Author(s): Suroso Suroso; Andras Bardossy
Linked Author(s):
Keywords: Precipitation; Singapore; Spatial dependence; Non-Gaussian copula
Abstract: Gaussian copula is a very popular and simple spatial model where dependence structure is determined completely by correlation coefficient matrix. Gaussian copula, however, obviously cannot capture a non-symmetric dependence structure of multivariate distributions. If in the real world data set shows asymmetric dependence structure, a non-Gaussian copula-based model is more realistic way to mimic this nature. The objective of this research is to investigate characteristic of spatial dependence of precipitation amounts using non-Gaussian and Gaussian copulas. A V-copula is used to represent a non-Gaussian copula constructed from a Gaussian copula through non-monotonic transformation using two additional parameters; parameters m and k. Both copulas are implemented in Singapore using precipitation fields at different temporal resolutions from hourly to monthly. Precipitation occurrences analyzed in this study are all selected time events on a given time step at which more than 0.7 of all gauge stations have precipitation depth of more than 0.1 mm. The parameters estimations are conducted using the maximum likelihood method incorporating zero precipitation amounts, which are treated as censored variables. Empirical evidence proves that spatial correlation lengths for V-copulas are systematically higher than Gaussian copulas. Precipitation amounts for most time events exhibit positive asymmetric spatial dependence structures where precipitation amounts with high values are more spatially correlated than low values. In addition, precipitation with high-intensity tends to exhibit a stronger positive asymmetric spatial dependence structure than low-intensity ones. This implies that precipitation with high intensity tends to occur in a spatially clustered manner due to local convective precipitation.
Year: 2018