INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer.
The quality of this reproduction is dependent upon the quality of the copy subm itted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps. Each original is also photographed in one exposure and is included in reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6” x 9” black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order. UMI' Bell & Howell Information and Learning 300 North Zeeb Road, Ann Artxar, Mi 48106-1346 USA 800-521-0600
PRECIPITATION OVER SOUTHERN AFRICA AND GLOBAL-SCALE
ATMOSPHERIC CIRCULATION DURING BOREAL WINTER
A Dissertation
Presented in Partial Fulfillment of the Requirements for
the Degree Doctor of Philosophy in the Graduate
School of The Ohio State University
By
Maurice J. McHugh
*****
The Ohio State University
1999
Dissertation Committee; Approved By:
Prof Jeffrey C. Rogers, Adviser
Prof A John Amfield
Prof J.S. Hobgood / ' AdvisW
Prof E. Mosley-Thompson Department of Geography UMI Number: 99413 83
UMI Microform 9941383 Copyright 1999, by UMI Company. Ail rights reserved.
This microform edition is protected against unauthorized copying under Title 17, United States Code.
UMI 300 North Zeeb Road Ann Arbor, MI 48103 ABSTRACT
Precipitation variability over Africa south of the equator is not very well understood and its variability has been poorly documented. Small populations, poor financial resources and relatively complicated meteorological conditions all contribute to our poor understanding of precipitation variability over these regions. The NCEP/NCAR reanalysis dataset, a new and unique dataset, is used to analyze the atmospheric circulation, and climatic conditions associated with precipitation variability over southeastern and southern Africa. Strong relationships are found between precipitation receipt and El Nino, and the North Atlantic
Oscillation over both southeastern and southern Africa. Regional precipitation indices were constructed to summarize the precipitation receipt at stations significantly correlated to the
North Atlantic Oscillation Index over both these regions.
Regression analyses are used to estimate climatic conditions associated with atmospheric circulation and precipitation indices over southeastern and southern Africa.
Decreased precipitation is recorded over eastern Africa, and increased precipitation over southern Africa during the positive phase of the North Atlantic Oscillation primarily due to near surface circulation anomalies. Conversely, increased precipitation is recorded over eastern Africa, and decreased precipitation over southern Africa during El Nino events, again.
u principally due to near surface circulation anomalies. Upper level dynamics are examined in relation to precipitation variability; but it is shown that there is no consistent upper level support for near surface level dynamics associated with precipitation variability. Specific humidity, vertically integrated precipitable water and outgoing longwave radiation are also examined in relation to precipitation variability and while they are shown to be closely related to the anomalous atmospheric circulation, they are shown to have no consistent relationship act to precipitation variability.
m DEDICATION
To my parents Manus and Margaret McHugh
IV ACKNOWLEDGMENTS
I would firstly like to thank my advisor Professor JefiB'ey C. Rogers for his unlimited patience, support, advice and encouragement over the past few years. All members of my committee contributed greatly to my dissertation and to my studies at OSU. I would also like to acknowledge the assistance of Drs. Chung-Chieh Wang and Ken Yetzer for their assistance in computing and mapping. Dr. S.E. Nicholson, Department of Meteorology, Florida State University, kindly provided me with the Afiican lake level data. Many thanks also to Drs. Firooza Pavii, Mike Holtzclaw, Irene Casas and Jamie Rulli for their fiiendship, sage advice and assistance in times of need. Mr. J. Michael Straczynski and Mr. Michael Garibaldi also assisted greatly. I am indebted to my parents, my brothers Michael, Brian, Eoin, Liam, Fergal and Ronan, and my many fiiends who have supported me spiritually and emotionally during my studies, go raibh maith agat. Lastly thanks to my fiancé, Meaghan Conte for giving me hope, love and a great reason to finish. VITA
April 15 1 9 7 1 ...... Bom - Dublin, Ireland
1989-1992 ...... Bachelor of Arts Degree (Honors) in Geography and Economics, University College Dublin, Ireland.
1992-1993 ...... Master of Arts Degree (Honors) in Geography, University College Dublin, Ireland.
1995-1998 ...... Graduate Research Associate in Geography at The Ohio State University, Columbus, Ohio.
1998-1999 ...... Graduate Teaching Associate in Geography at The Ohio State University, Columbus, Ohio.
FIELD OF STUDY
Major field: Geography
PUBLICATIONS
Rogers, J.C., C.-C. Wang, andM.J. McHugh, 1998: ‘Extreme persistent climate anomalies in the northwestern Atlantic: Role of decadal scale sea surface temperature anomalies’. GeophysicalResearch Letters, 25, 3971-3974.
VI TABLE OF CONTENTS
ABSTRACT...... ii
DEDICATION...... iv
ACKNOWLEDGMENTS...... v
VITA ...... vi
LIST OF FIGURES...... x
LIST OF TABLES ...... xxi
LIST OF ABBREVIATIONS ...... xxiii
CHAPTER 1 INTRODUCTION...... 1 LI Introduction ...... 1 1.2 Research outline...... 3
CHAPTER 2 LITERATURE REVIEW ...... 6 2.1 Introduction ...... 6 2.2 Climate variability associated with the North Atlantic Oscillation ...... 6 2.3 Interannual variability of precipitation over Africa ...... 22
CHAPTERS DATA AND METHODOLOGY...... 33 3.1 Introduction ...... 33 3.2 General description of the NCEP/NCAR reanalysis dataset...... 34 3.3 Precipitation d a ta...... 40 3.4 Indices representing atmospheric teleconnections ...... 45 3.5 Data analysis ...... 48 3.5.1 Correlation...... 48 3.5.2 Regression Analysis ...... 50
vu 3.5.3 Univariate and bivariate spectrum analysis and their relation to harmonic analysis ...... 53
CHAPTER 4 CLIMATOLOGY OF SOUTH AND SOUTHEASTERN AFRICA...... 61 4.1 Introduction ...... 61 4.2 A climatology of south and southeastern Africa ...... 61 4.3 The Mean Airflow over Africa...... 80
CHAPTER 5 RESULTS...... 91 5.1 The NAO and its relation to precipitation over tropical Africa...... 91 5.3 Southern African rainfall index (SAFR) ...... 102 5.4 Quasi-periodic behavior of SEAR, the NAOI, SAFR and the Nino4 SST index...... 106 5.5 Regression analysis between the NAOI and climatic fields over Africa...... 126 5.6 Regressions of the southeast Afiican rainfall index (SEAR) ...... 148 5.7 Regressions of the southern Afiican rainfall index (SAFR) ...... 167 5.8 Correlation analysis between global precipitation and El Nino - Southern Oscillation during the boreal winter ...... 185 5.9 Regressions of the Nino4 SST Index ...... 188 5.10 Eastern Afiican hydrological response to precipitation variability .... 210
CHAPTER 6 CONCLUSIONS AND FUTURE RESEARCH...... 216
BIBLIOGRAPHY...... 222
vm LIST OF nOURES
FIGURES PAGE
2.1 Climate anomalies during GB events; Sea Ice extent and SST anomalies are depicted by hatching (comparable to the NAO Positive phase) (after Barlow et al., 1993, based on Moses et. ai, 1 9 8 7 ) ...... 8
2.2 Climate anomalies during GA events; Sea Ice extent and SST anomalies are depicted by hatching (comparable to the NAO Negative phase) (after Barlow et al., 1993, based on Moses et. ai, 1987) ...... 9
2.3 Time series o f the standardized NAOI 1875-1995 ...... 12
2.4 Vectors of the transport of vertically integrated water vapor over the North Atlantic during the positive NAO phase (a) and negative NAO phase (b). Isopleth intervals of magnitudes is at 50 kg m'^ s'\ (after Hurrell, 1995) ...... 14
2.5 Différences in P-E over the North Atlantic between positive and negative NAO phases. Isopleth interval is 0.5 mm day'^; stippling indicates statistical significance at the 95% level (after Hurrell, 1995) ...... 15
2.6 Spectrum of the NAOI from 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis ...... 19
2.7 Political Map o f Africa (after Afiican Data Dissemination Service, 1999) ...... 23
2.8 Correlation coefficients o f the January Zimbabwean rainfall and OLR. Statistically significant correlations are indicated by large + and - signs (after Jury, 19 96)...... 28
2.9 Lagged correlation coefficients between November OLR and January rainfall over South Africa. Statistically signifrcant correlations are indicated by large + and - signs (after Jury, 1996) ...... 30
IX 3.1 Histogram of the number of GHCN precipitation stations by year from 1700-1990 (after Vose et a/., 1992) ...... 40
3.2 Locations of GHCN stations with at least 10 years of precipitation data (after Vose era/., 1992) ...... 41
3.3 Locations of GHCN stations with at least 50 years of precipitation data (after Vose era/., 1992) ...... 42
3.4 The standardized Nino4 SST index, 1899-1989. SSTs are averaged over the area 5“N-5°S, 150“W-90°W ...... 46
3.5 Shape of power spectra for time series having varying lag 1 autocorrelations, representing the degree o f Markhov persistence (after Mitchell era/., 1 9 6 6 ) ...... 60
4.1 Duration of rainfall of average with average monthly amounts greater than 100 mm. (after Thompson, 1975) ...... 62
4.2 Simplified climatological mean January SLP distribution (after Thompson, 1975). Units are hPa. Arrows indicate the direction of the predominant airflow. Dotted lines represent intertropical fronts ...... 64
4.3 Winter (DJF) climatological mean 1000 hPa Streamlines calculated from NCEP/NCAR Reanalysis data, 1958-1995 ...... 66
4.4 Idealized convergence zones over SEA, predominant airflow is indicated by double arrows, idealized frontal boundaries are indicated by broken lines (after Torrance, 1972) ...... 67
4.5 January mean precipitation receipt (after Thompson, 1975) ...... 69
4.6 Seasonality of precipitation receipts over SEA; units are mm. The year is from July to June to avoid splitting the rainy season. Zimbabwe is to the south, Zambia to the northwest, and Malawi to the east (after Torrance, 1972) ...... 71
4.7 Locations of GHCN precipitation stations having maximum seasonal precipitation receipt during December, January and February (D JF)...... 73
4.8 Locations of stations having DJF precipitation greater than 50% of the total annual rainfall...... 74 4.9 Locations of stations where DJF precipitation < 50% and > 25% of the average annual precipitation...... 75
4.10 Distribution of thunder days and hail days in South Africa (after Preston-Whyte and Tyson, 1988; based on Schulze, 1972) ...... 78
4.11 SLP Climatology calculated from NCEP/NCAR Reanalysis data, 1958-1995. Units are hPa, Isopleth intervals occur every 4 hP a ...... 81
4.12 850 hPa Climatological streamlines from the NCEP/NCAR reanalysis data, 1958-1995 ...... 82
4.13 300 hPa Climatological streamlines from the NCEP/NCAR reanalysis data, 1958-1995 ...... 84
4.14 Climatological mean 300-hPa u-wind component from the NCEP/NCAR reanalysis data, 1958-1995. Units are m s'\ Isopleth intervals are 10 m s'\ negative isolines are broken ...... 85
4.15 Idealized flow in an upper level easterly wave in the Southern Hemisphere (after Preston-Whyte and Tyson, 1988) ...... 87
4.16 Vertical and horizontal components of the Coriolis Force at two different
latitudes (<{)i and (j)2 , where
5.1 Significant correlations between the NAOI and global precipitation during DJF; + and - refer to the sign of the relationship; large signs represent stations significant at the 99% confidence level, small signs refer to stations significant at the 95% confidence level. Stations not significantly correlated are not shown ...... 92
5.2 Standardized southeast Afiican Rainfall Index (SEAR), 1899-1989 ...... 97
5.3 Locations of some African GHCN precipitation stations significantly correlated to the NAOI. Triangles represent stations contributing to SEAR; circles represent stations contributing to SAFR ...... 98
5.4 Covariance between the NAOI and SEAR, 1899-1989 ...... 99
5.5 The standardized South African regional rainfall index (SAFR), 1899-1989 ...... 104
XI 5.6 Covariance between SAFR and the NAOI, 1899-1989 ...... 105
5.7 The Spectrum of SEAR, 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis...... 107
5.8 The Spectrum of SAFR, 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y -a x is ...... 109
5.9 The Spectrum of the Nino4 SST index, 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis ...... 110
5.10 Cospectrum between Nino4 and SEAR, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis ...... I ll
5.11 Squared coherence between Nino4 and SEAR, 1899-1989. Periodicity is on the x-axis in years, and squared coherence is measured on the y-axis ...... 113
5.12 Cospectrum between the NAOI and Nino4, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis ...... 114
5.13 Squared coherence spectrum between the NAOI and Nino4. Periodicity is on the x-axis in years, and squared coherence is measured on the y-axis . . . 116
5.14 Cospectrum between the NAOI and SEAR, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis ...... 117
5.15 Squared coherence spectrum between the NAOI and SEAR. Periodicity is on the x-axis in years, and squared coherence is measured on the y -axis 118
5.16 Cospectrum between SAFR and the NAOI, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis ...... 120
5.17 Squared coherence spectrum between SAFR and the NAOI, 1899-1989. Periodicity is on the x-axis in years, and squared coherence is measured on the y-axis ...... 121
5.18 Cospectrum analysis between SAFR and Nino4, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis ...... 122
XU 5.19 Squared coherence between SAFR and Nino4, 1899-1989. Periodicity is on the x-axis in years, and squared coherence is measured on the y-axis 124
5.20 SLP standardized regression coefiBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 127
5.21 1000 hPa streamlines associated with the NAOI, 1958-1995. Light and dark shading represents the statistical significance of the average u- and v-wind coeflBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 129
5.22 1000 hPa u-wind standardized regression coeflBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-te s t ...... 130
5.23 1000 hPa v-wind standardized regression coefiBcients associated with the NAOI, 1958-1995. Negative iso lines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-te s t ...... 132
5.24 850 hPa Streamlines associated with the NAOI, 1958-1995. Light and dark shading represents the statistical significance of the average u-and v-wind coefiBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 133
5.25 850 hPa u-wind standardized regression coefiBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-te s t ...... 134
5.26 850 hPa v-wind standardized regression coefiBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 136
5.27 300 hPa Streamlines associated with the NAOI, 1958-1995. Light and dark shading represents the statistical significance of the average u- and v-wind coefiBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 137
xm 5.28 300 hPa u-wind standardized regression coeflBcients associated with the NAOI, 1958-1995 Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 139
5.29 300 hPa v-wind standardized regression coefiBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 141
5.30 850 hPa Specific Humidity standardized regression coeflBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-te st ...... 142
5.31 300 hPa Specific Humidity standardized regression coeflBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-te st ...... 143
5.32 PWAT standardized regression coeflBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 145
5.33 OLR standardized regression coeflBcients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 146
5.34 SLP standardized regression coeflBcients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 149
5.35 1000 hPa Streamlines associated with SEAR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coeflBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 150
XIV 5.36 1000 hPa u-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 152
5.37 850 hPa Streamlines associated with SEAR, 1958-1989. Light and dark shading represents the statistical significance of the average u-and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively .... 153
5.38 850 hPa u-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 155
5.39 300 hPa Streamlines associated with SEAR, 1958-1989. Light and dark shading represents the statistical significance o f the average u- and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively ...... 156
5.40 300 hPa u-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 158
5.41 300 hPa v-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 160
5.42 PWAT standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the Isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 161
5.43 1000 hPa Specific Humidity standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test.... 163
5.44 850 hPa Specific Humidity standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. ... 164
XV 5.45 OLR standardized regression coefiBcients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 165
5.46 SLP standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 168
5.47 1000 hPa Streamlines associated with SAFR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coefiBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 169
5.48 1000 hPa v-wind standardized regression coefiBcients associated with SAFR, 1958-1989. Negative iso lines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 170
5.49 1000 hPa u-wind standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 172
5.50 850 hPa Streamlines associated with SAFR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively ...... 173
5.51 850 hPa v-wind standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 175
5.52 850 hPa u-wind standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 176
x v i 5.53 300 hPa Streamlines associated with SAFR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coefiBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 177
5.54 300 hPa v-wind standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 179
5.55 300 hPa u-wind standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 180
5.56 PWAT standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 181
5.57 OLR standardized regression coefiBcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 182
5.58 Significant correlations between Nino4 and Global Precipitation during DJF; + and - refer to the sign of the relationship; large signs represent stations significant at the 99% confidence level, small signs refer to stations significant at the 95% confidence level. Stations not significantly correlated are not shown...... 187
5.59 1000 hPa Geopotential Height standardized regression coefiBcients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test.... 192
5.60 1000 hPa Streamlines associated with Nino4, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coefiBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 193
xvu 5.61 850 hPa Streamlines associated with Nino4, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coeflBcients’ t-scores at the 95% and 99% confidence levels respectively ...... 194
5.62 850 hPa u-wind standardized regression coeflBcients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 195
5.63 850 hPa v-wind standardized regression coeflBcients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 197
5.64 300 hPa Streamlines associated withNino4, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively ...... 198
5.65 300 hPa u-wind standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 201
5.66 300 hPa v-wind standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 202
5.67 1000 hPa Specific Humidity standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test . . . 203
5.68 PWAT standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 205
xvm 5.69 OLR standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test ...... 206
5.70 300 hPa Specific Humidity standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test.... 208
5.71 Standardized level of Lake Victoria, 1899-1989 ...... 212
5.72 Covariance between level o f Lake Victoria and the NAOI, 1899-1989 ...... 213
XIX LIST OF TABLES
TABLE PAGE
3.1 NCEP/NCAR Reanalysis variables used in the study, the levels for which they are available. Data reliability is best for class A variables, and worst for class C variables...... 39
5.1 Locations and names of GHCN Precipitation Stations significantly correlated to NAOI over SEA, with the magnitude of the correlation, and the statistical significance at the 95% or 99% confidence levels. Also shown are the magnitudes of the stations correlations to Nino4, and the statistical significance...... 94
5.2 Temporal dependence of the correlation coefiBcients between the NAOI and S E A R ...... 101
5.3 Locations and names of GHCN Precipitation Stations significantly correlated to NAOI over SAP, with the magnitude of the correlation, and statistical significance at the 95% or 99% confidence levels. Also shown are their correlation coefficients to Nino4, and statistical significance...... 103
5.4 Correlation coefiBcients between atmospheric circulation and precipitation indices. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined...... 125
5.5 Correlation coefficients between SEAR and NAOI stratified according to the Nino4 SST index. Statistically significant correlations and probabilities are highlighted in bold. The number of winter per category of Nino4 are also given. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined ...... 189
XX 5.6 Correlation coefficients between SAFR and NAOI stratified according to the Nino4. Statistically significant correlations and probabilities are highlighted in bold. The number of winter per category of NAOI are also given. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined...... 190
5.7 Correlations coefficients between East Afiican Lake Levels, the NAOI and the Afiican precipitation index, 1899-1989. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined. Data for Lakes Malawi and Tanganyika are only available for the period 1948-1978 inclusive ...... 214
XXI LIST OF ABBREVIATIONS
CO ADS Comprehensive Ocean-Atmosphere Dataset
DJF December, January and February.
