Predictability of vegetation cycles over the semi-arid region of Gourma (Mali) from forecasts of AVHRR-NDVI signals

► NDVI signal is taken as a proxy of the dynamics and used as a forecasting model. ► Predictability is estimated from the error growth of model forecasts. ► Horizon of predictability varies in time and with spatial scale. ► The model is especially efficient at intermediate scale (125 × 125 km2). ► At larger scale complexity is higher but overall behaviour becomes more predictable.