dc.description.abstract | The present study is a time series an~lysis of
the East African rainfall records. The first phase of
the study investigated the nature of fluctuations in
the rainfall records during the period 1922-1975. The
components examined included the trend, cyclical, seasonal,
and random fluctuations. The rainfall data used included
the pentad, monthly, and the annual records.
In examining the trend of rainfall in East Africa
both graphical and statistical methods were applied. The
statistical test applied was the Spearman rank correlation
test. The analysis of variance approach was further included
to detect any significant difference betw~en the
means of some standardperiods. The rainfall series were
smoothed using 5-term binomial coefficients in the graphical
approach.
The spectral analysis technique was applied to
describe the tendency for any of the rainfall records to
show oscillations at certain frequencies, while the sample
correlation tests were used in describing the random
variations and other types of persistence in the rainfall
records. In examining the seasonal fluctuations, the
Spearman rank correlation test was applied to the pentad
time series,The present study is a time series an~lysis of
the East African rainfall records. The first phase of
the study investigated the nature of fluctuations in
the rainfall records during the period 1922-1975. The
components examined included the trend, cyclical, seasonal,
and random fluctuations. The rainfall data used included
the pentad, monthly, and the annual records.
In examining the trend of rainfall in East Africa
both graphical and statistical methods were applied. The
statistical test applied was the Spearman rank correlation
test. The analysis of variance approach was further included
to detect any significant difference betw~en the
means of some standardperiods. The rainfall series were
smoothed using 5-term binomial coefficients in the graphical
approach.
The spectral analysis technique was applied to
describe the tendency for any of the rainfall records to
show oscillations at certain frequencies, while the sample
correlation tests were used in describing the random
variations and other types of persistence in the rainfall
records. In examining the seasonal fluctuations, the
Spearman rank correlation test was applied to the pentad
time series, and to the time series generated from the
yearly phase angles of the major harmonics observed
from the individual stations.
Many statistical tests assume that the data being
analysed is a sample from a single population (homogeneous).
Results of some statistical tests (e.g. significant trend)
may be a manifest of heterogeneity in the data which is
being used. Many statistical tests also assume that the
data is a sample from Gaussian (normal) distribution. Nonparametric
tests are used when the distribution is n6t·cr.lose
to normal. Before examining the nature of the fluctuations
in the rainfall series, the homogeneity of the rainfall
records were examined using one sample runs test. The
tests for the nature of the frequency distribution of
the pentad, monthly, seasonal, annual and regional rainfall
records included the Kolmogorov-Smirnov test and the
Fisher's gl and gz tests.
One of the main objects of this study was to examine
the possibility of using some stochastic models in describing
the mean deviations of somestandardized regional
'annual rainfall. Stochastic models were also fitted to
some dryness/wetness indices defined from empirical'
orthogonal functions. The stochastic models fitted included
Stepwise Trigonometric Regression models, Autoregressive
Integrated Moving Average (ARlMA) models,
and Combination models. In the combination models ARlM models were fitted to the residuals from the Trigonometric
Regression models in an attempt to find a ~ombination of
stochastic models which could explain substantial amount
of the initial variance of the regional annual series.
The method of Principal Component analysis was used in
delineating East Africa into homogeneous regions.
The results indicated homogeneity in all records
except four. The records of the four stations were
adjusted using Double Mass analysis. The frequency distribution
of twelve annual series with significantly
high values of Skewness and Kurtosis were not close to
normal. Most of these stations were located in dry
regions. The records of the wettest months and the wet
seasons were close to those of the annual records. The
pentad series, the records of dry months"and the dry
seasons were far from normal.
The results from trend analyses indicated an increasing
rainfall tendency in the dry belt. The fluctuations
in the pentad series and in the timing of the rainfall
maximum and minimum during the study were observed not
to be statistically different from those expected by
chance at 5% level. A family of four spectral peaks were
observed in most of the rainfall records. These were
the QBO, the 3.0-37, 4.8-6.0 and 10.0-12.5 year cycles.
The QBO was dominant in the coastal regions, while the
3.0-3.7 year cycle in the inland stations. The 4.8-6.0 year cycle appeared in most of the rainfall series, while
the few stations with significant 10.0-12.5 year cylces
were concentrated in Uganda.
The results from principal component analysis indicated
that the e~pirical orthogonal functions were capable
of representing East Africa into some simple and physically
recognizable regional patterns. Some composite indices
of dryness/wetness defined from this empiric~l orthogonal
functio~were also noted to satisfactorily describe the
fluctuations of the regional annual rainfall.
It was noted from the study that although some of
the fitted stochastic models could give some reasonable
patterns of the standardized mean annual regional rainfall,
these stochastic models generally under-estimated
the regional series | en |