DJFM December, January, February and March
EN El Nino
ENSO El Nino-Southem Oscillation
EOF Empirical orthogonal function
Magnitude of the Coriolis force
Coriolis component acting along the y-axis
Coriolis component acting along the z-axis
GA Greenland above
GB Greenland below
GHCN Global Historical Climatology Network hPa Hectopascals (equal to 1.0 millibar)
xxu ITCZ Intertropical convergence zone
JFM January, February and March
NAO North Atlantic Oscillation
NAOI An index representing the phase and intensity of the NAO
NCAR National Center for Atmospheric Research
NCEP National Center for Environmental Prediction, formerly the NMC
NET Northern limit o f the southeasterly trades blowing off the Indian Ocean onto southeastern Africa.
NMC National Meteorological Center
OLR Outgoing longwave radiation at the top of the atmosphere
PNA Pacific North American teleconnection pattern
P-E Precipitation minus evaporation
PWAT Precipitable water vertically integrated from 1000 hPa to 300 hPa.
QBO Quasi-biennial oscillation
xxm SAF A region of southern Africa including portions of northeast South Africa, Lesotho and Swaziland.
SAFR A rainfall index for the SAF region.
SEA A region of southeast Africa between the equator and 20°S, 20°E to 40“E.
SEAR A rainfall index for the SEA region.
SLM Southernmost limit of the northeast monsoon flow over southeastern Africa.
SLP Sea level pressure
SO Southern Oscillation
S O.I. Southern Oscillation Index
SST Sea surface temperature
S VD Singular value decomposition
u Zonal wind component
V Meridional wind component
ZAB Zaire air boundary
XXIV CHAPTER 1
INTRODUCTION
1.1 Introduction
Global climate and environmental variability has the potential to cause major changes in the structure of the world economy over the coming decades. Already unprecedented multilateral global action has minimized or eliminated the use of chlorofluorocarbons (CFCs) in response to depletion of stratospheric ozone and consequent health hazards. The manner in which human society uses and values natural resources, both renewable and non-renewable, is currently being examined and modified due to potential global environmental problems facing the inhabitants of this planet.
Less developed countries typically expend large amounts of gross national product in debt repayments leaving relatively little for investment in social programs. These marginalized societies alleviate their financial problems by increasingly permitting extraction of their natural resources, which contributes to local, regional and global environmental problems. Extensive social welfere programs are non-existent, and resources are not available to alleviate effects of drought and other widespread natural disasters. Consequently inhabitants of less developed countries are more vulnerable to natural hazards related to climate variability than populations of more developed countries. Precipitation is an important component of the global hydrological cycle and is also
vitally important to human activities and agriculture in particular. Rapid extraction of
groundwater is seriously depleting a scarce natural resource. Water rights and access to
water resources are quickly becoming topics of contention on global, national and local
scales. Precipitation receipt is vital to replenish fast dwindling groundwater resources and
effective water management policies are necessary to conserve this resource. An
understanding of the climate dynamics necessary for medium-range rainfall prediction is
critical for effective planning and management of water and other scarce resources. Drought
in the Sahel region of Africa illustrates the importance of water resources in the fragile human
economy and, in particular, to the marginalized peoples of the world whose living is closely
dependent on their ability to irrigate crops and provide water for their animals. Economies
of developed countries are more insulated from regional drought, and subsequent food supply
shortages, due to the global nature of the economic system and the ability to purchase foodstuffs and supplies necessary to makeup a local or regional scale shortfall. Those in less developed countries, with highly localized economies and few if any resources to purchase additional food and supplies, cannot afford the luxury of not knowing whether the rains will come.
The tropical regions of the Southern Hemisphere are important components o f the climate system but our knowledge of the climates and climatic variability of these regions and the contribution of these regions to the global climate system are poorly understood. The nature o f the population distribution and the largely underdeveloped economies in much of the tropics and Southern Hemisphere contribute to poor data availability in these areas and result in poor understanding of these important components of the climate system. The
advent of satellite derived climate data since the 1970s has helped remedy the paucity of data
for these areas and will prove extremely important to atmospheric scientists in forthcoming
decades once data are available over a relatively long time period.
1.2 Research Outline
The purpose of this dissertation is to describe precipitation variability over southern
and southeastern Africa associated with large scale atmospheric circulation anomalies.
Climate variability results from processes acting on large scales over relatively long (e.g.
monthly or seasonal) time scales coupled with local or regional scale processes acting on
similar or shorter timescales. Understanding of global scale atmospheric teleconnections
linking climate fluctuations of a region to atmospheric variability occurring in a remote location, e.g. large scale climate variability associated with the El Nino phenomenon, is essential in separating different spatial and temporal scales of variability. This study will describe associations between well known atmospheric teleconnections, the North Atlantic
Oscillation (NAO) and the coupled El Nino-Southem Oscillation system (ENSO), to precipitation variability over southern and southeastern Africa.
The NAO has long been known to influence Northern Hemisphere climates (e.g. Van
Loon and Rogers, 1978) but its relation to climate variability in the tropics and the Southern
Hemisphere is largely unknown. Climate variability associated with the NAO has been described in detail in observational studies but linkages to climate variability in the tropics and to the Hadley circulation are less well known. The atmospheric circulation associated with the NAO is hypothesized here to cause a significant proportion of precipitation variability over AJfrica south of the equator. Analysis of this hypothesis involves describing the atmospheric circulation over south and southeastern Afiica associated with NAO variability, followed by an analysis of the relation between precipitation variability associated with circulation anomalies throughout the atmosphere. The ENSO phenomenon is linked to precipitation and circulation anomalies on a global scale, including those over southern and southeastern Afiica (e.g. Trenberth, 1996). The relation between the anomalous ENSO circulation and precipitation anomalies over Afiica, south of the equator, will similarly be examined throughout various levels of the atmosphere.
This study utilizes the NCEP/NCAR reanalysis dataset, which is new and unique, with data available for the mandatory atmospheric levels in addition to the surface, thus enabling a thorough description of the atmospheric dynamics associated with precipitation variability.
Datasets having missing observations often compensate by replacing the missing data with an interpolated value fi-om neighboring points. This new dataset however, spreads information from regions having plentiful data to regions where it is sparse or incomplete using predicted values of existing observations output from a sophisticated climate model. A state of the art data assimilation system combines observations from many different sources (e.g. satellites, radiosondes, buoys and existing datasets), checks for erroneous values and then initializes data selected for use in the model. Data values are therefore dependent to some extent on the nature o f the climate model, but variables used in this study are closely related to observations and are not dependent to a large extent on model physics or parameterization schemes. CHAPTER 2
LITERATURE REVIEW
2.1 Introduction
This chapter reviews the role played by the NAO and ENSO in hydroclimatological variability over Africa. Section 2.2 emphasizes climate variability associated with the NAO,
Section 2.3 focuses on interannual variability of precipitation over Africa south of the equator in relation to ENSO and to interactions with South Atlantic and Indian Ocean SST variability.
The principal features of the climate system and of quasi-periodic modes of variability over
Africa south of the equator during boreal winter (DJF) are described.
2.2 Climate variability associated with the North Atlantic Oscillation
The NAO is a large scale exchange of atmospheric mass between Iceland and the
Azores that controls the pressure gradient and the strength o f westerlies over the North
Atlantic Ocean. The NAO exhibits variability on both interannual (Van Loon and Rogers,
1978; Rogers, 1984) and decadal timescales (Hurrell, 1995; Hurrell and Van Loon, 1997).
Eigenvector or principal component analysis shows that the NAO accounts for 33% of the variance in the winter SLP field over the North Atlantic Ocean (Cayan, 1992), is observed in all seasons (Rogers, 1990), and is one of the prominent teleconnection patterns in the
Northern Hemisphere (Wallace and Gutzler, 1981; Bamston and Livezy, 1987; Wallace et al.,
1990).
Wallace and Gutzler (1981), Rogers (1984) and Hurrell (1995) define the meridional pressure gradient between the North Atlantic centers of action as an index of both the strength of the NAO and the strength of the zonal wind over the North Atlantic Ocean. Each study uses a slightly dififerent index to characterize the NAO. Rogers’ (1984) NAO index
(NAOI) is based on SLP differences between Akureyri, Iceland and Ponta Delgada, Azores.
The winter (December-March) NAOI used by Hurrell (1995) and Hurrell and Van Loon
(1997) is based on normalized SLP differences between Stykkisholmur, Iceland and Lisbon,
Portugal. The signal to noise ratio for Hurrell’s NAOI is greater during the winter months than for Rogers’ (1984) index, but the latter has a greater signal to noise ratio during the remainder of the year and annually (Hurrell and Van Loon, 1997).
The positive (negative) phase of the NAO is characterized by anomalously low (high) pressure near Iceland and high (low) pressure over the Azores (Figs.2.1 and 2.2), resulting in a strong (weak) pressure gradient and very strong (weak) North Atlantic westerlies. During winter the mean westerlies into Europe are more than 8 m s'^ stronger during the positive
NAO phase than during the negative NAO phase (Hurrell, 1995). The North Atlantic centers of action intensify and move to the northeast of their climatological positions with an increase in the NAOI, and weaken and shift to the southwest with a decrease in the NAOI (Rogers,
1997; Kapala et ah, 1997). These circulation anomalies result in temperature oscillations over o
1011
IIGHER.SST
1020
Figure 2.1 Climate anomalies during GB events (comparable to the NAO Positive phase); Sea Ice extent and SST anomalies are depicted hy forward and hack hatching (after Barlow et aL, 1993, based on Moses et aL, 1987). 2Z
oz
0Z4
101Z
if 101S
.OWERS:
I01Z
tO K
Figure 2.2 Climate anomalies during GA events (comparable to the NAO Negative phase); Sea Ice extent and SST anomalies are depicted hy forward and hack hatching (after Barlow et aL, 1993, based on Moses et aL, 1987). certain regions between differing phases o f the NAO. The positive (negative) phase of the
NAO is associated with anomalous southwesteriy (northeasterly) flow over Northern Europe, northerly (warm southerly) flow over western Greenland and over eastern North America.
The temperature ‘seesaw’ between Jakobshavn, Greenland and Oslo, Norway is an aspect of the NAO (Van Loon and Rogers, 1978; Moses et aL, 1987) with modes named
‘Greenland Above’ (GA) for instances where the temperature at Jakobshavn exceeds that at
Oslo by 4.0° C and ‘Greenland Below’ (GB) when the opposite occurs. The temperature seesaw phases as defined here do not wholly correspond to, but are related to, the positive or negative phases of the NAOI. Positive (negative) extremes of the NAOI have northerly
(southerly) flow over western Greenland and southerly (northerly) flow over Scandinavia which is related to the GB (GA) mode of the seesaw. However, if the temperature difference between Jakobshavn and Oslo is less than 4.0°C a GB or GA event is not defined as having taken place, regardless of the NAO phase. The correlation coefBcient between the time series of the seesaw and the NAOI is relatively small, having r-0.5 (Rogers, pers. comm., 1999).
Eigenvector or principal component analyses reproduce the dipole characteristic of the NAO in SLP and geopotential height fields. Kutzbach (1970) and Wallace and Gutzler
(1981) show that the principal mode of January SLP variability closely resembles the NAO dipole, but has a subpolar center of action somewhat to the west of the climatological location of the Icelandic low. Wallace and Gutzler (1981) also show that the 1000-700-hPa thermal pattern associated with the NAO closely resembles that of the Pacific North American (PNA) teleconnection pattern, which occurs due to an out of phase relationship between the
Icelandic low and the Aleutian low.
10 The anomalous positive phase of the NAO since the mid-1970s (Fig.2.3) is related to the increased surface temperature over the Northern Hemisphere between 20"-80° N (Wallace et a i, 1995; Hurrell, 1996) accounting for 31% of the interaimual variance of hemispheric surfece temperature. Hurrell (1995) performed singular value decomposition (SVD) analysis between Northern Hemisphere surface temperatures and SLP over the period 1864-1995 to extract the principal modes of covariability. The first SVD mode shows covariability between
SLP and surface temperatures which appears to be closely associated with the NAO. The correlation between the temporal expansion coefficient of this mode and the NAOI over that period is 0.91, illustrating the relationship between the dynamics associated with the NAO with the recent warming in the Northern Hemisphere.
