E-WP5 - PROMISA: Process evaluation of the MiKlip decadal forecast system over tropical Africa

Rain-fed agriculture is the main industry in tropical Africa, and plays a crucial role in sustaining livelihoods and economic development. However, the absence of seasonal rainfall and interannual to decadal precipitation fluctuations provoke droughts or floodings which have an enormous impact on socio-economic activities. Therefore, understanding the mechanisms that produce this variability and developing both dynamical and statistical approaches for seasonal to decadal forecasts and climate projections is of great importance for policymakers and humanitarian aid agencies.

The MiKlip-PROMISA (Process Evaluation of the MiKlip Decadal Forecast System in Tropical Africa) study areas are the West Africa Monsoon (WAM) and the Greater Horn of Africa (GHA) regions. Both regions show large interannual to decadal rainfall variations, the Sahel drought in the 1970s and 1980s (Rodríguez-Fonseca et al. 2015) and the near absence of two consecutive rain-seasons in GHA regions presented a major climatic shock to the region, as witnessed by the resulting humanitarian crisis in 2011 (FEWS NET, 2011). Recent observational and modelling studies also indicated a drying trend of the March-April-May rainfall, the “long rains” in parts of the GHA (Yang et al. 2014). These interannual to decadal variations of rainfall over tropical Africa are often linked to sea surface temperatures anomalies in adjacent and remote ocean basins

Project aims

It is of great importance to evaluate the skill of the MiKlip decadal forecast system over tropical Africa. The process evaluation in MiKlip has shown deficiencies in representing atmospheric processes and also revealed that some teleconnections were underrepresented due to biases in the mean state of the WAM. In this regard, the PROMISA work package aims at enhancing and extending evaluation metrics for atmospheric processes and teleconnections governing the interannual to decadal rainfall variability over the WAM and the GHA regions. In addition, post-processed decadal ensemble hindcasts/forecasts, both from MiKlip II and other international model groups, will be evaluated using various process and teleconnection tools.  

Project structure

PROMISA belongs to module E and will be carried out at the Institute of Meteorology and Climate Research of the Karlsruhe Institute of Technology.

Tasks of the project

The PROMISA tasks include to:

  1. investigate and identify large-scale influences on the interannual to decadal rainfall variability over the GHA,
  2. develop novel metrics for atmospheric processes and teleconnections governing the inter-annual to decadal rainfall variability over the WAM, and particularly, the GHA region,
  3. post-process MiKlip-II and another center’s hindcasts using the CES (Central Evaluation System),
  4. augment and update the high-quality, daily rain gauge observational database for the WAM and GHA regions for validation purposes in the CES,
  5. derive probabilistic forecasts for droughts and floods (e.g. impact-relevant events) using relevant indices and teleconnections, and
  6. compare and evaluate inter-model and inter-experiment performance changes in teleconnections and processes.


The project will deliver African rain gauge data to the central MiKlip server and the CES. Process diagnostics and teleconnections tools will be added to CES. Finally, MiKlip hindcasts will be quantitatively assessed.

Progress so far

GHA teleconnections

Observational datasets from 1901 to 2013 were used to investigate interannual to decadal teleconnections during three major rainy seasons over the GHA. The strength and stationarity of linear statistical relationships between Sea Surface Temperature (SST)-based remote indices of climate variability and GHA rainfall indices were examined. The recent increase in drought frequency during the March-May (MAM) “long rains” is strongly associated with decadal variability in the Pacific Ocean. Our results also reveal that El Niño-Southern Oscillation (ENSO) is strongly related with June-September (JJAS) “Kiremt rains” over the Ethiopian Highland (dark blue curve in Figure 1). This correlation is significant, relatively stable over time, with percent variance explained (PVE) in recent decades larger than 50%, and is mainly based on high-frequency variability (<= 8 years). ENSO variations can be best represented by the Niño 3.4 index that showed some skill in MiKlip I and pre-operational MiKlip II hindcasts (Figure 2). Another interannual teleconnection to Kiremt rains, viz. related to the Atlantic Niño (represented by the ATL3 index) changed from significantly negative to insignificantly positive after the 1960s, thus suggesting a poor stability over time (magenta curve in Figure 1).

Fig. 1: Time evolution of the linear correlation coefficients between six indices of remote climate anomalies and the Standardised Precipitation Index of Ethiopian Highland Kiremt Rains (ETH-KR-SPI) using a 31-year running window. The correlation value for 1916, for instance, relates to the period 1901-1931. The horizontal dashed lines represent the 95% significance level according to the F-test using the standardised degree of freedom (SDOF), i.e., not taking serial correlation into account: Pacific Decadal Oscilations (PDO, green), El Niño 3.4 (Nino34, blue), Atlantic Meridional Oscillation (AMO, red), ATL3 (magenta), Indian Ocean Dipole Mode Index (DMI, orange), and Interdecadal Pacific Oscilation (IPO, light blue).

