Taking a systems approach to ecological systems - Wiley Online Library

3 downloads 0 Views 566KB Size Report
Increasingly, there is interest in a systems-level understanding of ecological problems, which requires the evaluation of more complex, causal hypotheses.
Journal of Vegetation Science 26 (2015) 1025–1027

COMMENTARY Taking a systems approach to ecological systems James B. Grace

Grace, J.B. ([email protected] ): U.S. Geological Survey, 700 Cajundome Blvd., Lafayette, LA, USA

Abstract Increasingly, there is interest in a systems-level understanding of ecological problems, which requires the evaluation of more complex, causal hypotheses. In this issue of the Journal of Vegetation Science, Soliveres et al. use structural equation modeling to test a causal network hypothesis about how tree canopies affect understorey communities. Historical analysis suggests structural equation modeling has been under-utilized in ecology.

Ecology and society are moving from a historical emphasis on individual processes to a concern about entire systems. The complexity of the world in which we live, along with the increased level of our ambitions, now cause us to want to understand systems and predict, as much as we can, their behaviour. Understanding systems requires approaches that permit both the discovery and extrapolation of system structure. Analytically, systems are often represented as networks of interacting elements; thus, the business of studying systems can be approached using methods for studying causal networks. One approach to studying causal networks is structural equation modeling (SEM). The SEM approach possesses certain characteristics that could be useful in many studies in plant ecology, some of which are illustrated in a recent paper by Soliveres et al. (see this issue of the Journal of Vegetation Science). As a point of contrast, it is easiest to describe SEM in comparison to traditional statistical modeling. The traditional statistical model allows for the explanation of an individual response variable y or vector of response variables Y as some function of a vector of explanatory variables X (i.e. Y = f(X)). In contrast, SEM models take the form Y = f(X, Y), thus permitting hypotheses to be in the form of networks. One result of the limited form of traditional models is an inflexibility that precludes specification of a causal explanation for why predictors might or might not be correlated (Ruffell et al. 2015). SEM, however, allows for more complete hypotheses including causal relations among predictors. SEM also involves a graphical representation of the model that accompanies the equations, whose purpose is to make explicitly clear the causal assumptions. There are two other properties that emerge from the use of networks: one is the ability to pose system hypotheses and the other the ability to investigate causal relationships. SEM is also flexible with regard to statistical

specifications, permitting a full range of the response and linkage forms of modern statistical models. In their paper, Soliveres et al. (2015) investigate a hypothesis about how tree canopies impact understorey plants. The authors were particularly interested in whether tree effects on soil properties might indirectly impact the understorey. They were also interested in how contingent factors, particularly rainfall and grazing, might alter understorey responses to trees. To address this, a sample of 2000 plots were arrayed across 100 sites spread along a gradient in rainfall ranging from 220 to 1400 mm located in southeastern Australia. At each site, two large Eucalyptus trees were selected and plots were arranged along micro-gradients extending from trunk, to mid-canopy, to canopy edge and in the open. In addition to measuring understorey cover, species composition and species richness, they also characterized 12 relevant soil attributes and quantified indicators of grazing by mammals. The authors chose SEM as the framework for their study. The starting point for their analyses was an a priori hypothesis about how soil characteristics varied with distance from the trunks of trees and how interactions with rainfall and grazing might alter those responses. They reported that approximately 50% of the effects of trees on understorey species composition could be attributed to impacts by trees on soil properties. In turn, they pointed out that an additional 50% of the effect of trees on the understorey plant community was not accounted for, and suggested several other mediators for inclusion in future models. Regarding contingent factors, they found impacts of mammalian grazers on understorey cover (reduced) and richness (increased), but these grazers did not modify the effects of trees. Rainfall amount, in contrast, did alter understorey responses to tree canopies; in particular, there was evidence for reduced facilitation in higher rainfall areas. A wealth of other interesting findings and important

Journal of Vegetation Science Doi: 10.1111/jvs.12340 Published 2015. This article is a U.S. Government work and is in the public domain in the USA

1025

Commentary

J.B. Grace et al.