SLP fluctuations associated with the NAO are related to changes in synoptic activity over the North Atlantic Ocean (Serreze et aL, 1997) and in the mean storm track (Rogers,
1997). The mean North Atlantic storm track is oriented towards the northeast North Atlantic during the positive phase and towards the Mediterranean Basin during the negative phase
(Rogers 1990, 1997; Hurrell and Van Loon, 1997; Serreze et aL, 1997). The positive NAO phase is associated with a higher fi'equency and intensity of low pressure systems passing close to the mean location of the Icelandic low (Serreze et aL, 1997). An increase in the fi'equency of blocking high pressure patterns in the extreme northeast Atlantic Ocean occurs during the negative phase, associated with higher than usual pressure in the vicinity of the
Icelandic Low (Moses et aL, 1987). There are fewer cyclones in the vicinity of the Icelandic low but cyclonic activity is increased further to the south (between 40°-60“ N) during the negative phase (Serreze a/., 1997).
11 3
2
1
0
- 1
-2
-3 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 Year
Figure 2.3 Time series of the standardized NAOI, 1875-1995.
12 Hurrell (1995) and Hurrell and van Loon (1997) describe precipitation variability over
Europe associated with the NAO. The orientation of the mean storm track controls the transport of vertically integrated water vapor over the North Atlantic (Fig.2.4) and differs between phases of the NAO (Hurrell, 1995). Composite analysis shows that the atmospheric moisture budget (Fig.2.5) given by precipitation minus evaporation, (P-E) is significantly negative (positive) over northern Europe, much of North Afiica and along the east coast of the U.S. during negative (positive) NAO events. Simultaneously central Europe, the northern and eastern Mediterranean, western Greenland and the Davis Strait have significant positive
(negative) P-E departures (Hurrell, 1995; Hurrell and Van Loon, 1997). This distribution is consistent with highly significant positive (negative) NAO-precipitation correlations observed over northern (southern) Europe in winter (DJFM)(Hurrell, 1995; Hurrell and van Loon,
1997). The P-E departures are consistent with those of Lamb and Peppier (1987) who show a statistically significant negative relationship between winter precipitation and the NAOI over
Morocco which appears to be strongly related to the variability in location and intensity of the Azores High that controls the movement of low pressure systems over this region.
The persistent strong positive NAOI evident since the late-1970s to the mid-1990s
(Fig.2.3) which induces cold dry northerly flow over western Greenland around the vigorous
Icelandic low, is linked to a decline in modeled precipitation over Greenland over the past two decades (Bromwich et al., 1993). Significant negative changes in regional precipitation variability occur over southern Europe and the Mediterranean Basin, while northern Europe has been wetter than normal over this period (Hurrell and Van Loon, 1997). Eigenvector analysis of North Atlantic precipitation has a signal resembling that of the NAO (Dai et al.,
13 .zaa .200
100 - 0 30-E
.200
20Q
90*W 0
Figure 2.4 Vectors of the transport of vertically integrated water vapor over the North Atlantic during the positive NAO phase (a) and negative NAO phase (b). Isopleth intervals of magnitudes is at 50 kg m^ s'^ (after Hurrell, 1995).
14 • ••»••• •••
30‘’W
Figure 2.5 Differences in P-E over the North Atlantic between positive and negative NAO phases. Isopleth interval is 0.5 mm day ; stippling indicates statistical significance at the 95% level (after Hurrell, 1995).
15 1997). This NAO-related precipitation EOF was also shown to have a dominant periodicity at about 7 years identified separately as an important periodicity in the NAO. Rogers and
Van Loon (1979) also find significant NAOI-precipitation relationships over Japan, China and
India during January.
The NAO signal is observed in studies emphasizing interannual climate variability related to atmosphere-ocean coupling. Wallace ei al. (1990) relate the NAO to SST variability over the North Atlantic Ocean, with the leading mode of SST variability resembling the NAO. It is shown that distribution of the correlation between the time series of the leading EOF of Atlantic Ocean SSTs to 500 mb heights closely resembles the NAO signal at that level. Eigenvector analysis shows that the NAO is the principal mode of Atlantic SLP variability over the North Atlantic Ocean and that the related air-sea energy exchanges over the North Atlantic Ocean are highly dependent upon wind direction (Cayan, 1992). The consequent energy exchange maxima over the Gulf Stream and the Labrador current are dipolar in structure and related to the phase of the NAO as is the saturation deficit over these regions (Cayan, 1992; Alexander and Scott, 1997). The positive (negative) phase of the NAO is associated with a positive (negative) saturation deficit and positive (negative) latent and sensible energy fluxes to the east of Labrador and south of Greenland and off the west coast of North Afiica and in small areas of the northern tropics (Cayan, 1992). Negative (positive) air-sea energy fluxes and negative (positive) saturation deficit to the east o f the United States immediately north of Bermuda and in the northeast Atlantic Ocean from Ireland to
Scandinavia also occur during the positive phase of the NAO (Cayan, 1992).
16 Increased sea ice concentrations in the Davis Strait and Labrador Sea are associated with anomalous northerly flow over western Greenland during the positive phase of the NAO.
During the negative phase sea ice extent is below normal in the Davis Strait/Labrador Sea region due to the anomalous southerly flow around the weak Icelandic Low (Rogers and Van
Loon, 1979; Kelly et al., 1987; Koslowski and Loewe, 1994). Sea ice concentrations tend to vary in opposition between the Baltic Sea and the Davis Strait, with the positive (negative) phase of the NAO coincident with light (heavy) concentrations in the Baltic Sea (Rogers and
Van Loon, 1979; Koslowski and Loewe, 1994).
Meehl (1978), Rogers and van Loon (1979), Meehl and van Loon (1979) and Lamb and Peppier (1987) have shown that certain climatic signals in the tropics and subtropics are significantly related to the NAO. Subtropical and tropical Pacific, Atlantic and Indian Ocean
SSTs are significantly lower (higher) during GB (GA) winters and the northeast trades over the North Atlantic are strong (weak) during GB (GA) winters (Rogers and van Loon, 1979;
Meehl and Van Loon 1979). The southeast trades over the South Atlantic and South Pacific
Oceans are strengthened (weakened) during GB (GA) winters (Rogers and van Loon, 1979).
Meehl and Van Loon (1979) also suggest that the ITCZ-related band of maximum precipitation shifts to the south over southern Afiica during GB winters. A band of negative precipitation differences between GB and GA phases is observed across Afiica between the equator and about 15° S, with positive departures to the south implying increased (decreased) rainfall over this regon during GB (GA) winters. It is also shown that the Gulf Stream flows faster (slower) during GA (GB) winters as indicated by fluctuations in relative sea level between pairs of hydrographic stations (Meehl and van Loon, 1979). However these
17 alterations could be due to dififerences in salinity and thermocline depth between the hydrographic stations and cannot be considered conclusive evidence of changes in the Gulf
Stream.
Rogers (1984) investigated the relationship between the NAO and the Southern
Oscillation (SO) in the Northern Hemisphere and showed that temperatures over southeast
North America are positively related to the interactions between these two interannual oscillations. Cold (warm) winters occur over this region when both the NAO and the SO are in their low (high) phase. A slight positive association was found between the strength of the
North Atlantic westerlies and the phase of the SO, with strong (weak) westerlies tending to occur during the high (low) phase of the SO. Rogers (1984) suggested that the SO signal in the Northern Hemisphere modulates the NAO signal over time, but could not produce strong evidence. The cospectrum of the NAOI and the normalized winter SLP anomalies at Darwin between 1900-1983 showed greatest power at 5.7 years.
The spectral characteristics of the NAOI for the period 1899 to 1989 (Fig.2.6) depicts a spectrum dominated by low frequency variability with a much smaller contribution at high frequencies. The NAOI spectrum has its largest powers at 7.58 years, between 3.7 to 3.9 years, at 2.6 and 2.1 years, between 5.0 to 5.3 years and at periods longer than 10 years. The peaks at longer than 90 years, at 2.67 years and at 7.6 years are all statistically significant, the latter of which agrees with Rogers’ (1984) statistically significant spectral peak at about 7.3 -
8.0 years. However evidence of the longest periodicities cannot be substantiated due to the relatively short length of record. Rogers (1984) also shows that 43% of the variance of the
NAOI occurs at periods greater than 6.7 years implying that the factors influencing the
18 0.7
0.6
0.5 (Ui_ S: O 0.4 CL o L_ 0.3 (DÜ CL CO 0.2
0.1
0.0 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period
Figure 2.6 Spectrum of the NAOI from 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis.
19 variability of the NAOI are primarily low frequency in nature. The cospectrum of January
temperatures at Oslo, Norway and Jakobshavn, Greenland has statistically significant
periodicities occurring between 2.2 and 2.5 years and around 6 years (Rogers and Van Loon,
1979).
The power spectrum of the Lisbon-Stykkisholmur NAOI (Hurrell and van Loon,
1997) shows significant variability at biennial periodicities, with low power occurring at 3-5
years and enhanced power at 6-10 years. The biennial contribution is strongest between
1865-1944 while the 6-10 year periodicity is strong between 1915-1974. The low frequency
contribution to the power spectra occurs from 1915 to the end of the record. In addition to
variability on interarmual timescales, the NAOI has a strong decadal component to its
variability (Hurrell, 1995). The interannual variability of the normalized DJF NAOI time series
(Fig.2.3) is characterized by mostly above normal wintertime values from about 1900 to the
mid-1920s, a decrease from the mid-1920s to a minimum during the mid-1960s and a
subsequent strong positive trend. Enhanced advection of warm, moist air masses into Europe
during the positive NAO phase appears related to increased surface temperatures over the
Northern Hemisphere between 20° and 80° N (Wallace et a i, 1995; Hurrell, 1996),
accounting for 31% of the interarmual variance of hemispheric surface temperature since the
1930s.
Shabbar et al. (1997) argue that a climate system change occurred in 1970 related to
changes in the interarmual variability of the NAO. Their pressure difference index between
BafiBn Island and the western Atlantic (their BWA index) measures the circulation strength associated with the western extremes of the NAO. High frequency variability dominates the
20 spectrum of the BWA index between 1947-1969 while low frequencies dominate from 1970-
1995. The shift between climate regimes appears to be related to a simultaneous strengthening of both the PNA and NAO. Wallace and Gutzler (1981) noted that both the NAO and the
PNA are related to winter temperatures over the eastern United States.
21 2.3 Interannual variability of precipitation over Africa south of the equator
Studies o f climatic variability over Southeastern Africa have primarily examined
relationships to global circulation features such as ENSO and associated SSTs of the oceans
surrounding Africa. Several studies show statistical relationships between eastern and
southern African regional rainfall receipt and ENSO (Tyson, 1986; Ropelewski and Halpert,
1987; Lindesay, 1988; Ogallo, 1988; Van Heerdeene/ar/., 1988; Jury and Pathack, 1993; Jury
et al., 1995; Nicholson, 1996; Ropelewski and Halpert, 1996; Kruger, 1999). Only the
exploratory work of Meehl and Van Loon (1979) shows any linkages between the NAO and
precipitation over tropical or southern Africa. They identified a southward movement of
rainfall associated with a possible southward displacement of the ITCZ over much of tropical
Africa between the equator and 15°S during GB winters.
The importance of moist westerly flow from Zaire to above normal precipitation receipt over equatorial eastern Africa is shown by Davies et al. (1985) and Nicholson (1996).
However Nicholson (1996) points out that some outbreaks of moist westerlies from Zaire
(Fig.2.7) have been associated with below normal precipitation receipt over eastern Kenya; no explanation is provided for this however. Recent research has focused on the interrelationships between African rainfall, tropical SSTs and ENSO (Nicholson, 1996).
Precipitation variability over much of equatorial eastern Africa is related to ENSO, with above normal precipitation during El Nino years and drought occurring the following year
(Nicholson, 1996). Equatorial African precipitation responds differently to ENSO forcing during the dififerent precipitation seasons. Nicholson (1996) shows that equatorial eastern
Afiican rain is more affected in April and May by ENSO than rainfall during October and
22 t œ t 2 9 0471471 M 3
Figure 2.7 Political Map of Africa (after African Data Dissemination Service, 1999)
23 November. Rainfall during October and November is positively related to the intensity of the
SST gradient in the Indian Ocean, and especially with positive SSTs in the western Indian
Ocean (Mcholson, 1996). Ogallo (1988) shows that seasonal rainfall is negatively related to
ENSO events over much of equatorial eastern Africa with ENSO forcing especially strong during the Boreal Winter.
The negative relationship between summer (DIE) rainfall over South Africa and
ENSO is statistically significant (Tyson, 1986; Ropelewski and Halpert, 1987; Lindesay,
1988; Van Heerdeen etal., 1988; Jury and Pathack, 1993; Jury et al., 1995; Ropelewski and
Halpert, 1996; Kruger, 1999) and is associated with near surface and upper level circulation anomalies which modulate the climate o f South Africa (Tyson, 1986; Lindesay, 1988; Van
Heerden, 1988; Kruger, 1999). While the negative correlation between the S O I. and Austral
Summer rainfall has been shown to be significant, there is considerable variability in the relationship with some El Nino events (e.g that of 1976/77) associated with above normal precipitation over large portions of South Afirica (Kruger, 1999).
Correlations between the NinoS SST index (Trenberth, 1996), a measure of El Nino, and rainfall appear to be dependent on the seasonality of precipitation over South Africa
(Kruger, 1999). Statistically significant negative correlations are found in the northeast of
South Africa, over Lesotho and southern Swaziland from October to December (Kruger,
1999; his Fig 2a), and move southwest over the summer with the advancing precipitation to cover most of central South Afiica during January to March (Kruger, 1999; his Fig 2b). This dependence of the distribution of the largest S.O.I. correlation coefficients on the timing of the rainfall maximum is also noted by Lindesay (1988) and by Van Heerden et al. (1988).
24 The relationship appears linked to increased tropical control over mid-latitude South Africa
which is at a maximum during Austral summer. Van Heerden et al. (1988) show that the
correlation maximum between ENSO and rainfall over Southern Africa is modulated by the
semi-annual change from a baroclinie atmosphere in December to a quasi-barotropic
atmosphere during January and February, with a shift back to a baroclinie regime in March.
The relationship between ENSO and South African rainfall is at a maximum during
December and March (Van Heerden et al., 1988), with much lower correlations during
January and February. This suggests that tropical controls on the climate of South Africa,
excluding ENSO, are greatest during the mid-summer months of January and February, while
extratropical influences are greater during the remainder of the year.
Circulation anomalies associated with ENSO include negative 500 hPa geopotential
height anomalies over and to the north of South Africa with positive 500 hPa heights to the
south of the continent (Van Heerden et a l, 1988). This pattern of geopotential height
anomalies produces weakening (strengthening) of mid-level westerlies coincident with
increased (decreased) precipitation over South Africa during El Nino (La Nina). An increase in the frequency of cut-off low pressure systems and near stationary troughs (Van Heerden et a i, 1988) is another possible cause of the link between decreased heights and decreased precipitation over South Africa. The significance of the relationship between circulation during ENSO events and rainfall over central (the Karoo) and northeastern (the Highveld)
South Africa is somewhat different. The S.O.I.- rainfall relationship is less strong over northeastern South Afiica than over the dry steppe of central South Afiica and it is suggested that this difference is possibly due to rainfall producing mechanisms. Rainfall over central
25 South Afiica is associated with passage o f large-scale synoptic systems while orographic thunderstorms occur along the eastern escarpment and convective air-mass storms occur over northeastern South Afiica. Invasions o f moist air fi*om the Indian Ocean and associated convection over northeastern South Afiica possibly mask the ENSO precipitation signal and contribute to lower correlations over northeastern South Afiica (Van Heerden et al., 1988).
