Generating hourly mean areal precipitation times series with an at-site weather generator in Switzerland

Continuous hydrological simulation is a powerful approach for generating long-term series of river discharges used for hydrological analyses. This approach requires as inputs precipitation time series generated by a stochastic weather generator (WGEN) to simulate discharge time series. For small catchments where a lumped hydrological model is suitable, the weather generator needs to generate time series of mean areal precipitation (MAP). Here we assess the ability of an at-site hybrid WGEN to generate time series of MAP for a set of test areas ranging from 9 to 1,089 km\(^2\). The generator is composed of a model based on a Markov chain model used to generate time series of daily MAP, and a multiplicative random cascade used to disaggregate them to an hourly resolution. The work is carried out at several test locations in Switzerland with different precipitation regimes. The parameters of the model are estimated on the observed MAP time series extracted from CombiPrecip, a 1 km\(^2\) resolution radar-gauge product of precipitation assimilating rain gauges and radar data. For each test location and each test area, 100-year time series are generated and compared with the observed MAP time series. Whatever the location and spatial scale considered, the performance of the WGEN is satisfactory. The model reproduces the observed standard statistics and extreme precipitation of observed MAP very well. At an hourly resolution, better results are obtained at larger spatial scales, while no difference is noticed at a daily resolution. The study shows that using this hybrid WGEN is possible to model and generate MAP for areas ranging from 9 to 1,089 km\(^2\). Moreover, this particular WGEN is easy to implement for end-user applications. The modelling approach is even more promising as high-resolution gridded precipitation data are expected to become increasingly available worldwide, offering a source of data to calibrate the hybrid model.