Validation of the MiKlip prediction system

The skills of all initialised MiKlip decadal experiments predicting SST-based remote indices of climate variability were evaluated. The result showed a significant predictability in the Nino 3.4 region in the first, and partly, in the second year, and the AMO in all 10 prediction years (Figure 2a). The initial validation using the current pre-operational MiKlip II model runs indicates no improvements compared to MiKlip I experiments. Pre-operational forecasts were also evaluated to what extent they reflect the recent decline in long rains and interannual variability over GHA. No exploitable skill was detected. Further evaluations of the pre-operational experiments regarding observed processes and interannual to decadal teleconnections to rainfall variations of the WAM and in the GHA region are currently underway with the goal to identify any predictive skill in the MiKlip II forecast systems.

Fig.2 Correlation between observed SST (HadISST) and simulated SST from all MiKlip model versions for (a) Niño3.4, (b) AMO. Solid lines show the ensemble means, dashed lines show the ensemble spread.
Evaluations with preop-HR

The preop and baseline1 MiKlip II experiments have been evaluated regarding their capability in representing observed inter-annual to decadal teleconnections to rainfall variations over the Greater Horn of Africa region. For this analysis, lead years+1 from each initialisation were concatenated to create monthly SST and precipitation time series for the period 1961–2013. In general, preop-HR produces excellent predictions of the spatial pattern of the observed Pacific Decadal Oscillation (PDO) at lead year+1. Although the observed percent variance explained in EOF1 is moderately represented both in preop and baseline1 experiments, the predictability of PDO-related teleconnection patterns over the tropical region is however weak in preop-LR and baseline1-LR. ENSO-derived inter-annual rainfall teleconnections illustrate that the MiKlip decadal hindcast predictive skill is typically very low. However, in contrast to preop, baseline1 shows better correlation skill for ENSO teleconnections over Australasia. In addition, the rainfall index over the Ethiopian Highlands is correlated with SST anomalies in the tropical oceans. Overall, the MiKlip decadal hindcasts have relatively little or no skill regarding rainfall predictions over western and eastern Africa. Further evaluations with upcoming preop-CMIP6 hindcasts, and the post-processed (using recalibration plug-in) preop-CMIP5 and preop-CMIP6 hindcasts are planned to check for skill improvements.


FEWS NET (2011), Past year one of the driest on record in the eastern Horn, Famine Early Warning System Network Report, June 14, 2011, U.S. Agency for International Development, Washington, D.C.

Yang, W., Seager, R., Cane, M. A., and Lyon, B.: The East African long rains in observations and models, J. Climate, 27, 7185–7202, 2014.

Rodríguez-Fonseca et al. 2015: Variability and predictability of West African droughts: A review of the role of sea surface temperature anomalies. J. Climate, 28, no. 10, 4034-4060, doi:10.1175/JCLI-D-14-00130.1.


Karlsruhe Institute of Technology (KIT)
Andreas H. Fink
Peter Knippertz
Titike Kassa Bahaga

Revisiting interannual to decadal teleconnections influencing seasonal rainfall in the Greater Horn of Africa during the 20th century

2019 - International Journal of Climatology

Titike K. Bahaga | Andreas H. Fink, Peter Knippertz

Interdecadal changes in the leading ocean forcing of Sahelian rainfall interannual variability: Atmospheric dynamics and role of multidecadal SST background.


Roberto Suárez-Moreno | Belén Rodríguez-Fonseca, Jesús A. Barroso, and Andreas H. Fink

Assessing recovery and change in West Africa’s rainfall regime from a 161-year record

2018 - Int J Climatol. 2018;1 – 17.

Sharon E. Nicholson | Andreas H. Fink, | Chris Funk

Rainfall over the African continent from the 19th through the 21st century

2017 - Global and Planetary Change, Volume 165, June 2018, Pages 114-127

Nicholson, S. E., | C. Funk, and A. H. Fink

A meteorological and 5chemical overview of the DACCIWA field campaign in West Africa in June–July 2016

2017 - Atmos. Chem. Phys., 17, 10893–10918

Knippertz, P. | Fink, A. H., Deroubaix, A., Morris, E., Tocquer, F., Evans, M. J., Flamant, C., Gaetani, M., Lavaysse, C., Mari, C., Marsham, J. H., Meynadier, R., Affo-Dogo, A., Bahaga, T., Brosse, F., Deetz, K., Guebsi, R., Latifou, I., Maranan, M., Rosenberg, P. D., and Schlueter, A.