时空相关气候水文要素的变异诊断方法研究

Hydrometeorological data are commonly spatial-temporally dependent and thereby deviate from the assumption of independence that underlies the conventional change detection methods. The presence of dependence will influence the significance of observed changes, leading to misunderstanding of regional water cycle processes, and brings risks to decision-making on water conservancy projects safety and efficient water resources allocation. This research focuses on mitigating the adverse effect of spatial-temporal dependence on change detection results. The scientific logic of the research is that: the fundamental reason why the significance of changes are over- or under-estimated is that the variance of the test statistic deviates from its normal state while data are not independent. Therefore, change detection methods can be improved by restoring the variance of the test statistic. It requires incorporating data persistence at temporal scale and physical-statistical cross-correlation at spatial scale as well. Specifically, serial change detection methods restore the variance of the test statistic by introducing long term and short term persistent parameters. Regional change detection methods restore the variance by coupling physical correlation of hydrometeorological stations and statistic correlation of observation data. The research also propose a new index to quantitatively assess the performance of change detection methods, which describing Type I and II errors simultaneously. In this way, a more robust technical system for serial and regional change detection will be developed that takes into account both temporal and spatial correlation.

气候水文要素固有的时空相关特性违背了传统变异诊断方法的独立性假设，极易导致错误的变异诊断结论，干扰对流域水循环过程的准确认识，为水利工程安全、水资源配置带来决策风险。本项目研究降低时空相关特性影响的气候水文要素变异诊断方法。采用的基本科学思路是，认为变异诊断的误判是源于时空相关序列的变异诊断统计量方差相对于独立序列发生偏离。因而，从时间尺度提取序列的长-短持续性信息、从空间尺度提取序列间的物理-统计相关性信息，通过理论推导提出变异诊断改进方法，有效提高变异诊断方法的性能。研究中探寻融合长-短持续性特征参数，修正序列变异诊断统计量方差的途径；首次融合气候水文站点的物理相关性与观测数据的统计相关性，减少对区域变异诊断统计量方差的过度修正；首次探讨构造变异诊断方法性能评价指标，定量刻画各诊断方法的错检、漏检概率。据此建立更加稳健的时空相关气候水文要素的变异诊断方法集成系统。

气候水文要素固有的时空相关特性违背了传统变异诊断方法的独立性假设，极易导致错误的变异诊断结论，干扰对流域水循环过程的准确认识，为水利工程安全、水资源配置带来决策风险。本项目研究降低时空相关特性影响的气候水文要素变异诊断方法。采用的基本科学思路是，认为变异诊断的误判是源于时空相关序列的变异诊断统计量方差相对于独立序列发生偏离。因而，从时间尺度提取序列的长-短持续性信息、从空间尺度提取序列间的物理-统计相关性信息，通过理论推导提出变异诊断改进方法，有效提高变异诊断方法的性能。研究中探寻融合长-短持续性特征参数，修正序列变异诊断统计量方差的途径。提出了基于统计量方差校正的Sen变异诊断方法，解决时序自相关序列的变异诊断问题。融合气候水文站点的空间相关性和观测数据的时序自相关性，修正区域变异诊断统计量方差。提出了区域Spearman秩次相关检验方法，用于区域时空相关气候水文要素的变异诊断。探讨构造变异诊断方法性能评价指标，提出了“有效诊断区分析法”探明变异诊断方法的适用范围，引入机器学习领域的“受试者工作特征曲线ROC”综合评价变异诊断方法的错漏检概率，明晰了诊断能力与气候水文要素各种固有统计属性间的相互作用。据此建立更加稳健的时空相关气候水文要素的变异诊断方法。新方法突破了传统突变与趋势诊断方法依赖于时空独立性假设的局限，提高了对气候水文要素时空变异特性的识别能力，在准确认识变化环境下水文响应、提高水文中长期预测预报精度等领域有较为广泛的理论研究价值与应用前景。