In the context of food security, the interdepdence of climate variables during the growing season at the country level plays a critical role in understanding the global price stabilization mechanisms, namely stockholding and international trade.
The commonly used measure of dependence is the covariance matrix. However, estimation of covariance matrix poses significant challenges because the length of time series is generally smaller than the number of countries being considered. As a result, the empirical estimate of covariance matrix is very unstable. While the spatial stationary assumption on covariance matrix, as commonly made in geostatistics, does not account for nonstationarity such as teleconnection patterns.
We therefore propose a spatial basis function approach by projecting the sample covariance matrix into a low-dimensional space produce by judiciously chosen spatial knot points. By this way, we can preserve the long range dependence structure observed in data while making the inference reliable by spatial regularization. We illustrate the usefulness of the proposed model by analyzing the country-wise temperature and precipitation.
|