Climate Model Parameterization is a Form of Statistical Modeling - Application to Land Surface Representation and Treatment of Hydrological Cycling

The concept of parameterization was introduced in the early days of climate modeling to describe simple rules that were assumed to treat scales of motion not resolved by climate models. Such rules have generally neglected that there are multiple scales being neglected. What is done is best understood as a form of statistical modeling that is designed to optimally relate statistics of small scale processes to state parameters on the scale resolved by the model. It is not inevitable that such parameterization should forget about what are the spatial scales over which smaller scale processes occur. The land surface component of climate models is a nice example of this assertion. It is essentially all parameterization as it involves describing processes at a local site scale and then averaging them to model resolved scales. However, multiple scales of land can be distinguished that need to be separately addressed. Satellite data can now look at the land surface at a scale as fine as 1-m and this information can be used to calibrate other satellites providing global coverage. This paper works through the formalism whereas land at fine scales can be included computationally in a climate model by appropriate statistical modeling. Some of the important scaling issues are described. Examples are given from work with the CLM3 land model coupled to the CAM3 (Community Climate Model components supported by NCAR