基于GAMLSS模型的水文干旱指数研究——以玛纳斯河流域为例

Study on hydrological drought index based on GAMLSS:Taking Manas River Basin as an example

传统的水文干旱指数是在一致性条件下确定的,而在非一致情况下水文干旱指数的识别精度受到质疑。特别是当全球气候发生剧烈变化的时候,应用合适的干旱指数可以增加干旱预警的精确度。一般在非一致情况下,通常会认为干旱指数的概率分布参数服从时间或者其他协变量的线性或者非线性变化。因此以玛纳斯河流域肯斯瓦特站为例,构建以时间为协变量的GAMLSS模型,建立非一致情况的标准化径流指数SRIns,并对比分析,讨论其适用性。结果表明:(1)肯斯瓦特站1957—2014年期间,径流量的变化趋势比降水和气温的变化趋势更为明显,径流发生明显变化主要集中在秋季和冬季,降水全年各月的变化趋势不明显,气温则在春季和夏季变化较剧烈。(2)通过对比研究区1957—2014年内有历史资料记载的历史干旱事件,SRIns对于研究区干旱事件的识别更准确,SRIns识别的严重干旱和极度干旱事件的发生频率要比SRIs高。(3)通过游程理论识别出干旱特征变量,将干旱特征变量采用均匀分布随机化处理可以提高干旱历时序列的拟合精度。干旱特征变量序列的最优分布均为对数正态分布。(4)SRIns和SRIs的干旱特征变量的二维联合分布的最优Copula函数均为joe函数。通过对比干旱特征变量二维联合概率和重现期,SRIns可以缩小风险区间,增加干旱风险预警的精度,因此更适用于该研究区的干旱预测与风险评估。

The traditional hydrological drought index is determined under the condition of nonstationary.However,the identification accuracy of this index is questioned,especially under the dramatical climate change,the application of appropriate drought index can increase the accuracy of drought warning.In general,under nonstationary conditions,the probability distribution parameters of drought index are generally considered to be subject to linear or nonlinear changes of time or other covariables.Therefore taking Kenswatt Station in the Manas River Basin as an example,we built a GAMLSS model with time as the covariable.The standardized runoff index SRInsin the case of nonstationary was established and compared with the standardized runoff index SRIsin the case of stationary,and the applicability of SRInswas discussed.The results show that:(1)During 1957-2014,the variation trend of runoff was more obvious than that of precipitation and temperature at Kanswatt Station.The change of runoff was mainly concentrated in autumn and winter,while the change trend of precipitation was not obvious in each month of the year,while the change of temperature was more fluctuant in spring and summer.(2)By comparing historical drought events recorded in the study area from1957 to 2014,SRInscan identify drought events more accurately.The frequency of severe drought and extreme drought events identified by SRInswas higher than that of SRIs.(3)Drought characteristic variables were identified by Run theory.The fitting accuracy of drought duration series can be improved by using uniform distribution and randomization of drought characteristic variables.The results of cumulative probability show that the optimal distribution functions of the drought characteristic variable series of SRInsand SRIsare lognormal distribution.(4)The optimal Copula function of the two-dimensional joint distribution of drought characteristic variables of SRInsand SRIsis Joe function.By comparing the twodimensional joint probability and return period of drought characteristic variables,SRINS can reduce the risk interval and increase the accuracy of drought risk warning,so it is more suitable for drought prediction and risk assessment in the study area.