Assessing the skill of the seasonal rainfall prediction over the greater horn of Africa using global models as multimodel ensemble
Abstract
The usefulness and limitations in seasonal forecasts are due to uncertainty inherent in the climate
system. The reduction of errors in the forecasts systems increases the reliability of the forecasts.
The improved seasonal rainfall prediction to reduce the climatic extreme events using dynamical
models with fewer uncertainties is important to the socio-economic development of the Greater
Horn of Africa (GHA).
In this study the overall objective of the study was to assess the skill and accuracy of the seasonal
rainfall forecasting using global models as multi-model ensemble during October to December
(OND) season over the study region. The data used in the study included the gridded rainfall data
from Climate Research Unit, University of East Anglia (CRU) and hindcast data from eight Global
Producing Center models (GPCs) for the period 1983 to 2001. The methodology employed included
spatial analysis, correlation analysis, Model output Statistics (MoS), regression analysis, time series
analysis, simple composite analyses, weighted average and categorical statistical skill score.
The spatial patterns of the individual models output from the models of Washington,
Montreal, Melbourne and model from Centre for weather forecasting and climate studies
(CPTEC) were closest to the observed rainfall patterns. The largest departure from observations in
this season was observed in the northern and southern sectors of the GHA. The spatial distribution of
rainfall anomalies of the observed and models output during extreme events showed that the
ensemble models were able to simulate El Niño (1997) and La Niña (2000) years. The models were
not able to capture the magnitude of the extreme events.
The skill of the ensemble model was higher than those of the individual member models in terms
of its ability to capture the rainfall peaks during the El Niño Southern Oscillations phenomena
(ENSO). The analysis for the correlation coefficients showed higher values for the ensemble model
output than for the individual models over the Equatorial region (5°N to 5°S). Comparatively, the
stations in the northern and southern sectors of the GHA had low skill. This is an indication that
the models have better skill and accuracy over the Equatorial region.
In general, the skill of the models was relatively higher during the onset of the ENSO event and
became low towards the decaying phase of ENSO period. Regarding the prediction of extreme
low and high values, the models generally indicated the direction of the anomalies but such
extremes were under-estimated or over- estimated in some cases.
Citation
Master of Science in Meteorology, Univesity of Nairobi, 2013.Publisher
University of Nairobi Department of Meteorology