Spatial delineation approach to weather derivatives with three multivariate manners

The delineation of natural hazard with adequate manners is an important factor and a dominant process in designing weather derivatives. This research conducts and compares three delineation techniques with real-time weather data from 2013 to 2015, collected from monitoring stations of the Central Weather Bureau and Environmental Protection Administration in Taiwan. Three multivariate analyses are performed, including principal component analysis (PCA), positive matrix factorization (PMF), and hierarchical cluster analysis (HCA). Results reveal the optimal number for temperature hazard delineation in Taiwan is three, composed of north of central Taiwan, the southwest, and southeast Taiwan; eigenvalues of the three sub-regions are greater than 1 and explanatory power of the variance accounted for temperature is 44.6%, 27.5%, and 24.7%, respectively. Though the regional delineation conducted by HCA is not as good as PCA, however, HCA is a useful technique to provide objective numbers of delineation. Analysis with PMF lacks uniqueness due to several stations having high factor loadings at distinct principal components. Additionally, PMF is not appropriate for high-dimension problems and non-uniform distributions of monitoring sites.