Our research combines high-level ecological theory and spatial analysis to address fundamental questions in ecology and to improve tools for applied ecology and conservation biology. We are particularly interested in the distribution and abundance of species over space – a fundamental question in ecology and vital information for species management. Our past and current research have addressed topics such as autocorrelation in species distributions, effects of dispersal on distributions and particular range edges, abundance-occupancy relationships, partitioning out sources of variation in species’ distributions, evaluation of habitat-based distribution models and effects of environmental change on species distributions. Our future research will build on this by investigating the cause of spatial patterns in species distributions using field-based and theoretical approaches. Population dynamics and dispersal will play a prominent role in this work. Understanding species distributions in such a causal manner is vital for threatened species management, climate change planning and protected area design.
Species distributions and distribution modelling
Our main interest is to understand why species occur where they do. This question ties into traditional biogeography practiced for more than two centuries. It could also be seen as a central question of ecology (see for example the books by Andrewartha and Birch, 1954 and 1984, and Krebs 1972), insofar as ecology is the study of the interactions between organisms and their environment and these interactions depend strongly on the location of the organism. The location and abundance of species also play a prominent role in conservation efforts.
The question is so tightly linked to ecology because the interaction of organisms with their environment is dependant on their location. The environmental conditions, and with them the interactions depend on the location of an organism. At the same time, however, the location of the organism depends at least in part on the environmental conditions. Additional complications are that most organisms move at least in one stage of their life-cycle, live, reproduce and die, and environmental conditions tend to fluctuate. In short, conditions and distributions are dynamic at many different temporal and spatial scales.
Niche-theory based habitat modeling for distributions has been moderately successful. Perhaps its success has been somewhat overstated. Pretty much all other modeling strategies have not been developed out of infancy, the most likely reason for this being the lack of adequate distribution data to start from. Dynamic models, spatial models, and in general complex models, require better data than we have for almost any organism, time span and location.
Limiting or correlating gradients?
The interest in distributions and their models started with Bahn's work on Marbled Murrelets and modelling their nest distributions. Already back then, Brian Cade made me aware of the faulty paradigm in distribution modelling, which is the use of models of central tendency rather than limiting effects. I started using his quantile least absolute deviation regressions instead and currently work mainly with regression trees that are also good at finding limiting rather than correlating gradients.
Spatial patterns in species distributions and ranges
During my PhD I became more interested in spatial patterns in model residuals. This is a good place to start looking for the inadequacies of habitat models: are there any patterns left in the residuals? I found strong spatial autocorrelation in the residuals and became interested in the sources of this autocorrelation. One is doubtlessly autocorrelation in the underlying resources and environmental conditions. However, at least in theory, if all important environmental predictors were included in the model, autocorrelation should disappear from the residuals. In most situations it doesn't. Therefore, we are either missing major predictors, or autocorrelation is also caused by something else. My research indicates that the second is the case and my main "suspect" is dispersal in the widest sense. I have not yet proven this empirically, mostly because long-distance dispersal data is sparse, but I'm working on it.
Given that habitat models do not work all that well, I became convinced that other mechanisms play a more important role in species distributions than is usually assumed. In particular, I believe that population dynamics and dispersal are underappreciated as pattern forming processes. This could partly be caused by the current paradigm of modeling presence/absence rather than abundances in distribution modeling. Again, this is likely mostly caused by lack of abundance data. My impression is, however, that presence/absence modeling masks other processes underlying species distributions such as population dynamics and autocorrelation caused by dispersal, which are most easily seen in abundance distributions.
I am currently looking for better data and/or different methods to investigate how population dynamics and dispersal influence distribution patterns within species (e.g., comparing core to range edges) and between species.
