Canopy structure and compensating biases in models of ecosystem function

Incorporating realistic canopy structure into models of the land surface and ecosystem functioning is important for numerous reasons. In particular it will improve predictions of the quantity and distribution of absorbed shortwave radiation. This in turn results in improved estimates of photosynthesis as well as better forward modelling of satellite observations for the purpose of data assimilation. Typical models use a one dimensional turbid medium representation, but vertically and horizontally varying structure adds significant realism to the canopy. However results from this study demonstrate that care must be taken when incorporating complex canopy schemes as they will likely reveal processes elsewhere in the model that are incorrectly parameterised to compensate for the lack of variability in the original canopy representation.

The Geometric Optic Radiative Transfer (GORT) model has previously been used as an observation operator for the Data Assimilation Linked ECosystem (DALEC) model to enable the assimilation of MODIS surface reflectance data. However the GORT model was not used to replace the light interception equations inside DALEC itself. The fraction of absorbed photosynthetically active radiation (fAPAR) predicted by GORT for a given canopy structure and leaf area index (LAI) will likely not be the same as that from DALEC (which uses a simple empirical relationship between fAPAR and LAI) for a corresponding canopy. Furthermore it is not possible to calibrate the DALEC fAPAR equation to fully reproduce the variability of the GORT fAPAR with respect to any given illumination conditions.

In this study the GORT model was used in place of the corresponding equations in DALEC. Previous experiments assimilating MODIS reflectance data were then repeated but results in terms of ecosystem scale carbon fluxes were not improved. Furthermore, as greater realism is added into the modelling experiments (for example the inclusion of realistic variation in sun angles) the prediction of carbon fluxes tended to get worse. This is due to biases within the model that compensate for missing processes, in particular accounting for the lack of seasonal variation in fAPAR in the original model formulation. Experiments examining individual parameters in the model highlight where those compensating biases exist.