Manuscript on machine learning approaches to predict seasonal precipitation.

Seasonal climate forecasts are commonly based on model runs from fully coupled forecasting

systems that use Earth system models to represent interactions between the atmosphere, ocean,

land, and other Earth-system components. Recently, machine learning methods are increasingly

being investigated for this task where large-scale climate variability is linked to local or regional

temperature or precipitation in a linear or non-linear fashion. This paper investigates the use of

machine learning methods to predict seasonal precipitation in the Greater Horn of Africa (GHA)

region through a flexible machine learning forecasting system. Indices of large-scale climate

variability–including the rate of change in individual indices as well as interactions between different

indices–are used as potential features to provide probabilistic predictions of precipitation fields.

Tercile forecasts from the interpretable machine learning algorithm are compared against the

ECMWF seasonal forecasting system (SEAS5) for three wet seasons, March to May (MAM), June to

September (JJAS) and October to December (OND), over the period 1993-2020. The machine learning

model yields significant positive skill compared to climatology in the OND season, a minimal positive

skill in the JJAS season and a skill equal to climatology in the MAM season. It is outperformed by the

ECMWF forecast in the OND season, while the ECMWF forecast is overconfident in the JJAS and

MAM seasons and receives negative skill compared to climatology. Several potential extensions of

the machine learning framework are discussed.