Project Outline:

Digital agriculture has the potential to enhance the profitability and sustainability of agriculture by supporting on-farm decision making within complex farm systems. Process based simulation models have existed since the 1950s. Advances in agricultural science and computational power in recent decades have supported the development of complex simulation models which can accurately predict crop growth under a range of climate conditions and farm types. The Agricultural Production Systems Simulator (APSIM) is one of the most advanced process driven crop simulation models in the world, being widely used for crop simulation research. However, the APSIM model is complex to set up, requiring accurate calibration and parameterisation, with computationally time-consuming simulation processes. These factors have limited the application of simulation models like APSIM in agronomy advisory services.

The use of modern machine learning methods to build predictive emulation models is an emerged field which has the potential to increase computational speed while maintaining model functionality and data fidelity, allowing for near real-time predictions for decision support. An important part to the development of any decision-support tool is the data that goes into it and the scale at which the data and model operate. While the APSIM model is primarily developed to perform well at the sub-field level, decision support may be needed at the paddock, farm, regional, or even national level. This project will attempt to bridge this gap by firstly, curating relevant spatio-temporal data for crop production (model simulations, remote sensing, in-situ field data) under different management and climate contexts, and secondly, developing methods for predicting harvest outputs at different spatial and temporal resolutions with uncertainties to support decision-making activities that could be captured in a farm simulation gaming tool similar to the FarmSim game. Specifically, this project will:

(1) build (and automate) a well-documented data cube that provides a robust training dataset to be used for the development of emulators; and

(2) create a farm emulator – one of the challenges that growers face is knowing which paddock to focus their attention on today. The APSIM crop model, when appropriately parameterised, can provide insights into the expected crop growth. These data can be reconciled against UAV imagery or satellite imagery, to assist growers to focus their attention on emerging problems. To create the informatic, the student will design a rapidly calibrated APSIM model / emulator that can help growers evaluate a field in real time, and act if needed. Ideally, the technique will scale to the entire farm, and assist farmers organise their effort.

(3) As a further stretch goal, this emulation activity could be captured in a gaming environment to explore farming decisions in a more generic environment that is fun and interactive for a range of end users and stakeholders.

The student will have a background in data science, statistics, or machine learning and be able to program (e.g. R  or python) skills are a good start.

The PhD student will develop an in-depth understanding of spatio-temporal modelling methods that scale and incorporate uncertainty with a focus on agricultural crop models.  They will also gain experience in computational approaches for emulation that use machine learning to increase simulation times.

The PhD project will be supported by a multidisciplinary team of researchers with experience in farming systems modelling, remote sensing, data science and machine learning, as well access to CSIRO’s broader team working in digital agriculture in the pursuit of innovative solutions for resilient farming.

The student will learn:

The successful candidate will be provided with:

1) Training in crop simulation modelling and associated data analysis

2) Training in deep learning applications to develop emulators

3) Exposure to, and understanding of, the latest crop sensing techniques.

4) Transferable core scientific skills including data management, analysis, presentations, paper writing, and peer review, as well as working in a multi-disciplinary team.

To register an expression of interest, click here. You will need to outline why you have selected the research project and how your skills, experience and/or knowledge meet the project requirements.