On December 3rd, we discussed the paper by Underwood et al. (2010). They take up an important issue related to systematic conservation planning, that is the distribution data used as a basis of reserve selection. If distribution data, as in this case, is relatively incomprehensive, reserve networks based on data may become very large (and thus very expensive), as there are many omissions and little flexibility. Using predicted distribution data aims at filling information gaps by detecting locations where a given species is likely to occur. This is supposed to add flexibility into the reserve selection and get rid of (at least a part of) the omissions in the original data. On the other hand, if there is large uncertainty in the predicted distributions of species and not the possibility to validate them, reserve selection with predicted distributions runs the risk of assigning priorities to locations where species do not actually occur.
Underwood et al. present two approaches to balance this tradeoff. First, at the modelling stage, they limit the predicted occurrences into the proximity of observed occurrences of species. They call these dispersal-limited predictions, as they constrain the predicted occurrences into locations which are considered to be within reach from the confirmed occurrences in the original data. Although this approach may reduce the risk of commission error, it leaves the predictions more sensitive to spatial bias in data collection than predictions made throughout the area. They compare the reserve networks based on dispersal-limited prediction data with ones based on the original observation data set and with predicted occurrences (Boolean vectors based on environmental variables) with no spatial constraints.
Second, they applied what they call the adaptive method in the reserve selection stage. The adaptive method is a nested selection scheme where 50% of conservation target for each species has to be met using the occurrence data, 30% is met using the dispersal-limited predicted data, and 20% can be met using predicted occurrences anywhere in the landscape. This is a very interesting approach, and we wished Underwood et al. had discussed the implications of their approach a bit more in depth. We were wondering if the occurrence data and predicted distributions Underwood et al. use are suitable for testing the approach further. As their data does not allow for confirming the predicted occurrences, it is not possible to compare the solutions in terms of their true efficiency. But it would be interesting and should be relatively easy to test their method with, for example, simulated data.
Underwood et al. show, importantly, that the type of data used in reserve selection has a great impact on the outcome, and the choice of whether one should use observation data or predicted occurrences is not a trivial one. Their reserve networks based on the alternative data sets had remarkably little overlap, although a small fraction of cells showing high irreplaceability were nearly always selected. Their results also show clearly, that when reserve selection is based on targets, the formulation of targets has a great impact on the available options from which the software can select conservation priorities from.
Full reference: Underwood, J.G., D’Agrosa, C. & Gerber, L.R. 2010. Identifying conservation areas on the basis of alternative distribution data sets. Conservation Biology 24: 162-170.
http://onlinelibrary.wiley.com/doi/10.1111/j.1523-1739.2009.01303.x/pdf