Evaluation of parameter sensitivity for different levels of land-surface model complexity

The importance of properly representing the land surface when modeling the general climate system is now widely recognized. Since the mid-1960s, and incorporation of the first "Bucket" model into General Circulation Models (GCMs), increasingly complex land-surface schemes have been introduced to better represent energy and water fluxes over the Earth’s surface. These Land Surface Schemes (LSS) have numerous parameters that must be estimated for the various vegetated and non-vegetated surfaces across the globe. Typical parameter estimation has been via a global land-surface classification scheme with standard values for various land-cover types. These "default" values are then either used as given or are sometimes manually adjusted (calibrated) via comparison of the off-line model performance against site-specific field data. Recently, studies have emerged to investigate the sensitivity of the global climate to the parameter values used in these land-cover schemes. Most of the LSS sensitivity studies have been based on the evaluation of a "one parameter at a time" effect on the land system. This study uses a recently developed systems-based approach to evaluate the sensitivity of energy and water fluxes to the parameter values used in LSS. Five land-surface types (grassland, boreal forest, semi-arid vegetation, agricultural cropland, and tropical rain forest) are analyzed using the Multi-Objective Generalized Sensitivity Analysis (MOGSA). Sensitivity of energy and water fluxes to parameters is evaluated in LSS with varying levels of complexity, from the simple Bucket model to the Biosphere-Atmosphere Transfer Scheme (BATS) and NOAH Land Surface Model (LSM). Random parameter sets are generated with a set percentage change in the baseline default values. Root mean squared error (from observed values), along with changes in latent and sensible heat fluxes, are evaluated for each parameter set. Preliminary results using the BATS2 model indicate varying sensitivity to parameters based on land-cover type. Semi-arid sites appear to have more sensitivity to parameter values, with larger variations in fluxes, while temperate and boreal regions appear to be less sensitive. Analysis of the sensitivity of parameters at various land-cover types will lead to better estimation of parameters at specific biomes around the world, increasing confidence in predictions of future climate from deforestation and other anthropogenic impacts on the land surface.