On the reliability of global seasonal forecasts: sensitivity to ensemble size, hindcast length and region definition

One of the key quality aspects in a probabilistic prediction is its reliability. However, this property is difficult to estimate in the case of seasonal forecasts due to the limited size of most of the hindcasts that are available nowadays. To shed light on this issue, this work presents a detailed analysis of how the ensemble size, the hindcast length and the number of points pooled together within a particular region affect the resulting reliability estimates. To do so, we build on 42 land reference regions recently defined for the IPCC-AR6 and assess the reliability of global seasonal forecasts of temperature and precipitation from the European Center for Medium Weather Forecasts SEAS5 prediction system, which is compared against its predecessor, System4. Our results indicate that whereas longer hindcasts and larger ensembles lead to increased reliability estimates, the number of points that are pooled together within a homogeneous climate region is much less relevant.This research has been partially supported by the AfriCultuReS (“Enhancing Food Security in African Agricultural Systems with the Support of Remote Sensing”) and FOCUS-Africa projects, which received funding from the European Union's Horizon 2020 Research and Innovation Framework Programme under grant agreements No. 77465 and 869575, respectively.Peer ReviewedPostprint (published version