Spatial considerations when monitoring reef fishes |
| |
Authors: | D Parker H Winker ATF Bernard MKS Smith A Götz |
| |
Institution: | 1. Branch: Fisheries Management, Department of Agriculture, Forestry and Fisheries (DAFF), Cape Town, South Africa;2. Centre for Statistics in Ecology, Environment and Conservation (SEEC), Department of Statistical Sciences, University of Cape Town, South Africa;3. South African Institute for Aquatic Biodiversity (SAIAB), Makhanda, South Africa;4. Department of Zoology and Entomology, Rhodes University, Makhanda, South Africa;5. South African National Parks, Rondevlei Scientific Services, Sedgefield, South Africa;6. South African Environmental Observation Network (SAEON), Elwandle Node, Port Elizabeth, South Africa |
| |
Abstract: | Spatial dependence can obscure relationships between response and explanatory variables because of structuring within the residuals reducing variance and biasing coefficient estimates. Here, we highlight the influence of the spatial component, in the presence of spatial dependence, on abundance trends. This is illustrated using abundance data for a Critically Endangered reef fish, dageraad Chrysoblephus cristiceps, which were obtained from a long-term monitoring programme in the Tsitsikamma National Park marine protected area, South Africa. Correlograms illustrate distinct spatial structuring in the abundance data, and spatial variables were determined as more important than temporal variables when ranked according to predictive power using a random forest analysis. A generalised additive model (GAM) that did not account for spatial dependencies was compared to a generalised additive mixed model (GAMM) that incorporated a spatial residual correlation structure. Results derived from the spatially explicit GAMM differed considerably from the GAM lacking a spatial component, with the latter deemed to produce over-precise and partially biased abundance trends. The study emphasises the importance of space in accurately modelling abundance estimates, particularly temporal trends, and provides an introduction to the minimal statistical requirements necessary to address the violations associated with spatial autocorrelation. |
| |
Keywords: | abundance trends Chrysoblephus cristiceps correlogram CPUE generalised additive model marine ecosystem spatial autocorrelation variogram |
|
|