【作者】 张和喜；

【导师】 迟道才；

【作者基本信息】 沈阳农业大学 ， 农业水土工程， 2013， 博士

【摘要】 近年来,随着全球性气候变化,旱灾发生趋于频繁,每年旱灾都造成了巨大的经济损失,严重的威胁着人类各种社会经济活动的正常进行。我国是一个水资源短缺的国家,干旱成为威胁我国经济发展与农业生产的主要制约因素,迫切需要对干旱特性、发生规律等进行深入研究。贵州省地处云贵高原东侧第二阶梯大斜坡上,境内地势西高东低,自中部向北、东、南三面倾斜。近年来,贵州省旱情发生频繁,程度不断加重,影响区域不断扩大,水资源供需矛盾更加突出,干旱已成为影响贵州省经济社会发展的重要因素。因此,对贵州喀斯特地区干旱发生的特征、规律进行分析和科学的预测,对防灾减灾具有十分重要的意义。本文运用贵州卡斯特地区多年气象资料,对该地区的旱灾发生及演变规律及降水预测方法进行了研究,主要成果如下：(1)贵州喀斯特地区干旱特征分析。以贵州喀斯特地区乌江水系、沅江水系、北盘江水系、红水河水系、赤水河綦江水系、柳江水系的1961年到2011年51年的降水量及温度等实际观测数据为基础,利用Palmer干旱指标、降水距平百分率法和Z指标,引入马利科夫判据的附加误差控制法,分析了贵州喀斯特地区逐年的年度旱情特征以及季度旱情特征,结果表明：Palmer指标对研究区域干旱强度的描述更准确,该方法综合考虑了降水量、温度、地表蒸腾、径流量等多方面因素,较适合贵州喀斯特地区岩溶发育,地表水汇水快,土壤蓄水能力差的干旱特征分析。(2)以蒙特卡洛算法及其分布函数为基础,运用P-Ⅲ型分布函数对降雨量进行模拟,将蒙特卡洛和NNBR模型相结合,提出了基于NNBR与蒙特卡洛算法相结合的降雨量预报模型。并利用回溯算法对预测降雨量序列进行回溯检测。结果表明,利用基于NNBR与蒙特卡洛算法相结合模型预测降水量更精确。(3)利用乌江流域1961-2010年共50年的降雨资料对马尔可夫、加权马尔科夫和趋势加权马尔科夫三种模型预测模型进行了验证和对比,结果表明,趋势加权马尔科夫预测效果较好,为提高降雨量预测精度提供了新途径。(4)运用BP神经网络、径向基函数(RBF)神经网络、Elman神经网络分别建立了研究区旱情预测模型,并进行三种预测模型的对比分析,结果表明,径向基函数(RBF)神经网络的预测效果较优。(5)以灰色预测模型为核心,建立小波分解—不同频率成分不同模型的预测架构,并引入波形预测,建立了降水量灰色预测优化方案。实际应用结果表明,本方法很好解决了频率震动大的问题,预测精度高,实现了方法创新。(6)根据1961-2010年贵州省19个气象台站的实测气温、降水、日照时间等数据为数据源,采用干旱综合指数法(CI),分析计算了贵州省近50年来干旱发生的频率、覆盖范围、干旱过程的持续日数、干旱强度；并应用ArcGIS9.3地理信息系统软件对不同发生强度、不同覆盖区域的干旱多年平均日数进行了空间分析,直观和形象的体现了研究区域不同等级旱情的空间分布、发展演变过程、覆盖范围及干旱强度等,为减灾防灾提供了科学依据。综上,本文的主要创新点如下：(1)首次尝试将基于最近邻抽样回归模型(NNBR)模型与蒙特卡洛滤波相结合,分析旱情的变化趋势、周期特性和突变特性。(2)将趋势加权马尔科夫模型运用到降雨量的预测之中,为降水量准确预测引入了新方法。(3)将灰色模型与小波相结合,将降水量数据进行小波分解,分别对不同频率成分分量进行预测,其中低频分量采用灰色模型进行预测；高频分量采用波形预测的方法进行预测,提出了基于小波分解的灰色预测模型。更多还原

