Downscaling of weather generator parameters to quantify hydrological impacts of climate change

Abstract

A major obstacle in quantifying the hydrological impacts of climate change is the mismatch between the coarse resolution of climate model outputs (general circulation models and regional climate models) and the fine resolution requirements of hydrological models. This research presents a statistical downscaling approach combining the attributes of both the stochastic weather generator (WG) and the change factor (CF) method to overcome this problem. It is further compared against the commonly used CF method in terms of quantifying the hydrological impacts of climate change over the next century for a Canadian watershed (Quebec Province). Both downscaling methods suggested increases in winter (November−April) discharge and decreases in summer (June−October), especially for those downscaled by the WG-based method. The WG-based method predicted higher peak discharges than the CF method. The 2 downscaling methods suggested significantly different increases in annual and seasonal discharges, particularly for low flows. Hydrology results show that precipitation and temperature variability play a very important role in the runoff generating process, and that neglecting to address this variability can lead to biased results. In particular, the WG-based method has a significant advantage in simulating low flows because it takes into account the change of precipitation occurrence. The results also outline the uncertainty linked to the choice of a downscaling method. KEY WORDS: Statistical downscaling · Stochastic weather generator · Change factor method · Hydrology · Climate change Resale or republication not permitted without written consent of the publisherClim Res 51: [185][186][187][188][189][190][191][192][193][194][195][196][197][198][199][200] 2012 too many small events resulting in an overestimation of the wet day frequency, as well as biases resulting in an under/overestimation of the mean annual precipitation (Rivington et al. 2008). Dynamical and statistical downscaling methods have been developed to resolve this issue. Dynamical downscaling was developed based on dynamic formulations using the initial and time-dependent lateral boundary conditions of GCMs to achieve a higher spatial resolution by nesting regional climate models (RCM) (Caya & Laprise 1999). The spatial resolution of RCMs are much higher than that of GCMs. However, the resolution remains too coarse for small and medium-sized watersheds as well as for site-specific impact studies. Moreover, RCM data is only available for limited regions, due to the large computational cost of running the models (Solman & Nunez 1999). To overcome this problem, statistical downscaling methods have been developed. These involve linking the states of some variables representing a large scale (GCM or RCM grid scale, predictors) to the states of some variables representing a much smaller scale (catchment or site scale, predictands). A range of statistical downscaling methods have been developed and can be classified into 3 categories: transfer function (Wi...

Stochastic generation of daily precipitation amounts: review and evaluation of different models

“…Long time series are used to obtain the true expectancy of a weather generator. Short time series could result in biases due to the random nature of the stochastic process (Chen et al 2012a). The precipitation amount of 0.1 mm is used as a threshold to determine whether a given day is wet or dry.…”

“…Over the last 2 decades, the most promising implementation of stochastic weather generators has been their use as downscaling tools for climate change impact studies, which is achieved by perturbing the weather generator parameters according to relative changes projected by global and regional climate models (e.g. Semenov & Barrow 1997, Wilks 1992, 1999a, Pruski & Nearing 2002, Zhang et al 2004, Qian et al 2005, Zhang 2005, Zhang & Liu 2005, Kilsby et al 2007, Chen et al 2012a). One advantage of using a stochastic weather generator relative to other statistical downscaling methods is that an ensemble of climate time series can be produced, thus permitting risk analysis of climate events and climate change impacts.…”

Combining Stochastic Weather Generation and Ensemble Weather Forecasts for Short-Term Streamflow Prediction

“…The Weather Generator of École de Technologie Supérieure (WeaGETS) (Chen et al 2012a), which is a WGEN-like (Richardson 1981;Richardson and Wright 1984) three-variate (precipitation, Tmax and Tmin) single-site stochastic weather generator programmed in Matlab, was used in this study. In WeaGETS, the parameters are estimated biweekly.…”

“…For a predicted rain day, a gamma distribution is used to generate daily precipitation amounts. The daily Tmax and Tmin are generated using a first-order linear Chen et al (2012a) and Richardson (1981) for complete details. WeaGETS has been tested extensively at several locations under various climates and found to be adequate at reproducing precipitation and temperature characteristics (Chen et al 2012a;Chen and Brissette 2014).…”

Validation of non-stationary precipitation series for site-specific impact assessment: comparison of two statistical downscaling techniques

“…(5), based on the variance of spatially downscaled monthly precipitation (σ 2 m ) (Wilks 1992(Wilks , 1999Chen et al 2012b). …”