Muita R. Richard*, S. Aura*, D. Mburu*, J. N. Kagenyi*, J. Muhindi**, A. Shaka***, J. G. Mungai****
*Institute for Meteorological Training and Research/Kenya Meteorological Department,
**National Meteorological Centre/Kenya Meteorological Department,
***Public Weather Service/Kenya Meteorological Department,
****Numerical Weather Prediction/Kenya Meteorological Department Corresponding Author
Richard. R. Muita
Institute for Meteorological Training & Research/
WMO Regional Training Centre, Nairobi
Kenya Meteorological Department
P.O. Box 30259 00100 Nairobi, Kenya
(Received 16 May 2016, Received in revised form 19 August 2016, Accepted 26 September 2016)
AbstractIn this study the skill of daily rainfall forecasts over various parts of Kenya during the October- December 2015 and October – December 2014 rainfall seasons is assessed.
Daily forecasts and rainfall observations from 6 climatic zones were analysed using the combined probability distribution of occurrence and non-occurrence of rainfall. The skill of the forecasts was assessed in two parts: 1) as deterministic binary events; and 2) as probabilistic spatial proportion of rainfall forecasted in a region. The contingency table and other standard verification scores were used to determine the forecasts’ skills.
The Brier score and other metrics indicated modest to skilful forecasts in both periods while the area under the ROC curve indicated skills of 46.3% – 73.6%. The highest forecast skills occurred in the highlands east, central, and west of the Rift Valley regions of Kenya in 2015 season while the lowest skill was indicated in the north-western region of Kenya. In 2014 period most of the semi arid regions exhibited higher forecast skills (>50%) relative to the ENSO season of 2015. For most locations in a given region, there was no evidence that the average forecasts skill was statistically different. This was attributed to the fact that the period under analysis coincided with El-Nino episode which brought about a large scale rainfall generating mechanism that influenced much of the country in 2015. The high skill for drier regions in 2014 may have been due to influence of the no meso-scale synoptic circulation effect during none ENSO years (2014)
Overall, the forecast skill is better where rainfall is higher (and with more observation stations) and well distributed in time and space during an El-Nino period but also higher in drier regions during non El-Nino period. Potentially, increased observation network may improve forecast skill in Kenya.
Download Full Text (PDF)