Parisa Noorbeh; Abbas Roozbahani; Hamid Kardan Moghaddam
Abstract
Introduction: During the last decades, runoff decreasing is observed in our country as many dam reservoirs face water supply crisis even in normal periods. This decreasing trend is mainly due to the uncontrolled withdrawals, lack of supply and demand management as well as droughts. Using different flow ...
Read More
Introduction: During the last decades, runoff decreasing is observed in our country as many dam reservoirs face water supply crisis even in normal periods. This decreasing trend is mainly due to the uncontrolled withdrawals, lack of supply and demand management as well as droughts. Using different flow prediction methods for surface water resources state analysis is important in water resources planning aspects. These methods can provide the possibility of planning for proper operation by using different factors to meet the needs of the region. Due to the stochastic nature of the hydrological processes, various models are used for prediction. Among these models, Bayesian Networks (BNs) probabilistic model has been considered by many researchers in recent years and it has shown the efficiency on these issues. Due to the growth of demand in different sectors and crises caused by drought of the water supply system that has put the basin under water stress, the water shortage has appeared in different sectors. Regarding to the strategic situation of Zayandeh Rood Dam in providing water resources for tap water, industry, agriculture and environmental water rights in Gavkhooni basin, this research presents the development of a model for prediction of Zayandeh Rood Dam annual inflow and hydrological wet and dry periods. Since the uncertainty of the predictions increase when the prediction horizon increases, this factor is the most important challenge of long-term prediction. Using Bayesian Network with reducing this uncertainty, provides the possibility of planning for water resources management, especially for optimal water allocation.
Materials and Methods: In this study for prediction of zayandeh Rood dam inflow five scenarios were defined by applying Bayesian Network Probabilistic approach. According to this, prediction of numerical annual dam inflow (scenario1), annual wet and dry hydrological periods (scenario 2, 3, 4) and range of annual inflow (scenario 5) were performed. For this purpose rainfall, runoff, snow, and discharge of transferred water to the basin from the first and the second tunnel of koohrang and Cheshmeh Langan tunnel were considered as predictor variables and the amount of Zayandeh Rood Dam inflow was selected as predictant for modeling and different conditions of input variable’s learning have been analyzed considering different patterns. Calibration and validation of the model have been done based on observed annual inflow data and the relevant predictors in scenario 1, by using SDI Hydrological drought index and long-term average of inflow to classify the runoff and clustering the other parameters in scenario 2, 3 and 4 and with classification of annual inflow data and other parameters by using clustering in scenario 5. To achieve this target, K-means method has been used for clustering and Davies-Bouldin and Silhouette Width has been used to determine optimal number of clusters.
Results and Discussion: The results of Bayesian Network modeling showed that the scenario 1 has a good potential to predict the dam inflow so that the best pattern of this scenario (considering discharge of first tunnel of Koohrang and Cheshmeh Langan tunnel, Zayandeh Rood natural inflow and rainfall with two years lag time as predictor variables), has had a correlation coefficient of 0.78 between observed and predicted dam inflow and relative error of 0.21 which shows an acceptable accuracy in prediction. Among scenarios 2, 3 and 4 for prediction of wet and dry hydrological periods, scenario 2 in which classification of runoff has been based on the long-term average, in the best pattern (with dam inflow with one-year lag predictor), is able to be predicted up to 75% accuracy. The analysis of the results showed that the scenario 5 is not very accurate in prediction of dam inflow’s range.
Conclusions: The results showed that the Bayesian Network model has a good efficiency to predict annual dam inflow numerically as well as hydrological dry and wet periods. Obtained results from prediction of hydrological dry and wet periods will be effective in better planning of water resources in order to considering possible ways of drought effect reduction. The overall results provide the possibility of water resources planning for the water authorities of this region. Systematic planning leads to optimal use of water and soil resources and helps considerably to analyze and modify the policy or rule curve of this dam for allocating water to downstream especially for agriculture and environment and industry sectors.
