AMIP II Evaluation of the Continental Component of the Global Energy and Water Cycles

We have analysed global land-surface energy and water budget components for the period 1979-1995 simulated by 10 AGCMs participating in the second phase of the Atmospheric Model Intercomparison Project (AMIP II). The simulations are compared with three reanalysis products (NCEP/DOE, NCEP/NCAR and ECMWF) and with the global land-surface simulations by the Variable Infiltration Capacity (VIC) land surface scheme. A lack of adequate observational data at the land surface is a limiting factor for evaluation of the land-surface climate in global climate models and it is anticipated that until such data becomes available, the model derived estimates such as are produced by the reanalysis is the best available option for model validation. Results underline a number of problems in some model simulations such as non-conservation of water and energy and problematic simulation of surface evapotranspiration This is in some degree related to poor initialisation and/or bad parameterisation of soil moisture. In a comparison of three reanalysis products and VIC land surface simulations we find large differences among the AMIP II models, among the reanalyses, and between models and reanalysis in simulating surface fluxes. The simulation of latent heat flux by different models relative to each other and relative to different reanalyses estimates vary in different climate zones and the differences among reanalyses are of the same order as differences between the model simulations and reanalyses' estimates. The mean latent heat flux over all land surfaces ranges from 42.1 W m^-2 to 56.8 W m^-2 among the validation data sets, comparable to that among the AMIP II models. Compared to ECMWF reanalysis, the mean absolute deviation latent heat flux of the two reanalyses is 6.8 W m^-2, much larger than the mean absolute deviation of 3.2 W m^-2 for the AMIP II models. Use of reanalyses for model evaluation of the continental component of the global water cycle is therefore problematical. Until reliable global land-surface observational data sets become available, it may be more appropriate to consider the ensemble of the available reanalysis data sets a the best validation tool.