Abstract ID: 161
Sub-seasonal to seasonal forecast in Senegal: Machine Learning approach
Lead Author: Dioumacor Faye
Ecole Supérieure Polytechnique de l’Université Cheikh Anta Diop de Dakar, Senegal
Keywords: seasonal to seasonal forecasting , , Machine Learning,, intra-seasonal atmospheric signals
Abstract: In the context of global warming, relatively frequent extreme floods and droughts can not only cause heavy economic damage, but also threaten people’s lives especially in developing countries like West Africa. Sub-seasonal to seasonal forecasting (S2S) with a time scale of 2 weeks to 2 months is essential as it is associated with crop selection, disaster reduction and human safety. In addition, S2S forecasting will bridge the gap between weather and climate forecasting ( Vitart et al., 2012).
Although the statistical methods and dynamic models that are commonly used in sub-seasonal forecasting, have shown high forecasting ability ( Li and Robertson, 2015 ;Zhu and Li, 2017 ), this sub-seasonal forecasting that depends on both local weather and global atmospheric circulations ( Robertson et al., 2015 ; Vitart et al., 2017 ) are called “”desert of predictability”” ( Vitart et al., 2012 ) and still remains a major challenge.
Senegal, like all West African countries, is frequently subject to rainstorms and flooding during the boreal summer monsoon. These climatic events have resulted in impacts on human socio-economic activities, including loss of life, water contamination, agricultural damage and disruption of transportation systems. Thus, reliable and accurate precipitation forecasts are essential as they can provide valuable information to mitigate these natural disaster risks. However, the origin of intraseasonal precipitation variability is of great complexity due to the mixed impact of tropical convection, Atlantic forcing, and mid to high latitude systems.
The study aims to develop a Machine Learning model to predict rainfall over West Africa, in particular over Senegal, using previous intra-seasonal atmospheric signals (from OLR, geopotential at 200, 500, 850hpa, wind 200 and 850 hpa). The model will use coupled covariance patterns between the previous atmospheric signals and precipitation to predict weekly mean precipitation and weekly mean precipitation anomalies over Senegal. The results of this model will be compared to the outputs of the ECMWF model (archived in the S2S database) to evaluate its performance. The goal of this study is to provide reliable and accurate rainfall forecasts to reduce the risk of natural disasters in the region.
Co-authors:
