Unfortunately, no...

I only did statistical validation (4 different measures) of my SD(Abundance)Ms using a 30% holdout. http://www.ace-eco.org/vol9/iss2/art7/ but you may find my paper useful b/c of the validation metrics used. It's hard to validate SAMs (cf SDMs).

Potentially relevant: https://esajournals.onlinelibrary.wiley.com/doi/full/10.1890/11-0826.1

Papers cited in this paper may point you to a few examples: https://pdfs.semanticscholar.org/0a31/1eaa99c20a1c88625cea0249f5430b895932.pdf

Ditto here: Article Assessing transferability of ecological models: An underappr...

I'm curious though: how do you see #2 as different from simply building the SDM in the first place?

Thanks! Am looking specifically for field-validation with independent dataset.

#2 (model calibration) is only relevant if you're trying to categorize predictions or set thresholds for a particular conservation objective. So, if you're trying to categorize abundance into levels of viability, or trying to categorize habitat suitability into classes (for example: 'poor, fair, good, excellent'). You need to calibrate the model to set these classes and it should correspond to something relating to actual habitat needs or viability of the species, and it is best if that dataset is independently collected from the dataset used to build the model.

I am not sure this will help, but we attempted to develop regulatory criteria for benthic invertebrate communities in Puget Sound, Washington, USA.

I've worked on fish habitat-suitability modeling, including transferability to other sites, in the following 2 pubs. that used statistical &/or GIS-mapping methods for cross-validation:

Rubec, P.J., and 15 coauthors. 2003. Development and evaluation of methods for habitat suitability modeling in Florida estuaries. Prepared for the U.S. Fish and Wildlife Service. Atlanta, GA. 69 pp. & app. (https://www.researchgate.net/publication/318116628).

Vadas, R.L. Jr., and D.J. Orth. 2001. Formulation of habitat-suitability models for stream-fish guilds: do the standard methods work? Transactions of the American Fisheries Society 130: 217-235 (http://www.ag.auburn.edu/auxiliary/alcfwru/fisheries/fish7380/vadas_orth2001.pdf).

-Bob Vadas, Jr.

These studies are a lot rarer than they should probably be but here is one: Article Field validation of habitat suitability models for vulnerabl...

Quite a specific topic but illustrate how sdm results should be considered carefully.

Hope it helps

This paper might be what you're looking for:

Article Field validation of an invasive species Maxent model

Dear all, I would like to share my recent paper where I analyzed part of such aspects. In particular, I focused my review and my study case on the reliability of SDMs for guiding the search of new localities of rare species. Test AUCs were based on presence records obtained after model-guided field surveys. I only used the most applied method for rare species (Maxent), which was, once again, revealed as a reliable tool. Nonetheless, I'm intrigued by the effect of using real absence data. So, next step would be comparing model techniques...

If you have some questions about this research, you can contact me.

Hope it would be of interest...

Article Using species distribution models at local scale to guide th...

Hi all,

You might be interested in a couple of papers we recently published on temporal transferability of SDMs:

Article Hindcasting the impacts of land‐use changes on bird communit...

Article Assessing the temporal transferability of raptor distributio...

Dear Alana,

in this paper:

(Article Comparing pseudo-absences generation techniques in Boosted R...

we used a presence/true absence dataset, using recently collected field data. We assumed the detectability as constant, because the field work was carried out, by the same people, in the context of a Natura 2000 project. We used these reliable data to assess the best algorithm to generate pseudo-absences in BRTs. We tested models' performance through AUC and two different TSS.

I am aware that it is not exactly what you were looking for, but i believe that it could help!

Cheers,

Mattia

Here is one: Article Field validation of an invasive species Maxent model

Try this one

A comparative evaluation of presence-only methods for modeling species distribution

• July 2007
• Diversity and Distributions 13(4):397-405
• DOI:
• 10.1111/j.1472-4642.2007.00346.x

Here's another one: Article Species distribution models predict rare species occurrences...

here is another one :

Article Future climate-driven in distribution of Calanus finmarchicus

Thank you everyone!!

want to share this

Article Genetic trials improve the transfer of Douglas-fir distribut...

Here, you can find another one:

Article Optimization of forest sampling strategies for woody plant s...

Dear Alana,

sorry for the late reply. We did something similar to steps 1-2-3 in an 'old' paper using a simple logistic regression model for a passerine species:

Article GIS-models work well, but are not enough: Habitat preference...

Good work, hope to read soon your paper!

Best regards to everyone,

Mattia

Dear Alana,

Try this

Article Evaluation by prediction of the natural range shrinkage of Q...

Article Contrasting climate risks predicted by dynamic vegetation an...

Dear Alana,

For a marine species example (humpback whales), our latest paper applied #1 and #2 from your list. Some of the #3 has been explored since then but is not published in the same paper.

Best wishes,

Solène

Another reference you may find useful is:

Fois et al. 2018. Using species distribution models at local scale to guide the search of poorly known species: Review, methodological issues and future directions. Ecological Modelling.

(and references therein)

Article Using species distribution models at local scale to guide th...

Ovaskainen, O., Tikhonov, G., Dunson, D., Grøtan, V., Engen, S., Sæther, B.-E. and Abrego, N. 2017b. How are species interactions structured in species rich communities? A new method for analysing time-series data. Proceedings of the Royal Society B: Biological Sciences, 284, 20170768.

Well, if you have 500 occurrence points, you take out 100 random points and you do your model training with 400 (including splitting it to a cal/training) then you have 100 independent samples.

The only caveat on this is if points were collected at the same time and therefore are part part of the same sampling effort. To avoid it, you could just remove those 100 points belonging to one single collection of data. Then they are independant.

All SDM are empirical input-output models and the SDM community should start moving towards knowledge on machine learning.

Dear Alana,

We just published a paper related to your question and especially the point #3 (niche based sampling).

Article Adaptive niche‐based sampling to improve ability to find rar...

One of the interests of the method is to allow for the kind of model checking you are raising in #3. We show such an evaluation in the results with a simple method. You can also check for the introduction where we discuss the rarity of #3 in the litterature, you will find 3-4 citations : the few papers we found doing it !

Hi Alana,

Please find our recent publication in Ecological Modelling at the link to follow (https://www.sciencedirect.com/science/article/abs/pii/S030438002200240X).

We validated our SDM models using independent data from another continent where our focal species has been intentionally introduced to see if native-range only GPS data could predict its establishment in the new geographic region.

Kind regards,

Guy