SEM. At present (2014), it is used in only 5% of JVS papers. It is interesting to reflect on causes for this apparent under-utilization. It is clearly not the case that vegetation scientists shy away from quantitative analyses, nor that their questions or data mismatch with the method. I believe the answer involves history. Figure 1 shows a map of the historical flow of information among different disciplines. A conspicuous feature of this map is its near donut shape. I would speculate from this, and based on the widths of the arrows, that information largely flows along well-established paths among familiar disciplines. Wright’s (1921) original description of SEM became very influential in disciplines

conclusions emerged from their evaluations, as detailed in their paper. Overall, their use of SEM not only created an interpretive framework for their study, but also for future studies of canopy–understorey interactions. SEM is used, at least to some degree, in virtually every field that employs statistical analysis, from economics to medicine, to artificial intelligence. The journal Structural Equation Modeling has been in existence since 1994. A bibliographic search of the Journal of Vegetation Science (JVS) papers, however, suggests that while there is increasing usage, there may be a much greater potential for the use of SEM in vegetation studies than seen thus far. Prior to 2006, only seven (of 1720) papers published in JVS used

Fluid Mechanics Material Engineering Circuits

Computer Science

Geosciences

Tribology

Astronomy & Astrophysics

Operations Research Computer Imaging

Power Systems

Mathematics

Physics

Telecommunication

Electromagnetic Engineering Control Theory Probability & Statistics

Economics

Chemistry

Business & Marketing

Geography

Environmental Chemistry & Microbiology

Chemical Engineering

Applied Acoustics

Analytic Chemistry

Psychology

Crop Science

Education

Sociology Pharmacology Law

Neuroscience

Political Science

Ecology & Evolution

Agriculture

Psychiatry

Environmental Health Anthropology

Medical Imaging Orthopedics

Molecular & Cell Biology

Veterinary

Dentistry

Medicine

Parasitology

Ophthalmology Gastroenterology

Citation flow from B to A

Citation flow within field

Otolaryngology Pathology

Urology Dermatology

Rheumatology

A Citation flow from A to B

B Citation flow out of field

Fig. 1. Citation map showing the historical flow of knowledge among disciplines based on cross-disciplinary citations (reprinted with permission from eigenfactor.org).

1026

Journal of Vegetation Science Doi: 10.1111/jvs.12340 Published 2015. This article is a U.S. Government work and is in the public domain in the USA

Commentary

J.B. Grace et al.

that focused on system behaviour, like economics and sociology. In contrast, the proximity of ecology to crop science in the citation map reflects an influence from the reductionist emphasis in agriculture that seeks to isolate individual effects and ‘control for’, either experimentally or statistically, a few factors of interest. The result: the natural scientist has to reach across unfamiliar disciplinary boundaries to find information related to SEM. Most of that information is embedded in example situations quite different from those encountered by those studying natural systems. Some attempted translations for natural scientists include Shipley (2000), Pugesek et al. (2003) and Grace (2006). However, most of the exposure vegetation scientists have to quantitative analysis is derived from the field of statistics, which is itself reductionist and descriptive. I would suggest that it might be worth the effort for vegetation scientists to give SEM a closer, more sincere examination because of the possibilities it presents, many of which are nicely illustrated in the paper of Soliveres et al.

Acknowledgements John Morgan, Meelis P€artel and James Cronin provided helpful suggestions for the manuscript. Supported by the

USGS Ecosystems and Climate and Land use Change Programs. The use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

References Grace, J.B. 2006. Structural equation modeling and natural systems. Cambridge University Press, Cambridge, UK. Pugesek, B.H., Tomer, A. & von Eye, A. 2003. Structural equation modeling: applications in ecological and evolutionary biology. Cambridge University Press, Cambridge, UK. Ruffell, J., Banks-Leite, C. & Didham, R.K. 2015. Accounting for the causal basis of collinearity when measuring the effects of habitat loss versus habitat fragmentation. Oikos doi: 10.1111/ oik.01948 (early online). Shipley, B. 2000. Cause and correlation in biology. Cambridge University Press, Cambridge, UK. Soliveres, S., Eldridge, D.J., M€ uller, J.D., Hemmings, F. & Throop, H.L. 2015. On the interaction between tree canopy position and environmental effects on soil attributes and plant communities. Journal of Vegetation Science 26: 1030– 1042. Wright, S. 1921. Correlation and causation. Journal of Agricultural Research 20: 557–585.

Journal of Vegetation Science Doi: 10.1111/jvs.12340 Published 2015. This article is a U.S. Government work and is in the public domain in the USA

1027