It has been hypothesized that the link between ENSO and Afiican precipitation is through tropical SSTs in the oceans surrounding Afiica (Nicholson and Entekhabi, 1986,
1987; Nicholson, 1996). Nicholson and Entekhabi (1987) and Nicholson (1996) show that the spectrum of Indian and Atlantic Oceans SSTs has significant power between 5 - 6 years, which is also evident in a spectrum of regional east Afiican precipitation, and in the S.O.I.
Nicholson and Entekhabi (1986, 1987) describe the periodic components of Afiican precipitation showing the significant spectral peak at 5.0-6.3 years at stations in southeastern
Afiica. Nicholson (1996) hypothesizes that the common periodicities hint at a linkage between ENSO, SSTs in the Indian and South Atlantic Oceans and eastern Afiican rainfall.
East African precipitation stations fi'om the equator to about 15°S approximately between latitudes 25°-35°E have dominant spectral peaks between 2.2 to 2.4 years and between 3.3 and 3.8 years (Nicholson and Entekhabi, 1986). Precipitation over the southern part of this region (10°-15°S, 25°-35°E) is dominated by longer periodicities with largest spectral power between 5.0 to 6.3 years (Rodhe and Viiji, 1976 Nicholson and Entekhabi, 1986).
Interannual rainfall variability over South Afiica has several large quasi-periodic components and spectral peaks with periods of about 18 years, 10 to 12 years and between
2.0 and 2.3 years (Tyson, 1986; Mason, 1990; Mason and Tyson, 1992; Kruger, 1999). The
26 18 year oscillation is prominent in South Afiica rainfall spectra but is most prominent in northeastern South Afiica which experiences large amounts of summer (DJF) rainfall. Wet and dry climate epochs (Kruger, 1999) for the period 1930 to 1995 seem to recur every 9 years (Tyson, 1986; his Table 4.1) substantiating the 18 year periodicity in South Afiican rainfall. The rainfall spectrum of central South Afiica is dominated by a 2.3 year periodicity and the coastal region in the southwest is dominated by periodicities great than 20 years
(Tyson, 1986). The predominant periodicity of the summer (DJF) rainfall spectrum of the southern Cape region occurs between 10 to 12 years but is confined to this relatively small region (Tyson, 1986).
Further analysis of atmospheric teleconnections related to a South Afiican rainfall index during January shows a large positive relationship ( r > 0.6 for the period 1955-1990) to central Indian Ocean SSTs (Jury, 1996), a region which has been shown to have a strong response to ENSO forcing (Jury eta l, 1994). Correlation between SSTs and January rainfall over Zimbabwe shows coefiBcients of slightly greater magnitude over the central Indian Ocean
(Jury, 1996). Both Zimbabwean and South Afiican summer rainfall appear to be forced by similar mechanisms, highly correlated with a r = 0.62 (Jury, 1996). This is possibly related to tropical circulation controls on the summer climate and precipitation regime while mid latitude forcing dominates at other times. The distribution of the correlation coefiBcients between January Zimbabwean rainM and outgoing longwave radiation (OLR) ( Fig.2.8; Jury,
1996; his Fig 7c), shows large significant negative correlations over the Gulf of Guinea extending on both sides of the equator to South America. This implies that a reduction in the deep tropical convection associated with the ITCZ over this region is associated with an
27 O ê
-10 1 0 ao 110 voHotnxx
Figure 2.8 Correlation coefficients of the January Zimbabwean rainfall and OLR. Statistically signiffcant correlations are indicated by large + and - signs (after Jury, 1996).
28 increase in Zimbabwean rainfall. Similar correlations between South African rainfall and
January OLR exhibit a strong dipole between South Africa and Madagascar (Jury, 1996; his
Fig.7a) substantiating the hypothesized overturning circulation between the regions, possibly
related to a branch of the Walker circulation (Juryet al., 1994). The January atmospheric circulation exhibits easterly flow at several levels over South Africa associated with rainfall.
Evidence o f a possible lagged relationship between South African rainfall and the Indian monsoon (Fig.2.9; Jury, 1996; his Fig. 6a) appears in OLR (stretching from India to eastern
Africa) correlations to January precipitation.
A strong relationship has been established between SSTs of the upwelling Benguela current of western Africa and precipitation over southern and eastern Africa (Nicholson and
Entekhabi, 1987) with annual rainfall receipt increasing (decreasing) during years with anomalously warm (cold) SSTs along the Benguela coast. The enhancement (reduction) of annual precipitation during warm (cold) years is shown to be particularly strong in eastern
Africa between the equator and 10°S. A weakening and equatorward movement of the South
Atlantic subtropical high pressure cell, and resultant weakening of the southeast trades, is possibly related to the SST variability along the Benguela coast and also to the interannual variability of precipitation over southern and eastern Africa.
Ocean-atmosphere interactions play an important role in determining rainfall over
South Africa during the Austral summer (Walker, 1990; Jury, 1995; Mason, 1995; Mason,
1998). Lower (higher) SSTs in the tropical Indian Ocean (latitude < 20° S) are related to lower (higher) than usual rainfell over eastern South Africa in JFM (Walker, 1990). Positive relationships occur between SSTs in the vicinity of the Agulhas and Benguela currents, to the
29 2 0
'o
- 10- O i
o
-40 -7 0 -10 10 7 0 •0
Figure 2.9 Lagged correlation coefficients between November OLR and January rainfall over South Africa. Statistically significant correlations are indicated by large + and - signs (after Jury, 1996).
30 southeast of South Afiica and precipitation over eastern South Afiica (Walker, 1990).
Relationships between these SSTs and South Afiican rainfall are quite different when the
ENSO signal is removed (Walker, 1990). Walker (1990) performs composite analysis of wet
and dry extremes in South Afiica using only those extremes with small S.O.I. values, eflTectively eliminating the ENSO signal. Removal of ENSO influences almost eliminates the significance o f the negative correlation between rainfall and tropical Indian Ocean SSTs.
A strong anomalous atmospheric anticyclone is present to the southeast of South
Afiica during wet summers with higher than normal SSTs to the south of Afiica, and is associated with enhanced poleward flow in the Agulhas current (Walker, 1990). This enhanced poleward transport of warm water increases the positive SST anomaly, and warms and moistens the marine boundary layer to the south of Afiica, completing a positive air-sea feedback mechanism (Walker, 1990). The strong anticyclone enhances moist onshore
(easterly) winds and convergence over the interior of eastern South Afiica. It is also hypothesized (Walker, 1990) that the enhanced SST gradients, and large latent and sensible heat fluxes to the atmosphere, upwind of South Afiica produces a region favorable for cyclogenesis.
A statistically significant relationship between precipitation and solar activity is observed over South Afiica after stratifying the quasi-biennial oscillation (QBO) in equatorial stratospheric winds into east and westerly phases (Mason and Tyson, 1992). When the QBO is in its easterly (westerly) phase an increase (decrease) in solar activity during the early summer (October to December) and positive (negative) SSTs in the southwest Indian Ocean are associated with an increase (decrease) in precipitation over South Afiica. An opposite
31 relationship between the QBO, rainfall, SSTs and solar activity is observed during late summer season (January to March). The precipitation decrease is related to a possible reduction of tropical-extratropical troughs that connect tropical systems over central South
Africa to mid-latitude depressions to the south of the continent (Mason and Tyson, 1992).
These tropical-extratropical troughs are the principal determinants of rainfall anomalies in the summer rainfall regime of South Afiica (Harrison, 1984) and are affected by atmospheric variability related to ENSO (Lindesay, 1988).
An antiphase relationship between Austral summer OLR over South Afiica and over the western Indian Ocean is related to uplift over South Afiica and descent over Madagascar during summers when the QBO was in the westerly phase (Jury et al., 1994). These relationships between the QBO phase and climatic conditions in the Southern Hemisphere have been suggested by Labitzke and Van Loon (1989). They show that geostrophic winds over a large extent of the Southern Ocean between 100° W to about 100° E have a large and significant response to solar activity during the east phase of the QBO and produces strengthened geostrophic midlatitude westerlies. It is known that SSTs in the South Atlantic and Indian Oceans have a strong spectral peak at about 11 years hinting at a possible solar linkage in the strength o f the westerlies and thus to precipitation (Tyson, 1986; Mason and
Tyson, 1992).
32 CHAPTERS
DATA AND METHODOLOGY
3.1 Introduction
The data, and methodologies used to analyze them, are described in this chapter. Due to the unique nature of the National Center for Environmental Prediction/National Center for
Atmospheric Research (NCEP/NCAR) reanalysis dataset, the data assimilation and quality control schemes will be described along with the assumptions made within the reanalysis scheme. Data reliability and its dependence on the parameterization of processes such as cumulus convection, latent and sensible heat fluxes will also be outlined in Section 3.2.
Precipitation data used in this study are discussed with particular regard to their quality and spatial and temporal distribution in Section 3.3 Data used to characterize large scale atmospheric teleconnections will be described in Section 3.4. Data analysis methodologies are presented in Section 3.5.
33 3.2 General description of the NCEP/NCAR Reanalysis Dataset.
The principal dataset used in this study is the NCEP/NCAR reanalysis described in
detail by Kalnay et aL (1996). This dataset was created using a state of the art assimilation
system similar to one operationally used by NCEP to assimilate vast quantities of data from
a wide variety of sources and instruments over varying spatial and temporal scales.
Previously available global climatological sets, such as the Comprehensive Ocean-Atmosphere
Dataset (COADS), have been incorporated into the data assimilation system in addition to radiosonde, satellite, aircraft, merchant shipping and meteorological station observations.
Consistent usage of the global data assimilation system in the reanalysis results in a climate record that avoids any major analysis discontinuities due to incorporation of a new data source (Kalnay et ai, 1996). The data are then extensively quality controlled using a variety of spatial and temporal analyses.
Reanalysis of the combined datasets is performed using an operational spectral forecast model with T62 resolution, 28 vertical levels, and a grid resolution of approximately
210 km (Kalnay et aL, 1996). With the exception of the horizontal resolution this model is identical to the NCEP global operational model. Boundary conditions such as SST, snow cover, sea ice distribution, albedo, surface roughness and soil wetness are prescribed from existing climatologies and analyses (Kalnay et aL, 1996). The model has five levels in the boundary layer and about seven levels above 100 mb. The model parameterizes major physical processes including convection, large scale precipitation, shallow convection, gravity wave drag, radiation with a diurnal cycle, interactions with clouds, boundary layer physics, interactive surface hydrology and horizontal and vertical diftusion processes (Kalnay et aL,
34 1996). The operational model uses a simplified Arakawa-Schubert convective parameterization scheme, improving the representation o f rainfall in the tropics and over the
United States compared to the previously used Kuo scheme (Kalnay et aL, 1996).
Convection is produced when a cloud work function exceeds a certain threshold. The cloud work function is a fimction of temperature and moisture in each air colunm corresponding to a grid point (Phillips, 1995). In a modification of the Arakawa-Schubert scheme, only the deepest cloud is considered, and the spectrum of clouds are ignored. A downdraft mechanism is incorporated, in addition to evaporation of precipitation. Entrainment of the updraft and detrainment of the downdraft in the sub-cloud layers are also included (Phillips, 1995). Two other improvements include a better diagnostic cloud scheme, producing OLR closely agreeing with observations, and an improved soil model scheme that produces more realistic surface temperatures and better forecast skill (Kalnay et aL, 1996). Further details of the model physics and dynamics are described in NOAA/NMC Development Division (1988),
Kanamitsu (1989) and Kanamitsu et aL (1991).
Data compiled from a variety of sources and sensors worldwide are rigorously screened by quality control methods designed to eliminate errors produced by instrumental, human or communications mistakes (Kalnay et aL, 1996). Rawinsonde heights and temperatures are checked using the hydrostatic equation in addition to horizontal and vertical interpolation checks to ensure consistency, and temporal interpolation between adjacent 12 hour periods are also used to check data quality. Observational data are compared to climatological means in order to detect problematic observations. Optimal interpolation quality control performed on all available data includes a three-dimensional multivariate
35 statistical interpolation scheme to obtain comparison values to compare neighboring
observations. Accurate data whose representation would cause the model to have problems
are also eliminated from the analysis by the assimilation scheme. Output data are also
monitored for inconsistencies using three dimensional analysis and temporal comparisons to
identify extreme deviations from climatology (Kalnay et aL, 1996).
The aim of the reanalysis is to produce a 40 year dataset of daily atmospheric
observations (1957 to 1996). Input data are assimilated into the model and a 6 hour forecast
is performed with these and data from the previous 6 hour period. “The analysis cycle, with the use of the 6 hour forecast as a first guess, is able to transport information from data-rich to data-poor regions, so that even in relatively data-void areas the reanalysis can estimate the
evolution of the atmosphere over both synoptic and climatological timescales” (Kalnay et ai,
1996, p.453). Model output are subjected to an automated quality control check for every six hour forecast period. Some variables, such as precipitation and surface flux fields, are wholly determined by parameterizations of the model’s physics and labeled Class C variables.
Other fields such as atmospheric moisture fields, are determined partially by the model and partly by observations of that variable, and are classified as Class B fields. Class A fields are the most reliable and the least influenced by the model. Upper air temperature data, for example, “ ... are generally well defined by the observations and, given the statistical interpolation of observations and first guess, provide an estimate of the state of the atmosphere better than would be obtained using observations alone” (Kalnay et aL, 1996, p.453).
36 Comparisons between reanalysis output data and observations show that the
differences are quite small even for the class C fields (Kalnay et aL, 1996). Zonal mean
precipitation or radiative fluxes, for example, compare extremely well to observational data,
but the spatial distribution of that data can be somewhat less reliable. Comparison between
the global precipitation data containing monthly precipitation estimates obtained fi'om a
microwave sounding unit (Kalnay et al, 1996), shows small differences with the reanalysis
underestimating tropical precipitation somewhat. Cullather and Bromwich (1996) identified
a spurious wavetrain observed near the south pole to the south of 75°S. These suspicious
wave-like results are characterized by large maxima and appear to be correlated to the
spectral noise of the model at high latitudes. For a variety of locations in Antarctica the
reanalysis output is greater than the observations by a factor of 10 (Cullather and Bromwich,
1996). However a time series of the observations and the reanalysis precipitation data for
Antarctica are well correlated. Precipitation over the southeast United States is
overestimated almost by a factor of 2 (Kalnay et al, 1996); however, the daily variability of the precipitation analysis compares well to station variability (Kalnay et ai, 1996).
Reanalysis data used in this study, and their reliability are given in Table 3.1. Data are obtained for all mandatory pressure levels, but only a few levels are used in this analysis as results are often very similar at adjacent levels. Only data fi'om the surface to 300 hPa are utilized, but the outgoing longwave radiation (OLR) is calculated for the top of the atmosphere. Precipitable water (PWAT) is calculated as vertically integrated specific humidity from 1000 hPa to 300 hPa. Variables available at the Earth’s surface include surface
37 temperature, and sensible and latent heat fluxes. Other variables such as geopotential height, specific humidity and wind speeds are available at the mandatory pressure levels.
