[Truncated abstract] This thesis is presented as a series of four papers which describe new procedures for developing simple statistical models for accurately predicting perennial pasture production, and more generally agricultural production or yield, such as for crops, fruits, or pastures, based on relatively simple inputs. The models currently available for this purpose range from data-driven or empirical models to process-driven or mechanistic models. Difficulties with empirical models arise in situations where there is a scarcity of real or experimental data, as may occur in cases of newly developed crop and pasture species or cultivars. The disadvantages of mechanistic models are their complexity and resulting opacity, their high computational time requirements and the difficulties of extending them to new soils and sites and updating them with new trial data. Developing an alternative model which avoids such difficulties was the motivation behind the work presented in this thesis. Specifically, a simple statistical model or emulator was developed utilising extensive data obtained from an existing comprehensive mechanistic agricultural production model. The question addressed by this study was whether emulators can be generated and successfully used to accurately predict agricultural production. In this work, the biomass production of the perennial pasture species lucerne (Medicago sativa) was used as a case study in developing the emulator and testing its performance. The original lucerne production data for model creation and statistical validation were obtained from APSIM (the Agricultural Production Systems sIMulator) and graphically examined to ascertain general trends and features (Chapter 2). The resulting linear and cubic spline models were found to provide relatively simple emulators of APSIM lucerne biomass production (Chapter 3).
|Qualification||Doctor of Philosophy|
|Publication status||Unpublished - 2012|