A numerical study on effects of land-surface heterogeneity from "combined approach" on atmospheric process part I: Principle and method

A method based on Giorgi (1997a, 1997b) and referred to as 'combined approa ch', which is a combination of mosaic approach and analytical-statistical-d ynamical approach, is proposed. Compared with those of other approaches, th e main advantage of the combined approach is that it not only can represent both interpatch and intrapatch variability, but also cost less computation al time when the land surface heterogeneity is considered. Because the inde pendent variable of probability density function (PDF) is extended to the s ingle valued function of basic meteorological characteristic quantities, wh ich is much more universal, the analytical expressions of the characteristi c quantities (e.g., drag coefficient, snow coverage, leaf surface aerodynam ical resistance) affected by; roughness length are derived, when the roughn ess length(and / or the zero plane displacement) heterogeneity has been mai nly taken into account with the approach. On the basis of the rule which the PDF parameters should follow, we choose a function y of the roughness length z(0) as the PDF independent variable, and set different values of the two parameters width ratio alpha(n) and hei ght ratio gamma of PDF (here a linear, symmetric PDF is applied) for sensit ivity experiments, from which some conclusions can be drawn, e.g., relevant characteristic terms show different sensitivities to the heterogeneous cha racteristic (i.e., roughness length), which suggests that we should conside r the heterogeneities of the more sensitive terms in our model instead of t he heterogeneities of the rest, and which also implies that when the land s urface scheme is coupled into the global or regional atmospheric model, sen sitivity tests against the distribution of the heterogeneous characteristic are very necessary; when the parameter alpha(n), is close to zero, little heterogeneity is represented, and alpha(n) differs with cases, which have a n upper limit of about 0.6; in the reasonable range of alpha(n), a peak-lik e distribution of roughness length can be depicted by a small value of gamm a, etc..