基于集合卡尔曼滤波的土壤温湿度同化试验

以通用陆面模式CLM 3.0(Community Land Model 3.0)为模型算子,基于集合卡尔曼滤波(Ensemble Kalman Filter,En KF)发展了一个土壤温湿度同化系统,主要用于改进模式对土壤温湿度和地表水热通量的模拟精度,并考察集合样本数、同化频率及不同观测量的组合对同化效果的影响。该系统同化了FLUXNET两个站点(阿柔和Bondville)不同土壤深度、不同时间频率的土壤温度和湿度数据。通过对阿柔站不同集合样本数的设计,综合考虑计算成本和计算精度,最终将集合样本数设置为40。通过分析三种同化方案对同化频率的敏感性得出,同化土壤温度最为敏感,同时同化土壤温湿度次之,同化土壤湿度最不敏感。对于阿柔站点,同化系统对不同土壤深度温度和湿度的模拟精度均能提高90%,潜热通量的均方根误差由94.0 W·m~(-2)降为46.3 W·m~(-2),感热通量均方根误差由55.9 W·m~(-2)降为24.6 W·m~(-2)。Bondville站点浅层土壤温度的改进在30%左右,深层土壤温度改进达到60%,对土壤湿度的改进均在70%以上,潜热通量和感热通量的均方根误差分别从57.4 W·m~(-2)和54.4 W·m~(-2)降为51.0 W·m~(-2)和42.5 W·m~(-2)。试验结果表明,同化站点土壤温湿度数据对土壤水热状况及通量的模拟改进非常有效,同时也验证了同化土壤水分遥感产品的可行性和必要性。

Experiments of soil temperature and moisture assimilation system based on Ensemble Kalman Filter

In this work,with the Community Land Model Version 3.0 (hereafter CLM) as a model operator, a soil temperature and moisture data assimilation scheme was developed, which was based on Ensemble Kalman Filter. The system was mainly used to improve the estimations of the soil temperature, moisture, sensible and latent heat fluxes. Also, the ensemble size, assimilation frequency, three assimilation programs with different observations were executed to evaluate the improvements of land surface process simulation.Soil temperature and moisture in different layers and time frequencies from two flux observing sites (Arou and Bondville) were assimilated into CLM.With considering the computational cost and accuracy, the optimal ensemble size was set as 40. The sensitivity analysis of three programs to assimilation frequencies showed that assimilating soil temperature was the most sensitive to assimilation frequency, then assimilating soil temperature and moisture simultaneously, and assimilating soil moisture was the least sensitive to assimilation frequency. At Arou site, assimilation system could improve the soil temperature and moisture accuracy in different layers by 90%. The RMSE of latent and sensible heat flux reduced from 94.0 W·m^-2 to 46.3 W·m^-2 and from 55.9 W·m^-2 to 24.6 W·m^-2 respectively. At Bondville site, the improvements of shallow layer and deep layer soil temperature were about 30 % and 60%respectively. The improvement of soil moisture was above 70%. The RMSE of latent and sensible heat flux dropped from 57.4 W·m^-2 to 51.0 W·m^-2 and from 54.4 W·m^-2 to 42.5 W·m^-2. The results indicated that assimilating the soil temperature and moisture site data could improve the simulation of land surface process effectively. It was feasible and significant to assimilate soil moisture remote sensing products.