Dear community, I would like to ask for advice to apply a maxent in a particular case. Sampling was carried out in previously selected stations, environmental variables were measured, the stations were equidistant from each other and the target species was Octopus vulgaris.
Here are my doubts about this particular case, for example, in some stations, there is more than one presence, the presence was recorded with the coordinates of the station, for example, 4 presences would be reported in a coordinate value. I think this could create a problem of spatial autocorrelation, should I use only 1 presence per station?
Likewise, I created the ASCII files for the environmental variables, however, the selection of the pixel size is very important tough, Is there any criteria to select the pixel size, any paper that could help?
Finally all the variables are correlated, I could perform the maxent and only look at the final results, but I could not distinguish the impact that each individual variable has. Is there any solution that you can recommend me for this problem, maybe using a PCA instead? Thank you for your time, and sorry for my english.
@Ilyas: correct
Correlated environmental predictor variables are the rule. Ecosystems are indeed systems. There are at least four different methods to remove variables with a high correlation in the literature. I prefer the step-wise removal (in maxent and GLM; for various reasons) from the SDM to reduce the number of predictor variables (a.k.a. reduction of dimensionality/multi-collinearity). Please consult my bear and endemics articles on this site.
Spatial autocorrelation is the rule as well. First a question. From your description I guess that you have carried out a systematic Octopus survey and therefore you have both presence and absence observation points. Correct?
Dear luis
as for your first question Maxent will automatically removes duplicate points that fall into a single environmental cell, so even if the occurrences are not in a single point but within the spatial resolution of your environmental variable, they will be removed. Therefore in your case, 4 presences in one point will just be counted as one presence.
see the details here :
- Boria, R. A., Olson, L. E., Goodman, S. M., & Anderson, R. P. (2014). http://doi.org/10.1016/j.ecolmodel.2013.12.012
- Fourcade, Y., Engler, J. O., Rödder, D., & Secondi, J. (2014).
http://doi.org/10.1371/journal.pone.0097122
- Bystriakova, N., Peregrym, M., Erkens, R. H. J., Bezsmertna, O., & Schneider, H. (2012). http://doi.org/10.1080/14772000.2012.705357
- Kramer-Schadt, S., Niedballa, J., Pilgrim, J. D., Schröder, B., Lindenborn, J., Reinfelder, V.,Wilting, A. (2013). http://doi.org/10.1111/ddi.12096
as for your second question, i may suggest you read it here...
DOI: 10.1080/11263504.2012.685194
DOI: 10.1016/j.jnc.2013.08.001
@Ilyas: correct
Correlated environmental predictor variables are the rule. Ecosystems are indeed systems. There are at least four different methods to remove variables with a high correlation in the literature. I prefer the step-wise removal (in maxent and GLM; for various reasons) from the SDM to reduce the number of predictor variables (a.k.a. reduction of dimensionality/multi-collinearity). Please consult my bear and endemics articles on this site.
Spatial autocorrelation is the rule as well. First a question. From your description I guess that you have carried out a systematic Octopus survey and therefore you have both presence and absence observation points. Correct?
Thank you Ilyas, I will check it, and yes Hein, is was a systematic sample, I believe that a method like GAM or GLM would be a better approach, tough I was asked to apply maxent too, my guess is that they want to see the different results from different models, so Im trying to do a maxent the best I can.
this link may help you
Article Mapping Species Distributions with MAXENT Using a Geographic...
But there's no actual problem if you report spatial autocorrelation (SA) since it's just a feature of natural ecosystems. SA is a measure of the degree to which a set of spatial features and their associated data values tend to be clustered together in space (positive spatial autocorrelation) or dispersed (negative spatial autocorrelation).
Thank you guys for your comments, I will consider them. Cheers.
Dear Luis, review this papers:
Brodie et al. 2017. Seasonal forecasting of dolphinfish distribution in eastern Australia to aid recreational fishers and managers.
Brodie et al. 2015. Modelling the oceanic habitats of two pelagic species using recreational fisheries data.
Marin/Enriquez et al. 2017. Environmental modeling of occurrence of dolphinfish (Coryphaena spp.) in the Pacific Ocean off Mexico reveals seasonality in abundance, hot spots and migration patterns.
Rubec et al. 2015.Using Delta-Generalized Additive Models to Predict Spatial Distributions and Population Abundance of Juvenile Pink Shrimp in Tampa Bay, Florida.
Su et al. 2012. Environmental and spatial effects on the distribution of blue marlin (Makaira nigricans) as inferred from data for longline fisheries in the Pacific Ocean. Try using GAM or GAMM instead of MAXENT. In GAM you can include the spatial (and/or temporal) autocorrelation (review gam of mgcv pakage and gls package in R). As a simple way in GAM, To account for potential spatial autocorrelation you can include the geographical position, i.e. the latitude-longitude interaction term in the analyses.
https://www.fromthebottomoftheheap.net/2014/05/09/modelling-seasonal-data-with-gam/
http://bbolker.github.io/mixedmodels-misc/notes/corr_braindump.html
- Badges
- Science topic