| Summary: | Studies on climate changes relies on the results generated from general circulation models, which is normally presented in a coarse time scale to the hydrological applications which requires a fine time scale. Simulation studies for design and operation purposes studies the hydrological systems with multiple sequences of rainfall while flood simulations uses detailed storm hyetograph with known characteristics. Disaggregation is the generation of data using stochastic model involving two time scales, typically from a coarse time scale to a fine time scale. There is a difference in models used for the two time scales, although the fine time scale must be consistent with the coarse time scale. The more commonly used models include the HyetosR’s Bartlett-Lewis model, K-Nearest Neighbours (KNN) model and WeaGETS’s weather generator’s Markov models. Each model has its own advantages depending on the purpose of study and for this project, the focus will be on disaggregation using HyetosR and comparing the results with the other types of models. HyetosR supports different versions of the Bartlett-Lewis model, such as the original Bartlett-Lewis model, Bartlett-Lewis rectangular pulse mode and random parameter Bartlett-Lewis model. Whereas KNN algorithm is a non-parametric model used for pattern recognition. There are different uses for the KNN algorithm such as classification and regression. The input depends on the k closest training examples in the feature space and the output is determined by whether it is utilized for classification or regression. For WeaGets Stochastic Weather Generator, it can be used as a downscaling tool changing the parameters to support any of the predicted changes in the amount of rainfall. The results after comparing show that HyetosR model can reproduce the essential characteristics of the observed data. It also produces the autocorrelations and dry probabilities. Overall, HyetosR is an appropriate model to be used for designing storms in flood studies however the parameter estimation can be improved. As compared to HyetosR, KNN could keep a better basic statistics like STD, AC1 and Pwet. Although WeaGets can be used to generate any stations independently, this model cannot maintain important interstation correlations of variables. Therefore, employing KNN, the spatial dependence can be maintained by resampling the day’s weather simultaneously.
|
|---|