This dissertation addresses the need for improved estimates of land-atmosphere carbon dioxide and energy exchange at two scales: that of a homogeneous canopy and that of a heterogeneous landscape composed of multiple cover types. At the scale of a homogeneous canopy, a large array of data now exists from the growing network of eddy covariance flux towers (ie. FluxNet). These data provide both a temporally dense observational record with which to analyze ecosystem processes, as well as a rich information source to constrain land surface models. The scale of a heterogeneous landscape, from which the integration of surface fluxes into estimates that realistically account for landscape heterogeneity, and from which regional budgeting is possible, provides a challenge as to the method of spatial integration and appropriate constraints on this process.

Significant improvement in estimation skill will require the synthesis of land surface models with multiple observation types and across scales. The goal of the work contained in this dissertation is to fill several gaps in our ability to merge the information contained in eddy covariance flux measurements, canopy-scale and atmospheric boundary layer concentration and temperature observations, and models of canopy-atmosphere interactions to improve our ability to estimate land-atmosphere fluxes. To achieve this goal, models including a detailed multi-layer canopy model, a low-dimensional formulation coupling canopy-atmosphere carbon dioxide and vapor exchange, and a coupled atmospheric boundary layer—canopy model are all applied. Several techniques are developed to merge observations and models at the appropriate scale. The results demonstrate that improvements in model skill can be made by the careful consideration of model characteristics and the type of information contained in unique observation sources. Examples of these improvements include the identification of model error sources in multi-layer canopy models, methods for applying and evaluating the impact of multiple flux constraints (ie. CO ₂ and H₂O flux) on land surface models, and landscape-scale identification of surface properties and improved flux estimation by spatial extension of the model control volume in a data assimilation context.