基于核熵成分分析的综合干旱指数的构建与应用

针对传统单变量干旱指数难以全面表征干旱及部分综合干旱指数难以反映多变量之间的非线性关系等问题，采用标准化降水蒸散发指数（SPEI）、标准化径流指数（SRI）及标准化土壤湿度指数（SSMI）3个单变量指数分别表征气象干旱、水文干旱和农业干旱，利用核熵成分分析法（KECA）构造综合干旱指数（SMDI），采用M-K趋势检验、小波分析及典型历史旱情验证等方法分析干旱的时空变化特征以及干旱指数的适用性。以黑河流域中上游为例，结果表明：研究区全年77.6%的区域表现为干旱不显著加重的趋势；在流域尺度上，干旱存在43 a的长周期，15~23 a的中周期，3~8 a的短周期；20世纪90年代夏、秋两季及21世纪以来春、冬两季干旱发生频率较高，且整体夏旱发生频率最高；1969年春、1997年秋和2009年冬的典型历史旱情验证表明SMDI优于其他3种单变量干旱指数。说明基于KECA构建的SMDI是一种有效的干旱监测指数，在黑河流域中上游干旱监测中有好的适用性。

Construction and application of comprehensive drought index based on kernel entropy component analysis

In view of the cavity that the traditional single variable drought indices are difficult to characterize the comprehensive drought situation, and some existed composite drought indices reflect poorly on the nonlinear relationship among multiple variables. We adopted kernel entropy component analysis (KECA) to construct a comprehensive drought index (SMDI) by considering three various single drought indices: the standardized precipitation evaporation index (SPEI) to represent meteorological drought, standardized runoff index (SRI) to characterize hydrological drought, and standardized soil moisture index (SSMI) to show agricultural drought. Taking the upper and middle reaches of Heihe River Basin as an example to analyze the drought spatiotemporal variation and the applicability of SMDI, the M-K trend test, wavelet analysis and typical historical drought events validation were utilized. The results showed that 77.6% of the grids of the study area presented an insignificant worsening trend of drought in the whole year. Drought had a long period of 43 a, a medium period of 15~23 a and a short period of 3~8 a on the watershed scale. In the season of summer and autumn of 1990s as well as in spring and winter since the 21st century, the drought frequency was higher. Moreover, the overall frequency of summer drought was virtually the highest. The typical historical drought events test in the spring 1969, autumn 1997, and winter 2009 also argued that SMDI was superior to the other three univariate drought indices. As such, the KECA-based SMDI was an effective drought monitoring index and had better applicability in the upper and middle Heihe River Basin.