Nonparametric Trend Analysis And Prediction Of Hydro-meteorological Elements

Hydro-meteorological system is a complex and dynamic circulatory system.Influenced by climate change and human activities,hydrometeorological processes show obvious non-parametric statistical characteristics such as nonlinear,non-stationary,multi-scale and randomness in time.It is beneficial for water resources management,climate assessment and environmental protection to study the evolution law of hydrometeorological elements,grasp the change trend and improve the prediction accuracy.Therefore,the trend analysis and prediction of hydrometeorological elements based on non-parametric statistics have theoretical and applied significance.This dissertation systematically studies the research progress of non-parametric trend analysis methods at home and abroad,including Mann-Kendall(MK)method,Sen's Innovertive Trend Analysis(ITA)method and multiple ITA method.In addition,the theoretical basis of combined model about Ensemble Empirical Mode Decomposition(EEMD),Recurrent Neural Network(RNN),and Long short-term Memory(LSTM)are reviewed.Combined with their application status in hydrometeorological elements,the problems and shortcomings of these classical algorithms are analyzed and summarized,thus forming the research focus of this dissertation.The main research work and achievements of this dissertation are as follows:(1)Aiming at the shortcomings of the ITA method,an modified ITA trend analysis(MITA)method is proposed.Through numerical simulation analysis about Sen's ITA method,it is found that results of trend type and its significance test are related to the length and standard deviation of the time series.The trend detection result of ITA is worse when the length of time series is smaller or its standard deviation is larger.Based on the idea of ITA method,a relative quantity is proposed as the index of average trend.Combined with a non-parametric statistical method,namely bootstrap,significance test was realized,and finally an improved trend analysis method was proposed.Monte Carlo numerical simulation shows that the MITA method is less affected by length and standard deviation of sequences,and the consistency rate of trend significance test with MK method is better than the original ITA method in short time series.(2)A Trend Stability Analysis(TSA)method is proposed for the first time.This method is based on the idea of multiple ITA method,designs a quantitative index to quantify trend stability on the basis of MITA,and completes the non-parametric test method of significance level of trend stability with the help of bootlifting method.(3)The proposed MITA and TSA are applied to nine hydro-meteorological element sequences involving precipitation,air temperature and runoff in the Xilin River Basin to analysis their trends and stability,and the validity of the method is verified.A series of detailed conclusions about their trends and stabilities are obtained.(4)The EEMD-LSTM-PSO combined model for long-term prediction of hydrometeorological elements is established.This model adopts the idea of decompositionprediction-optimization reconstruction,introduces the adaptive EEMD method to decompose the time series into several component sequences,and then establishes the LSTM neural networks model to predict them respectively.Finally,the PSO algorithm is used to optimize the reconstruction coefficients,and the final prediction is obtained by the weighted sum.Four evaluation indexes including root mean square error(RMSE),mean absolute percentage error(MAPE),correlation coefficient(R)and Nash coefficient(NSE)and voting principle are used to evaluate the models.The long-term prediction of annual runoff series in the upper reaches of Heihe River,annual mean precipitation series in India and monthly precipitation series in xilin River basin are studied.The practical application shows that the method proposed in this thesis is effective in long-term prediction of precipitation and runoff.The improved trend analysis method and stability test method proposed in this dissertation can be effectively applied to the trend analysis of hydrometeorological elements.The combined prediction model constructed in this study has a great prospect in the long-term prediction of hydrometeorological elements.