Abstract:

It is crucial to the regional climate and hydrological and ecological applications that how to get a better understanding of the spatiotemporal distribution of precipitation. With the rapid development of satellite remote sensing technology and applications, the estimation of precipitation make it possible for the precipitation forecast by using remote sensing complements the data from ground gauge observation in the areas where the number of observation stations is not enough.Due to the artificial neural network characterized by a description of the nonlinear relationship,it can be used to describe the nonlinear information of precipitation and its regional differences and thus fit the nonlinear relationship between values of precipitation estimation and gauge observation. The artificial neural network is powerful due to its capability in processing data by using parallel,selfadaptive and selflearning methodology and mapping the nonlinear patterns. Thus, artificial neural network technology can make a new and valuable method for the estimates of the precipitation information.In this paper,the typical northern Tibetan Plateau is taken as the study area.The data mainly includes gauge observation data and a long time series of FY2C imagery.Based on the analysis of observation data,the multispectral characteristics and features of clouds,the parameters for the description of the process of precipitation in the study area were acquired and normalized.Then,the parameters for precipitation simulation and the intensity of precipitation were used as the input and the output of the neural network for the estimation of precipitation by using satellite remote sensing data.Finally,a combination of qualitative and quantitative indicator evaluation system was employed to test the accuracy of model estimation.Multiindicator system was developed for analyzing the accuracy of precipitation simulation.The results show that the infrared band could be used to effectively indicate precipitation of clouds.By using high temporal resolution satellite remote sensing imagery,the subtle changes of clouds could be captured and the parameters of precipitation simulation,which reflect precipitation characteristics,could be acquired.The nonlinear features of precipitation in this area could be well described by using artificial neural network.The precipitation estimation model used in the paper was stable,efficient,and globally representative.The model could better disclose the precipitation characteristics of the study area in the northern Tibet Plateau.The results also demonstrate that correlation coefficient between values of modeling estimation and gauge observation data was 0.57.Moreover,the systematic bias of the estimates was small,indicating that the model could predict the precipitation of weak intensity more accurately in the site.The values of precipitation estimation were less than that of gauge observation data,which had a better indication of the weak intensity of precipitation.The precipitation estimation results could provide valuable parameters for various hydrologic researches.Due to the significant variation in terrain of the north Tibetan Plateau,whether or not the precipitation simulation parameters and the model by using FY2C data can be used in other parts of the Tibetan Plateau is to be answered by further researches