Improving NWP post-processing at the Bureau of Meteorology

Statistical post-processing of numerical weather prediction (NWP) guidance is a crucial element of the forecast value chain, enabling the provision of reliable forecast-ready guidance tailored to the needs of users.   For over a decade, the Bureau of Meteorology has relied on its own Gridded Operational Consensus Forecast (GOCF) system to provide calibrated, consensus-based guidance from a mix of ensemble and deterministic input models. While GOCF has been instrumental in enhancing the skill of the Bureau's forecasts and supporting increased automation of routine forecast production, this system has reached the limit of its utility in both technical execution and scientific capability. Hence the need to pivot to a new post-processing solution, particularly one that aligns with the shift towards ensemble models and probability-based forecasts.  In 2019, the Bureau of Meteorology entered a partnership with the UK Met Office to adopt its Integrated Model Post-PROcessing and VERification (IMPROVER) system as the replacement post-processing platform. Since this time, staff at the Bureau have worked towards building a robust, extensible post-processing platform using the IMPROVER framework to replace our legacy system. In this talk we present an overview of the Bureau's implementation of IMPROVER and then reflect on the journey thus far.  We explore our usage and adaptations to enable effective post-processing over the vast Australian domain where a greater reliance on global models is required. We highlight our incorporation of machine learning based calibration for rainfall, use of statistical downscaling, and multi-model blending to produce accurate forecasts across Australia. We present verification results benchmarking IMPROVER against our legacy system and include some discussion on future directions.