The core monsoon season in Southern Africa (5°S to 30°S) stretches from December to February, the Southern Hemisphere summer. Precipitation becomes increasingly seasonal as we move towards higher latitudes, much like any other monsoon region. Rainfed agriculture is the mainstay of economies across Southern Africa, and therefore, accurate monsoon season prediction is critical for sustainable agricultural practices.
Figure 1. (Left) Color contours: December-January-February (DJF) surface temperature (ocean) and precipitation (land), line contours: DJF precipitation standard deviation (red, blue, white: 1, 2, 3 mm/day), vectors: winds at 850 hPa. The white box indicates the core monsoon area. (Right) The monthly zonal average (5°E–52°E, land points) of precipitation. The figure has been taken from Horan et al. 2024.
The scientific community recognizes El Niño-Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) as primary mechanisms that influence the precipitation variability in the Southern African monsoon season; however, their interdependence complicates understanding their independent roles.
We have been advocating for investigating the leading mode of precipitation variability in the Indian Ocean and its role in global teleconnections. Our research has revealed that ENSO's teleconnection in some regions, such as Central Southwest Asia, is primarily mediated through this precipitation mode in the Indian Ocean. Therefore, our curiosity has led us to consider its contribution to monsoon precipitation variability over Southern Africa, especially due to its proximity.
So, we constructed an empirical model for predicting Southern Africa's monsoon precipitation using three forcings: ENSO, IOD, and the leading precipitation mode in the Indian Ocean. Given its spatial characteristics, we call it the Indian Ocean Precipitation Dipole or IOPD.
Interestingly but not surprisingly, our empirical model can predict austral summer monsoon precipitation in Southern Africa up to 5 months in advance.
It is possible to predict monsoon season anomalies in Southern Africa at longer lead times due to the characteristics of the natural forcings that affect precipitation variability in the region. All three forcings (ENSO, IOD, IOPD) are variability modes related to tropical oceans that reach their strength at or before the onset of the rainy season in Southern Africa. While IOD and IOPD peak in October, ENSO peaks in December. In most years, these phenomena persist in the same phase during the fall and winter months, driving the predictability of southern African monsoon rains at several months' lead. However, it is crucial to note that none of these natural modes operate independently. IOPD strongly couples with IOD in the Southern Hemisphere spring (October, November) and ENSO during the Southern Hemisphere summer (December-February).
(Top) Climatological monthly strength of each index (ENSO, IOD, IOPD) with respect to its maximum. (Bottom) The persistence of each index is shown as the correlation of the December value with the preceding months.
Returning to our empirical model, we compared it with two prominent seasonal forecasting systems: the Geophysical Fluid Dynamics Laboratory (GFDL) Seamless System for Prediction and Earth System Research (SPEAR) and the European Center for Medium-Range Weather Forecasts (ECMWF) fifth-generation seasonal forecasting system (SEAS5) over 30 years. And guess what? Our model outperforms both seasonal forecasting systems with up to five months lead times.
SEAS5 and SPEAR accurately represent the interplay of the three forcings but show varying skills in representing their teleconnection over Southern Africa. Unsurprisingly, both exhibit the overly strong influence of ENSO over the region and the weaker influence of IOD and IOPD. These biases consequently impact prediction skills in these models. SPEAR suffers from these issues more than SEAS5 among the two.
At longer lead times, all three forcings (ENSO, IOD, and IOPD) contribute to predicting summer monsoons over Southern Africa. However, as we approach the monsoon season, strong coupling between ENSO and IOPD reduces ENSO's independent impact as IOPD acts as a mediator. As a result, IOD and IOPD become the primary factors influencing precipitation variability over the region as the season progresses. This mediation by IOPD is problematic for models that rely heavily on ENSO teleconnections for their prediction accuracy but fail to capture IOPD's role in the region.
Understanding the independent teleconnections of the three forcings is also essential, as all three tend to have a dipolar influence, with positive phases of each bringing more precipitation in the northern belt and less precipitation in southern latitudes. However, after we account for their interdependences, IOD's negative influence mostly diminishes, while the dipolar influence persists in the case of ENSO and IOPD. SEAS5 and SPEAR lack representation of these latitudinal variations in their teleconnections characteristics.
To explain how our empirical model compares with the two forecasting systems, let us look at their prediction skills with the shortest lead time (December preconditions; at lead zero). With December preconditioning, our empirical model explains up to 80% of monsoon (December-January-February) precipitation variance over the northern belt and >55% over southern latitudes. In comparison, SEAS5 (SPEAR) explains ~67% (~22%) of precipitation variance over the northern belt and ~31% (~25%) over southern latitudes.
The mean area-averaged precipitation over northern (5°S to 12°S; left) and southern (17°S to 25°S; right) parts of Southern Africa in ERA5 (black), empirical model (green), SPEAR (orange) ensemble mean, and SEAS5 (violet) ensemble mean. The empirical model is based on December IOD, IOPD, and ENSO indexes, while the dynamical models are based on December initializations. Light circles indicate ensemble members in SPEAR and SEAS5. The figure has been taken from Horan et al. 2024.
We have tested our model in predicting the 2023/24 monsoon season in Southern Africa, comparing it with SPEAR and SEAS5. This year's peak monsoon season has passed, so the numbers are now in.
All three forcings experienced a positive phase during the Southern Hemisphere spring to summer months of 2023/24. The El Niño phase of ENSO was one of its strongest in recent decades. Although the IOD usually weakens as winter approaches, it remained unusually strong this year. The IOPD was also relatively strong. Therefore, above-normal precipitation in the northern areas, while below-normal precipitation in the southern latitudes, was in anticipation, a pattern that eventually prevailed at the end of the season.
Although both dynamical models and our empirical model could predict the dipole precipitation anomalies over Southern Africa this monsoon season, our empirical model performed better than the two dynamical forecasting systems in capturing the magnitudes of anomalies at every lead time. Likely, SEAS5 and SPEAR's limited ability to represent IOD and IOPD teleconnections over southern Africa persisted, making them less effective in predicting monsoon precipitation than the empirical model. As expected, among the two, SEAS5 outperformed SPEAR, given its relative superiority.
This table shows the 2023/24 seasonal (DJF) average precipitation anomaly in ERA5 (actual) and predictions (empirical model, SEAS5, and SPEAR) at different lead times. It also shows the strength of ENSO, IOD, and IOPD forcings.
Spatial pattern of the 2023/24 seasonal (DJF) average precipitation anomaly in ERA5 (actual) and predictions (empirical model, SEAS5, and SPEAR) at different lead times.
The promising results shown by our empirical model indicate that the potential predictability in many regions across the globe may be much greater than what is typically perceived based on dynamic forecasting systems. However, the key to unlocking the full potential of predictability lies in a path currently not followed in the modeling world. Overreliance on ENSO limits the predictability skills in seasonal forecasting systems. It is high time that models improve the representation of other oceanic and atmospheric modes of variability that impact sub-seasonal to seasonal climate variability.