A coupled land surface-boundary layer model and its adjoint

In this paper, a simple coupled land surface-boundary layer model and its a djoint are presented. The primary goal is to demonstrate the capabilities o f the adjoint model as a general tool for sensitivity analysis and data ass imilation. The adjoint method was chosen primarily for two reasons: 1) the adjoint model can be used not only to obtain parameter sensitivities with g reater efficiency but, more important, to provide added insight into the se nsitivities as compared with that obtained with traditional simulation tech niques (e.g., pathways, time variations in sensitivity) and 2) the adjoint model can be used in a variational data assimilation framework to combine m easurements and the model of the physical system optimally in order to esti mate state variables and fluxes. Two simple examples are presented to illus trate how the framework can be used for performing both diagnostic sensitiv ity experiments and hydrologic data assimilation. In the sensitivity experi ment, temporal patterns and total influence of model states and parameters on average daily ground temperature are shown. In the synthetic data assimi lation example, the adjoint model is used as an estimation tool to initiali ze the coupled model through assimilation of ground temperature observation s. As a result, great improvement was gained in simulation of model states and surface fluxes based only on a minimal set of basic land temperature me asurements and the auxiliary parameters: incident solar radiation, large-sc ale wind speed, and free atmosphere profiles of temperature and humidity. F orthcoming studies will use the framework developed here to examine thoroug hly the consequences of using uncoupled versus coupled models of the land a nd the atmospheric boundary layer (ABL). In assimilation mode, the coupled surface-ABL model and its adjoint will be used to estimate surface fluxes a nd micrometeorological conditions based on remote sensing measurements of l and temperature and minimal auxiliary data.