Abstract:
Significant changes are being experienced in the climate system due to the unprecedented rate of
global warming. This has resulted in the increased frequency of weather extreme events such as
heatwave occurrence in the Northern Nigeria. In order to mitigate the effects of heatwaves, early
warning systems are needed to be implemented. Insufficient knowledge about the performance of
the models is partly a factor that hinders the development of such systems. This study thus,
addresses the gap by assessing the predictability skills of sub-seasonal to seasonal numerical
weather model over different time lead and as well improves the predictability skills through the
incorporation of deep learning to post-process the model output at a 30-day lead period.
The Excess Heat Index (EHI) was used to detect heatwave occurrence over the study area, using
both observational and forecast data from selected S2S models at 5 -, 7 -, 15 -, and 30 – days lead
time. Metrics employed to evaluate the skills of the models are; the Anomaly correlation
coefficient (ACC), and Symmetric External Dependency Index (SEDI) with each evaluating
different strength of the models.
The result of the analysis shows that the three models considered in this study overestimated the
heat wave frequency in the region. This results in reduced reliability of the models in the region.
Further analyses shows that the use of deep learning to bias correct the model output increases the
forecast reliability in the region significantly.
