AI-Enhanced Seasonal Climate Forecasts for Agriculture: Generating Skillful Daily Precipitation Ensembles with Advanced Generative Models
This project leverages cutting-edge generative AI models to transform global climate data into realistic daily rainfall forecasts. By capturing rainfall patterns and extremes, it enables early warnings for agricultural droughts, guiding crop planning and water management seasons in advance.
Project Outline:
Building on previous successes in high-resolution precipitation downscaling using generative models, extreme rainfall forecasting, soil moisture prediction, and long-term time series forecasts, this project aims to harness advanced generative deep learning models—such as GANs and diffusion approaches—to transform Global Climate Model (GCM) outputs into realistic daily precipitation ensemble forecasts. The objective is to develop early warning systems for agriculture and agricultural drought (with a focus on soil moisture deficits) one to two seasons in advance. Beyond predicting total rainfall, the system will capture key rainfall characteristics such as spatial distribution, dry-day frequency, persistence of dry spells, and the occurrence of extreme events—critical information for agricultural planning and risk management.
The research will focus on a core set of GCM predictors most relevant to rainfall variability, including raw precipitation forecasts, sea surface temperatures (SST), geopotential height, specific humidity, and zonal and meridional winds. To enhance robustness and capture uncertainty, multiple GCMs (e.g., ACCESS-S2, ECMWF SEAS6, GloSea5) will be incorporated. The generative models will be trained to map these predictors into skilful precipitation ensembles, benchmarked against observed rainfall datasets, extreme events, and agricultural drought indices. By leveraging prior experience in high-resolution downscaling and long-term time series forecasting, this project aims to deliver a next-generation early warning system capable of informing crop planting, irrigation scheduling, and drought preparedness under a changing climate.
From engaging in this project the student will learn how to:
Gain hands-on experience with high-performance computing.
Master seasonal climate forecasts from modern climate models, including their limitations for agricultural applications.
Learn generative AI methods (GANs, diffusion) to transform GCM outputs into realistic rainfall ensembles.
Develop applied research skills in interdisciplinary collaboration, early warning system design, and scientific communication.
Questions about this project can be directed to warren.jin@data61.csiro.au
To register an expression of interest, click here. You will need to outline why you have selected the research project and how your skills, experience and/or knowledge meet the project requirements.