km resolution under socio-climatic scenarios. Sci. Data 9 (1), 125. https://doi.
tune both models for individual species. org/10.1038/s41597-022-01208-6.
Elith, J., Leathwick, J.R., 2009. Species distribution models: ecological explanation and
Data is expensive. It is relatively easier to get occurrence records;
prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40 (1), 677–697.
temporal replication of survey is more resource intensive. But is it https://doi.org/10.1146/annurev.ecolsys.110308.120159.
worth? It has been said that performance of a model should be assessed Elith, Jane, Hastie, Trevor, Dudík, Miroslav, Chee, Yung En, Yates, Colin, 2011.
with respect to its objectives or else the comparative studies may fail to A statistical explanation of MAXENT for ecologists. Diversity and Distributions 17,
43–57. https://doi.org/10.1111/j.1472-4642.2010.00725.x.
reveal the benefits of accounting for detectability (Lahoz-Monfort et al., Elith, J., Graham, C.H., Anderson, R.P., Dudík, M., Ferrier, S., Guisan, A., et al., 2006.
2014). Use of occupancy models is unlikely to result in improved pre­ Novel methods improve prediction of species’ distributions from occurrence data.
dictive performance when detection probability is relatively homoge­ Ecography 29, 129–151.
Engler, J.O., Stiels, D., Schidelko, K., Strubbe, D., Quillfeldt, P., Brambilla, M., 2017.
neous across sites (Rota et al., 2011). Our study shows that for majority Avian SDMs: current state, challenges, and opportunities. J. Avian Biol. 48,
of the species, presence-only MaxEnt model performs as good as 1483–1504.
occurrence model. Future research can explore the effect of tuning Godsoe, W., 2014. Inferring the similarity of species distributions using species’
distribution models. –Ecography 36, 130–136.
variable and parameters, correction for bias in occurrences, and cor­ Gormley, A.M., Forsyth, D.M., Griffioen, P., Lindeman, M., Ramsey, D.S.L., Scroggie, M.
rections to reduce overfitting on consistency between species richness P., Woodford, L., 2011. Using presence-only and presence-absence data to estimate
maps from MaxEnt and occupancy models. Not occupancy, but MaxEnt the current and potential distributions of established invasive species. J. Appl. Ecol.
48 (1), 25–34. https://doi.org/10.1111/j.1365-2664.2010.01911.x.
model perform well with small sample sizes which is particularly the Gu, W.D., Swihart, R.K., 2004. Absent or undetected? Effects of non-detection of species
case for rare species. Hence, for studies involving rare species and when occurrence on wildlife-habitat models. ? Biol. Conserv. 116, 195–203.
dealing with multiple species data, MaxEnt should suffice. For Guillera-Arroita, G., et al., 2015. Is my species distribution model fit for purpose?
Matching data and models to applications. Glob. Ecol. Biogeography 24, 276–292.
species-specific studies, occupancy modelling can be a better approach.
Guisan, A., Zimmermann, N.E., 2000. Predictive habitat distribution models in ecology.
This information is important in a conservation context. Ecol. Model. 135, 147–186.
Hijmans R.J., Phillips S., Leathwick J. and Elith J. (2020) dismo: species distribution
modelling. R package version 1.3-3. https://CRAN.R-project.org/package=dismo.
Author’s contribution
Fiske, I., Chandler, R, 2011. unmarked: an R package for fitting hierarchical models of
wildlife occurrence and abundance. J. Stat. Softw. 43 (10), 1–23. URL. https://www.
All authors contributed to the study conception and design. Data jstatsoft.org/v43/i10/.
collection was done by Praveen J and P. O. Nameer. Data analysis was Jha, A., Vasudevan, K., 2020. Environmental niche modelling of globally threatened
Yellow-throated bulbul for conservation prospects in the Deccan Peninsula, India.
performed by Ashish Jha, the first draft of the manuscript was written by Current Sci. 119 (11), 1815–1823.
Ashish Jha. All authors commented on previous versions of the manu­ Karger, D.N., Conrad, O., Böhner, J., Kawohl, T., Kreft, H., Soria-Auza, R.W.,
script. All authors read and approved the final manuscript. Ashish Jha: Zimmermann, N.E., Linder, H.P., Kessler, M., 2017. Climatologies at high resolution
for the earth’s land surface areas. Sci. Data 4, 170122. https://doi.org/10.1038/
Conceptualization, Methodology, Software, Writing- Original draft sdata.2017.122.
preparation Praveen J: Data collection and curation, Project adminis­ Ko, Chia-Ying, Ruey-Shing, Lin, Tzung-Su, Ding, Chih-Hao, Hsieh, Pei-Fen, Lee, 2009.
tration, Supervision P. O. Nameer: Data collection and curation, Funding Identifying biodiversity hotspots by predictive models: a case study using Taiwan’s
endemic bird species. Zoological Stud. 48 (3), 418–431.
acquisition, Supervision, Resources Lahoz-Monfort, José J., Guillera-Arroita, Gurutzeta, Wintle, Brendan A., 2014. Imperfect
detection impacts the performance of species distribution models. Glob. Ecol.
Declaration of Competing Interest Biogeography 23 (4), 504–515. https://doi.org/10.1111/geb.12138.

