| The role of soil moisture is widely recognised as a key paremeter in global energy and water cycle and numerous environmental studies, including not only in meteorology and climatology but also in hydrology, ecology and agriculture, et al. Lots of field and similation experiments approached to understand the importance of soil moisture, and the remote sensing data provided a great chance to improve soil moisture estimation. This thesis mainly addresses the simulation of soil moisture by using different land surface models (LSMs) and the estimation of soil moisture by using ensemble Kalman filter (EnKF) data assimilation scheme wih the background provided by each single-model or multimodel. The main contents are as follows:Firstly, there are various kinds of Land surface models in different nations and institutions. To evaluate the porformance of different models, and ultimately to improve the parameterization of land models, many land model intercomparision projects have been arranged to this aim.The first part of work in this thesis is to simulate soil moisture by using different numerical discretization schemes. The 1-D Richards equation for soil moisture is highly nonlinear and it is impossible to find out an analytical solution with general initial and boundary condition. Therefore, numerical approximations are typically used to solve the equation. To prevent the influence of the parameterization schemes of upper boundary(eg. evaporation and run-off), three kinds of ideal upper boundary conditions (BCs) are adopted, including fixed evaporation rate, fixed infiltration rate and fixed near-surface soil moisture. The main purpose is to evaluate the influence of the numbers of soil layers and the heterogeneity of soil on the prediction of soil moisture. The results show that if the soil along depth is homogeneous (i.e. one soil type with fixed percentages of sand, clay and loam) and the large number of soil layers is used, the different schemes will produce almost similar results. By comparing the soil moisture profiles produced with more layers and less layers, we found that the soil moisture predicton becomes worse with less layers than with more layers at deep soil layers. The results also shows that if the soil along the depth is heterogeneous (i.e. still one soil type but with different percentages of sand, clay and loam in each layer), the different schemes will produce very different solutions, and the main difference is the continuity of the soil moisture profile. The rms difference of the soil moisture predictions by the fine grids (i.e., more soil layers) and coarse grids (i.e., less soil layers) is not consistent among differet LSMs. In order to quantitatively evaluate the impact of linearization on the modeling of soil moisture, an interative scheme with no linearization is introduced and its results are used to compare with those from the different LSMs, showing that the profile with the scheme of CoLM is similar with the interative scheme. The second chapter lists the different numerical discretization schemes. The third chapter gives the result of the experiments.The Second part of work in this thesis is to simulate the soil moisture, soil temperature and surface heat fluxes by the forcings of the meteorological observations. The dataset is from the Soil Moisture Experiments in 2003 (SMEX03) and the GEWEX Asia Monsoon Experiment over the Tibetan Plateau (GAME/Tibet) and the three LSMs are CoLM, CABLE and Noah. The result shows the models can not correctly simulate the soil temperature and soil moisture at near-surface layer, however, they all do a fairly good job in reproducing the soil moisture anomalies. The soil temperature and soil moisture simulated by different models have large differences at near-surface layer, but are very similar at deep soil layers. For latent and sensible heat fluxes, The fluxes from CoLM are relatively larger and those from Noah LSM are smaller than the observations, however, the fluxes from CABLE are close to the measurements. This content appears in the fourth chapter.Secondly, Soil moisture observation, especially remote sensing techniques, can provide additional information about near-surface soil moisture (generally less than 10cm) at large scales. Land data assimilation scheme can be designed to estimate soil moisture profile and fluxes by assimilating different data sources. Multi-model ensembles have been found to perform significantly better than a single-model system in weather and seasonal climate forecasts, however, data assimilation rarely used in multi-model forecasts.The last part of work in this thesis is estimating the soil moisture profile by the land data assimilation schemes. In this part of work, the ensemble Kalman filter (EnKF) is used in land data assimilation system. The soil moisture and fluxes are estimated with three single models by assimilating near-surface soil moisture observations. Random errors are added in initial soil moisture profile, forcing datasets and soil parameters are used to evaluate the influence of different sources of errors. This work appears in chapter five.Then, we carry out a multi-model EnKF data assimilation experiments by assimilating surface soil moisture observation for the retrieval of soil moisture profile. Two algorithms, i.e., the simple model average (SMA) and the weighted average method (WAM), are investigated for estimating the multi-model background superensemble mean and the corresponding multi-model background superensemble error covariance matrix. The two algorithms are tested and compared in terms of their abilities to retrieve the true soil moisture profile by respectively assimilating both synthetically generated and actual near-surface soil moisture measurements into the multi-model. The results from the experiment show the SMA does not help to improve the estimates of soil moisture at the deep layers, even worse than with single model. On the contrary, the results from the WAM are better than those from any single model and SMA. This part of wok appears in chapter six.Finally, all of work in this doctoral thesis is summarized in chapter seven, and further research is also included.
|
PDF Full Text Request