Research On The Model And Method Of Distributed Rainfall Estimation

Rainfall has always been a focus of academic research in the field of hydrology and water resources because it is not only an important input parameter in hydrological model, but also an important consideration for the issue of flood control, disaster mitigation, runoff forecast, irrigation, etc. With the increasingly requirements of fine spatial scale and time scale in the research of distributed rainfall estimation, the average rainfall that obtained from the discrete rainfall observation sites in the region is difficult to meet the requirements of practical work effectively. However, the rapid development of spatial information technical methods and the non-linear modeling methods are combined with the traditional science of hydrology and water resources, a new way of solving these new problems is brought forward. How to apply the remote sensing (RS) and geography information system(GIS) to build a theoretical model is an problem need to be solved urgently , which is able to estimate the distributed rainfall in different spatial scales of the river basin. And it can provide scientific theories and technical support to spatial interpolation and remote sensing retrieval of distributed rainfall estimation. In this paper, distributed rainfall estimation model is discussed and studied based on the spatial interpolation theory and remote sensing retrieval theory. And the main contents of this paper are as follows:1. Digital terrain model (DTM) of basin and digital elevation model (DEM) derived from DTM is studied, and combined with the study object of this paper, digital rainfall model (DRM) is used to represent the distribution of rainfall in a river basin, and which is suitable for applying in geographic information systems analysis and management.2. The traditional spatial interpolation methods are studied and is combined with concrete examples, the scope of the Hubei Province is divided into several study areas based on different size of spatial scales, and then the rainfall in the study areas depending on rainfall intensity is divided into several levels. In this paper, Inverse Distance Weighted Method(IDWM), Kriging method, Back Propagation Neural Network(BPNN) model and BPNN model optimized by genetic algorithm are used to establish spatial interpolation estimation model respectively.3. The main meteorological parameters which influencing the rainfall can be distilled from the MODIS satellite cloud imagery, and these meteorological parameters are combined with the actual observed rainfall data which is obtained from ground-based rainfall site correspondingly. The remote sensing retrieval model is established respectively based on the BP neural network and GA-BP neural network, and a better effect of error precision estimation is obtained. It's also proved that the high spectral resolution of MODIS data products used in distributed rainfall estimation of river basin is very practical significant.4. Aiming at the problem of the hidden layer nodes of BP neural network is difficult to determine, in this paper, an improved algorithm formula used to confirm the hidden layer nodes is brought forward. It is proved that the improved formula is very practicality and effectiveness. In addition, BP neural network has inherent defects that the initial connection weights and thresholds is determined randomly and is prone to local convergence easily, Due to the genetic algorithm (GA) has good performance in global search and optimize, so GA is used in this paper to optimize the connection weights and thresholds of BPANN. And these two models are used to construct the spatial interpolation model and retrieval model of rainfall estimation respectively.Through the research about estimating missing rainfall on River Basin in Hubei Province, the experimental results show that:1. In some spatial interpolation precipitation models, the traditional methods of estimation interpolation have their own limitations and applied range. The results of estimation on the study about Hubei Province for the different spatial scales of interpolation also show that in small regional or balanced density distribution of rainfall area, the interpolation is better to estimate in the effect of large area or uneven distribution of rainfall area. And from another point of view in interpolation model, Comparing with other traditional spatial interpolation based on statistical model, BP neural network model based on nonlinear optimization is better in the estimated accuracy. At the same time, the estimation error of spatial interpolation model largely affect by the scope of spatial scale, topography and the precipitation. Under different conditions, the error precision estimates vary, and the mean relative error in the range of roughly between 30% -80%.2.Using meteorological satellite data to estimate basin inversion of remote sensing research distributed rainfall results shows that BP neural network model to build estimates of the inversion and the BP neural network model optimized using of genetic algorithms, the results in estimating rainfall are better. The estimate of the accuracy of the model selected only with the relevant meteorological factors. The relevance with the distribution of rainfall stations,the range of research area and the rainfall are not too great. This article accessed seven meteorological factors closely related with rainfall intensity from the cloud MOEDS product. For the basin in Hubei province within different spatial scales and different rainfall intensity, this study built different models for each cases. And the estimated mean relative error was between 20% -25%.