Harvard Forest >

Harvard Forest Symposium Abstract 2011

• Title: Testing causal/correlative chains in large-scale ecological datasets; development and application of Bayesian hierarchical path analysis
• Primary Author: Simon Smart (CENTRE FOR ECOLOGY AND HYDROLOGY)
• Additional Authors: Aaron Ellison (Harvard University); David Foster (Harvard University); Peter Henrys (CENTRE FOR ECOLOGY AND HYDROLOGY); Keith Kirby (Not Specified); Rob Marrs (Not Specified); Andy Scott (Centre for Ecology & Hydrology)
• Abstract:

  Simon M Smart1, W.Andy Scott1, Peter Henrys1, Rob Marrs2, Keith Kirby3, David Foster4, Aaron Ellison4
  
  
  
  1 CEH Lancaster, UK; 2 University of Liverpool, UK; 3 Natural England, UK; 4 Harvard Forest
  
  
  
  What is Bayesian hierarchical path analysis? Modern Bayesian methods enable model parameters to be estimated by repeated computer-intensive sampling and refining of a joint probability surface defined by initial, often uninformative, parameter estimates that are updated by observed data. Hierarchical means that the probability surface explicitly accounts for nested random effects such as samples from within a random selection of forests across a state. Path analysis allows explanatory variables to be linked in a plausible causal chain. For example, reduced atmospheric sulphur deposition drives a recovery of soil pH which in turn drives changes in species richness and composition.
  
  
  Why do it? These methods offer an appropriate statistical treatment of the complexity of multi-scale ecological observations. Importantly, they allow variables to be linked together to test hypothesized causal chains based on a mechanistic cascade of effects, with one factor driving the next which drives the next. As well as testing for the indirect effects of a driver on a response via another variable it is also possible to simultaneously quantify any direct effects.
  
  
  Example 1) We have been able to quantify the direct and indirect effects of land-use and pollution drivers on spatial variation of ecosystem service indicators across potential biodiversity refugia in lowland Britain. As an example, Figure 1 shows that all significant effects suppressed richness of high conservation value plants (a cultural ecosystem service indicator). Succession to woodland and scrub in small remnant fragments of habitat (Hf) had the largest effect. The direct effect of intensive land use was also high in habitat fragments and on linear features (Ln) includings river/stream banks, roadsides and field boundaries.
  
  
  Example 2) The technique has also been used to quantify the impact of global change phenomena on changes in soils, tree canopy and ground flora in temperate broadleaved woodlands. The path diagram (Fig 2) shows the size direction and significance of direct and indirect effects on observed plant species richness change across British forests between 1971 and 2002. In this example, reductions in woodland management have a direct correlation with reduced ground flora richness but there is also an indirect effect via increase in tree basal area as canopies close and shade increases. Increasing tree growth was also correlated with a reduction in soil pH in turn correlated with reduced ground flora species richness.
  
  
  The boxes indicate site-level factors, the ellipses show plot-level factors. The numbers on the vertices are standardized regression coefficients.
  
  

• Research Category: Biodiversity Studies
  Conservation and Management
  Ecological Informatics and Modelling


• Figures:
U:Simon_SmartLTER_symp_posterFig_1.jpg
U:Simon_SmartLTER_symp_posterFig_2.jpg