COS 12-8 - Joint species distribution modeling of plant species in an agricultural landscape

Monday, August 7, 2017: 4:00 PM
E145, Oregon Convention Center
Knut Anders Hovstad, Department of Landscape and Biodiversity, NIBIO, Stjørdal, Norway
Background/Question/Methods

Species distribution modelling is often targeting single species and correlations in spatial patterns and environmental responses across species are usually not included in the models. This may be about to change as new methods for joint species distribution modelling are being developed. In this presentation, I will explore some of these possibilities using data on small-scale distribution in common plant species in an agricultural landscape. In a study area of 5 x 6 km in central Norway, presence-absence data on eight plant species characteristic for semi-natural grasslands were recorded in 930 plots of 10 x 10 meters.  The distribution of these species was then examined using a probit regression model in which the joint probabilities of occurrence could be estimated using the correlation matrix of a multivariate normal distribution. The approach was first developed by Pollock et al. (2014) and makes it possible to separate correlation in occurrence between species due to environmental variables from residual correlation. Finally, the outcome of the joint distribution model is compared with results from logistic regression models developed for each species and some of the differences between these approaches are highlighted.


Results/Conclusions

The studied species generally had a high correlation in occurrence caused by similarities in their responses to environmental variation. The presence of semi-natural grassland, litter cover, soil moisture and soil fertility were consistently important to predict the occurrence of all plant species included in the study. The estimated residual correlation in co-occurrence varied more among pairs of species. Sources of residual correlations in joint species distribution models can be biological interactions among the studied species, and environmental variables that influence occurrence but for some reason have not been included in the model. In this study, missing information on grassland management in the model may be a source of residual correlation. The results highlights the importance of semi-natural grasslands not only for plant species diversity but also for an ecosystem service like insect pollination as several of the study species provide important feeding resources for pollinators.