Sensitivity analysis and calibration of multi energy balance land surface model parameters

Abstract: Flows of energy between the atmosphere, the oceans and the land surfaces drive weather and climate on Earth. Increased understanding of these processes is crucial to successfully predict and address the challenges of climate change. Land surface models (LSM) are mathematical models designed to mimic natural processes and evolution of land surfaces with the basic task to simulate surface-atmosphere energy flows. Within the SURFace EXternalisée modeling platform (SURFEX), developed by Météo-France and a suite of international partners, a new LSM called the Interaction Soil Biosphere Atmosphere model - Multi Energy Balance (ISBA-MEB) has been developed. There are however still uncertainties in how to accurately prescribe model parameters used to numerically define the physiography and natural processes of modelled land surfaces which consequently results in uncertainties in modelled outputs. In the present study, Quasi-Monte Carlo simulations based on Sobol sensitivity analysis was applied to explore the uncertainty contribution of individual parameters to modelled surface-atmosphere turbulent sensible and latent heat fluxes in forest environments. Those parameters to which modelled fluxes were identified as significantly sensitive were then calibrated by generating multiple sets of parameter values with the Latin Hypercube sampling technique on which the model was run to identify what parameter values generated the least amount of model output bias and to evaluate how much model output uncertainty could be reduced. To explore variations in parameter sensitivity and optimal parameter prescriptions between forest environments, four separate forest areas with varying vegetation types and climate classifications were modelled. Results disclose that the level of uncertainty contribution of individual parameters varies between forest environments. Three parameters were however identified to contribute with significantly output uncertainty; 1) the ration between roughness length of momentum and thermal roughness length, 2) the heat capacity of vegetation and soil and 3) the leaf orientation at canopy bottom. Calibrating these parameters marginally reduced model output uncertainty at all study areas.