Study On The Time Scale Of Medium And Long-term Runoff Prediction

The medium and long-term prediction of river runoff can provide technical support for the rational development of water resources and the allocation of water resources.It is not only a major demand for national social-economic development but also a frontier issue of hydrology science.At present,the accuracy study of medium and long-term runoff prediction is often concentrated on a single time scale,and horizontal comparison studies of medium and long-term runoff forecast at different time scales are rare.Four methods are forecast unoff series from 1980 to 2012 in the Yangtze River Basin in order to study the influence of time scales on the forecasting accuracy of medium and long-term runoff prediction.The main contents and conclusions are as follows:(1)Forecasting method based on time series is more suitable for medium and long-term runoff prediction in the Yangtze River Basin.Forecasting methods based on time series(i.e.the combined model based on wavelet analysis and nearest neighbor bootstrapping regressive model),and forecasting methods based on hydrometeorological factors(i.e.the stepwise regression-artificial neural network method and the stepwise regression-support vector machine method)are used to forecast ten-day and month runoff series of three main stations in the Yangtze River Basin from 1980 to 2012.Results show that prediction accuracy of methods based on time series are better than those based on hydrometeorological factors,so it is considered that the forecasting method based on time series is more suitable for medium and long-term runoff prediction in the Yangtze River Basin.It is found that the prediction accuracy of ten-day scale is higher than that of month scale,so it is considered that runoff series of different time scales will affect the accuracy of runoff prediction accuracy.(2)The accuracy of medium and long-term runoff prediction decreases first and then increases with the increase of the forecasting time scale.The prediction accuracy is the worst on month scale and bimonthly scale,and is relatively good on the three-day and year scale.To further study the relationship between time scales and prediction accuracy,research stations were increased to 6,and the daily runoff data from 1980 to 2012 were converted into 10 different time scales: three-days,week,ten-day,half month,month,bimonthly,season,half year,nine-month and year.The combination model based on wavelet analysis and nearest neighbor bootstrapping regressive model were used to forecast the runoff series.Results show that with the increase of the forecast time scale,prediction accuracy decreases first and then increases.The prediction accuracy is the worst on month scale and bimonthly scale,and is relatively good on three-day and year scale.It is considered that the most disadvantageous time scale for medium and long-term runoff prediction is the monthly scale.(3)The time scale accuracy evaluation index of medium and long-term runoff prediction,prediction accuracy index,is put forward.To analyze the causes of the most disadvantageous time scale,the influence of time scales on the accuracy of runoff prediction is analyzed from persistence,discrete degree and periodicity.By synthesizing the influence of three factors,a synthetic index,prediction accuracy index,is obtained to indicate runoff prediction accuracy at different time scales.Compared with the actual prediction results,it is found that prediction accuracy of different time scales is well reflected by the prediction accuracy index.The prediction accuracy index is larger on the three-day scale and year scale,and the value near month scale is smaller.The minimum value appears on the month scale,confirming that the most disadvantageous time scale for medium and long-term runoff prediction is month scale.