Towards operational forecasting of agricultural soil water in Australia

Abstract

Australia has large and diverse agricultural industries, in addition to a highly variable climate. Recent government policy has focussed on increasing farm productivity. One of the challenges for achieving this, especially for rain-fed agriculture, is improving local-scale information on current and forecast future soil water status. Reliable soil water forecasts will allow users to better predict likely future water availability and make better informed farming decisions over timescales from days to seasons. This paper explores the state-of-the-art of the components that need to be brought together to transition our research knowledge into operational soil water forecasts, and the challenges that lie therein. Achieving operational agricultural soil water forecasts requires bringing together a diverse range of components, including: Long-term soil water monitoring - high quality profile soil water monitoring datasets extend to 10 years for some sites, while remote sensing data for surface (<5cm) soil water extends to fifteen years and beyond. These provide fundamental data against which soil water simulation 'hindcasts' and forecasting methods can be developed and tested. Remote sensing and data assimilation - the resolution, frequency, quantity, and inter-product consistency of remote observations relevant to modelling of soil water is increasing. This includes dedicated satellites for soil water. Current technology, techniques, and infrastructure provide the capability to assimilate such data into soil water models on an ongoing operational basis. These datasets can improve both forecasting methods, by bounding hindcast predictions, and operational forecasts, by providing good initial conditions for predictions. Rainfall and temperature forecasts - multi-week to multi-season climate forecasts are a critical driver of soil water forecasts. These are currently produced weekly as daily ensemble (probabilistic) forecasts over a 250 km grid out to 270 days ahead, with significantly diminishing skill (accuracy) beyond about 3 months. Limited research to increase the resolution and skill of these forecasts is underway, with a focus on agricultural applications. Operational improvements, including increased spatial resolution, will occur in the near future. Soil water modelling - Operational earth system forecasting models and crop growth models are both capable of producing soil water as outputs. Each operates at extremes of the global spatial scale (landscape versus point-scale). Their skill to produce explicit probabilistic soil water forecasts as ensembles has yet to be fully evaluated. There is also a challenge to determine the best approaches to take advantage of, and integrate, rainfall forecasts, remote observations, and local scale soil and management information. Local scale inputs - The ability of soil to store water and make it available to plants varies widely. Local information on soil type and properties, soil water observations, and crop management increases the relevance of forecasts. They can also increase forecast accuracy by tuning the down-scaling of broad- scale forecasts to local conditions. Farm-scale monitoring and communication technologies have now matured to the point where they offer significant potential as a key input to the localisation of forecasts. Integrating and maturing these components in a structured and strategic manner is required for us to achieve the goal of useful operational seasonal forecasting of agricultural soil water in Australia.