This is a final research report from Hort Innovation’s historical archives. Please note that as these reports may date back as far as the 1990s, the content and recommendations within them may be superseded by more recent research.

What was it all about?

This project (‘Better Macadamia Forecasting Part 2’) ran from August 2009 until April 2015. It followed on from the earlier macadamia forecasting projects, which commenced in 2000. This report integrated and covered methods and results for the overall project, rather than focusing specifically on ‘Part 2’.

Accurate crop forecasts for the Australian macadamia industry were required each year, to facilitate planning and marketing. This project had produced crop predictions for the whole industry since 2001, and more recently for each of the six separate production regions. These forecasts have been forwarded to the Australian Macadamia Society (AMS) in March each year, forecasting that year’s crop (which was then harvested throughout the remainder of the year).

There were two complementary levels of forecasting:

Firstly, the overall longer-term forecast was based on tree census data of growers in the AMS, scaled up to include the non-AMS orchards. Expected yields were based on historical data, with a nonlinear regression model incorporating tree age, variety, year, region and tree spacing. This long-term model forecasts expected production for six to 10 years into the future.

The second level of crop prediction was an annual climate-based adjustment of these overall long-term estimates, taking into account the expected effects of the previous year’s climate on production. The dominant climatic variables were observed temperature, rainfall and solar radiation, and modelled water stress. Based on the proven forecasting success of boosted regression trees and ‘random forests’, the average forecast from an ensemble of regression models was adopted (rather than using a single best-fit model). Exploratory multivariate analyses and nearest-neighbour methods were also used to investigate the patterns in the data.

In parallel, a survey of growers and pest scouts was also conducted early each year, with their estimates of the coming crop being integrated into regional and then overall totals.

Real-world problems, including flooding rain during harvest and the destructive winds of ex-tropical cyclone Oswald, have obviously affected the accuracy of the forecasts that were made early each year. There were also major problems between 2008 and 2011, when industry yields in some regions were well below the levels previously achieved, and this was attributed mainly to management problems caused by lower prices.

These forecasting methods have been evolving over the past decade, and the recent years have shown average absolute error rates of 6.8% for the growers forecast, and 8.6% for the climate forecasts. These were within the targeted ±10%, and also compare well with other crop forecasting applications around the world.

The resources required to continue the growers’ forecasts were quite minimal, and it was recommended that these continue. The long-term forecasts were based on the now-somewhat-dated census data of the AMS, and these (or valid alternatives) would need to be updated and revised for this aspect to continue.