The land (or ‘terrestrial’) biosphere strongly influences the exchange of carbon, energy and water between the land surface and the atmosphere. The size of the land carbon store and the magnitude of the interannual variability of the carbon exchange make models of the land surface a vital component in climate models. This thesis addresses two aspects of land surface modelling: soil respiration and phenology modelling, using different techniques with the goal of improving model representation of land-atmosphere interaction. The release of heat associated with soil respiration is neglected in the vast majority of large-scale models but may be critically important under certain circumstances. In this thesis, the effect of this heat release is considered in two ways. Firstly, a deliberately simple model for soil temperature and soil carbon, including biological heating, is constructed to investigate the effect of thermal energy generated by microbial respiration on soil temperature and soil carbon stocks, specifically in organic soils. Secondly, the mechanism for biological self-heating is implemented in the Joint UK Land Environment Simulator (JULES), in order to investigate the impacts of the extra feedback in a complex model. With the intention of improving estimates of the parameters governing modelled land surface processes, a data assimilation system based on the JULES land surface model is presented. The ADJULES data assimilation system uses information from the derivative of JULES (or adjoint) to search for a locally optimum parameter set by calibrating against observations. In this thesis, ADJULES is used with satellite-derived vegetation indices to improve the modelling of phenology in JULES.

The scale-bound view of landsurfaces, being an assemblage of certain landforms, occurring within limited scale ranges, has been challenged by the scale-free characteristics of fractal geometry. This thesis assesses the fractal model by examining the irregularity of landsurface form, for the self-affine behaviour present in fractional Brownian surfaces. Different methods for detecting self-affine behaviour in surfaces are considered and of these the variogram technique is shown to be the most effective. It produces the best results of two methods tested on simulated surfaces, with known fractal properties. The algorithm used has been adapted to consider log (altitude variance) over a sample of log (distances) for: complete surfaces; subareas within surfaces; separate directions within surfaces. Twenty seven digital elevation models of landsurfaces arc re-examined for self- affine behaviour. The variogram results for complete surfaces show that none of these are self-affine over the scale range considered. This is because of dominant slope lengths and regular valley, spacing within areas. For similar reasons subarea analysis produces the non-fractal behaviour of markedly different variograms for separate subareas. The linearity of landforms in many areas, is detected by the variograms for separate directions. This indicates that the roughness of landsurfaces is anisotropic, unlike that of fractal surfaces. Because of difficulties in extracting particular landforms from their landsurfaces, no clear links between fractal behaviour, and landform size distribution could be established. A comparative study shows the geomorphometric parameters of fractal surfaces to vary with fractal dimension, while the geomorphometry of landsurfaces varies with the landforms present. Fractal dimensions estimated from landsurfaces do not correlate with geomorphometric parameters. From the results of this study, real landsurfaces would not appear to be scale- free. Therefore, a scale-bound approach towards landsurfaces would seem to be more appropriate to geomorphology than the fractal alternative.