Multi-objective conditioning of a simple SVAT model

It has previously been argued that current Soil Vegetation atmosphere Transfer (SVAT) models are over-parameterised given the calibration data typically available. Using the Generalised Likelihood Uncertainty Estimation (GLUE) methodology, multiple feasible model parameter sets are here conditioned on latent heat fluxes and then additionally on the sensible and ground heat fluxes at a single site in Amazonia. The model conditioning schemes were then evaluated with a further data set collected at the same site according to their ability to reproduce the latent, sensible and ground heat fluxes. The results indicate that conditioning the model on only the latent heat flux component of the energy balance does not constrain satisfactorily the predictions of the other components of the energy balance. When conditioning on all heat flux objectives, significant additional constraint of the feasible parameter space is achieved with a consequent reduction in the predictive uncertainty. There are still, however, many parameter sets that adequately reproduce the calibration/validation data, leading to significant predictive uncertainty. Surface temperature measurements, whilst also subject to uncertainty, may be employed usefully in a multi-objective calibration of SVAT models.