Extreme events are of interest for design and planning problems in engineering. Two hydrological extreme events, floods and droughts, are of concern for the effective development and management of water resources. Flood events have been studied extensively, while less work has been done on drought events due in part to the relative complexity of drought characteristics.

The main objective of this research is to investigate a method of bivariate frequency analysis of drought characteristics to better characterize the risk of natural disasters resulting from drought events. The drought characteristics of severity, duration and magnitude are derived from streamflow data using a variable truncation level. Pooled frequency analysis is applied to combine information of hydrologic drought events for a group of sites. The L-moment method is used to test the homogeneity and to derive the regional marginal distribution of drought characteristics for each pooling group. Synthetic drought characteristics derived from generated streamflow are also applied to validate the marginal distribution of drought characteristics. The dependence structure of drought characteristics is modelled via a 2-copula function. Simulation is used for model evaluation to check whether the model can reproduce the general pattern of the original correlated drought characteristics.

The method is applied to a section of the Athabasca River, a sub-basin of the Mackenzie River basin. Based on the marginal distribution and the selected copula function, the joint and conditional probability distribution are derived. Results indicate that the model is suitable for representing the joint distribution of the drought characteristics.