Abstract

The East African region is highly susceptible to severe floods and persistent droughts, which greatly impact the livelihood of millions of people. Early warnings, at least a few seasons in advance, would help implement mitigation measures. However, most prediction systems using dynamical models perform poorly at long lead times. In this study, we propose a statistical deep learning approach based on a convolutional neural network (CNN) to predict extreme floods and droughts during the short rains season (October–December). The proposed CNN model captures the phase of extreme floods and droughts two to three seasons ahead, except for a few cases. By diagnosing the model’s skills using heatmaps, we find that predicted extreme floods and droughts are linked with the sea surface temperature anomalies of the Indian Ocean Dipole at shorter leads and with western and southern Indian Ocean, equatorial Pacific, and southern Atlantic Ocean at longer leads. Although there were a few poorly predicted exceptions, the superior skill of our CNN-based predictions at longer leads provides a significant advantage in developing mitigation measures.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call