38 Variable Name Levels Available Reliability U-wind (m s^) All pressure levels A V-wind (m s^) All pressure levels A Specific Humidity (kg kg'^) 1000 - 300 hPa only B Geopotential Heights (gpm) All pressure levels A Sensible Heat Flux (W m^) Surface C Latent Heat Flux (W m^) Surface C Precipitable Water (kg m^) Surface B Sea Level Pressure (Pa) Surface A Surface Temperature (K) Surface B Outgoing longwave radiation (W m^) Surface C Precipitation Rate (kg m ^ s^) Surface C
Table 3.1 NCEP/NCAR Reanalysis variables used in the study, the levels for which they are available. Data reliability is best for class A variables, and worst for class C variables.
39 3.3 Precipitation Data
Station data are from the Global Historical Climatology Network (GHCN) dataset containing long term monthly mean precipitation for 7533 meteorological stations across the globe having data available up to and including 1990 (Vose et ai, 1992). The number of stations contributing data to the dataset (Fig.3.1) increases dramatically during the early decades of the 1900s, peaking during the 1950s and 1960s but then decreasing after about
1971. The distribution, available time series, and number of stations in Africa are poor in comparison to the data rich regions of Northern Europe and eastern North America, but are comparable to the coverage over much of the world. The distribution of stations with at least
10 years of precipitation data over Africa (Fig.3.2) depicts a relatively (for Africa) high concentration of stations in eastern Africa south of the equator with especially high densities occurring from Kenya through Tanzania, Zimbabwe and Zambia to South Africa. The coastal areas of Namibia and Angola also have relatively high data densities but inland toward the
Kalahari Desert and Botswana an absence of stations is evident. Station coverage is relatively good across the African continent south of about 20°N with coverage sparse over the Sahara
Desert and over North Africa with the exception of the stations along the coastal regions of the Mediterranean.
The distribution of stations with at least 50 years o f precipitation data is relatively poor across Africa (Fig.3.3) but small clusters of stations exist in Zimbabwe, Tanzania along
Lake Tanganyika, Botswana and in northeastern South Africa. Sub-Sahelian west Africa is relatively well covered in comparison to the average for the continent but the remainder of the continent has poor coverage. African precipitation data are consequently largely
40 7500 3
I 6000:
« 4500 - o o 3000- E Â = 1500-4
I' ' I ' r ' " I ' . " I " I 1 ■ '1 i ■ I f 0 -3 —T—r . I . . 1700 1750 1ÎB00 1850 1900 1950 2000 Year
Figure 3.1 Histogram of the number of GHCN precipitation stations by year from 1700-1990 (after Vose et aL, 1992).
41 Figure 3.2 Locations of GHCN stations with at least 10 years of precipitation data (after Vose etoL, 1992).
42 Figure 3.3 Locations of GHCN stations with at least 50 years of precipitation data (after Vose et aL, 1992).
43 comprised of a relatively sparse network of stations having comparatively short records.
However eastern Africa south of the equator from Kenya to South Africa is relatively well
covered by a number of stations with relatively long time series in comparison to the
remainder of the continent.
For inclusion in the GHCN a station had to have at least 10 years of data for one of
the four variables (precipitation, sea-level pressure, station pressure and temperature),
resulting in an uneven distribution o f stations over the globe (Vose et aL, 1992). No
quantitative analysis of data homogeneity has been made for the station data, but gross errors
and discontinuities have been flagged in the dataset. Errors that may have been introduced
include data measured using the imperial scale being rounded to the nearest integer and then
converted into metric. Instrumentation change over time, changes in recording and
observational protocols over time, and procedural differences between countries also
introduce dataset errors and biases (Vose et ai, 1992).
GHCN data are subjected to extensive quality control procedures before distribution
(Vose et al, 1992). Metadata (station identifiers, elevation, latitude and longitude
coordinates) associated with each station were digitally compared to data available from the
latest WMO publication and corrected accordingly. The data were screened for extremes and
errors, for example negative precipitation, and set to “missing”. Time series plots were used to check for large and obvious errors in the data and revized if possible. Duplicate records from differing data sources were merged to form a time series of maximum completeness and length of record (Vose et a/., 1992).
44 3.4 Indices representing atmospheric teleconnections
The North Atlantic Oscillation Index (NAOI) (Rogers, 1984) measures the strength
of the mid-latitude westeriies over the North Atlantic and is calculated from normalized values
of monthly averaged sea level pressure near the NAG centers o f action at the Azores and
Iceland. Monthly pressures at Ponta Delgada, Azores and Akureyri, Iceland form the index and are available since 1875. Positive (negative) values of the NAOI indicate that pressures are simultaneously greater (lower) than normal over the Azores and lower (greater) than normal over Iceland and represent an increase (decrease) in the North Atlantic Ocean pressure gradient and zonal wind speed.
The Nino4 SST index (Fig.3.4) is an areally averaged SST record for the region
5°N-5°S, 160PE-15CPW (Trenberth, 1996) and is an indicator of the phase and intensity of the
ENSO phenomenon over the central Pacific. ENSO is a complex phenomenon which is comprised of the El Nino (La Nina) event in the Pacific Ocean which is manifested as an extensive warming (cooling) of tropical Pacific SSTs, and the Southern Oscillation (SO) which is an atmospheric oscillation related to the basin-wide change in Pacific SSTs
(Trenberth, 1996). The SO is a global scale atmospheric teleconnection pattern and represents the inverse relationship between SLP in Darwin, Australia and Tahiti in the South
Pacific Ocean (Chen, 1982; Trenberth, 1996). The SO is related to basin-wide Pacific SSTs by the zonal Walker circulation producing sinking (rising) motion in the eastern tropical
Pacific and rising (sinking) motion in the western tropical Pacific during La Nina (El Nino) events (Trenberth, 1996). An ENSO event occurs when higher than normal SLPs occur at
Darwin and lower than normal SLPs at Tahiti with a simultaneous warm EN event in the
45 3
2
1
0
- 1
-2
_ 3 I ' I ' I I 1____1 I 1------1------1------1------1------1------1------1------1------1— 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 Year
Figure 3.4 The standardized Nino4 SST index, 1899-1989. SSTs are averaged over the area 5°N-5°S, 150“W-90“W.
46 tropical Pacific (Trenberth, 1996). El Nine (EN) and SO events can occur separately and are not necessarily linked on a consistent basis (Trenberth and Shea, 1987; Deser and Wallace,
1987). Various indices using tropical Pacific SSTs averaged over somewhat different regions are used to describe the phase and intensity of EN (trenberth, 1996) but there is no guarantee that the SO is occurring simultaneously. Regression coefficients associated with a 1.0 standard deviation increase in the Nino4 and Nino3 SST indices were compared on a global scale, and over Afiica in particular, producing very similar results. The Nino4 SST index is representative of a relatively large region of the central equatorial Pacific rather than the
Nino3 region of the eastern equatorial Pacific.
An optimally interpolated SST dataset extending fi'om 1856 to 1991, is used to calculate theMno4 index (Kaplan et al, 1998). Positive (negative) values of the Nino4 index indicate warmer (colder) than normal SSTs in the central Pacific Ocean consistent with the
El Nino (La Nina) events in the Pacific. It is assumed that the atmospheric SO is occurring in conjunction with the EN events.
47 3.5 Data Analysis
Statistical methods used in the data analyses, including methods for calculating
significance and the physical interpretation of the results, are described in the following
sections. Correlation analysis is initially described, followed by regression analysis and the tests for statistical significance used. Spectrum and cospectrum analysis and the methods for
estimating significance are described along with the means for best interpretation of the results.
3.5.1 Correlation
Pearson’s product moment correlation coefiBcients are used to describe the strength and direction of the statistical relationship between two variables. The correlation coefiBcient is a standardized measure of the covariance between two variables and consequently is dimensionless and independent o f the units of measurement. The formula for Pearson’s product moment correlation coefiBcient for a known population, p, is
where C(X,Y) is the covariance between variables X and Y, and 0 % Gy are the population standard deviations for X and Y respectively (Burt and Barber, 1996). The formula for
Pearson’s product moment correlation coefficient (r) for a sample with n observations is
48 S Y
where X and Y are observations of variables X and Y; X and 7 are the means of X and Y;
Sx and Sy are the sample standard deviations (Burt and Barber, 1996). The numerator of the
correlation coeflBcient is the covariance between two variables X and Y, while the
denominator is the product of the sample standard deviations. Covariance is the average
product of two variable’s deviations from their means and can range in value from +«> to -«>.
Dividing the covariance by the product of the standard deviations of each variable limits its
range of values to +1.0 to -1.0.
The strength of the same relationship between Y and X is called the coefiBcient of
determination, or r^. It is exactly equal to the square of Pearson’s product moment correlation coefiBcient and is the ratio of the explained variance to the unexplained. Unlike the intercept and slope in regression analysis (section 3.5.2 below) which are asymmetrical, the r^ is symmetrical meaning that the amount of variance in Y accounted for by X is equal to the amount o f variance in X accounted for by Y.
The statistical significance of the correlation coefiBcient is tested under the null hypothesis that the population correlation coefiBcient, p, is equal to zero (Burt and Barber,
1996). The sample correlation coefiBcient, r, is divided by the estimated standard error s^.
[3.3]
49 where the estimated standard error is
The magnitude of the t-statistic is then compared to published t-test tables for appropriate
degrees of freedom and confidence levels.
3.5.2 Regression Analysis.
Ordinary least squares (OLS) regression analysis is a method o f statistical analysis in
which the causal linkage between two or more variables is prespecified. This is quite different
from non-causal statistical methods, such as correlation analysis, in which the results of the
analysis do not depend on the nature of the causal relationship. The causal model in
regression specifies that the variance of a dependent variable, Y, is caused by one or more
independent variables, Xi, Xj,..., X^. In the case of a only a single independent variable the
regression model for a known population would be:
7 = a + pJSr+e [3.5]
where Y is a linear function, p, of X, a constant, a, and an error term e (McClendon, 1994;
Burt and Barber, 1996). The error term is assumed to be normally distributed and constant for all the values of Y; these are the Gaussian assumptions under which the sum of the
50 squared errors are minimized (McClendon, 1994; Burt and Barber, 1996). For a sample population the OLS regression model is:
[3.6] Y=a+bX+ e
where the population estimators, a, P, and e are replaced by the sample estimators, a, b and e (McClendon, 1994; Burt and Barber, 1996).
The meaning of the parameters a and b is easily interpreted for variables measured on a ratio scale. The intercept, a, is the value that variable Y would have if variable were zero. Sometimes a meaningful physical interpretation of the intercept is difBcult. For example, if sea level pressure were regressed onto sea surface temperature
SST=a+bSLP+ e [3.7]
the intercept, a, is interpreted as the surface temperature that would occur if sea level pressure were zero. The slope, b, is the unit change in Y given a one unit change in X. Using the same example as above, this represents the change in K of surface temperature given a 1 hPa change in sea level pressure. The slope is a measure of the direction and magnitude of the causal relationship between Y and X. The units of measurement of the slope are units of Y per unit of X (K per hPa for example).
If the data are standardized before input to the OLS regression, the units of measurement of the slope are changed from units of Y per unit of X, to a dimensionless ratio.
51 The standardized slope, as b is now known, is interpreted quite dijfferentiy from the unstandardized slope. The standardized slope is interpreted as the standard deviation change in Y per one standard deviation change in X (McClendon, 1994). Data are standardized so that the dimensionless results can be compared between different levels o f the atmosphere, over different regions of the Earth and between different variables having quite different variances.
The Students t-test statistic is used to test the null hypothesis, Hq, that the population slope is not significantly different from zero. The alternative hypothesis, H^, states that the population slope is significantly different from zero, which is a two tailed hypothesis and is tested as such. It is assumed that the sample variances are unequal. The t-statistic for H<, =
0 is calculated assuming that the null hypothesis is true, and is simply the ratio of the slope to the estimated standard error of the slope
, _ i - ^ [3.8]
where b is the slope coefficient and Sy is the estimated standard error o f the slope. Sy is calculated as
S - ^ b - r ------~ [3.9] n n r2
;=1 1=1
52 In this case the t-statistic is also exactly equal to the square root of the F-statistic which is another commonly used test of statistical significance in regression analyses (McCelendon,
1994).
3.5.3 Univariate and Bivariate Spectrum Analysis and their relation to Harmonic
Analysis
Spectral analysis is used to detect recurrent non-random signals in data with no a priori assumptions made concerning the nature of the non-random signal (Mtchell et ai,
1966). Power spectrum analysis is based on harmonic analysis (Warner, 1998) which is described first, followed by descriptions of the univariate and bivariate power spectra and interpretation of the results.
Harmonic analysis assumes that a time series can be represented by a recurrent trend
(or period) in the time series (Warner, 1998). Time series data with a simple linear trend or period could therefore be written as
X = a+ bt [3.10]
where X, are the observations of X at time t, a and b are the intercept and slope coefiBcients in a conventional ordinary least squares (OLS) regression model (Warner, 1998). A time series of length n is represented in a harmonic analysis as a sinusoidal waveform fitted to the time series
53 = |i+i?COS(o)/+({))+6j [3-11]
where Xj are the observations o f X at time t; n is the mean of the time series; R is the amplitude of the waveform; (j> is the phase, or the location o f the spectral peaks relative to time zero; e^are residuals from the fitted curve; t is the observation number (0,1,2,...,N); w is in radians and equal to où = 2 îr/T r, where ris the period o f the cosine fimction being fitted
(Warner, 1998). However this assumes that the time series is composed of a finite number of oscillations each with a discrete wavelength. To overcome this limitation, spectral analysis is based on the assumption that time series are composed o f a nearly infinite number of small oscillations spanning a continuous distribution of wavelengths (Mitchell eial., 1966). The spectrum is a measure of the distribution of variance over the domain of all possible wavelengths in a time series. The wavelengths that are represented in the spectrum range from infinity to twice the interval between successive observations.
Another way of representing [3.11] in a more computationally efiBcient form is
X^ = \s.+Acos(o)t)+Bsm((ùt)+e^. [3.12]
By varying the magnitude of A and B terms it is possible to produce all possible sinusoids of period x, phase ({), mean p and amplitude R (Warner, 1998). For some period, t, the remaining parameters (p, A and B) can be calculated using OLS multiple regression methods
(Warner, 1998). Estimates of p, Â, Bare calculated using
54 [3-13] jV f=i
X)cos((ùt) [3.14] 2\l [=l
2 " B = —X) sin(o)0 [3.15] N f=i
Periodogram analysis is an extension of harmonic analysis and is a form of ANOVA
which partitions the variance in a time series of length N into the variance accounted for by
N/2 periods (Warner, 1998). The model used to represent a time series of length N can be
written as
= for i=\,2,2,...,N/2 [3.16]
where X, are the values of the X time series; t is the time or number of the observation; q ^ is the cosine function of frequency to.; S;, is the sine function of frequency to.. The set of frequencies is given (in radians) by 2ni/N; i = 1,2,3,. ,.,N the number of observations in the series (Warner, 1998). The periodogram parameters p., Â, B can be evaluated using
N f=i [3.17]
55 Â=^è(^r-^cos(G),0, for i = \,l,3,...,q [3.18] iV t=l
B, = ^T,(X^-X)sm(çù.t), for i=l,2,3,...,q [3.19] N t=i
^ = Y if N is even [3.20]
A separate estimate of each Âand .6 pair is made for each of the Fourier frequencies, 1/N,
2/N, 3/N,...q/N. The last observation q/N corresponds to a frequency of 0.5 or a period of
2 times the separation between two adjacent observations. The longest Fourier frequency has
the same length as the time series and the shortest has length of 2 observations, and the cycles
in between these extremes are equally spaced and are mathematically orthogonal so no
variance overlaps between periods (Warner, 1998). Analysis of longer time series is
performed using Fast Fourier Transforms which are more computationally efficient.
