多源观测在水文数据同化中的应用

Application of Multivariate Observations in Hydrological Data Assimilation

为理解观测信息对水文数据同化性能的作用,建立了基于分布式水文模型SWAT的集合卡尔曼滤波(EnKF)算法,通过虚拟实验讨论径流、土壤水、地下水位等观测的影响。研究结果表明:当以流域出口点的径流为唯一观测时,可以较为准确地估计径流和参数CN_2,但对深层土壤水估计效果有限;浅层含水量观测可以用于估计暴雨径流,但模拟垂向水分运动时会产生偏差,不能改善整个径流过程估计;由于地下水位的变化可以在一定程度上反映土壤的垂向运动信息,补充地下水位观测可以提高土壤水的估计效果。因此,采用多源观测能更为精确地模拟径流和土壤水分状态,可以考虑使用成本低、易获取的地下水观测来反映深层土壤水的变化情况。

Aims to investigate how observation datasets affect performance of data assimilation. The Ensemble Kalman Filter （EnKF） is coupled with the distributed hydrologic model Soil Water Assessment Tool （SWAT） to assimilate observations including streamflow, soil moisture, and water table. Results shows are as follows： when adding downstream streamflow separately, predictions of streamflow and CN2 are accurate, but soil moisture estimates of deep layers are biased. Adding surface soil moisture observation could predict storm runoff, but distort vertical moisture movement and base flow. Complementary groundwater information could reflect vertical movement of soil moisture, and improve estimates of soil moisture. Therefore, multivariate observations lead to more accurate model states, and low-cost and available groundwater observation is moderately useful when assimilating soil water.