DEnKF-based Hybrid Data Assimilation Of Multi-Source Snow Data Observations Into The Common Land Model For Snow Simulation

As one component of the Earth System, snow plays a key role in the Earth's hydrological processes, land surface energy balance, and climate changes. Located in the hinterland of the Eurasian continent, the Altay region of Northern Xinjiang is an important seasonal snow area of China and the base of animal husbandry of Xinjiang. The seasonal snowmelt water flows to the rivers in the Altay region. Long-time and heavy snow may cause snow disasters and even lead to snowmelt flood. The serious snow disasters exert negative influences on transportation, electric power, communication, agriculture and animal husbandry, threaten people's lives and property, and destroy the socioeconomic construction. To investigate spatiotemporal distribution of snow in the Altay region, it is crucial for us to accurately estimate snow parameters of the snow system.In the land surface model, mathematical physical equations of the snow process model are built in principles of the conservations and transformations of snow mass, energy and momentum. It can simulate and predict snow parameters theoretically in any temporal and spatial scales, typically snow parameters of snow cover fraction, snow depth, and snow water equivalent. At present, the land surface model with snow sub-processed is widely applied to explore snow's impact on climate, hydrology and snowmelt runoff forecast. However, the errors of the simplified parameterization schemes, meteorological forcing data and initial values of these input parameters are accumulated and transported into the simulation results of snow albedo, SD and SCF in the land surface model. Some snow data assimilation (DA) methods based on the direct insertion (DI) and stochastic ensemble Kalman filter (EnKF) methods are proposed to assimilate remote sensing data into the land surface model for snow process simulations. In most cases, observations are assumed to be perfect and have no errors in the DI data assimilation. The stochastic EnKFs need observation perturbations, which may introduce sample errors of observations. The ensemble-based assimilation results are affected jointly by the ensemble member size, snow observation operator and observation error covariance. To overcome these issues, some advanced snow data assimilation technologies of intergrating multi-source remote sensing and Common Land Model (CoLM) are developed in this thesis.To reduce the errors from observation perturbations, a new SCF DA method extending the deterministic ensemble Kalman filter (DEnKF) is proposed. The MOD10A1 SCF is assimilated into the CoLM using the proposed method for SD simulations. Physically, snow albedo has great effect on snow simulations. However, the estimates of albedo in the CoLM are usually underestimated. A DI method is proposed to assimilate the black- and white-sky albedos into the CoLM for updating the black- and white-sky albedos in the CoLM. The SD simulations in the CoLM are indirectly improved. First, the MODIS-based albedo is calculated with the MODIS Bidirectional Reflectance Distribution Function (BRDF) model parameters product (MCD43B1) and the solar zenith angle as estimated in the CoLM for each time step. Then, the MODIS-based albedos are assimilated into the CoLM using the DI DA method. Based on the two DA methods above, a new DEnKF-Albedo DA method of integrating the DI and DEnKF DA methods is proposed. The assimilation results are validated against in-situ SD observations from November 2008 to March 2009 at five sites in the Altay region. The results show that three DA methods can improve SD simulations in general, and the DEnKF-Albedo DA method achieves the best analysis performance in particular. This is due to that the DEnKF-Albedo DA method significantly reduces the bias and root mean square error (RMSE) during the snow accumulation and ablation periods at all sites except for the Fuyun site. For the Fuyun site, the DEnKF-Albedo DA method tends to overestimate the SD accumulation with the maximum bias and RMSE values because of the large positive innovation (i.e., the difference between observation and forecast). The SCF assimilation via DEnKF produces better results than the albedo assimilation via DI. It is implied the SCF assimilation, directly updates the SD state variable, is more efficient than the indirect albedo assimilation.The above results have shown that the assimilation of MODIS SCF only using the DEnKF method with a sample ensemble size of 30 cannot effectively improve SD simulations. To avoid underestimation of analysis ensemble mean and background error covariance of DEnKF with smaller ensembles, a hybrid DEnVar method of DEnKF-Variational data assimilation (DA) is proposed to assimilate MODIS SCF observations into the CoLM for SD and SWE simulations. Coupling the DEnKF method with a one-dimensional variational DA method (1DVar), DEnVar without observation perturbations is a two-step DA method. That is, the analysis ensemble mean and analysis error covariance of DEnKF are introduced into the 1DVar hybrid cost function, and the analysis mean of DEnKF is assigned direclty through 1DVar analysis. The analysis performance of DEnVar is experimentally compared with DEnKF, 1DVar, and EnVar (hybrid ensemble-variational DA) at five sites in the Altay region from November 2008 to March 2009. The results show that the four DA experiments can improve snow simulations at most sites. The DEnVar experiment using the hybrid error covariance achieves the best analysis performance among the four DA experiments at most sites. Furthermore, a sensitivity tests of DEnVar to weighting coefficients and MODIS SCF observation errors founds that DEnVar is slightly sensitive to the weighting coefficient determined with ensemble-and (National Meteorological Center) NMC-based error covariances, and is highly sensitive to the observation error.During the SCF data assimilation, the simplified observation operator and the presence of cloud cover cause large errors of the assimilation results. To reduce these errors, a new snow cover depletion curve (SDC), known as an observation operator in DA, is statistically fitted to in-situ SD observations and MODIS SCF from January 2004 to October 2008. Further, using this new SDC, a four-dimensional ensemble variational data assimilation method (4DEnVar) of integrating the DEnKF and 4DVar DA methods is proposed. The MODIS SCF is assimilated into the CoLM using 4DEnVar for SD simulations. The assimilation results are also validated against in-situ snow depth observations from November 2008 to March 2009 at five sites in the Altay region. The results show that the 4DEnVar DA method shows the best analysis performance. This is due to that the bias and RMSE are reduced during the snow accumulation and ablation periods at all sites except for the Qinghe site. Also, it is due to that more MODIS SCF observations are assimilated in with the 4DEnVar method, which produces more innovations than the DEnKF method with only one MODIS SCF observation. To some extent, the problems of cloud cover and overestimation are sloved with the 4DEnVar. During a long period of unavailable MODIS SCF, e.g., during 2/8/2009 to 2/20/2009 at Jimunai site, the 4DEnVar DA method can still improve SD simulations well. Furthermore, a sensitivity test of 4DEnVar to assimilation window length founds that 4DEnVar is sensitive to the assimilation window length. Given reasonable the observation operator and the assimilation window length,4DEnVar without tangent linear and adjoint of the CoLM model achieves better analysis performance as the assimilation window length increases.To accurately estimate the spatiotemporal distribution of SD in the Altay region, a new observation operator with snow density, modeling sub-grid heterogeneity, is designed for the DEnKF-Albedo DA. With these techniques, MODIS-derived albedo and SCF are assimilated into the CoLM for regional SD simulations. The assimilation results are validated subsequently with in-situ SD observations and AMSR-E-retrieved SD product provided by the Environmental and Ecological Science Data Center for West China (WestDC) in the Altay region. The proposed DEnKF-Albedo DA method significantly improves the simulation of SD spatiotemporal distribution over the Altay region. It is found that snow in Altai Mountains are mostly distributed from northwest to southeast, deep snow is in the high altitude of the north Altay region, and shallow snow is in the central plain area of the Altay region.