非饱和土壤水的集合卡尔曼滤波Ⅰ:状态向量与非饱和流算法的选择

土壤水运动是水分循环中的基本过程,但土壤水预测面临着参数获取难、预测精度差等挑战。数据同化技术为土壤水参数估计和精确预报提供了一种新的方法。本文建立了基于3种不同非饱和水流求解方法的集合卡尔曼滤波(En KF)算法,针对状态向量的选择和正演模型的选择两个问题,研究了非饱和土壤水En KF的计算性能。研究结果表明:对于非线性非饱和水流问题,同时更新水头和参数比仅仅更新水头能够取得更好的预测效果,特别是当多参数未知时;En KF本质上是Monte Carlo方法,极端样本容易导致Picard-h和Picard-mix算法的崩溃,因此传统的HYDRUS程序与复杂非饱和土壤水的数据同化兼容性不佳;当同时同化水头和参数时,如果极端的样本值能够快速得以更新,Picard-h和Picard-mix算法在数据同化模拟中的适用性能得以提升;但由于观测信息对参数的校正能力取决于特定的问题和条件,Ross算法是执行非饱和土壤水数据同化模拟的更好选择。

Ensemble Kalman filter for unsaturated soil water flow Ⅰ:selection of state vector and unsaturated flow algorithm

Soil water movement is one fundamental process of hydrological cycle. However, soil water pre-diction is challenging due to the difficulty of parameter acquisition and poor simulation accuracy. Data as-similation technique provides a new approach to soil water parameter estimation and precise prediction. This paper presents the ensemble Kalman filter (EnKF) based on three different algorithms of unsaturated flow. To address the selection issue of state vector and forward model, the performance of EnKF for unsaturated soil water flow is investigated under different situations. The results show that for nonlinear problem, aug-mented state vector of water head and parameters leads to better prediction than that from head state vec-tor, especially when multiple parameters are to be estimated. EnKF is essentially a Monte Carlo method, and extreme samples may cause the collapse of Picard-h and Picard-mix algorithm. The traditional HY-DRUS code is prone to failure in the data assimilation problem of complex soil water flow. The applicabili-ty of Picard-h and Picard-mix algorithm may improve when the head and parameters are assimilated simul-taneously, and when the samples with extreme values can be updated quickly. However, due to strong dry-wet alternating phenomenon in unsaturated flow, and the fact that the correction capability of measure-ment information to parameter depends on specific problem and associated conditions,Ross algorithm is sug-gested to be a better choice during the implementation of unsaturated flow data assimilation.