【Abstract】 In recently years, along with the global climate change, drought tend to be frequently, every year, the drought caused enormous economic loss, and serious threat to normal man’s various social economic activities. China is a country with water storage, drought has become main restrictive factor that threaten to the economic development and the agricultural production, there is an urgent need to research into the characteristics and regularity of drought. Guizhou province is located in the second step slope, which is in the east of Yunnan-Guizhou plateau. The terrain is higher in the west and lower in the east, lying tilt from central to north, east and south. In recently, drought occurs frequently in Guizhou. The degree aggravating and the influence area expanding constantly, and the contradiction between supply and demand of water resources become more prominent. Drought has become important factor that affect the economic and social development in Guizhou. Therefore, analyze and scientific forecast the characteristics and regularities of the drought in Guizhou karst regions is of great significance for disaster prevention and mitigation. This paper researched the drought occurrence and evolution regularities and the precipitation prediction method by using years of meteorological data in Guizhou karst regions. The main results were as following:(1) Analysis of drought characteristics in Guizhou karst regions. Based on precipitation and temperature observed data from1961to2011of WuJiang, YuanJiang, BeiPanJiang, HongShui River, ChiShui River, QiJiang and LiuLiang water system in Guizhou karst regions, using Palmer drought index, precipitation percentage and Z index, and introduced Malikov Criterion Additional Error Control Method, this paper analyzed the annual and quarter drought characteristics in Guizhou karst regions. The results show that Palmer index described the research region drought intensity more accurate, the method considers multiple factors such as precipitation, temperature, surface transpiration and runoff synthetically, it is suitable for drought characteristics analysis in Guizhou karst regions where Karst development, fast surface water catchment and poor soil water storage capacity.(2) Based on the Monte Carlo Algorithms and its distribution function, using P-III type distribution function to simulate the precipitation, and combining the Monte Carlo Algorithms and NNBR model, this paper proposed precipitation forecast model which is based on the combination of Monte Carlo Algorithms and NNBR model, and used backtracking algorithm to test the predict rainfall sequence. Results show that based on NNBR model combined with Monte Carlo Algorithms to predict rainfall is more accurate.(3) Using precipitation data in1961-2010of Wujiang Drainage Area to verify and compare three prediction model Markov, weighted Markov and trend weighted Markov, and the results show that the trend weighted Markov prediction effect is better than the other two models, it provides a new way for improving the rainfall forecasting precision.(4) Using BP neural network, Radial Basis Function(RBF) neural network and Elman neural network established drought forecasting model of the study area respectively, and compared and analyzed of the three kinds of prediction model, the results show that the Radial Basis Function(RBF) neural network has a better prediction effect.(5) With gray prediction model as the core, established wavelet decomposition- different frequency components with different model architecture, and introduced waveform prediction, and rainfall gray prediction optimization scheme is established. Practical application results show that the method is very good to solve large vibration frequency problems, and has high prediction accuracy, it implements the method innovation.(6) According to the measured temperature, precipitation and sunshine time data in1961-2010of19meteorological offices and stations in Guizhou province, and by using comprehensive drought index (CI), the drought frequency, coverage, last days of process and intensity in Guizhou province in recent50years are analyzed and calculated. Spatial analyses for drought years average number of different strength and different coverage area are realized by geographic information system software ArcGIS9.3, different grade drought, evolution process and the range and intensity of drought in the research area are reflected intuitively and visually, it provides scientific basis for disaster prevention and reduction.In conclusion, the main innovation points are as follows:(1) The first attempts to combine model based on the nearest neighbor sampling regression (NNBR) with Monte Carlo Algorithms to analyze the variation trend, cycle characteristics and mutation features of drought.(2) Using trend weighted Markov model in rainfall forecast, thus introduced a new method for accurately precipitation predictions.(3) Combined gray model with wavelet, took the precipitation data wavelet decomposition, forecast different frequencies components respectively, of which using gray model to the low frequency component forecast and waveform prediction method to high frequency component forecast, and put forward the gray forecasting model based on wavelet decomposition.更多还原