Abstract
In line with the objectives of the Impact-Based Forecast, hydrological predictions are used to support water resources management, mitigation of natural disasters, and climate variability impacts. The Artificial Neural Network algorithm was applied to build the prediction model for the Saguling Reservoir monthly inflow by utilizing observation data; the monthly rainfall (P) and inflow (Q) as the predictors in different lag time; t (recent), t-1 (previous month), and t-11 (11 months ago). The predictors are simulated in three variations of hidden layer numbers (2, 6, and 10). The best model under the normal period is model nine with RMSE 31.23 and R 0.88. This model is also the best model under La Nina's condition with RMSE 30.01 and R 0.83. For the El Nino period. We found that model five is the best model with the highest accuracy at the level of generalization in both the training process and predict extreme conditions at the validation stage. Overall, this model has a good performance and high potential usage from a practical point of view and costs but needs further simulation to make it more reliable and robust in any climate conditions.