Discrete-continuous downscaling model for generating daily precipitation time series
Wei Yang · OPUS Publication Server of the University of Stuttgart (University of Stuttgart) · 2008
This work was aimed to improve the generation of daily precipitation time series with information from atmosphere. The motivation is to develop a conditional stochastic downscaling model to well describe temporal and spatial behavior of local precipitation, in particular, the extreme rainfall events. Thereafter, the generated daily precipitation can be applied to integrate with other water-related models for climate impact studies. Global warming has been concerned since the end of last century. The increase of the temperature may result in various consequences, which may further cause the changes in agricultural production, water resources distributions and so on. Therefore, the impact studies must be carried out to understand the influence of the changing climate and predict its possible consequences in order to mitigate and adapt to the changing climate. In the field of hydrology, detailed information such as the situation of landuse, the condition of local climate and others is always required to describe the hydrological processes. Precipitation, especially, is of great concern due to its spatial and temporal variability. Basically, precipitation is a product of atmospheric motions and physical processes in the atmosphere on one hand and an important driving force in the land-atmosphere interactions on the other hand. It is the result of atmospheric movement. Therefore, it is reasonable to derive information about precipitation from the atmospheric studies. The global climate models (GCMs) do produce daily precipitation time series. However, due to their coarse resolutions and incomplete understanding of climate science, the outputs of GCMs can not properly describe the processes in detail for the local regions. A certain method is required to match the mismatch between two different scales and this method is “Downscaling”. In this thesis work, three statistical downscaling methods were explored: a CP- and Regression-based downscaling approach, a CP- and Copula-based downscaling approach and a multi-site weather generator. The first two methods were developed based on the circulation patterns. The circulation patterns can be obtained either from professional knowledge (subjective classification) or from statistical characteristics derived from the observations (objective classification). The scheme proposed and investigated here is based on fuzzy-rule logic. It is a method that works on the concept of fuzzy sets, describing the atmospheric circulations using imprecise statements. The circulation patterns are useful to capture the information at a large-scale, however, they are weak in capturing the continuity of the whole natural system. The additional predictors are therefore required. A combined term, moisture flux, is introduced into the pure CP-based downscaling model. It describes convey of the water vapor by the wind field. It was proved to be highly correlated to the local rainfall events in terms of rainfall probabilities and rainfall amounts and enhance the performance of the CP-based downscaling model. A CP- and Copula-based downscaling model is a further development of the previous downscaling model. The relationship between daily precipitation and moisture flux was described by a joint distribution based on the concept of copula instead of regression method. By using the concept of copula, the dependence is represented by a copula function that couples one-dimensional uniform distribution functions to a multivariate joint distribution. All the marginal distributions of studied variables are uniformed in the space (0, 1). Therefore, any appropriate marginal distribution is allowed to be selected. Furthermore, the copula function is able to represent the various dependence structures between the different quantiles of the variables, which makes it possible to fully reproduce the dependence structure identified from the observations. The CP-based downscaling model is suitable for the regions located in the higher latitudes, where the Coriolis force is quite dominant in forming anticyclones and cyclones. For other regions near to the equator, where the Coriolis force is weaker, the same methodology does not work properly anymore. To downscale daily precipitation for those lower latitude regions, a multi-site weather generator was developed. The developed model is a stochastic statistical downscaling model. It is able to simultaneously generate the reasonable daily rainfall time series. The models have been successfully applied to the different river basins located in the contrasting climate zones. The critical CPs for specific river basins were identified and they could be used to explain the large rainfall events. The generated daily precipitation are comparable to historical observations and can be used as input to other water-related model for the river basin management studies under the impact of climate change.