Open Access

Manuel Rauch, Jan Bliefernicht, Marlon Maranan, Andreas H. Fink, Harald Kunstmann

• The West African Monsoon, known for its significant rainfall variability, led to the Sahel drought from 1968 to the 1990s, followed by a recovery in rainfall since the 1990s. In response to such variability, this study introduces a statistical approach for reconstructing interannual rainfall variability across seven rainfall regimes, each representing unique climatic zones in West Africa. Initially, a robust catalogue of daily atmospheric circulation pattern classifications over West Africa is established, based on pre-selected variables from the ERA5 reanalysis and using k-means clustering. Subsequently, the annual occurrence frequencies of these circulation pattern classifications, along with the annual rainfall conditions in the rainfall regimes, serve as inputs in a multi-class logistic regression model. This model is designed to identify dry, normal, and wet years, relative to the climatology. The rainfall regimes are determined using k-means clustering and a quality-controlledThe West African Monsoon, known for its significant rainfall variability, led to the Sahel drought from 1968 to the 1990s, followed by a recovery in rainfall since the 1990s. In response to such variability, this study introduces a statistical approach for reconstructing interannual rainfall variability across seven rainfall regimes, each representing unique climatic zones in West Africa. Initially, a robust catalogue of daily atmospheric circulation pattern classifications over West Africa is established, based on pre-selected variables from the ERA5 reanalysis and using k-means clustering. Subsequently, the annual occurrence frequencies of these circulation pattern classifications, along with the annual rainfall conditions in the rainfall regimes, serve as inputs in a multi-class logistic regression model. This model is designed to identify dry, normal, and wet years, relative to the climatology. The rainfall regimes are determined using k-means clustering and a quality-controlled dataset from 971 rainfall stations, with daily observations ranging from 1959 to 2010. These regimes vary from the Sahelian belt, characterised by a short rainy season, to tropical regions exhibiting a bimodal rainfall regime. After comprehensive predictor screening of specific West African Monsoon patterns, such as the Tropical Easterly Jet and the African Easterly Jet, six variables at four different pressure levels under a running split-sampling cross-validation, the best models achieve an average proportion correct of 0.57 and a positive Peirce skill score for all regions over West Africa. This shows the performance in reconstructing dry, normal, and wet years for the different rainfall regimes in West Africa. Therefore, this study provides a statistical tool for the reconstruction of annual rainfall anomalies in this challenging region.…