The authors declare that they have no known competing financial

6
A. Jha et al. Ecological Modelling 472 (2022) 110105

Liu, C., White, M., Newell, G., 2011. Measuring and comparing the accuracy of species Román, M.O., Wang, Z., Sun, Q., Kalb, V., Miller, S.D., Molthan, A., Schultz, L., Bell, J.,
distribution models with presence– absence data. Ecography 34 (2), 232–243. Stokes, E.C., Pandey, B., Seto, K.C., et al., 2018. NASA’s black marble nighttime
https://doi.org/10.1111/j.1600-0587.2010.06354.x. lights product suite. Remote Sens. Environ. 210, 113–143. https://doi.org/10.1016/
MacKenzie, D., Nichols, J., Royle, J., Pollock, K., Bailey, L., et al., 2006. Occupancy j.rse.2018.03.017.
Estimation and modelling: Inferring patterns and Dynamics of Species Occurrence. Rota, C.T., Fletcher, R.J., Evans, J.M., Hutto, R.L., 2011. Does accounting for imperfect
Academic Press, Burlington, MA, p. 324. detection improve species distribution models? Ecography 34 (4), 659–670. https://
MacKenzie, D.I., et al., 2002. Estimating site occupancy rates when detection doi.org/10.1111/j.1600-0587.2010.06433.x.
probabilities are less than one. Ecology 83, 2248–2255. Rushing, C.S., Royle, J.A., Ziolkowski, D.J., Pardieck, K.L., 2019. Modelling spatially and
Margules, C.R., Pressey, R.L., 2000. Systematic conservation plan-ning. Nature405 temporally complex range dynamics when detection is imperfect. Sci. Rep. 9, 12805.
243–253. https://doi.org/10.1038/s41598-019-48851-5.
Merow, C., Smith, M.J., Silander Jr, J.A., 2013. A practical guide to MaxEnt for Sadoti, G., Albright, T.P., Johnson, K., 2017. Applying dynamic species distribution
modelling species’ distributions: what it does, and why inputs and settings matter. modelling to lek-mating species. J. Biogeogr 44 (1), 75–87. https://doi.org/
Ecography 36, 1058–1069. 10.1111/jbi.12886.
Olson, G.S., Anthony, R.G., Forsman, E.D., Ackers, S.H., Loschl, P.J., Reid, J.A., Sing, T., Sander, O., Beerenwinkel, N., Lengauer, T., 2005. ROCR: visualizing classifier
Dugger, K.M., Glenn, E.M., WJ, 2005. Ripple Modelling of site occupancy dynamics performance in R. Bioinformatics 21 (20), 7881. <URL: http://rocr.bioinf.mpi-sb.
for northern spotted owls, with emphasis on the effects of barred owls. J. Wildlife mpg.de. >.
Manag. 69, 918–932. Sreekumar, E.R., Nameer, P.O., 2021. Impact of climate change on two high-altitude
Perkins-Taylor, Ian E., Frey, Jennifer K., 2020. Predicting the distribution of a rare restricted and endemic flycatchers of the Western Ghats, India. Current Sci. 121,
chipmunk (Neotamias Quadrivittatus Oscuraensis): comparing MaxEnt and 1335–1342. https://doi.org/10.18520/cs/v121/i10/1335-1342.
occupancy models. J. Mammal. 101 (4), 1035–1048. https://doi.org/10.1093/ SoIB, 2020. State of India’s birds, range, trends and conservation status. The SoIB
jmammal/gyaa057. Partnership 50.
Peterson, A.T., Soberon, R.G., Pearson, R.P., Anderson, E., Martinez-Meyer, M., Tantipisanuh, N., Gale, G.A., 2018. Identification of biodiversity hotspot in national level
Nakamura, Araujo, M.B, 2011. Ecological Niches and Geographic Distributions. - Importance of unpublished data. Glob. Ecol. Conserv. 13 https://doi.org/10.1016/
Princeton University Press, Princeton, New Jersey, USA. j.gecco.2018.e00377.
Phillips, S.J., Robert, P.A., Robert, E.S., 2006. Maximum entropy modelling of species Tyre, A.J., Tenhumberg, B., Field, S.A., Niejalke, D., Parris, K., Possingham, H.P., 2003.
geographic distributions. Ecol. Modell. 190 (3–4), 231–259. https://doi.org/ Improving precision and reducing bias in biological surveys: estimating false-
10.1016/j.ecolmodel.2005.03.026. ISSN 0304-3800. negative error rates. ? Ecol. Appl. 13, 1790–1801.
Praveen, J., Nameer, P.O., Jha, A., et al., 2022. Kerala bird atlas 2015-2020: features, Valavi, R., Guillera-Arroita, G., Lahoz-Monfort, J.J., Elith, J., 2022. Predictive
outcomes and implications of a citizen-science project. Curr. Sci. 122 (3), 298–309. performance of presence-only species distribution models: a benchmark study with
https://doi.org/10.18520/cs/v122/i3/298-309. reproducible code. Ecol. Monogr. 92 (1), e01486. https://doi.org/10.1002/
Raman, S., Shameer, T.T., Pooja, U., Hughes, A.C., 2022. Identifying priority areas for bat ecm.1486.
conservation in the Western Ghats mountain range, peninsular India. J. Mammal. Velazco, S.J.E., Ribeiro, B.R., Laureto, L.M.O., Júnior, P.D.M., 2020. Overprediction of
gyac060. https://doi.org/10.1093/jmammal/gyac060. species distribution models in conservation planning: a still neglected issue with
R Core Team, 2022. R: A language and Environment For Statistical Computing, 2020. R strong effects. Biol. Conserv. 252, 108822 https://doi.org/10.1016/j.
Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/. biocon.2020.108822.
Rahman, A.A.A., Mohamed, M., Tokiman, L., Sanget, M.S.M., 2019. Species distribution Wilson, P.D., 2011. Distance-based methods for the analysis of maps produced by species
modelling to assist biodiversity and conservation management in Malaysia. IOP distribution models. Methods in Ecol. Evolution 2, 623–633. https://doi.org/
Conf. Ser.: Earth Environ. Sci. 269, 012041 https://doi.org/10.1088/1755-1315/ 10.1111/j.2041-210X.2011.00115.x.
269/1/012041.

View publication stats