A Comparative Analysis Of Two Land Surfac Models In The Context Of Simulation And Assimilation

Land surface model (LSM) is an important part of a weather forecasting model and/or a climate model, which has a large impact on the accuracy in the weather forecasting and climate prediction. Because of large number of land surface models, the international community has carried out lots of projects for intercomparison of land surface parameterization schemes by evaluating the outputs of these models and parameterization schemes. Despite this, due to the rapid progresses in the field land surface observation experiment, the more knowledge are acquired about the land surface processes, the land surface parameterization scheme has become more and more perfect with a new version of LSM frequently releasing out, so more work on LSM intercomparison is very necessary. Under these backgrounds, the dissertation is going to evaluate a widely used LSM by using numerical simulation and data assimilation (DA) technique when the LSM is driven offline by the meteorological observations or coupled with the atmospheric boundary layer (ABL), i.e., an offline NOAH LSM (hereinafter referred to as LSM) and a single column model (SCM). Our comparative study may make a contribution to model development and provide a good suggestion for the model developers.For this purpose, three experimental settings with different weather and land surface conditions are designed. The first one is that the bare soil becomes drying out with no precipitation, and the second one is that the land is coved by grass and there are rainfall processes. Under these two settings, the two models are all driven by the outputs from the mesoscale Weather Research and Forecasting (WRF) model so the experimentation belongs to one kind of Observing System Simulation Experiments (OSSEs) with no model errors. The third one is that the two models are driven by meteorological observations and moreover the model outputs are compared with observations.Under the first situation, there is a difference between model outputs from the two models but the differences of the surface soil temperature and soil moisture from the two models are very small; also the two surface heat fluxes are different and their difference changes quickly with time. Under the second situation, there is a large change in the partitioning of available surface energy between the sensible and latent heat fluxes with the sensible heat fluxes decreasing and latent heat fluxes largely increasing, and also the fluxes difference becomes large from the two models with the largest difference appearing when the air stability change.Under the third situation, the tests show that there is a certain deviation between the simulated and observed due to uncertainties in boundary conditions and model parameters, but the outputs still reflect the atmospheric characteristics and trends in the near-surface layer. The SCM simulated soil moisture is better than that with the LSM; for the simulation of soil temperature, both models do not have a good performance; for2-m air temperature and humidity and10-m wind, LSM cannot simulate, while SCM has a relatively good performance and the simulated temperature is best among three model states, but still has a relatively large error; the simulation of surface heat fluxes with SCM is better, however, it becomes worse when the ABL states change; LSM cannot simulate the ABL state profiles while the performance of SCM is not consistent with a better simulation of the ABL wind.OSSE shows that the errors of estimation can be greatly reduced by assimilating the near-surface soil moisture and temperature observations, but the assimilation effects are different between the two models. Assimilating more types of observations by SCM will further improve the estimates of surface heat flux, and meanwhile effectively improve the boundary layer. More soil layers or smaller assimilation time interval can effectively increase the assimilation effect. Localization or inflation of background error covariance matrix will improve the performance of data assimilation. The sample size and its forms of background ensemble have a very large impact on the data assimilation; if the sample has a good representative, better estimates can be obtained even if the ensemble number is small. The ABL states estimated with SCM are more sensitive to the increase while are not to the decrease of observational errors in the near-surface atmospheric states.Assimilating the real observations into SCM once every6hours can effectively improve the estimation of soil moisture profile and two mid-layer soil temperature estimates, but does not significantly improve the estimates of the near-surface soil temperature,2-m air temperature and humidity,10-m wind, surface heat flux estimates, and the ABL states. The reason is that the time interval is6hours between two successive observations so the DA frequency is small. Besides, unlike the OSSE, the prediction error is so large that seriously influences the DA effects.In summary, the difference of soil moisture and soil temperature simulated with two models is small while that of surface heat fluxes is large which are greatly influenced by the interaction between the land and atmosphere. In the context of comparison between two-model DA performance, the soil moisture estimates by LSM is better than that by SCM but the soil temperature estimates becomes worse; on the average, the estimates of surface heat fluxes by LSM is better than that by SCM; SCM can be combined with DA method to estimate the ABL states while LSM cannot do by assimilating the near-surface atmospheric observations. However, due to large prediction error and long observational time interval, the estimates of ABL states are not good in comparison with the corresponding observations. Therefore, if the model is used to simulate and estimate the land surface states, we suggest adopting LSM; If the model is used to provide initial ABL states, we suggest using SCM.