Towards an operational seasonal forecasting system for the Arabian Peninsula

Seasonal forecasting is a key tool for managing climate risks in arid regions, yet its performance over the Arabian Peninsula (AP) remains insufficiently characterized. In this study, we developed and evaluated a multi-model seasonal forecasting system for the AP, comparing simulations from the Copernicus Climate Change Service (C3S), the North American Multimodel Ensemble (NMME), and a combined ensemble (CMME). Forecasts were post-processed using two alternative methods, the extended logistic regression (ELR) and the counting approach, and verified using probabilistic skill metrics. Results show that no single system or method consistently outperforms the others. The counting method generally yields higher skill than the ELR across most regions, but also produces some areas with negative skill that are not observed in the ELR results. The C3S and CMME ensembles show comparable performance, while the NMME exhibits localized skill in predicting certain extreme events. A sensitivity study indicates that the spatial resolution of seasonal forecast models plays a significant role in improving retrospective skill. Analysis of heavy rainfall episodes suggests that seasonal forecasts can potentially provide early indications of extreme precipitation in the region, though predictability varies spatially and seasonally. These findings underscore the need to communicate forecast skill alongside probabilistic outcomes to ensure informed use by decision-makers. The proposed forecasting framework will form the basis of an operational seasonal prediction system for the AP, supporting climate-resilient planning in sectors such as water management, agriculture, and infrastructure. Future developments will focus on downscaling approaches to enhance local-scale prediction of extreme rainfall events.