The lake scheme of the Common Land Model and its performance evaluation

Abstract

The Common Land Model (CoLM) has been updated to a new version, but the lake scheme in CoLM (CoLM-Lake) has never been evaluated. This paper introduces the structure and physical processes of CoLM-Lake, and evaluates its simulation performance through the observations from 10 lakes over different regions. It also discusses the sensitivities of simulated results to some important parameters in the model. Results show that CoLM-Lake performs very well over the three shallow lakes (Kossenblatter, Taihu and Sparkling Lake) where the model accurately captures the magnitudes and seasonal variations of lake surface temperature, turbulent fluxes, and vertical thermal structure. The freeze-thaw cycle of Sparkling Lake is reproduced at a reasonable level as well. Also, CoLM-Lake show acceptable performance in simulating vertical structures and their variations for deep lakes, although the vertical mixing strength remains underestimated. The biases of surface temperatures in the Great Lakes of North America are relatively larger, but the magnitudes and seasonal variations in temperature fall within a reasonable range. In general, CoLM-Lake is suitable for the simulation of lake physical processes on the global scale. The simulated results have strong sensitivities to some parameters with values that have large uncertainties. The surface roughness lengths determine the amounts of turbulent fluxes which are emitted into the atmosphere, and thus affecting the simulated lake surface temperature. Although the surface roughness based on the dynamic diagnosis of lake properties in the model has made the simulated turbulent fluxes very accurate, the further adjustment of surface roughness can improve the realism of the simulated turbulent fluxes and temperature on lake surface. Lake depth, optical extinction coefficient and thermal diffusivity affect the simulated lake thermal structure from different angles. Lake depth determines the amount of water in vertical mixing; optical extinction coefficient determines the distribution of solar radiation at different lake depths; and thermal diffusivity determines the simulated vertical mixing strength that a lake can achieve. Either a larger lake depth, a smaller extinction coefficient, or a larger thermal diffusivity enables a lake to transfer and store more heat internally, thereby increasing lake thermal inertia, reducing daily and seasonal variations of lake water temperatures, and causing late freeze-up dates in winter. Therefore, the biases of simulated lake thermal structure can be corrected by adjusting the above three parameters. Note that lake depth and optical extinction coefficient are lake inherent attributes, and thus it is important to collect high-resolution and high-precision data for these two parameters in the future. The modification of thermal diffusivity should focus on increasing simulated vertical mixing strength for deep lakes, while the relatively accurate simulation for shallow lakes should be maintained, and thus more complex processes in deep lakes need to be considered in the model than simply multiplying the background thermal diffusivity of entire lake. The temperature at which lake water reaches maximum density is affected by lake salinity, and adjusting this temperature also improves simulated results. Hence, the effects of lake salinity on lake water physical properties should be considered in the development of future lake schemes. As three-dimensional lake schemes become more sophisticated, it is increasingly important to couple them into regional and global climate models. However, the description of important processes such as lake vertical mixing, phase change and snow hydrology are still too simple in one-dimensional schemes, and thus the upper limit of accuracy that one-dimensional schemes can achieve in deep lake simulations remains unclear, and the necessity of coupling the three-dimensional schemes with climate models is under dispute. If one-dimensional schemes are found to have better performance in simulating the characteristics of deep lakes, then the large amount of computational resources that need to be consumed to run three-dimensional models can be saved. Therefore, developing one-dimensional lake schemes at current stage is extremely important.