Performance of multisite stochastic precipitation models for a tropical monsoon region

Abstract

Stochastic weather generator (SWG) produces synthetic time series of weather data based on the statistical characteristics of observed weather for a given location. Although SWG models are extensively evaluated and applied in different hydro-climate related studies, they often ignore the spatial correlation between weather patterns observed at multiple locations. This can limit the value of some spatial impact assessments such as flood modeling, agricultural crop modeling, water resources management and urban infrastructure design. To address such limitations, multisite SWG models are implemented to preserve the spatial characteristics of weather variables. In this study, we compared the performance of three multisite stochastic precipitation models, which includes modified Wilks model (modWilks), RainSim V3 (RSIM) and perturbed K-Nearest Neighbor (pKNN) models. The performances of these models are investigated for a study area located in the tropical monsoon climate region over Central Highland, Vietnam. The models are evaluated based on their performance for simulating precipitation occurrence and amount statistics on a wet day, extreme cumulative wet/dry days, transition and joint probability of wet/dry state, cross-correlation across all sites as well as the behavior of precipitation amount in relation to neighboring station state. The performance of model depends on the type of the precipitation characteristics, for example, the RSIM model performed well in term of the mean precipitation intensity. Overall, the pKNN model outperformed other models in term of temporal statistics, spatial characteristics, as well as extreme events measured based on Intensity–Duration–Frequency (IDF) curves.