This investigation presents a new method for real time forecasting in operation of reservoirs. Potential impacts of climate variability on hydrology and consequently its influence on water resource systems planning such as reservoir operations is the core of this research. The central part of Iran water system and supplying its water demands is highly dependent on a short seasonal precipitation and hence has great potential to suffer from unbalanced distribution of both demand and supply sides of water. Reservoirs in this region are the main water structures to reduce the shortages and help to balance the monthly water needs based on limited supply. Historical data indicates that the region is susceptible to hydrological drought periods where consecutive years of below-average annual streamflow occur. A developed integrated data driven simulation method with an evolutionary optimization algorithm constituted a hybrid model to determine suitable reservoir operating policies during uncertain periods. Reservoir operation is evaluated under different climate conditions and the impacts on downstream water needs are investigated. Using the Harmony Search (HS) algorithm, the optimal strategy for reservoir releases are determined based on actual monthly inflows for different climatic conditions. Support Vector Machines (SVMs) method for regression modeling, is applied as a state of the art modeling tool. The proposed methodology is applied to Zayandeh-rud reservoirs system and the application explored the model’s accuracy and its potential for real time prediction of reservoir monthly releases for adaptive planning and management. Based on the research objectives, two real time operating models are developed in which the fundamental differences are in the consideration of climate variation. Effective information to train suggested models include inflow runoff, monthly demand and initial volume of reservoirs. In terms of accuracy and generalization performances, both models, non-climatic and climatic approaches, indicate very small RMSE and close agreement between training and testing phases of modeling. Considering particular variables in the input vector, to identify the climatic features of the system reduces the RMSE and hence the improvement of model forecasting. To compare the performances of the proposed methodologies with the results of previous study using average ruled-curve method of operation, monthly releases forecasted by SVM models were simulated for period of 44 years and the results indicated 15 and 11 percent improvement for climatic and non-climatic approached, respectively, of meeting the demands over the average ruled-curve situation.