Machine learning assisted identification of sea surface temperature patterns related to extreme rainfall events over Cameroon: Machine learning assisted identification of Sea Surface temperature patterns related to Extreme Rainfall events over Cameroon

Cameroon has seen a rise in extreme rainfall events in recent decades, leading to significant socio-economic and environmental impacts. To shed light on the causes of the events, we analyzed rainfall data from 608 locations and sea-surface-temperature (SST) data from ERA5 (1993–2022). We defined extreme events as days with a standardized rainfall anomaly ≥ 2 σ over at least 5% of Cameroon’s area.We further split the seasonality of extreme events into two time frames—March to May and June to November—based on the frequency of occurrences among the top 500 events. The overall goal was to identify large-scale forcing factors associated with the occurrence of such extreme events. For this purpose, we applied the K-Means clustering algorithm to the global SSTs of extreme event days. The results show that the K-Means clustering can effectively identify well-known large-scale forcing factors including the El Niño–Southern Oscillation, Atlantic Niño, South Atlantic Ocean Dipole and Indian Ocean Dipole. Notably, some years are observed to exhibit above-average frequency of extreme rainfall even when none of the well-known forcing factors are active, highlighting alternative configurations in SSTs that may lead to extreme events. The machine learning further revealed such SST patterns underpinning extreme events, including for instance, abnormal cooling of the Oman Sea while the area from the Maldives to the Andaman Sea is abnormally warm, and a north (cold) - south (warm) dipole-like pattern of oceans. These newly identified patterns may serve as valuable proxies for forecasters, helping them to detect and anticipate extreme rainfall events in Cameroon.