Monday, December 8, 2014

Identifying the correct spatial scale, a work in progress

It’s a truism in ecology that there is a spatial scale at which to each ecological process and interaction occurs. Competition often occurs at local scales, speciation generally occurs at biogeographic scales. Empirical evidence seems to support this - the relationship between, for example, forest cover and beetle abundance changes from strong to nonexistent as the spatial scale of analysis increases, suggesting small scales are most meaningful for the relationship (Holland et al 2004).

But do most ecological studies choose the appropriate scale for data collection and analysis? A new meta-analysis from Heather Bird Johnson and Lenore Fahrig suggests that ecological studies, even multi-scale studies, rarely do. Multi-scale studies can show how a relationship changes in strength with spatial scale, and so should be ideal for identifying the “intrinsic scale” or the “scale of effect” – the spatial extent that best predicts a particular ecological process. (Figure below)
From Jackson and Fahrig 2014. The scale of effect illustrated using a multi-scale study: the strength of the relationship between abundance and spatial scale is strongest at 4 km (radius).
Identifying the appropriate spatial scale for a question and system is of course ideal for a researcher. Researchers can then collect appropriate data, choose to focus on interactions with processes occurring at the same scale, and to correctly analyze data. However, the appropriate spatial scale may not be easy to identify: appropriate spatial scales must be included a multi-scale study. If a study includes spatial scales that cover too small or too large an extent, or has divides the extent into too few sub-scales, simply having a study with multiple scales may still be insufficient.

Theory suggests that species' traits--e.g. dispersal distances and reproductive rates--should be strongly related to the scale of effect, but empirical evidence is surprisingly inconclusive. If studies are already successfully identifying the scale of effect, the authors hypothesized that the observed scale of effect (the scale of prediction at which results are strongest) should vary with taxonomic group, body size, species’ mobility, reproductive rates, and species function. On the other hand, if the scale of effect is being inappropriately determined, perhaps due to decisions about the number of scales to include, and the minimum and maximum spatial scale considered, then these may be the primary correlate of the scale of effect.

To determine whether multi-scale studies were successfully identifying the scale of effect, the authors performed a literature review and meta-analysis. They identified studies that featured abundance or occurrence, which was measured at multiple nested spatial scales, for multiple taxonomic groups. The scales considered in these studies ranged from 10m to 100km.

The results were rather disappointing. By far the strongest predictors of the scale of effect in a study were the largest or smallest scales they considered. This suggests that the true scale of effect might be outside the scales considered by such studies. Worse, differences between taxonomic groups disappeared when you controlled for the minimum and maximum spatial scales used in a study. Where the same species appeared in several different studies, their scales of effect from each study were no more similar than if you had chosen any random species in the same taxonomic group.

From Jackson and Fahrig 2014. There were no significant differences between the observed scale of effect and taxonomic groups. Instead, the largest and smallest spatial scales evaluated in the study drove the conclusion about the scale of effect.
The good news is that the more scales that were considered in a study, the less likely it was that the minimum or maximum scales considered appeared to be driving the results.
From Jackson and Fahrig 2014. The more scales evaluated, the less likely that choice of minimum or maximum scale was driving the result.
In addition, the authors found that the relationship between observed scale of effect and species traits was weak to non-existent in most studies. This is particularly unfortunate given the recent focus on species traits as useful predictors of ecological relationships. The inability to correctly identify the scale of effect certainly may limit our ability associate spatial scale and traits. It is also likely that context modifies the effect of traits (for example, body size may have different effects on dispersal depending on the type of matrix and the environment), further weakening the observed relationship.

One of the largest issues Jackson and Fahrig identified is that in many of the papers, choice of scales was driving by methodological (data availability, precedent, etc.) issues rather than biological justifications. Questions about trait-scale relationships may well be unanswerable until studies have the data for a necessary range of spatial scales. Until then, Jackson and Fahrig recommend that studies be more forthright about their limitations, something this paper will hopefully draw attention to.