Climate and weather forecasting

To help farmers make paddock-level decisions based on current-season as well as multi-year weather and climate forecasts

Most agricultural sectors are exposed to the vagaries of weather, which directly affects their productivity and economic return and indirectly affects them through global changes in prices. Next gen weather and climate forecasting capability and tools will help Australia’s agricultural and land management industries with enterprise planning, investment and profitable decision-making.

Decadal climate forecasting

What could the agricultural sector do if they had much longer term climate forecasts? We are harnessing our work developing a multi-year climate forecasting system to include key relevant Digiscape activities in beef, grains and carbon. We’re using our Climate Analysis Forecast Ensemble (CAFE) system to deploy advanced data assimilation and ensemble generation methods for high tech climate forecasting. We’re also generating a new set of forecasts every year for the World Meteorological Organisation’s World Climate Research Programme Decadal Climate Prediction Project.

A key challenge in advancing state-of-the-art climate forecast systems is to predict the onset of key climate modes of variability.

The primary aim of this project is to understand the basis of this climate variability and to forecast enough variability to be useful to sophisticated users of climate information such as land sector decision makers.

 

Paddock-scale and in-season forecasting

Each farming application responds to different features of the weather and climate as they unfold year by year. The success of a wheat crop, for example, is sensitive to the timing of rainfall through the year and whether it is accompanied by optimal temperatures and sunlight. On the other hand, the onset of the wet season is key to land managers in northern Australia. Having season-level forecasts could help farmers implement better management decisions for their particular enterprise.

Climate forecasts currently exist at ‘low res’ large scales (hundreds of kilometres) but we are turning them into paddock-scale, application-relevant information. A grain grower, for example, doesn’t just need to know that 70% above average rainfall is forecast for the year. But translating that into what variety of wheat to plant or how much nitrogen fertiliser to apply is where Digiscape is aiming to help.

We are integrating our expertise in weather and climate forecasting with individual Digiscape real-world use cases such as sugarcane, grains and aquaculture. Using seasonal climate forecast (SCF) downscaling techniques, we will explore what different weather and climate features each are sensitive to and quantify the confidence we can have in forecasts at the season-scale relevant to them.

This work has developed innovative climate downscaling techniques, building on CSIRO’s existing modelling techniques. We’ve been able to generate extreme-aware downscaling techniques and spatio-temporal modelling techniques to downscale SCFs to locations even without any historical weather observation data. More about CSIRO’s modelling techniques is here.

We’ve also used and developed other up-to-date techniques to create innovative downscaling solutions for SCFs. For example, we’ve extended Bayesian model averaging to generate a consensus SCF from multiple climate models with higher forecast skills, and you can see visualisation examples here. We’ve also successfully used deep learning for downscaling SCFs for the first time, which opens a promising door for future operational SCFs.

 

Find out more about Digiscape’s climate and weather forecasting 

Watch team member Guobin Fu present the science about ‘coherent ensemble forecasts of surface and groundwater availability’.

Select publications are below.

Beischer TA, Gregory P, Dayal K, Brown JR, Charles AN, Wang WXD, Brown JN (2021) Scope for predicting seasonal variation of the SPCZ with ACCESS-S1. Climate Dynamics 56, 1519–1540. https://dx.doi.org/10.1007/s00382-020-05550-6

Herath P, Thatcher M, Jin HD, Bai XM (2021) Effectiveness of urban surface characteristics as mitigation strategies for the excessive summer heat in cities. Sustainable Cities and Society 72, 103072. https://dx.doi.org/10.1016/j.scs.2021.103072

Bakar KS, Biddle N, Kokic P, Jin HD (2020) A Bayesian spatial categorical model for prediction to overlapping geographical areas in sample surveys. Journal of the Royal Statistical Society Series A 183, 535-563. https://doi.org/10.1111/rssa.12526

Chen WJ, Xu C, Zou B, Jin HD, Xu J (2019) Kernelized elastic net regularization based on Markov selective sampling. Knowledge-Based Systems 163, 57-68. https://dx.doi.org/10.1016/j.knosys.2018.08.013

Huang X, Song JY, Jin HD (2020) The casualty prediction of earthquake disaster based on Extreme Learning Machine method. Natural Hazards 102, 873-886. doi:10.1007/s11069-020-03937-6

Li M, Jin HD (2020) Development of a postprocessing system of daily rainfall forecasts for seasonal crop prediction in Australia. Theoretical and Applied Climatology 141, 1331-1349. doi:10.1007/s00704-020-03268-3

Li M, Jin HD, Brown JN (2020) Making the output of seasonal climate models more palatable to agriculture: a copula-based post-processing method. Journal of Applied Meteorology and Climatology 59, 497-515. doi:10.1175/JAMC-D-19-0093.1

O’Kane TJ, Squire DT, Sandery PA, Kitsios V, Matear RJ, Moore TS, Risbey JS, Watterson IG (2020) Enhanced ENSO prediction via augmentation of multimodel ensembles with initial thermocline perturbations. Journal of Climate 33, 2281–2293. doi:10.1175/JCLI-D-19-0444.1

Bakar KS (2019) Interpolation of daily rainfall data using censored Bayesian spatially varying model. Computational Statistics 35, 135-152. doi:10.1007/s00180-019-00911-0

Crimp SJ, Jin HD, Kokic, P, Bakar KS, Nicholls N (2019) Possible future changes in South East Australian frost frequency – an inter-comparison of statistical downscaling approaches. Climate Dynamics 52, 1247-1262. doi:10.1007/s00382-018-4188-1