Identification, Estimation And Optimization Of Key Surface Parameters In Land Surface Models And Its Application In Regional Modeling

Energy, water and mass exchange are the main approaches of land-atmosphere interaction process. Large number of parameters involved in surface physical, biological and soil water/heat process are included in land surface model to describe above interaction process. Because of lack of field observations and insufficient knowledge of the land surface processes, the parameters in current land surface models (LSMs) cannot entirely describe land surface characteristics, which severely restrict the improvement of model capacity for successfully simulation of surface-atmosphere interaction. Therefore, identifying, estimation and optimization the key parameters which are most influential in simulation surface sensible and latent heat fluxes, is a hopefully way to improve the integral performance of LSMs.To achieve this goal, we firstly employed extended Fourier amplitude sensitivity test (EFAST) to identify the most sensitivity parameters of two LSMs, i.e., Common Land Model (CoLM) and Simplified Simple Biosphere model (SSiB2), in simulation sensible heat and latent heat fluxes, based on the field observations over arid/semi-arid regions in Northern China. CoLM includes 43 tunable parameters while there are 32 in SSiB2. EFAST can present both first order (without the consideration of parameter interaction) and total order (with the consideration of parameter interaction) sensitivity of parameters. The results shows the of top 10 sensitive parameters can explain 90% of the variance of objective variable. The sensitivity order of those parameters can be roughly written as:roughness length (z0m), leaf area index (LAI), parameter "b" in soil water process, root depth, hydraulic conductivity at saturation, vegetation cover fraction, aerodynamic resistance coefficient (canopy to air and soil to canopy) are related to energy balance, evaporation, evapotranspiration and turbulence interaction.In order to improve the ability of LSMs in representing the land surface processes through better key parameters, two approaches were applied to estimate and optimize the key parameters at ID LSMs application. One approach is estimating z0m with the combination of land surface field observations and boundary layer theory. Three estimation methods, i.e., independent method, optimal method and fitting method, were employed to estimate z0m of SACOL 2006～2008 and Tongyu degrade grass and cropland 2003～2008. There are subtle discrepancy in the z0ms estimated by different methods. The estimated z0m shows distinct seasonal and inter-annual variation which are not represented properly in current LSMs. Using the estimated z0m in LSM, it is found that the simulations of sensible heat flux have been greatly improved. Further application of estimation methods is determining z0m with the field observations of 15 sites over arid/semi-arid region of China in July, August and September,2008. The results show that the z0ms for the sites over the same plant functional type (PFT) share the same order of magnitudes. It suggests that the parameter estimated at point study can be applied to regional study.Another approach is parameter optimization to determine the parameters which cannot (or difficult to) be obtained from field observation. The genetic algorithm (GA) were applied to optimize the key parameters, i.e., parameter "b" in soil water process, root depth, hydraulic conductivity at saturation, to obtain the best match of heat fluxes between simulation and observations. Based on the traditional GA, we modified the evolution strategy in this study for higher evolution efficacy and saving computing cost. Although z0m and LAI can be determined by observation, they were also included as physical criteria to supply the gap of mathematical optimization and its physical explanations in GA. The results show that GA can find the parameters which have the equivalent performance of default parameters in LSM. With the help of the auxiliary criterion of land surface temperature, same PFT sites cross validation, and the physical criteria, the remaining parameters are suitable for all the sites in same PFT in this study and have better performance than the default ones in heat fluxes simulation for most sites.For the regional scale study, satellite data provides spatial and temporal continuous land surface parameters. We introduced MODIS land surface product and Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences LAI products into WRF_NMM/SSiB2, the result shows better simulation of precipitation amount and distribution, especially for monsoon precipitation seasonal evolution:The simulation with default parameters produces rain band much northern than observation, overestimating the precipitation in northern China and underestimating in Huang-Huai area; within satellite retrieved LAI, the model produces larger gradient in temperature from South to North over East Asia, leading to stronger west wind due to thermal wind adjustment, which restricts the northward march of subtropical high, therefore, restricts the fake northward march of rain band, and fixes the bias in precipitation simulation.In this study, we also discussed the impacts of land surface parameters in dynamic global vegetation models (DGVMs). We found that DGVM SSiB4/TRIFFID represents a better global vegetation distribution through tuning C3 plant photosynthesis optimal temperature. The negative bias in modeling LAI in eastern U.S. is also fixed by adding the PFT for deciduous broadleaf trees.