| Runoff and sediment, key factors in a river system, play an important role in shaping fluvial morphology and fluvial geomorphology. Besides, those supplying water resources and maintaining regional environmental and ecological systems. The Lancang-Mekong River is a famous international river in Asia. Recently, the development of cascade dams on the mainstream of Lancang River has aroused widespread concern at home and abroad on cross-border hydrological, environmental, ecological and social problems. Nowadays, most dams have been built, thereupon their hydrological and ecological effects began to highlight in the Buyuanjiang River, which is one of the main tributary in the lower part of the Lancang River. So further understanding on water and sediment changes in the basin over the past few decades and the future forecasting is necessary for science and practical application.Used the lower reaches of Yangtze river basin the monthly average flow from1959-2008and monthly average sediment concentration from1993-2008at the Manan Gauging Station, and five meteorological stations rainfall data from1959-2007, this paper analyzed the runoff and sediment characteristics and evolution trend, then established the forecasting models for mid-long term hydrological forecasting.Based on the data analysis and result deduction, detailed conclusions have been drawn as following:(1) The inter-annual variability range of runoff was low during1960to2008, the coefficient of variation of annual runoff and inter-annual extreme value proportion at the Manan Gauging Station was0.21and2.72respectively; But from1993to2008, the inter-annual variability range of annual sediment load was high, and the variation coefficient of annual sediment load was0.59, and inter-annual extreme value proportion reach to7.1. The seasonal annual distribution of runoff and sediment was fairly concentrated, mainly comes from June to November within the flood season. And the intra-annual nonuniform coefficients of runoff and sediment load were high up to0.87and1.65. Therefore, seasonal changes of runoff and sediment were obvious.The annual average incoming sediment coefficient is only0.006kg/(s.m6) at Manan Gauging Station from1993to2008, so the sediment concentration per unit flow of Buyuanjiang River was low. The average monthy flow and sediment concentration was significantly positive power function correlation during1993-2008; And during this period, the regression decision coefficient was0.6334, which indicated that the fitting degree was high; Compared with annual runoff and sediment load relationship, the regression decision coefficient of the relationship between month runoff and sediment load was0.8137, which indicated that the fitting degree was obvious. From1993to2008, the average annual sediment load and runoff process were showed as similar declining trends as precipitation process. During the period from1996to2008, changes in the Manan cross section is not obvious, showing relatively stable in riverbed morphology.(2) Based on hydrological data at Manan hydrological station, this study built two autoregressive moving average models to forecast annual discharge and sediment concentration series. As a result, ARMA (5,4) and ARMA (2,3) regression model were elected to predict average flow of flood season and dry season respectively, the results showed dry season average flow would be49.4455,52.0674,49.9494,51.9686,48.2954m3/s and flood season average flow would be316.0015,174.8079,288.0196,342.2409,283.9677m3/s from2009-2013. And ARIMA (1,1,1), ARIMA (1,1,2) regression model were chose to predict mean sediment concentration of flood season and dry season respectively, the results indicated dry season mean sediment concentration would be0.0179,0.0219,0.0263kg/m3and flood season mean sediment concentration would be0.8232,0.8029,0.7826kg/m3during2009～2013.(3) With Matlab’s Neural Network Toolbox, two classic three-layer BP neural network models were built to simulate the annual runoff and sediment load respectively at Manan hydrological station. For Runoff forecasting model, chose annual rainfall and the rainy season precipitation as two input layer nodes. The number of hidden layer neurons, which selected base on the empirical formula and used the "trial and error" method, according to the fitting results, was set to8in the end. The output layer has one neuron:annual runoff. Sediment forecasting model ultimately preferred a classic three-layer BP network which contain the five inputs:flood seasom flow, annual flow, average monthly flow from July to September, annual precipition and Rainy season precipitation; one output:annual sediment load; and one hidden layer which contained7nodes. Overall, the fitting degrees of those two models are very high, and achieved the desired effect. The fitting results, whether qualitative or quantitative, are in accordance with requirements of "Standard for hydrological information and hydrological forecasting". |
PDF Full Text Request