Long-lead precipitation forecasting using large-scale climate signals and downscaling models,Case study: Zayandehroud dam basin
Many techniques and methods used in water resources management and planning directly or indirectly are related to the prediction of rainfall and the general climatology prediction. In recent decades, large-scale climatic signals as hydrological predictors are identified as a merging tool. Long-leg projections has many benefits for reservoir operation that water store and release decisions can be more dynamic and flexible in dealing with particular circumstances and can lead to more profit in water resources operation.
In this thesis, two categories of rainfall prediction model are developed in the basin of Zayandehroud dam. The first group of models based on large-scale climatic signals has been developed. Time scale for this group of models is seasonal one (two dry and wet seasons). Considering the low historical information, models with less possible complexity should be used.
For this purpose, principal component analysis is used to examine changes of transmission information versus the number of signals. In order to determine the inputs for rainfall prediction model, the correlation method and gamma test have been used. Comparison of modeling results utilizing the evaluation criteria show the better performance of gamma test model in choosing input variables than correlation method.
In this study, Support Vector Machine (SVM) model for predicting rainfall in the region has been used and its results are compared with the benchmark models such as K-Nearest Neighbors and Artificial Neural Network (ANN). The results show better performance of SVM model at testing stage. Thus, SVM model is used for rainfall projection in the future based on the climate change scenarios A2 and B2.
In the second category models, statistical downscaling model is used. In this model, using the outputs from GCM, the precipitation of Zayandehroud dam is projected. Most effective variables have been identified among 26 predictor variables. Then, based on observed data, the model was calibrated and verified, and daily rainfall in future periods under climate change scenarios is generated..
The results obtained from the first and second groups of models show that SVM has less error in estimation of the precipitation. In the observation period for the years 1998 to 2008, precipitation in the region is more closely to the B2 scenario. Thus in order to estimate the impacts of climate change on the precipitation of the region, scenario B2 has been used. The results show that the precipitation in the future wet periods are more than historical values and in the dry periods, it is lower than historical values.