SCOPE Project on Tree-Grass Dynamics

Document 4, March 1999

Modelling Workshop 1 Summary Report:

Modelling the Mulga data, South Africa, November 1998

As part of the ongoing work of the SCOPE tree-grass group, the first modelling workshop was held in South Africa in September 1998 to begin the "Level 2" modelling analyses.

The aims of this workshop were to

- ensure all models were working well and producing reasonable reaults

- improve understanding of the different models and their approaches (controls and drivers)

- test the models against real data

- begin to examine some of the key savanna theory questions (eg. effect of increasing "treeness" on grass production and total ecosystem production - see Working Paper 1)

The four models involved were GRASP (Joe Scanlan, Robert Wicks Research Centre, Australia), CENTURY-SAVANNA (Bill Parton, NREL, USA), SAVANNA (Mike Coughenour, NREL, USA) and MUSE-TreeGrass (Jaques Gignoux, ENS, France & Xavier Le Roux, INRA, France), they each differ in complexity and modelling approach and are at different stages of development (for model descriptions see Santa Barbara meeting report and upcoming papers). The meeting was organised and hosted by Bob Scholes (CSIR, S. Africa). Compilation and analysis of results was carried out by Jo House (King’s College London). Other local participants were Steve Higgins (University of Witwatersrand); Thiambi Netshilurhi (CSIR), Charlie Shackleton (CSIR) and Greg Kiker (CCWR). The work was continued by the modelling groups involved and consolidated in a meeting of Bill Parton and Jo House at King’s College London in February 1999. The aims of the workshop were achieved and are being written up as a scientific paper. Further areas of work were identified.

The models were parameterised using an arid Mulga (Acacia anuera) shrubland site in Australia which had tree and grass biomass data for three different tree density treatments. Each model was run for each of the tree densities over a sixteen year period using real climate data. Each model was also run over the same period with no trees to provide a control against which the different density treatments could be compared. CENTURY and SAVANNA "grew" trees within the model while GRASP and MUSE were forced with increasing treeness data. A variety of analyses were carried out. The detailed results can be found in the more detailed draft Working Paper 2 (in preparation).

Each of the models showed similar seasonal patterns of grass production over time, MUSE predicted around twice as much as GRASP. CENTURY was highly correlated with rainfall and all the models were highly correlated with transpiration except GRASP. Standing crop at different tree densities was reasonably predicted by all models with fairly similar root mean squares, although some showed bias, eg. CENTURY tended to overestimate at all treatment densities, SAVANNA underestimated at low density and overestimated at high density, while MUSE overestimated at low and medium densities. Tree basal areas were well captured by SAVANNA and CENTURY. The preliminary results are encouraging. Further model testing and model validation should be carried out using other data sets.

The models produced different results for the theory driven analyses. SAVANNA and MUSE indicated a virtually linear decline of grass NPP with increasing treeness, while GRASP and CENTURY showed a more exponential decline. SAVANNA showed an increasing trend in total ecosystem production with increasing treeness, while no clear trend was observed for the other model results. Finally, GRASP indicated that grasses did better in higher production years compared to trees whereas CENTURY and SAVANNA indicated they did better in lower production years. Further modelling experiments are needed to test these theories further for this particular site and to test if and how these relationships change with different site conditions (eg. climate and soils).