This paper presents a methodology for the representation of variability in land surface fluxes across a given domain. Landsat thematic mapper (TM) data are employed to estimate the pixel-scale variability of the energy partition at the time of a TM overpass. Multiple realizations of the TOPUP soil vegetation-atmosphere transfer (SVAT) model are run deductively to define a space of model responses over a longer period of time. This allows the models to be classified into different functional types. The uncertain pixel estimates are then used to condition or map the estimates of the local landscape fluxes into the space of model functional types using fuzzy measures. The identified fuzzy weights may then be employed to derive time series of the mean areal latent heat flux and quantiles to represent the range of the flux variability. This scheme has the advantages that the resulting model is simple enough to be used directly as a representation of a heterogeneous land surface in an atmospheric model and that the fuzzy weights may be updated with additional data.