How coherent is rainfall in northern tropical Africa in time and space – and why?

Rainfall forecasts over northern tropical Africa are potentially beneficial for a wide range of applications from agriculture to health, but current numerical weather prediction models have very limited skill over this region, even in a probabilistic sense. A recent study by Vogel et al. (2021, doi: 10.1029/2020GL091022) has demonstrated for the summer monsoon season July-September that even a relatively simple statistical forecast model based on past time-space rank correlations in rainfall estimates from Tropical Rainfall Measuring Mission can outperform numerical models.
Here we extend the correlation part of the Vogel et al. study in various ways: (a) we use time lags up to the previous three days, (b) we use a larger geographic area spanning several thousand kilometers over northern tropical Africa, the Atlantic Ocean and the Indian Ocean (c) we consider five different seasons per year, (d) we use the more recent satellite-based globally gridded Integrated Multi-satellitE Retrievals for Global Precipitation Measurement final-version product from 2001–2019, and (e) we link the detected correlation patterns to known meteorological features such as African Easterly waves.
Our results show that significant correlations can be found for all lags from one to three days in all seasons along the fringes of the climatological rainbelt over tropical Africa. We attribute this to the large-scale drivers that trigger and organize rainfall, which in turn causes coherent spatio-temporal anomalies. On the contrary, low correlations are observed within the rainbelt at all time lags, indicating the lack of a single dominant forcing, high stochasticity, or both. To quantify the coherence of the forcings identified in each season, we introduce a new metric called coherence-factor. It is computed at every grid-point and summarizes the extent to which the lagged correlations reflect a propagation with a constant phase speed and direction. High values of the coherence-factor combined with healthy levels of correlations over the three days considered indicate physically interpretable, stable relationships that potentially translate into high potential predictability. For example, high coherence over the Sahel region in the July-September season shows the dominance of AEWs in triggering and organising rainfall. In contrast, the December-February season shows a very different picture with high coherence only over the equatorial oceanic region. The May–June season closely resembles July–September, indicating early stages of AEWs activity. March–April and October–November seasons show features characteristic of those they are transitioning between.
In the future, the coherent features identified in this study will be used as predictors for testing several statistical and hybrid (i.e., additionally including predictors from numerical weather prediction) models in every season to forecast rainfall over northern tropical Africa.