Re-imagining standard timescales in forecasting precipitation events for Queensland’s grazing enterprises

The typical presentation of precipitation and climate information is organised according to the Gregorian calendar defined by the months and, in alignment with the temperate, savanna and desert climate zones of Australia, the three monthly seasons. While this is sufficient for many human-centred operations and the currency acceptable norm, grazing land managers (and potentially workers in other agricultural based enterprises) are reportedly restricted in their use of precipitation and climate information presented in this form. Compounding this issue with standard temporal packaging of climate information is the lack of reliability of the forecasts and spatial resolution capacity in the existing precipitation prediction tools that are being promoted to the graziers. Due to a lack of temporally and spatially robust information, graziers are managing their operations in the presence of significant risks, threatening their contributions to Australia’s $17 billion red meat and other industries. Grazing land managers require access to near real-time drought data that will enable the timely and informed decisions to be made about the movement of stock from the cattle stations to the grazing and/or growing properties. Providing graziers with this information having appropriate temporal resolution of the drought status for specific locations will enable the most productive use of the available grazing lands to grow the cattle to specified weights before slaughter. Hence a novel forecasting approach using data driven models of the monthly rainfall decile drought index (RDDI) is being considered in this paper, where the calculation of relative monthly indices are updated on a running weekly basis, providing land managers with spatially and temporally refined information. A similar process is proposed for the determination of relative seasonal rainfall indices, in addition to the consideration of the alternative definitions of Australian seasons as identified in existing literature. In future development of this research work, this approach will be used to forecast precipitation patterns, where the machine learning models’ architecture will be trained and evaluated with historical records of precipitation and other significant climate variables from a selection of sites relevant to the cattle industry around Queensland, Australia. The forecasts are to be derived from the novel implementation of a data intensive hierarchical categorizing support vector machine framework (or alternatively a regression-based data intelligent model) which is being proposed to deliver the graziers with the appropriate information to plan their operations within the stochastic nature of Australia’s climate. When compared with the seasonal and calendar monthly deciles, the more frequent forecast feeds (i.e., over weekly updated drought status, yet utilizing the concept of decile-based drought) presents a more detailed and robustly reported distribution of rain over the future seasons at specific sites, whilst catering to the graziers’ reported decision making processes. The more temporally refined presentation of the predicted rainfall events, for specified sites, has the potential to provide graziers (and other agricultural ventures), with the most relevant information to allow for more confident and profitable management of their enterprises