CSIRO intern - Sam MacDonald

Immediately at commencement of my internship with CSIRO, I was made to feel part of the team. Their collaborative culture made it easy to resource ideas and correct myself when astray, helping me learn and contribute the best I could. The experience will always be remembered as a key milestone in my development.

My objective was to forecast nitrate levels in a riverine environment to assist sugarcane farmers reduce nitrate runoff, whilst maximising yield. Using a long-short term memory (LSTM) neural network, nitrate levels were predicted up to eight hours into the future using input features of upstream rainfall, local rainfall, water level, turbidity, and nitrate concentration. Basic statistical inference and a small literature review elucidated the modal interplay of the source-mobilisation-delivery (SMD)-continuum of nutrients in catchments. This insight directed the feature design to restrict LSTM learners to periods of relatively constant hydrological conditions, which in turn significantly improved the predictive performance. On reflection, it was interesting to see how integral the basic statistics were to applied deep learning; I since have reshaped my study plan, accordingly.

I'd highly recommend everyone to apply. Like me, you may be surprised by the outcome. My advice would be to identify what experiences you have in common with the employer’s problem setting. Perhaps make that a point of emphasis when applying for a position.