Assessment of a process-based model to predict the growth and yield of Eucalyptus grandis plantations in South Africa.

Abstract

It is believed that the process-based model 3-PG (Physiological Principles Predicting Growth; Landsberg and Waring, 1997) can potentially play a useful role within South African forestry, both as an operational and a strategic tool. Strategic applications may include the prediction of potential productivity on a site-by-site basis; broadscale productivity estimates based on remote sensing and the spatial application of 3-PG; identification of production constraints; and estimation of carbon fluxes to help address sustainability issues. Operationally, 3-PG could complement empiricallybased models or be used in conjunction with them as a hybridised product. The challenges of this study were therefore to see whether it is possible to adapt 3-PG to predict the growth and yield of E. grandis under South African conditions, test that model predictions are consistent with observed growth data and are biologically reasonable, and to assess the practicality of using 3-PG as either a strategic or operational tool. The main emphasis of this study was to understand the internal logic of 3-PG and how physiological processes are represented, and to develop methods to objectively parameterise and initialise the model. Thereafter a detailed model validation study was performed, ending off with selected potential applications of 3-PG within the South African context. The sensitivity of predicted stand volume (SV) and leaf area index (LAI) to the values of the species-specific parameters in 3-PG was examined. These analyses enabled the development of three distinct parameter sensitivity classes: insensitive parameters (i.e. those that can be varied widely without affecting the outputs studied), sensitive parameters (i.e. those whose value strongly affects the outputs, and non-linear parameters (i.e. those for which the outputs depend in a non-linear manner on the parameter value). Minimum data requirements for the parameterisation and initialisation of 3-PG are considered in detail. Conventional methods used for the parameterisation of models, specifically 3-PG, are reflected upon. An automated parameter estimation technique was examined and used for the estimation of selected parameters. Species-specific parameters were categorised according to data source estimation and sensitivity classes. Parameters describing allometric and age-dependent relationships were assigned values using observed data from biomass harvests. Critical parameters that could not be directly assigned using observed data were the ratio of foliage to stem allocation (i.e. P2 and p2o), allocation of net primary production (NPP) to roots (TJRX and T]Rn), optimum temperature for growth (7^,) and maximum canopy quantum efficiency (acx)- These were estimated using Parameter ESTimation, by fitting model output to corresponding observed growth data. As well as species-specific parameter values, mandatory inputs required by 3-PG include weather data, site-specific factors such as site fertility (FR) and physical properties of the soils, and stand initialisation data. Objective techniques to determine these site-specific factors and the initial values for the biomass pools were proposed. Most of these data are readily available for sites where experimental trials exist, or where monitoring networks are in place. However, this is the exception rather than the rule, so alternative data and information sources are required. These, together with the need for accurate weather inputs (especially monthly rainfall) and physical properties (especially soil texture, maximum available soil water and FR) of the sites being modelled were explored. 3-PG was validated using four simple tests by comparing predicted versus observed SV. Results showed that 3-PG predictions are relatively consistent with observed stand data. Analyses performed using time-series data showed model predictions accurately tracked observed growth in response to erratic and fluctuating weather conditions. Results from the initial model validation showed production on high and low productivity sites was under- and over-predicted, respectively. Further results presented here show a similar, but less marked trend (i.e. over- and under-predictions are not as extreme), and individual biases are less than those from predictions made using another locally developed parameter set. The application of 3-PG showed that the model is able to make estimates of tree growth that are consistent with those used within the forestry site classification. This showed the considerable potential 3-PG has for strategic planning by the forest industry (i.e. projected wood supplies etc) and in research planning (refining existing site classifications). The model could be useful in predicting growth in various areas where E. grandis is not grown, assisting in future decision making. 3-PG was able to identify growth constraints on a site-by-site basis and distinguish among them, and was able to identify environmental and site limitations to plantation growth, and how they vary in space and time. These results together with predictions of site productivity demonstrate the potential for 3-PG to be used to improve existing forest site classifications. The model comparison study between empirically-based models and 3-PG showed that although the empirical models made accurate predictions of volume under static climatic conditions, under fluctuating weather conditions empirical estimates of volume were less accurate than those made with 3-PG. 3-PG can therefore be used operationally with minimum input data to predict growth using enumeration data. This is useful in saving time and cutting costs. The use of process-based models (PBMs) in general, and 3-PG in particular, needs to be "championed'' to the South African forest industry. This is necessary for two reasons. Firstly, the model and the manner with which it describes physiological processes of growth need to be explained in layman's terms. This will demonstrate how and why a process-based model can work better in a fluctuating environment than empirically based models. Secondly the comparison between 3-PG and the local empirical models needs to be presented as an example of how 3-PG can be applied on an operational basis. It is accepted that much convincing is still required.