Periodogram analysis produces peaks with relatively large sampling errors. A power
spectrum is a smoothed version of the periodogram designed to eliminate or minimize the
sampling errors and produces spectral estimates which are smoother (Warner, 1998). Various
smoothing procedures are used to modify the periodogram, having variations in their window
size (the number of adjacent frequencies smoothed), and the magnitude of their weights
(Warner, 1998). Spectrum analysis does have the shortcoming of not representing the exact periodicities as clearly as in classic harmonic analysis.
56 The significance of the spectral estimates can be calculated using the sample lag-one
correlation coefiBcient (rj, assumed to be an unbiased estimator of Pi, the population lag-one
correlation coefiBcient. The following is evaluated for the frequencies between i=0 and i=N/2.
I-'-' 5,. = 5 [3.21] 1 2 « TtZ 1 +r, -2r, cos----- ^ ^ M2
In this spectrum s is the average of all N/2 smoothed spectral estimates . The chi square statistic (x^ divided by the degrees of fi'eedom is used to scale the null continuum to the sample spectral estimates computed (Mitchell et al, 1966). The degrees of freedom are calculated from
2N-m/2 v = . m [3.22]
where typically m = N/2 (Mitchell et al, 1966). The resulting values of x^/^ are multiplied by the null continuum calculated above and plotted against the smoothed spectral estimates to estimate the statistical significance of the estimates at each frequency. If none of the spectral estimates surpasses the threshold significance at any frequency then the estimated spectrum is not significantly different from the null continuum described (Mitchell et al.,
1966). The shape and magnitude of the null continuum described is dependent on the lag-one correlation coefficient as shown in Fig.3.5. Bivariate spectrum analysis can be performed to examine periodic elements in two individual spectra. The cross-spectrum is essentially a
57 lagged cross correlation function between two variables upon which a Fourier analysis is
performed and the resultant complex numbers are converted back into estimates of coherence and phase (Warner, 1996). Squared coherence, analogous to the r^ of a correlation analysis, describes the proportion of shared variance between the variables at a particular frequency and lies between 0.0 and 1.0. Phase represents the timing of the peaks in the Y time series compared to peaks in the X time series at a given frequency. The coherence at certain frequencies is a useful way of describing the correlation between the two time series over the whole range of frequencies. The coherence spectrum has to be understood with reference to the individual spectra of the variables so that meaning can be attached to large spectral power or large coherence at certain frequencies (Warner, 1998). Coherence between two series at any given frequency is not meaningful if there is no spectral power in either of the univariate spectra; conversely large spectral power in the univariate spectra is meaningless if there is low coherence between the series at that particular frequency.
Similarly, unless the coherence between the variables is relatively high, the interpretation of the phase relationship is meaningless. Phase is usually interpreted as meaning the lag (in radians) of phase between the two variables and is measured from + 7t to -it or 0.5 periods to - 0.5 periods. If the phase is zero then the peaks are synchronous, otherwise one variable lags and the other leads. If the phase is close to + 7t to -n the phase relationship is interpreted as being nearly out of phase at that frequency. A full period (or cycle) is given as
2n radians; to convert from radians to periodicities the phase data are multiplied by this constant. For a particular frequency known in terms of cycles, years in this study, it is possible to quantify the lag/lead of the variables. However in this study the interpretability of the
58 phase data is meaningless. The data length is at best from 1899 to 1989, a time series of 91 seasonal means, and it is impossible to say with confidence, for example, that X leads Y by
45 years, as only two cycles are possible in the data. The interpretability of phase data with higher frequencies becomes somewhat more reliable as increasing number of cycles are available in the data for evaluation, but phase data with lower frequencies are meaningless.
No mention of the phase relationship between variables will be made in this study due to this difiBculty.
59 8 =.60
6 =.60
Is.lO 0 = .C O
C J .2 FRCCUENCY
Figure 3.5 Shape of power spectra for time series having varying lag 1 autocorrelations, representing the degree of Markhov persistence (after Mitchell et oL^ 1966).
60 CHAPTER 4
CLIMATOLOGY OF SOUTH AND SOUTHEASTERN AFRICA
4.1 Introduction
The climatology of south and southeastern Africa is presented in this chapter including descriptions of the principal features of the climatological mean atmospheric circulation during DJF and of the principal determinants of the precipitation climatology of southeastern
Africa (SAP). Section 4.2 outlines the importance of the large scale Hadley circulation to the climates of these regions. The climatological situations of these regions during boreal winter
(DJF), including the circulation of the atmosphere, are presented in section 4.3.
4.2 A climatology of south and southeastern Africa.
Thompson (1975) illustrates the seasonality of African rainfall maxima (Fig.4.1) depicting a broad region over central and eastern Africa between the equator and 10° - 15°S with rainfall maximum between November and March. Much of southeastern Africa occupies an upland area with elevations typically between 1,000 and 1,500 m with several major river valleys cutting deeply into the mountains (Torrance, 1972). Its DJF climate is
61 Nov.-March ..Nov-r March
December
Insignificant Rainfall
July—September : A ug ust
iii^.M binly r.-ÿer>:insigni5 r f ic a n t M arch— ra in fa ll N ovem ber March Nov Aug.-Nov March-July a y -J u n e Sept.- Oct. M arch-June Feb.-Mar *r4 P ri I. fir Nov.- Dec.
O ct-1 X-.v/Nov- o '"
°'Dec- July
0 500 1 000 1500 Miles 00 Kilometres
to w
Figure 4.1 Distribution of average monthly rainfall with amounts greater than 100 mm (after Thompson, 1975).
62 strongly influenced by the large scale features of the thermally direct Hadley circulation,
primarily the latitude of the intertropical convergence zone (TTCZ).
The mean January SLP distribution and the resultant atmospheric circulation (Fig.4.2)
depicts strong northeasterly trade winds across most of North Africa converging into what
Thompson (1975) calls the North Equatorial Trough along the Gulf of Guinea and over
central Equatorial Africa. The high pressure center over the South Atlantic is responsible for
the southeasterly trade winds and southerly flow along the west coast of southern Africa
which also converges into the North Equatorial Trough. Northeasterly flow about the
western extreme of the high pressure system in the southern Indian Ocean advects moist air
masses from the Indian Ocean in the easterly flow across southern Afiica. Equatorial westerlies to the north, and the easterlies and southeasterlies to the south converge, into the
South Equatorial trough which is located between 10° and 20° S across southern Africa. The northeast monsoonal flow is observed along the coast of eastern Africa, from Eritrea and
Somalia, over Kenya, recurving to become northwesterlies to the south of the equator and converging into the South Equatorial trough.
Climatological streamlines were calculated from seasonally averaged NCEP/NCAR reanalysis data from 1958-1995. Inspection of the simplified climatological surface airflow
(Fig.4.2) and the calculated climatological streamlines at 1000 mb (Fig.4.3) indicates that there is no well defined location of the Southern Hemisphere ITCZ over eastern Afiica, in comparison to the well defined ITCZ over west-central Afiica in the Northern Hemisphere.
Thompson’s (1975) analysis separates the ITCZ over tropical Africa into the Northern and
Southern Equatorial Troughs (Fig.4.2). Convergent flow into the mean low over eastern
63 Trough y
High latiU jde__W jfg/^V ^^
Isobars at mean sea-level Run of isobars at 1,500 metres Intertropical front Predominant airflows between surface and 1,500 metres
500 1.000 1.500 Miles 800 1,600I------2,400 1 Kilometres , I
Figure 4.2 Simplified climatological mean January SLP distribution (after Thompson, 1975). Units are hPa. Arrows indicate the direction of the predominant airflow. Dotted lines represent intertropical fronts.
64 Angola (at about 20° S, 20° E in Fig.4.2) during DJF occurs at a location that would be over
the western end of the southern equatorial trough in Thompson’s scheme. It must be noted
that the calculated climatological streamlines (Fig.4.3) shows the northeastern monsoon being
more prevalent and the equatorial westerlies less prevalent than in Thompson’s scheme
(Fig.4.2).
The elevated equatorial highlands o f Kenya, Tanzania, Uganda and the lakes in this
region and further to the south, help break up the classic ITCZ configuration (GrifiBth, 1972
a). Torrance (1972; his Fig.2) conceptualizes the Southern ITCZ region slightly differently
than Thompson, replacing the equatorial trough with three air mass boundaries (Fig.4.4). The
Zairean or Congo Air Boundary (ZAB) is the leading edge of maritime air masses from the
Atlantic Ocean. The second boundary demarcates the southernmost limit o f the northeast
monsoon flow (SLM), and the third boundary delineates the northern limit of the south
easterly trades (NLT) blowing off"the Indian Ocean. The fiow-arrows in Torrance’s (1972)
diagram are in good agreement with those of Fig.4.3 based on the NCEP/NCAR reanalysis data from 1958-1995. The defining characteristics of these air masses are hard to distinguish over the rift valley region resulting in the occasionally discontinuous nature and indeterminate location of the ITCZ. The frontal boundaries vary frequently and can be quite inactive at times when wind directions are similar on both sides, and very active at other times when convergence occurs along the boundary. The ITCZ and its associated boundaries tend to return to a latitude of about 15°-17°S after a temporary displacement, but can move as far south as the Limpopo Valley (below 20°S) (Torrance, 1972).
65 60N SON
20N 20N
20S 320S
60S 60S
180W 140W lOOW 60W 20W 20E 60E lOOE 140E 180E
Figure 4.3 Winter (DJF) climatological mean 1000 hPa Streamlines calculated from NCEP/NCAR Reanalysis data, 1958-1995. f<
20
3C-S
20 X
Figure 4.4 Idealized convergence zones over SEA, predominant airflow is indicated by double arrows, idealized frontal boundaries are indicated by broken lines (after Torrance, 1972).
67 Zairean air masses associated with precipitation over equatorial eastern Afiica
(Thompson, 1975; Nicholson, 1996) are comprised of moist Atlantic air adverted fi-om the
west and northwest over the Congo and Zaire, cooling adiabatically due to uplift and
becoming nearly saturated as a result (Torrance, 1972). Monsoonal flow fi’om the north and
northeast is increasingly altered as it moves further south (its leading edge being the SLM).
Its characteristics largely depend on whether it had a trajectory over the dry Afiican continent
or over the northwestern Indian Ocean. The monsoonal flow generally has a steep lapse rate
favoring convective activity (Torrance, 1972). The southeast trades and their leading edge,
the NLT, are usually moist throughout a limited depth and are relatively dry in comparison
to the other air masses (Torrance, 1972). The strong negative relation between the NAG and
southeastern Afiican precipitation may be linked to latitudinal variations in the location of the
three convergence zones comprising the south equatorial trough.
A small climatological heat low over eastern Angola and in the vicinity o f its borders
with Zambia, Botswana and Namibia is caused primarily by a local positive net radiation
maximum during DJF and is the focus for the convergent winds flowing clockwise into the low as noted here on the 1000-hPa streamlines (Fig.4.3) and as noted in the text by Preston-
Whyte and Tyson (1988; their Fig. 10.15). The ZAB is typically found oriented meridionally between Lake Malawi and the west coast (Torrance, 1972; Preston-Whyte and Tyson, 1988) with its location fluctuating on a daily basis. The ZAB is evident fi-om a persistent trough separating the moist air over Zambia from the drier easterlies over Zimbabwe and is associated with fi-ontal activity and precipitation (Torrance, 1972). This is a region favorable
68 Above 1,600 600 — 1 600 400 — 80 0
- Y/\ 200 — 400 t-'.-'A 100 — 200 Below 100
1,500 Miles 1,600 2.400 Kilometres 10 w
Figure 4.5 January precipitation receipt over Africa; units are mm (after Torrance, 1972).
69 for cyclogenesis, with lows forming that tend to be deeper than typical heat lows which form elsewhere over tropical Afiica (Preston-Whyte and Tyson, 1988).
Average January precipitation (Fig.4.5) over SEA range fi"om in excess of 300 mm over southern Tanzania to less than 100 mm along the equator in eastern Afiica. Maximum precipitation during DJF is located between 10° - 15° S. Precipitation receipts have a single maximum during DJF (Fig.4.6) with the exception of the stations of eastern Malawi when a smaller secondary maximum is evident in March due to the increased flow of moist air masses fi-om the southeast over the elevated coastal plateau and the resultant orographic precipitation
(Torrance, 1972). Northward moving pools of cold air often move along the elevated eastern escarpment of the South Afiican coastline, with the gap in the mountains along the coast of southern Mozambique providing a route for the cool air to advance inland along the major river valleys (Torrance, 1972). These migratory anticyclones result in clouds and precipitation along the eastern border of Zimbabwe, southern Mozambique and over the highland areas of Malawi. Cold air fi-om the southeast undercutting moist Zaire air is associated with intense precipitation receipt over southern Zimbabwe. The low pressure center present over Angola and Botswana is often characterized by a trough extending toward low pressure systems off the south coast of Afiica which subsequently moves inland bringing cool southerly air and frontal activity (Torrance, 1972). It has been shown that these tropical- extratropical troughs, which connect tropical systems over central South Africa, and further inland, to mid-latitude depressions to the south of the continent are the principal determinants of rainfall anomalies in the summer (DJF) rainfall regime of South Afiica (Harrison, 1984).
70 M w inilunço
• Kas«fnaa
Miant»
f Wntoi - ÎCH-
Figure 4.6 Seasonality of precipitation receipts over SEA; units are mm. The year is from July to June to avoid splitting the rainy season. Zimbabwe is to the south, Zambia to the northwest, and Malawi to the east, see Fig.2.7 for country (after Torrance, 1972).
71 The precipitation climatology of southern and eastern Africa is largely determined by
the passage o f the ITCZ and localized convergence zones (e.g. the ZAB). Multiple rainfall
maxima are related to the repeated passage of the ITCZ over regions close to the equator and
produce the long and short rainy seasons over Kenya, northern Tanzania, Uganda, Zaire and
also over West Africa. The timing of the rainfall maximum is related to the location of the
region in relation to the position of the seasonal mean ITCZ. The seasonality o f precipitation
receipt is shown in Fig.4.6 over Zimbabwe, Zambia and Malawi. The distribution of meteorological stations over Southern Africa must be considered when examining these distributions. The Kalahari Desert has relatively few stations (Fig.3.2) and the absence of DJF precipitation maxima over the Kalahari cannot be attributed solely to the timing or the magnitude of the precipitation maxima due to the dearth of meteorological stations with long term precipitation data. Precipitation maxima observed over SEA during the Austral summer
(DJF) (Fig.4.6) correspond well with calculated seasonal average precipitation maxima over
Africa during DJF (Fig.4.7). Meteorological stations with greatest seasonal precipitation receipt during DJF (Fig.4.7) closely associated with the latitude of the ITCZ lie to the south of the equator over Eastern Africa and exclude the regions to the west including much of
Northern Namibia, Angola and southeastern Zaire. Further analysis of the seasonality of precipitation over southern Africa (Fig.4.8) shows the contribution of the DJF season to total annual rainfall receipt is greatest over Eastern Africa and increases away from the equator over much of subtropical southeastern Africa with the exception of mid-latitude southern
Africa. Stations where DJF rainfall is less than 50% and greater than 25% o f the average annual rainfall receipts are depicted in Fig.4.9. Regions along coastal and near-equatorial
72 20S
,2QS
ZCZ
73 20X
zas
60S
ICO: uoz tacs
Figure 4.8 Locations of stations having DJF precipitation greater than 50% of the total annual rainfall.
