The Study On Distributed Precipitation Estimation Model & Method Based On MODIS & Artificial Neural Networks

Precipitation is not only an essential parameter in hydrology and in the research of water resources, but also important for the flood control of the valley and the preparation against it, as well as the construction of hydraulic and hydropower engineering. However, the conventional monitoring approaches of precipitation involve many disadvantages, such as limited observing range, high cost and only-one-point precipitation observation. Consequently, how to get the precipitation of any part of the valley attracts more and more attention. With the progress of the science and technology, the satellite based precipitation observation, which can cover a wide range, and is not restricted by terrain, displays its merit in this field. Growing concern has been devoted into the observation system of this approach. MODIS is a new generation satellite sensor, which combines image with spectrum. It can provide substantial data on the rainfall. If it is applied to estimate precipitation, the accuracy of the estimation must be improved, and the development of the precipitation estimating technology would be promoted too. This thesis set up a regional scale estimation model of precipitation, and analyzed the relation between the data of MODIS with rainfall, based on the data of satellite, obtained with the MODIS sensor. The precipitation of any unit of the valley can be acquired through this model. Firstly, the process and the physical mechanics of rainfall are analyzed. Then parameters of cloud and atmosphere, which are directly involved with rainfall, are chosen as the estimating factors. On the basis of analyzing the data processing flow of MODIS, a data set of parameter factors is hereby established by reading the cloud product data of MODIS, which, together with data obtained by surface measurement of rainfall, compose the original sample. At last, the modeling method of artificial neural networks is adopted; and the network model is tested with the original sample. Thus a regional scale precipitation estimating model has been established. In order to train and test the model, large amount of raw data are required. The precipitation data in this thesis is borrowed from observation data of rainfall in the Qingjiang valley, Hubei Province, in the first half of the year, 2004. The MODIS satellite data product, received and processed by the MODIS ground station of Digital Valley Engineering Center of Huazhong University of Sci. & Tech., are adopted for the satellite data in this thesis. For the artificial neural network, BP network is utilized. Through examining the model, it is found that its estimating ability is good, which can reflect the actual rainfall in spite of some deviation. The study of this thesis reveals that it bears practical significance to apply the high spectral resolution and rich data products of MODIS to satellite estimation of precipitation, which can develop the application field of MODIS. Meanwhile, the study also shows the excellent capability of adaptive study and nonlinear mapping of artificial neural networks, which is significant for improving the accuracy of precipitation estimation.