Analysis And Simulation Of Rainfall-runoff Process In The Typical Small Watershed In Loess Plateau Of Eastern Gansu, China

Rainfall and runoff in Zhifanggou watershed, Pingliang city, Gansu P.R.C was analyzed and simulated by combing field observation with indoor simulation and by combining slope scale with watershed scale. The main research approach of this paper was "hydrological features analysis-study periods division-model parameters selection-model structure designation-model simulation". Firstly, analyzed and discussed on historical hydrological rainfall and runoff data from1955to2004, and divided research periods (before1986and after1986) as the base of rainfall-runoff model. Secondly, the response of soil moistures to slope degree and climate elements (rainfall, temperature, etc.) was analyzed under slope scale. At last, rainfall-runoff was simulated by using the improved BP neural network based on historical data and current research, as well as adopting the approach of transferring research scale from slope to small watershed and selecting model parameters. The simulation effect is preferable.The main results of this paper are shown as below:(1) The concentration ratios of inter-annual rainfall and runoff were both high in the watershed. Rainfall and runoff were both concentrated in flood season. The changing trends of rainfall and runoff were both obvious. The relationship between rainfall and runoff was very close, while the relationship gradually divided after1986.(2) By analyzing on the measured data in2012, it indicated that each hydrological element, meteorological elements and soil moisture element on the underlying surface had a strong response to climate change in flood season. Especially for evaporation and temperature, with the correlation coefficient between them was0.81. The soil moistures in different depths in each runoff plot also had a strong response to climate change.(3) The response of soil moistures in each runoff plot to climate change was significant. By analyzing on changing characteristics in each month, it could be known that the variation degree in August was the biggest and smaller in September and October. There was a negative correlation between water absorbing capacity of vegetation roots and variation coefficient of soil moisture. The vegetable roots in slopes with potato plants and natural grass had much better capacity on water absorb and storage. These results are important to make research on water and soil conservation and agricultural structure adjustment in small watershed.(4) Based on the non-ideal simulation effect by using BP neural network with single factor, in this paper, annual rainfall-runoff characteristics and process were analysed and simulated by using improved BP neural network with double hidden layer. By taking rainfall in flood season (P), drought index, and water and soil conservation index as input parameters of the model, and establishing an improved BP network model with a3-3-3-1topological structure to make rainfall-runoff simulation, we obtain an ideal effect. It can be concluded from our results that the artificial neural network method is suitable to rainfall-runoff simulation in the Loess Plateau region.