【作者】 杨帆；

【导师】 段兴武；

【作者基本信息】 云南大学 ， 自然地理学， 2021， 硕士

【摘要】 干旱是世界上危害最大的自然灾害之一。干旱传播,是指从一种干旱传播到另一种干旱的过程。当前对气象、水文以及农业干旱三者间传播关系的空间分异机制仍不明晰,限制着干旱防治和生态恢复。西南干热河谷区干旱频发,地形复杂,是我国最为典型的脆弱生态系统,在该区开展干旱时空分布特征、干旱传播规律和干旱预测模型构建研究,对缓解和防治干旱具有重要的科技指导意义。本研究以云南省龙川江流域为研究区,基于流域内31个气象站的降水量数据、4个水文站的径流量数据和Landsat系列遥感影像,计算气象、水文和农业干旱指标,研究干旱的强度、频率和空间分布范围,阐明龙川江流域三种干旱的时空分异特征;进而通过时滞互相关分析法计算干旱间的时滞无偏互相关系数,判断流域内三种干旱的时滞时间及发生顺序,总结干旱传播规律;最后基于支持向量机(Support Vector Machine,SVM)建立干旱等级预测训练模型并进行精度验证,优化干旱预测模型,得到适用于龙川江流域的干旱等级预测模型。主要结果如下:1.龙川江流域内不同干旱类型的时空分异特征:在干旱强度上,2000-2015年间气象干旱和水文干旱均呈增强趋势,而农业干旱呈先加强后减弱的趋势。在干旱频率上,水文干旱频率总体高于气象干旱频率。在干旱发生的空间范围上,发生气象干旱和水文干旱的空间范围均呈扩大趋势,而发生农业干旱的空间分布范围呈缩小趋势。改变耕作方式和种植低耗水性作物可降低龙川江流域水文干旱和农业干旱发生风险。2.龙川江流域不同干旱类型的传播规律:在海拔较高的林地地区,气象干旱经历2个月传播至农业干旱,而后经过2个月传播至水文干旱;在海拔较低的耕地地区,气象干旱和水文干旱均经历1个月传播至农业干旱。在流域内种植低耗水树种可延缓或者阻止气象干旱传播到水文干旱;推广高效灌溉技术可减轻水文干旱。3.适用于龙川江流域不同干旱类型的干旱等级预测模型:气象干旱等级预测模型为基于C-classification分类方法和Radial核函数,各旱涝等级权重相同的SVM模型;水文干旱等级预测模型为基于C-classification分类方法和Linear核函数,各旱涝等级权重相同的SVM模型;农业干旱等级预测模型为基于C-classification分类方法和Radial核函数,“正常”类权重为“偏旱”类权重的1.1-1.2倍的SVM模型。更多还原

【Abstract】 Droughts are one of the most damaging natural disasters in the world.The transition from one type of drought to another is known as drought propagation.At present,the spatial differentiation mechanism of the propagation relationship among meteorology,hydrology and agricultural drought is still unclear,which limits drought prevention and ecological restoration.The dry-hot valley in Southwest China is the most typical fragile ecosystem in China due to its frequent drought and complex terrain.The research on the spatio-temporal distribution characteristics of drought,the research on drought propagation relationship and the construction of drought prediction model in this region have important scientific and technological guiding significance for alleviating and preventing drought.In this study,the Longjiang River Basin(LRB)was taken as the study area.Based on the precipitation data of 31 meteorological stations,the runoff data of 4 hydrological stations and the Landsat remote sensing images,the meteorological,hydrological and agricultural drought indexes in the LRB were calculated to study the intensity,frequency and coverage of drought,and to clarify the spatio-temporal differentiation characteristics of three droughts in the Longchuan River Basin.Then,the time-lagged unbiased cross correlation coefficient between droughts was calculated by the time-lagged cross-correlation analysis method,the lagged time and occurrence order of three droughts in the basin were judged,and then the drought propagation patterns in the basin was summarized.Finally,based on Support Vector Machine(SVM),the drought level prediction training model was established and the accuracy was evaluated,and then the drought prediction model was optimized to obtain the drought level prediction model suitable for LRB.The main results were as follows:1.The spatio-temporal differentiation characteristics of different types of drought in Longchuan River Basin: in term of drought intensity,the meteorological drought and hydrological drought intensity showed an increasing trend while the agricultural drought intensity first increased and then decreased during 2000-2015.In term of drought frequency,overall frequency of hydrological drought was higher than that of meteorological drought.In term of drought coverage,the coverage of meteorological drought and hydrological drought was expanding,while that of agricultural drought was narrowing.Changing farming methods and planting low water consumption crops can reduce the risk of hydrological drought and agricultural drought in Longchuan River Basin.2.The drought propagation relationships among different types of drought in Longchuan River Basin: at higher altitudes with more forests,meteorological drought propagated to agricultural drought over 2 months,and then to hydrological drought over 2 more months.At lower altitudes with more arable lands,the meteorological and hydrological drought both propagated to agricultural drought over 1 months.Planting low water consumption plants in the basin can delay or prevent the propagation of meteorological drought to hydrological drought.Promoting efficient irrigation techniques can reduce hydrological drought.3.Prediction model of drought level for different types of drought in Longchuan River Basin: the prediction model of meteorological drought level is a SVM model based on C-classification classification method and Radial kernel function,with the same weight of each drought level.The prediction model of hydrological drought level is a SVM model based on C-classification classification method and Linear kernel function,with the same weight of each drought level.The prediction model of agricultural drought level is a SVM model based on C-classification classification method and Radial kernel function,and the weight of ‘normal’ level is 1.1-1.2 times that of ‘partial drought’ level.更多还原