SUNSET: Addressing key challenges for the successful provision of climate services

Climate services leverage state-of-the-art knowledge, data and tools from the climate sciences to tailor services to user needs. To do so, climate service scientists ensure scientifically robust and traceable analyses to address specific, co-produced applications in sectors such as energy, agriculture and health. However, the diversity of post-processing methodologies applied to climate datasets, together with the variety of data sources (e.g. reanalyses, in situ observations, and climate predictions across different forecast horizons) and heterogeneous user requirements, makes the development and long-term maintenance of the required software a major challenge for the timely delivery of climate products that fulfill those user needs.

The SUbseasoNal to decadal climate forecast post-processing and asSEssmenT (SUNSET) is a software suite developed by the Earth Sciences Department at the Barcelona Supercomputing Center, building on extensive expertise in state-of-the-art climate science, climate service co-production and software development for HPC environments. SUNSET integrates in-house R-based software packages, including CSTools, CSDownscale and s2dv, which implement established methodologies for climate forecast post-processing, such as bias adjustment, statistical downscaling, verification and visualisation. The suite addresses key challenges commonly faced by climate service scientists, including the management of multiple forecast systems and reference datasets, the alignment of temporal dimensions (e.g. initialisation dates and forecast lead times with respect to reference datasets), and the consistent handling of hindcasts and observations to enable robust and comparable verification through cross-validation approaches. When requested, SUNSET uses the Autosubmit workflow manager to parallelise and orchestrate multiple workflows, ensuring efficient use of computational resources and the timely generation of climate products.

SUNSET currently delivers near real-time operational climate products, including the probability of the most likely tercile, the probability above or below specific percentiles, and absolute thresholds for essential climate variables. These products can also be tailored to sector-specific indicators, such as the growing degree days required by the agriculture sector. SUNSET verification workflows support the evaluation of the next generation of Copernicus Climate Change Service seasonal forecast systems within the CERISE project by providing comprehensive skill metrics and scorecard summaries. Together, these capabilities ensure successful research and service delivery in several projects, including ASPECT, BigPrediData, and BOREAS.

SUNSET is open-source and hosted in a public GitLab repository, following a structured development strategy with regular releases, continuous integration, and a dedicated conda environment to ensure reproducibility and long-term sustainability. Ongoing and future developments focus on extending methodological capabilities, improving usability, and optimising memory usage and workflow multi-node parallelisation for efficient execution on HPC systems.