Chapter 7 - A Multimethod Global Sensitivity Analysis Approach to Support the Calibration and Evaluation of Land Surface Models

Abstract

Land surface models (LSMs) are used to predict and understand processes at the Earth surface. The parameters of such models are often calibrated and validated using Earth observation (EO) data such as soil moisture content and surface energy fluxes. Due to the complexity of LSMs, which encompass many interacting components and parameters, manual approaches to calibration and validation are impracticable and formal approaches must be employed. Global sensitivity analysis (GSA) provides a structured methodology to investigate uncertainty propagation through complex simulation models, and in particular to quantify the relative influence of model input factors, e.g., parameters, on simulation outputs. This information can be used to focus computationally intensive calibration procedures on the most influential parameters, and more broadly to evaluate the consistency between the model's behavior and the physical processes that it should reproduce. In this work, we demonstrate the use of GSA to support the calibration and evaluation of the Joint UK Land Environment Simulator (JULES), a global LSM developed and currently employed by the UK Met Office. We investigate the influence of nine parameters and three initial conditions on a set of model outputs capturing the model's accuracy in reproducing EO data like soil moisture and heat fluxes. We illustrate and apply a multimethod approach that includes several state-of-the-art GSA methods, namely, regional sensitivity analysis and variance-Based (Sobol') analysis, and a novel density-based approach, named PAWN. The multimethod approach provides information about model sensitivities from different angles, with no additional computing cost with respect to the application of an individual method.