Modeling ecological processes at global scales is complicated by the fact that available underlying spatial and temporal data often incorporate enormous uncertainty. For example, the WorldClim data set (which has been cited over 2000 times) offers 1-kilometer resolution globally whether the pixel contains a weather station or the nearest station is hundreds of kilometers away (Hijmans et al., 2005). Furthermore, most climate data used in ecological models are coarse temporal aggregations such as monthly means rather than variables that are more relevant to the process under study such as consecutive dry days or absolute minimum temperature. This is problematic in light of growing awareness that organisms may respond more to climate extremes and other climate metrics than mean values (Gutschick and BassiriRad, 2003; Jackson et al., 2009; Trnka et al., 2011).
YCEI postdoc Adam Wilson’s recent publication in International Climatology addressed this problem, demonstrating a Bayesian procedure that accounts for the uncertainties introduced by interpolation between data points. The process was computationally demanding – processing years of data from over 700 weather stations required over 20 processor-days to complete – but resulted in high predictive accuracies for key metrics used to predict plant germination and survival in remote areas of South Africa’s 90,000 square kilometer Cape Floristic Region, the focus of the study. The methodology has the potential for wide application in ecological models capable of incorporating and propagating data uncertainty through to model output and results. This will lead to projections with wider prediction intervals and a higher degree of confidence.
Wilson, Adam M., and John A. Silander. 2013. “Estimating Uncertainty in Daily Weather Interpolations: A Bayesian Framework for Developing Climate Surfaces.” International Journal of Climatology doi:10.1002/joc.3859.