陆面模式中土壤和植被经验参数随机误差的传播研究

Soil and Plant Parameters-Related Stochastic Uncertainty Propagation in the Common Land Model

数据质量问题和模式参数化方案的非完备性是陆面模拟中不确定性的主要来源。本文将高斯误差传播原理(Gaussian Error Propagation,GEP)应用于通用陆面模式(the Common Land Model,CoLM),研究关键的植被和土壤属性参数随机误差在模式中的传播,确定由此类误差导致的CoLM模拟的不确定性。结果表明:(1)基于本研究给定的土壤和植被参数的不确定性,CoLM模拟的表层土壤温度、土壤湿度和植被蒸散通量(植冠蒸腾+地表蒸发)的相对误差分别为0.11%、34.07%和5.58%;砂土和稀疏森林上模拟效果最差。土壤参数随机误差对CoLM模拟的影响高于植被参数,而土壤水文参数(孔隙率、饱和基质势、气孔尺寸分布指数和饱和导水率)对各模拟量不确定性的贡献率均远大于热力参数(饱和反照率和热容)。对于本研究涉及的所有模拟变量而言,最关键的参数均是气孔尺寸分布指数b,这可能与描述基质势与体积水含量关系的函数有关,其次重要的是砂土的孔隙度和粘土的饱和导水率。混交森林上的根深分布和苔原上的动力学粗糙度对蒸散通量贡献显著。本身相对误差大的经验参数对CoLM模拟不确定性的贡献不一定多。(2)干燥条件下(表层液态水饱和度小于0.1)土壤温度的不确定性大;相变发生时刻附近(表层土壤温度在0℃附近且表层液态水含量大于0)土壤湿度不确定性显著;蒸散通量的不确定性随本身绝对值的增大而增大,在相对温暖干燥环境中(表层土壤温度高于280K且表层液态水饱和度小于0.3)其不确定性最高。研究证实,GEP能够辨识CoLM中需优先提高观测精度的关键参数和关键参数化过程,对陆面模拟的参数选定、不确定性评估和模式完善具有重要意义。

Bad data quality and imperfect parameterizations are major sources of uncertainties in the land surface modeling.Gaussian Error Propagation（GEP）principle is used to study the propagation of key plant and soil parameters-burden random errors in the Common Land Model（CoLM）,and to quantify the resultant uncertainties in the CoLM modeling.Results show that：（1）Based on the uncertainties of soil and plant parameters specified in this study,the relative errors of surface soil temperature and moisture as well as plant evapotranspiration（canopy transpiration plus ground evaporation）simulated by the CoLM are 0.11%,34.07%,and 5.58%,respectively.The highest uncertainties exist in simulations of sandy and sparsely vegetated areas.Compared to the random error of plant parameters,that of soil parameters affects the CoLM＇s simulation more remarkably;moreover,soil hydraulic parameters（porosity,saturated matrix potential,pore-size distribution index,and saturated hydraulic conductivity）contribute more（saturated soil albedo and volumetric heat capacity）to the modeling uncertainties than thermal parameters.As to all the simulated physical quantities in this study,the pore-size distribution index is always the most critical,which is probably associated with the function describing the relationship between matrix potential and volumetric water content.Porosity of sand and saturated hydraulic conductivity of clay is secondly important.The standard deviations of root distribution on the underlying surface of mixed forest and aerodynamic roughness length on the underlying surface of tundra make an appreciable contribution to evapotranspiration.The empirical parameters with higher relative errors are not necessarily greater contributors to the standard deviation of the predicted physical quantities.（2）Under dry soil conditions（the surface soil liquid water saturation degree is below 0.1）,the standard deviation of soil temperature is typically the highest.The soil moisture uncertainties are higher in the soil experiencing phase changes（the surface soil temperature is near 0 ℃ and the surface soil liquid water content is above 0）.The stochastic error of evapotranspiration grows with increasing of the absolute value of flux itself,and is much more significant in relatively warm and dry environments（the surface soil temperature is above 280 K and the soil liquid water saturation degree is below 0.3）.The research verifies that GEP is able to identify the critical parameters and parameterizations of the CoLM,thus is significant for parameter determination,uncertainty analysis as well as model improvement.