The research described above has kept statistics high on my list of interests. I see statistics as a tool and not an end in itself. However, I am interested in staying up-to-date in statistics (without necessarily participating in every fad), so that I can select the most appropriate tool for the purpose and wield it efficiently. I mostly use R and S-PLUS®. My use of regression trees (and now also regression forests and boosting) is a legacy of my late PhD advisor Raymond O'Connor. As mentioned above, in addition to the regular array of statistical tools in ecology (correlations, regressions, t-tests, ANOVA, PCA, etc), quantile regressions were one of the earlier advanced techniques that caught my interest. Later, I developed a strong interest in spatial statistics (variograms, kriging, general linear models, conditional autoregressive regressions etc.). In addition, I have mistrusted traditional null hypothesis tests from early on in my statistics career and now rely on Akaike's Information Criterion, bootstrapped confidence intervals, and cross-validation (or reserved data).
The scope of my interest in marbled murrelets is fairly easy to explain: I would like to determine the characteristics of breeding habitat they a) select for, and b) have the highest fitness in under natural conditions. These are deceptively simple sounding questions to which a decade of intensive research throughout the Pacific coast of Canada and the US has contributed surprisingly little.
While many nests of this elusive bird have been found by now, most have been found by non-random searches in fragmented landscapes. The microhabitat requirements of murrelets have been fairly well explored through these nests - they mostly nest on platforms formed by large branches of large and old trees and cushioned by moss - but macrohabitat requirements and preferences as well as breeding success in different habitats is still poorly known.
Several reasons are responsible for the difficulty of describing "optimal" breeding habitat for marbled murrelets. Initially and possibly still most importantly, the elusiveness of the nests and the cryptic behaviour of the murrelets were the most significant obstacles to habitat requirements research. Murrelet nests are simply very difficult to locate. This has led to biased sampling (search for nests were researchers suspected them to be) and low sample sizes. A second difficulty is the nest selecting behaviour of murrelets. They do not seem to concentrate nesting efforts in specific habitats. Rather, I came to the impression that murrelets purposefully space their nests, possibly for predator avoidance, and thereby occupy a large range of different habitats, as long as they can find the necessary nesting structures. A final difficulty is that they fly large distances and to a wide range of altitudes to locate nest sites. This means that habitat characteristics need to be mapped for very large tracts of land in the order of thousands of square km. Mappings at such a scale typically do not contain detailed information on the vegetation cover. Relating murrelet nesting to coarse habitat characteristics such as distance to forest edge, distance to ocean, elevation, slope, exposition, and similar mostly topographic information, doesn't seem to lead to very good models. (However, again it is unclear which of the above reasons is most responsible for the relatively poor models.)
With no indication in sight that murrelets concentrate in "optimal" breeding habitat, I believe that conservation must focus on saving large quantities of habitat over a whole range of conditions and locations. Unfortunately, this goal is in substantial conflict with the commercial interests in the murrelets' native forests. These commercial, and with them governmental, interests want to see the murrelets protected but only if it can be done without impact on the land base for forest exploitation. For reasons given above, such a conservation scheme (as in B.C.'s Identified Wildlife Management Strategy) can only fail and, if not improved, in the long term we will have to get used to the thought of losing most of the murrelet population of the world and retaining only critically endangered populations in a few larger conserved areas, as is now the case in California and Oregon.
If I would go back into murrelet research, I would try to find more nests in Clayoquot Sound, using radio telemetry similar to the 32 nests found by the SFU researchers and used in their paper (Zharikov 2006) and my paper (Bahn and Lank in prep.). I think that a sample size of about 200 could lead to really good results in this pristine area, and as shown in Desolation Sound, such a sample size is not unrealistic. So why not simply use the Desolation Sound data? In Desolation Sound a good proportion of the lower forests were logged so that nesting patterns there could be biased away from low elevations. An area correction for the lost habitat can hardly make up for this loss, because it is unclear whether murrelets that used to nest in low habitats now bunch in the remaining low habitats, or if they simply moved upward to maintain low nesting densities, or if they possibly even fled the disrupting activities in the valley and aggregated in the higher elevation habitats. In addition, Clayoquot Sound has better habitat mapping available (Vegetation Resource Inventory).
The analytical problem with murrelet nesting habitat is well known in the field of distribution modeling: No information on true absences is available. Given the wide range of habitats that murrelets seem to use, this lack of true absences weighs in especially heavily.