Evaluating Stochastic Precipitation Generators for Climate Change Impact Studies of New York City’s Primary Water Supply

Abstract

Abstract Watersheds located in the Catskill Mountains of southeastern New York State contribute about 90% of the water to the New York City water supply system. Recent studies show that this region is experiencing increasing trends in total precipitation and extreme precipitation events. To assess the impact of this and other possible climatic changes on the water supply, there is a need to develop future climate scenarios that can be used as input to hydrological and reservoir models. Recently, stochastic weather generators (SWGs) have been used in climate change adaptation studies because of their ability to produce synthetic weather time series. This study examines the performance of a set of SWGs with varying levels of complexity to simulate daily precipitation characteristics, with a focus on extreme events. To generate precipitation occurrence, three Markov chain models (first, second, and third orders) were evaluated in terms of simulating average and extreme wet days and dry/wet spell lengths. For precipitation magnitude, seven models were investigated, including five parametric distributions, one resampling technique, and a polynomial-based curve fitting technique. The methodology applied here to evaluate SWGs combines several different types of metrics that are not typically combined in a single analysis. It is found that the first-order Markov chain performs as well as higher orders for simulating precipitation occurrence, and two parametric distribution models (skewed normal and mixed exponential) are deemed best for simulating precipitation magnitudes. The specific models that were found to be most applicable to the region may be valuable in bottom-up vulnerability studies for the watershed, as well as for other nearby basins.