Land surface modelling and Earth Observation of land/atmosphere interactions in African savannahs

Land/atmosphere feedback processes play a significant role in determining climate forcing on monthly to decadal timescales. Considerable uncertainty however exists in land surface model representation of these processes. This investigation represents an innovative approach to understanding key land surface processes in African savannahs in the framework of the UK‘s most important land surface model – the Joint UK Land Environment Simulator (JULES). Findings from an investigation into the carbon balance of Africa for a 25-year period from 1982 to 2006 inclusive show that JULES estimated Africa to behave as a carbon sink for most of the 1980‘s and 1990‘s punctuated by three periods as a carbon source, which coincided with the three strongest El Niño events of the period. From 2002 until 2006 the continent was also estimated to be a source of carbon. Overall, the JULES simulation suggests a weakening of the African terrestrial carbon sink during this period primarily caused by hot and dry conditions in savannahs. Applying the model further, land surface temperature (LST) displayed large uncertainty with respect to savannah field measurements from Kruger National Park, South Africa, and JULES systematically underestimated LST with respect to Earth Observation data continent-wide. The postulation was that a reduction in the uncertainty of surface-to-atmosphere heat and water fluxes could be achieved by constraining JULES simulations with satellite-derived LST using an Ensemble Kalman Filter. Findings show statistically significant reductions in root mean square errors with data assimilation than without; for heat flux simulations when compared with Eddy Covariance measurements, and surface soil moisture when compared with derivations from microwave scatterometers. The improved representation of LST was applied to map daily fuel moisture content – one of the most important wildfire determinants - over the mixed tree/grass landscapes of Africa, whereby values were strongly correlated with field measurements acquired from three savannah locations.