74 scs
2C.V
icoir 2 0 S SCS
Figure 4.9 Locations of stations where DJF precipitation > 25% and < 50% of the average annual precipitation.
75 southern Africa, and South Africa are shown to have large and important rainfall receipts during DJF.
A substantial number of tropical cyclones that form over the Indian Ocean enter the
Mozambique channel from the north, contributing to precipitation receipt inland over eastern
Africa (Torrance, 1972; Preston-Whyte and Tyson, 1988). The cyclone season over the southern Indian Ocean is between December and March (Torrance, 1972). Moist equatorial
Zaire air masses strongly associated with increased precipitation are advected by the tropical cyclones from the north and west into SEA producing rainfall over eastern Zambia and the southern highlands of Malawi (Torrance, 1972). The circulation associated with tropical cyclones is clockwise about the center, and is associated with increased advection of Zaire air from the northwest, and maritime air masses from the southeast (Torrance, 1972).
The climates of southern Africa are largely controlled by the seasonal variations in the large scale features of the Hadley circulation, principally the subtropical high pressure centers flanking the region located over the South Atlantic and South Indian Oceans. The mean flow in DJF is consequently southwesterly to the west and northeasterly to the east of South Aflica around the eastern and western margins of the subtropical highs over the South Atlantic and
Indian Ocean respectively (Preston-Whyte and Tyson, 1988). Flow about the subtropical highs intensify during summer and this produces enhanced advection of relatively cold water into the Benguela and warm water into the Agulhas Currents. Consequently the SSTs observed in the Benguela Current are lowest along the Afiican coast during summer when the
South Atlantic high is strongest (Preston-Whyte and Tyson, 1988). However synoptic scale variability over South Aflica is largely independent of the Hadley Circulation and is driven by
76 seasonally varying synoptic and smaller scale features of the atmospheric circulation (Preston-
Whyte and Tyson, 1988).
Synoptic scale weather producing features over South Afiica originate in disturbances in the tropical, subtropical and temperate features of the atmosphere (Preston-Whyte and
Tyson, 1988). During winter, tropical and subtropical control over the weather producing atmospheric disturbances is at a maximum and the importance of temperate controls to South
African weather is at a minimum (Preston-Whyte and Tyson, 1988). Tropical controls on
South African weather largely occur through wave disturbances and lows propagating through the tropical easterlies, while temperate control is exerted through wavelike disturbances and lows propagating through the mid-latitude westerlies (Preston-Whyte and
Tyson, 1988). The principal disturbances associated with summer (DJF) precipitation over
South Africa include easterly waves and easterly lows (Preston-Whyte and Tyson, 1988). A strong annual cycle is evident in the frequency of occurrence of both of these features with maxima occurring during the summer and rarely occur during winter when tropical control is at minimum (Preston-Whyte and Tyson, 1988). The atmospheric dynamics associated with easterly waves are described in more detail in section 4.3. The surface trough of an easterly wave is often associated with a well defined boundary between dry air to the southwest and moist air to the northeast (Preston-Whyte and Tyson, 1988). Synoptic and small scale disturbances such as thunderstorms and squall lines often form ahead of the moisture discontinuity (Preston-Whyte and Tyson, 1988). Much of the precipitation over the summer rainfall area of northeastern South Africa occurs in the form of convective activity. The distributions of thunder days and hail days (Fig.4.10) are concentrated over this region due
77 Thunderdays Hail days
0. L
Figure 4.10 Generalized distribution of thunder days and hail days in South Africa (after Preston-Whyte and Tyson, 1988; based on Schulze, 1972).
78 to such factors as the diurnal cycle of surface heating and the consequent atmospheric instability (Preston-Whyte and Tyson, 1988). The frequency of hail days is at maximum over this region during late spring when lapse rates are steep and temperatures relatively high
(Preston-Whyte and Tyson, 1988). Rainfall intensities are greatest over the IDghveld region of northeastern South Africa and the eastern escarpment where synoptic-scale dynamics and local factors, including orography and intense surface heating, combine to produce vigorous convective activity (Preston-Whyte and Tyson, 1988).
79 4.3 The Mean Airflow over Africa.
The climatological streamlines at 1000-hPa over southern Africa (Fig.4.3) converge
into the weak climatological low pressure center over eastern Angola, an area of convergence
that has no significant corresponding feature in the SLP climatology calculated from
NCEP/NCAR reanalysis data from 1958-1995 (Fig.4.11). Monsoonal flow from the western
Indian Ocean occurs over eastern Africa to the south of the equator (Fig.4.3), resulting in
streamlines vdth a substantial easteriy component to the flow. West and south of Madagascar
the flow is from the southeast and the trades supply warm moist air to SEA contributing
somewhat to the occurrence of the precipitation maxima during DJF over Mozambique and
the upland regions o f southern Malawi (Torrance, 1972). Increased moisture supply is not
the principal determinant of precipitation receipt over SEA, but is largely determined by the
convergence of air masses, by vigorous disturbances resulting in convergence and ascent within the moist equatorial westerlies, by rainfall associated with easterly waves and by
convective uplift. The southerly outflow from the South Atlantic high pressure system is parallel to the coastline of western Afiica and converges strongly into the climatological low over eastern Angola and Botswana between 10° - 20° S.
Climatological streamlines at the 850-hPa level, as calculated from the NCEP/NCAR reanalysis data (Fig.4.12), are dominated by the convergent flow into the climatological low pressure center over Botswana, Namibia and Angola. Convergent streamlines also occur over
Lake Victoria and over north-central South Afiica brought about by westerly flow over western Afiica and easterly flow over eastern Afiica just south of the equator. Convergent streamlines are observed at this level immediately to the north of the equator over eastern
80 SON SON
20N —” — 20N
20S — —c — 20S
60S Î 60S
laoff 140W lOOW sow 20W 20E SOE lOOE 140E 180E
Figure 4.11 SLP Climatology calculated from NCEP/NCAR Reanalysis data, ' 1958-1995. Units are hPa, isopleth intervals occur every 4 hPa.
81 60N 60N
20N 2QN
20S 20S
60S 60S
180W 140W lOOW 60W 20W 20E 60E lOOE 140E 180E
Figure 4.12 Winter (DJF) climatological mean 850 hPa Streamlines calculated from NCEP/NCAR Reanalysis data, 1958-1995. Zaire, Uganda and Kenya. Outflow from the South Atlantic high results in southerly flow along the coast but with a stronger easterly component to the flow than is observed at 1000- hPa (Fig.4.3); this outflow is also observed to contribute to the strong almost due westerly flow immediately south of South Afiica.
Southern Afiica is dominated by the high pressure system at 300 hPa (Fig.4.13), with the outflow associated with strong westerly flow over the continent to the south of about
20°S. The anticyclone located over Ethiopia, Kenya and Somalia interacts with the high over southern Afiica to produce strong easterlies between 5°N and IO°S. Both of these upper level systems are strongly zonal in orientation in keeping with the thermal gradient of the atmosphere.
The Jetstream can be observed more clearly in the mean 300-hPa u component of the wind (Fig.4.14). The climatological location of the Jetstream during DJF is located at about
40°S, to the southwest of Afiica. The equatorial region is dominated by easterly flow of relatively small magnitude between about 5°N and 15°S, with the flow becoming strongly westerly to the south and north. The intensification of the westerly component of the flow is particularly abrupt to the north of the equator where the isopleths are tightly packed. The core of the northern subtropical Jetstream is located over Egypt but the whole of North Afiica is dominated by strong westerly flow at this level. The isopleths are predominantly meridional in orientation although they tend to curve to the south over the relatively cold Atlantic Ocean.
The impact of dynamically induced transient mid-latitude weather systems associated with the subtropical jet on the near equatorial climates of Afiica is limited during DJF as the subtropical Jetstream is for to the south, with the relatively weak core of the Jet found between
83 60Nk$ 60N
20N 20N
20S 20S
60S 60S
laow 140W lOOW 60W 20W 20E 60E lOOE 140E 180E
Figure 4.13 300 hPa Climatological streamlines from the NCEP/NCAR reanalysis data, 1958-1995.
84 60N 60N
SON SON
SOS SOS
■eei 60S 60S
140W lOOW SOE60E lOOE 140E 180E
Figure 4.14 Climatological mean 300-hPa u-wind component from the NCEP/NCAR reanalysis data, 1958-1995. Units are m s*^; isopleth intervals are 5 m s '; negative isolines are broken.
85 300- and 150-hPa (Preston-Whyte and Tyson, 1988). The jet is strongly associated with mid
latitude transient weather systems that have a large impact on the surface climates of Southern
Africa's mid-latitudes. Dynamically induced tropical transient systems are important determinants of weather and climate in the central African rift valley where easterly waves propagating through the tropical easterlies, and disturbances in the equatorial westerlies result in much rainfall over this region (Torrance, 1972). Tropical easterly waves preferentially form in the regions o f convergence between Zaire air, monsoonal air and southeast trade winds (Preston-Whyte and Tyson, 1988). Convergence, large horizontal shear and differing moisture characteristics combine with relatively unstable westerly flow to produce regions where easterly waves develop (Preston-Whyte and Tyson, 1988). Typically easterly waves are semi-stationary in nature and form within deep easterly flow near an easterly Jetstream
(Preston-Whyte and Tyson, 1988). Easterly waves are primarily tropical in origin and thus are most likely to influence weather in mid-latitude southern Africa during DJF (Preston-
Whyte and Tyson, 1988). At any level of the atmosphere in the Southern Hemisphere, flow within an easterly wave (Fig.4.15) having a large northerly component is associated with convergence and flow having a large southerly component is associated with divergence
(Hastenrath, 1985; Preston-Whyte and Tyson, 1988). Near surface easterly waves in the
Southern Hemisphere produce convergence (large northerly component to the flow) and uplift to the east of the surface low and divergence (large southerly component to the flow) and subsidence to the west of the surface low (Hastenrath, 1985; Preston-Whyte and Tyson,
1988). This balance of forces in the easterly propagating wave produces uplift upstream of the low resulting in cloudiness and precipitation, with descending air downstream associated
86 North
Equator Equatorial Low Î Supergeostrophic Superg.eostrophic C.Fi J».G.F. C.Fi JP.G.F.
,996 hPa 00 996 hPa 1000 1000 ■hPa hPa C.R
Subgeostrophic
Figure 4.15 Idealized balance of forces in a frictionless easterly wave in the Southern Hemisphere. C.F. represents the Centrifugal force; P.G.F. the pressure gradient force; f the Coriolis Force; CONV represents convergence; DIV divergence; H represents high pressure areas. The length of the arrows is proportional to the relative magnitude of the forces. with clear skies and fair weather. 300 hPa upper level easterly waves produce divergence ahead of the upper level trough and convergence behind it.
