This paper created train sample input and expectation output data relying on Neural Networks tool boxes GUI based on Matlab with surface water resource quantity time series data as train sample basic,then make forecasts based a model of temporal serial Neural Network.The merits of the model is that it can simulate numerous variable and obtain mapped connection between them by train Back Propagation Neural Networks without needing to assume intricate connection input variable,and without needing to know connection between input and output variable.Lastly,the feasibility and validity of the model was validated with the past years surface water resource quantity time series data from Kenswat Station on Xinjiang Manas River.