ideas for visualisation of model output

[timadriaens]

@amyjsdavis @DiederikStrubbe as discussed during the core group meeting, here are some ideas for visualisation of the usual model output (on top of projections for current climate and rcp scenarios 2.6 and 4.5) that could serve the risk assessment of alien species in Belgium, based on Beckmann et al. 2019 Projection of climatic suitability for Pycnonotus cafer establishment. Unpublished. (SDMs prepared for the EC PRA project, based on the method of Chapman et al. 2019.

• map with uncertainty on model predictions for Belgium (standard deviation in projected suitability). Example:

• bar plots showing the proportion of grid cells classified as suitable under current climate and projected climate. For Belgium, I would use Atlantic and continental bioregion, as well as the Belgian ecoregions. This can e.g. inform areas for targeted surveillance. Example:

but then with these regions:

• bar plots showing the proportion of grid cells classified as suitable under current climate and projected climate inside and outside Natura 2000. This could provide support to the risk assessor on questions relating to occurrence in protected areas. To do this, you can use the grid codes prepared by @damianooldoni who did an intersect with the protected areas (see #54)

For the EU PRA's, projections were classified into suitable and unsuitable regions using the ‘minROCdist’ method, which minimizes the distance between the ROC plot and the upper left corner of the plot (point (0,1)). No idea how that works though, perhaps there are other methods.

Further ideas:

• a limiting factor map following [Elith et al. (2010)](Elith J, Kearney M, Phillips S (2010) The art of modelling range-shifting species. Methods in Ecology and Evolution, 1, 330-342). Such map shows the most strongly limiting factor for establishment estimated by the model. I think this one is less 'important' for PRA-ing but it is informative.

• tabular summaries of variable importance, an example:

[qgroom]

Thanks Tim, I like the bar chart approach. One thought... Are uncertainties more-or-less the same within a species group? For example, because we know which grid cells birds were well surveyed in we know more about the uncertainty of bird recording, than we know about the present-absence of any one species of bird. Would it therefore be better to produce a single uncertainty map for all birds, rather than a multitude of ones for each model?

[timadriaens]

I assume uncertainties are species-specific. In the EU modelling approach, the uncertainty in the ensemble projections is expressed as the among-algorithm standard deviation in predicted suitability, averaged across 10 datasets (10 samples of 5000 randomly sampled pseudo-absences weighted by recording effort). How is this done in the models @amyjsdavis ?

[amyjsdavis]

I need some time to review this approach and respond.

On Mon, Oct 21, 2019 at 10:52 AM Tim Adriaens notifications@github.com wrote:

I assume uncertainties are species-specific. In the EU modelling approach, the uncertainty in the ensemble projections is expressed as the among-algorithm standard deviation in predicted suitability, averaged across 10 datasets (10 samples of 5000 randomly sampled pseudo-absences weighted by recording effort). How is this done in the models @amyjsdavis https://github.com/amyjsdavis ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/trias-project/risk-modelling-and-mapping/issues/5?email_source=notifications&email_token=AC4KXBXFAUL5HJCEQCGJ423QPVUWNA5CNFSM4JCIRMD2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEBZSHPI#issuecomment-544416701, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC4KXBTMYSXMXNRCIS4IML3QPVUWNANCNFSM4JCIRMDQ .

-- Dr. Amy J.S. Davis Data-driven solutions to invasive species and biodiversity conservation

Terrestrial Ecology Unit Department of Biology Ghent University K. L. Ledeganckstraat 35 B-9000 Ghent Belgium http://amyjsdavis.com

[qgroom]

When you have standard surveys that give present/absence data then all the errors for every species are the same. I'm not clear in this case, but my feeling with presence only data is that the uncertainty decreases with the number of occupied grid cells. Perhaps @amyjsdavis knows whether there is a cut off where the number of grid cells is too small to give a reasonable model i.e the uncertainty is just too high for the model to be useful. Also, is the relationship between numbers of presences and uncertainty linear? I suspect it plateaus at some value and it would be nice to know where this is. Quentin