Application of Bayesian spatial smoothing models to assess agricultural self-sufficiency

With the rising oil prices, climate change, and the ever increasing burden of nutrition-related disease, food security is of growing research interest in academic disciplines spanning agronomy to epidemiology to urban planning. Some governments have developed progressive policies encouraging individuals to consume locally produced foods in order to support local economies, improve agricultural sustainability and community access to food, and to plan and prepare for adverse environmental impacts on food security. However, fundamental methods are lacking for conducting research on food security across these various disciplines. In this article, we first present a method to measure agricultural self-sufficiency, which we refer to as our self-sufficiency index (SSI) for the province of British Columbia, Canada. We then present a Bayesian autoregressive framework utilizing readily available agricultural data to develop predictive smoothing models for the SSI. We find that regional capital investment in agriculture and cropland acreage is the strong predictor of SSI. To accommodate spatial variability, we compare linear regression models with spatially correlated errors to less traditional spatially varying coefficient models, and find that the former class results in better model fit. The smoothed maps suggest that relatively strong self-sufficiency exists only in subset clusters in the Okanagan, Peace River, and lower mainland regions. In spite of policy to promote local food, the existing local agricultural system is insufficient to support a large-scale shift to local diets. Our approach to estimating neighborhood-based self-sufficiency with a predictive model can be extended for use in other regions where limited data are available to directly assess local agriculture and benefit from explicit consideration of spatial structure in the local food system.