Precipitation over equatorial Africa is very strongly associated with westerly flow
(Thompson; 1975; Nicholson 1996) since vertical motion tends to develop strongly if the westerlies are sufBciently deep. This is largely due to the effect of the vertical component of the Coriolis parameter (Q which is potentially of sufiBcient magnitude to induce instability in a neutral or weakly stable air in the near-equatorial atmosphere. The magnitude of the
Coriolis force, f acting along the x-axis can be written as the sum of two components acting perpendicularly to each other (Fig.4.16),
f=fy^fz [4.1]
with ^ the Coriolis parameter acting along the y-axis and 4 the component acting along the z-axis. The Coriolis force acts radially outwards from the axis o f rotation on a parcel of air moving eastward (u>0), and radially inwards from the axis o f rotation on a parcel moving westward (u<0) (Wallace and Hobbs, 1977). The Coriolis force acting on a parcel of air with a horizontal velocity, u, can be decomposed into two components,
/=2QMsin(|)+2Qmcos(|) [4 2]
where 2Q u sin (() is the horizontal component, and the vertical component, ^ has a magnitude oflQ u cos 88 90“N 00 VO Equator Figure 4.16 Horizontal (f^) and vertical (fjcomponents of the Coriolis Force (f) at two different latitudes ((|)j and ()>2, where ({>] < (I);). Arrows are proportional to the relative magnitude of the forces. on the cosine of the latitude results in an increase of the magnitude of this component towards the equator and dépendance of ^ on the sine of latitude results in a decrease of the magnitude of this component towards the equator. It is apparent from [4.2] that westerly (easterly) flow produces an upward (downward) directed acceleration. After scale analysis, 4 is typically assumed to be negligible in comparison to the other forces acting upon the vertical motion of an air parcel and is assumed to be zero (Wallace and Hobbs, 1977). However the f^ component may induce instability in a neutral atmosphere. The strong association between westerly equatorial flow and precipitation over SEA (Thompson, 1975; Nicholson, 1996) suggests a possibility that this process helps to establish the strong relationship between near equatorial westerlies and rainfall receipt. 90 CHAPTERS RESULTS 5.1 The NAO and its relation to Precipitation over Tropical Africa. Large and highly significant correlations are found between the NAOI and precipitation observed at meteorological stations over the globe (Fig.5.1) during DJF. The area with the largest and most coherent correlations between Afiican precipitation and the NAOI occur in Tanzania, Zambia, southern Kenya, Malawi, Mozambique, Uganda, Zimbabwe and Rwanda, bounded by the equator and 20°S, 20°E to 40“E. Comparison of Figs. 4.1 and 5.1 shows that the annual precipitation maximum is typically unimodal and occurs between December to Febniary over this area. Other parts of equatorial Afiica either have precipitation maxima at other times of the year, are not correlated to the NAO, do not have an adequate precipitation data record to evaluate the relationship, or the DJF period is a dry or transitory period. A large cluster of significant negative correlations in Morocco substantiate Lamb and Peppier’s (1987) finding of a strong negative NAO signal in precipitation in that region. This negative NAO-precipitation relationship is strongest for the northernmost and coastal stations in Morocco and is related to the location and strength of the subtropical high during winter 91 60N 60N 20N 20N wVO 20Ê 203 60S 60S 180W MOW lOOW (low 20W 20E 60E lOOE MOE 160E Figure 5.1 Significant correlations between the NAOI and global precipitation during DJF; + and - refer to the sign of the relationship; large signs represent stations significant at the 99% confidence level, small signs refer to stations significant at the 95% confidence level. Stations not significantly correlated are not shown. (Lamb and Peppier, 1987). During the negative NAO phase the westerlies are stronger across the Mediterranean Basin than during the positive phase and the mean storm track is oriented over the Mediterranean Basin. The positive correlations to the NAOI noted in Fig.5.1 across northern Europe, and the negative correlations across Southern Europe, have been discussed by Hurrell (1995) and Hurrell and Van Loon (1997) and can be attributed to differences in the orientation o f the seasonal mean storm track during differing NAO phases (Rogers, 1997) and also appear in precipitable water differences between NAO phases (Hurrell, 1995). A large cluster of African stations exhibiting large, negative and highly significant precipitation correlations are observed in a region roughly bordering the latitude/longitude boundaries of 0“ - 20°S, 20“ - 40“E and includes portions of Uganda, Rwanda, Burundi, Kenya, Tanzania, Mozambique, Zambia, Zaire, and Malawi (Fig.2.7). This region includes the central Afiican rift valley. Lake Victoria, Lake Malawi (formerly known as Lake Nyasa) and Lake Tanganyika. Table 5.1 lists the meteorological stations from this region of Afiica having significant negative correlations with the NAO. The cluster of stations in this region exhibit significant negative correlations between local precipitation and the NAOI, consequently this region will be referred to as southeast Afiica (SEA). 93 GHCN Precipitation Lat Lon r(NAOI, sig % r(Nino4, sig % Station Name P r e c in l P r e c in l NAIVASHA -0.40 36.30 -0.43 95 0.42 95 NAROK -1.13 35.83 -0.28 95 MACHAKOS -1.50 37.20 -0.24 95 NGARA -2.40 30.60 -0.54 99 MUSOMA -1.50 33.80 -0.34 95 TARIME -1.40 34.40 -0.38 95 USHIROMBO MISSION -3.50 32.00 -0.61 95 NGUDU -2.90 33.30 -0.37 95 DONGOBESH MISSION -4.10 35.40 -0.42 95 0.34 95 MONDULI -3.30 36.50 -0.51 95 0.42 95 KJBONDO MISSION -3.60 30.70 -0.42 95 TABORA AIRPORT -5.08 32.83 -0.23 95 MAZUMBAI ESTATE -4.80 38.50 -0.42 95 AMBANGULU ESTATE -5.10 38.40 -0.43 95 NGOMENI -5.20 38.90 -0.58 99 MANYONI, D O. -5.70 34.80 -0.29 95 MPWAPWA VET. OFF. -6.30 36.50 -0.32 95 SINGE)A, D O. -4.80 34.80 -0.27 95 KELOSA AGRIC. OFF. -6.80 37.00 -0.28 95 0.31 95 MAHENGE -8.60 36.70 -0.31 95 KABALE -1.25 29.98 -0.37 95 0.42 99 TSHIBINDA -2.30 28.70 -0.61 95 ANKORO -6.70 26.90 -0.49 95 RULINDO -1.70 29.90 -0.50 95 QUELIMANE -17.88 36.88 -0.28 95 Sm W A NGANDU -11.10 31.70 -0.31 95 KA.SEMPA -13.53 25.85 -0.27 95 MPONGWE MISSION -13.50 28.20 -0.29 95 KABWE -14.45 28.47 -0.27 95 MCHINEBOMA -13.80 32.90 -0.27 95 MOUNT DARWIN -16.78 31.58 -0.21 95 Table 5.1 Locations and names of GHCN Precipitation Stations significantly correlated to NAOI over SEA, with the magnitude of the correlation, and the statistical significance at the 95% or 99% confidence levels. Lat and Ion are the latitude and longitude of the stations. Also shown are the magnitudes of the correlations to Nino4 where a significant correlation exists, sig% represents the statistical significance in %. 94 5.2 The Southeast Africa Rainfall Index (SEAR). This section describes the mechanisms resulting in the negative relationship between precipitation and the NAO over SEA. To illustrate the nature o f the precipitation anomalies associated with the NAO, the regional precipitation index is regressed onto several variables from the NCEP Reanalysis for the 32 winters between 1958 to 1989. GHCN precipitation data are available up to and including 1990, limiting the length of the regional precipitation index to the 32 winters. The standardized beta coefiBcients from these analyses are obtained for each grid point and represent the change, measured in standard deviations, of the variable in question given a change of one standard deviation in the SEA precipitation index. Standardized beta coefficients are plotted because the variance of any one variable differs markedly across the globe, between high and low latitudes, between different levels of the atmosphere, and from land to sea. The standardized coefficients provide a basis for signal comparison between these very different regions and levels. The response of any one variable can be compared to any other as the coefficients have units of measurement that are equal and provide an accurate basis for comparison of the signal strength. An advantage of the regression analysis is that all 32 winters of the overlapping time period between the precipitation and reanalysis datasets are used in the analysis, as opposed to only using the most positive and negative extremes from that period in a typical composite analysis. An additional advantage of regression analysis is that only the relationship between X and Y is estimated, all other sources of precipitation variability are statistically held constant (McClendon, 1994). The association of the NAO index to SEA regional rainfall is described ffirst. 95 A regional index of precipitation over SEA was created to evaluate potential links to the NAO and the atmospheric circulation. In creating this index, precipitation data for 31 regional meteorological stations having significant correlations to the NAOI at the 95% confidence level or greater (Table 5.1), are normalized by dividing the DJF departures from their long term mean by the DJF standard deviation for each station as per Jones and Hulme (1995). The 31 time series of standardized anomalies are averaged over each winter producing a precipitation index for SEA which is then standardized, and referred to as SEAR (Fig.5.2) subsequently used to characterize the regional DJF rainfall. The locations of the stations contributing to SEAR are plotted in Fig.5.3. Due to differing record lengths among the stations and missing data, the number of stations contributing to each temporal element in the index varies. The reliability of the index increases as more stations contribute and its variance subsequently decreases. Large variability in the first 10 years occurs because only 2 stations were contributing to the index. The long-term (1895-1989) correlation between the precipitation index and the NAOI is -0.48 (Table 5.2), statistically significant at the 99% confidence level. The short term (1958-1989) correlation over the time period of the reanalysis data is -0.70 (Table 5.2), statistically significant at the 99.9% confidence level. In order to further evaluate the negative correlations, the covariance between the NAOI and SEAR index is plotted over the period of record (Fig.5.4). The large negative values throughout the length of the record are the outstanding feature o f the covariance. With the exception of the relatively large positive contribution in 1899, the period from 1895 to 1923 is dominated by extremely large negative covariances resulting in a correlation of - 96 3 r - 1 -2 - 3 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 Year Figure 5.2 Standardized southeast African Rainfall Index (SEAR), 1899-1989. 97 20N 20N EQ EQ 20S 20S 40S 40S OE 20E 40E 60E Figure 5.3 Locations of some African GHCN precipitation stations significantly correlated to the NAOL Triangles represent stations contributing to SAFR; circles represent stations contributing to SEAR. 98 3 2 1 0 - 1 -2 - 3 — 4 L _ i ------1890 1900 1910 920 1930 1940 1950 1960 1970 1980 1990 Year Figure 5.4 Covariance between the NAOI and SEAR, 1899-1989. 99 ■ 0.29 (Table 5.2). Covariances then decrease to relatively small negative values during the rather lengthy period 1924 to 1960, with the exception of 1936 and 1949 which had exceptionally large negative covariance. This produces a correlation of -0.42 for the entire period 1924-1960 and of -0.35 excluding those two specific years. The correlations are surprisingly large for a period with small covariances but it must be noted that during this 38 year period only 1953 has a positive covariance with magnitude greater than 0.5. The period with the largest individual contributions to the negative correlation between the NAOI and SEAR starts around 1958. From 1958 to 1989, the period for which NCEP/NCAR reanalysis data are available, the covariances are extremely negative resulting in a correlation of -0.70. There appear to be three individual periods which characterize the NAOI-precipitation relationship, fi-om the start of the record to 1923, fi"om 1924 to the late 1950s and from the late 1950s to 1989. Several studies suggest that a shift in the climate system, related to interdecadal variability occurred during the 1920s, and again in the late 1950s and early 1960s (e.g. Deser and Blackmon, 1993; Kushnir, 1994; Kruger, 1999). The period since the 1950s is one of prolonged droughts in the Sub-Sahelian region of northern Africa (Nicholson, 1993; 1996). The drought in the Sahel and surrounding regions of Afiica is related to a prolonged precipitation decrease during boreal summer and appears to have started in the early 1960s and continued to the 1990s, with a wet period between the mid-1970s and early-1980s (Nicholson, 1993). Kruger (1999) shows interdecadal modulation of the precipitation-ENSO relationship over South Afiica and describes alternating wet and dry climate epochs similar to those described above, including a change in rainfall receipt that occurred about 1959/60 which appears related to change in the climate state over Southern Afiica. 100 Years Correlation Coefficient 1895 - 1989 -0.48 1895 -1923 -0.29 1924 - 1962 -0.42 1958 - 1989 -0.70 Table 5.2 Temporal dependence of the correlation coefficients between the NAOI and SEAR. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined. 101 5.3 Southern African Rainfall Index (SAFR) The correlation between the NAOI and precipitation over South Africa, Swaziland and Lesotho (Fig.5.1), referred to as SAF, will also be examined in relation to atmospheric circulation anomalies. Stations with significant correlations over Namibia and southern Botswana (Fig.5.1) close to the west coast o f southern Afiica are not included because their inclusion would produce an index that is not regional in nature. A regional precipitation index is created to characterize summer rainfall receipt among Southern Afiican stations significantly correlated to the NAO at least at the 95% confidence level. Standardized rainfall data for 11 meteorological stations in SAF (Fig.5.3), significantly correlated to the NAOI at the 95% confidence level or greater (Table 5.3), are averaged to produce a South Afiican regional rainfall index (SAFR) (Fig.5.5). The index shows relatively large summer (DJF) rainfall receipt from 1893 to 1920, a period with below average rainfall occurs between 1927 and 1966. SAF rainfall from 1967 to the end of the record in 1989 is characterized by some extremely large departures of both signs. The covariance between SAFR and the NAOI (Fig.5.6) is largely positive throughout the length of record consistent with the positive correlation between the indices (r=G.43, significant at the 99% level). The covariance Is dominated by large positive covariance from 1961 to 1989 with the exception of the extremely large negative covariance in 1983, the correlation for this period is 0.54 but the removal of 1983 increases the correlation to 0.66. The strength of the relationship between the indices is therefore relatively consistent over time, but an increase in the magnitude of the relationship is observed since 1960. 102 GHCN Precipitation Lat Lon r(NAOI, sig r(Nino4, sig - ..... Station Name Precinl % Precinl % KLERSDORP -26.90 26.60 0.28 95 PEETRETIEF -27.03 30.80 0.44 99 BOSHOF -28.50 25.20 0.30 95 -0.36 95 FICKSBEURG-TNK/MUN -28.90 27.90 0.36 99 -0.37 99 LADYSMITH -28.57 29.77 0.34 95 UMTATA -31.53 28.67 0.30 95 PORT ST JOHNS -31.63 29.55 0.30 95 MANZINI/MATSAPA AIRPO -26.53 31.30 0.24 95 MPISI-1 -26.40 31.60 0.28 95 -0.27 95 NHLANGANO -27.10 31.20 0.30 95 BUTHABUTHE -28.80 28.20 0.32 99 -0.27 95 Table 5.3 Locations and names of GHCN Precipitation Stations significantly correlated to NAOI over SAFR, with the magnitude of the correlation, and the statistical significance at the 95% or 99% confidence levels. Lat and Ion are the latitude and longitude of the stations. Also shown are the magnitudes of the correlations to Nino4 where a significant correlation exists, sig% represents the statistical significance in %. 103 4 r 3 2 1 0 -1 -2 - 3 - 4 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 Year Figure 5.5 The standardized South African regional rainfall index (SAFR), 1899-1989. 104 - 1 -2 - 3 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 Year Figure 5.6 Covariance between SAFR and the NAOI, 1899-1989. 105 5.4 Quasi-Periodic behavior of SEAR, the NAOI, SAFR and the Nino4 SST index. This section describes the quasi-periodic components of the covariability of the various seasonally averaged precipitation and circulation indices over the 91 year period 1899-1989. Cross-spectral analysis (Mitchell et al, 1966; Warner, 1998) is used to describe dominant quasi-periodicities in the covariability of SEAR, SAFR, the NAOI and the Nino4 SST index. The shared periodicities are then compared to prominent periodicities evident in climate data for southern Afiica. The interannual variability of precipitation over southern and eastern Afiica, fi"om the equatorial regions of Kenya, Zaire and Uganda through Tanzania to mid-latitude South Africa, is strongly related to the ENSO phenomenon (Ogallo, 1987; Nicholson, 1996) and is shown to be significantly correlated to the NAO in this study. The intensity and phase of El Nino events are represented by the Nino4 SST index (Fig.3.4) from the central equatorial Pacific Ocean over the area 5°N-5°S, 160°E-150°W (Trenberth, 1996). The spectrum of the NAOI is presented in Fig.2.6 and described in section 2.2. The spectrum of Nino4, SEAR and SAFR indices are described prior to presenting the results from the cross-spectral analysis. Spectral analysis of SEAR (Fig.5.7) shows a spectrum dominated by the peak at 7.6 years and smaller peaks with relatively large spectral variances occurring at about 5.5 years, and about 3.25 years. However only the spectral variances at 7.6 years and 3.25 years are statistically significant at the 95% confidence level. These spectral peaks are comparable to the spectral periods known to dominate rainfall over southern and eastern Africa (Tyson, 1986; Nicholson and Entekhabi, 1986, 1987; Mason and Tyson, 1992). The large peak at 7.6 106 0.8 0.7 0.5 i_ !U ? 0.5 O CL 0.4 0.3 0.2 0.1 0.0 1 8 .2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.7 The Spectrum of SEAR , 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis. 107 years is also encompassed by the largest peak in the spectrum of the NAOI (Rogers, 1984; Hurrell and Van Loon, 1997) in Fig.2.6. The spectrum of SAFR is examined and compared among the many quasi-periodic components known to exist (Nicholson and Entekhabi, 1986; Tyson, 1986; Nicholson and Entekhabi, 1987; Mason and Tyson, 1992), but no relationship to the NAO has been described in the literature (see section 2.2). The spectrum of SAFR (Fig.5.8) is dominated by spectral power at periodicities less than 3.5 years including statistically significant spectral peaks between 2.67-2.75 years and at the high fi-equencies at about 2 years. The spectral power decreases and becomes relatively small in the intermediate periodicities and very weak at longer periodicities. The spectrum of the Nino4 SST index (Fig. 5.9) has large and statistically significant spectral peaks at the 95% confidence level at 5.7 years, 2.75 years, and at 2.3 years. The Southern Oscillation Index (S.O.I.; Chen, 1982; Trenberth, 1996) has been used in many studies to represent the phase and intensity of ENSO and is known to have a comparable spectmm to theNino4 index, with dominant powers at 5.3 to 6.6 years, 3.8 years, 2.8 years and between 2.3 to 2.4 years (Nicholson and Entekhabi, 1986). These periodicities are similar to those identified in Afiican rainfall south of the equator (Nicholson and Entekhabi, 1986) and are examined below in the cospectrum. Rogers (1984) performed spectral analysis on normalized winter Darwin SLP anomalies and finds significant spectral power only occurring at 5.7 years. The cospectrum between the Nino4 index and SEAR (Fig. 5.10) shows the largest positive contribution to the correlation between the precipitation and the Nino4 indices 108 0.6 0.5 0.4 ID 5 o CL 0.3 o QJ CL 0.2 (/) 0.1 0.0 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.8 The Spectrum of SAFR , 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis. 109 0.25 0.20 5 3 ? O 0.15 CL O o 0.10 5 3 CL 00 0.05 0.00 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.9 The Spectrum of the Nino4 SST index , 1899-1989, statistical significance at the 95% confidence level is indicated by the thick line. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis. 110 0.1 2 0.10 0.08 0.06 (U 5 O CL 0.02 0.00 u 03 Û. - 0.02 GO -0 .0 4 -0 .0 6 -0 .0 8 - 0.10 15.1667 7.5833 5.0556 3.7917 3.0333 2.5278 2.1667 Period Figure 5.10 Cospectrum between Nino4 and SEAR, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis. I l l occurring between 4.5 and 5.4 years, with other prominent contributions observed at 3.6 years, 3.0 years, 2.75 years and at 2.3 years. A large negative contribution is evident between 7.6 to 8.3 years, which is one of the dominant powers in the spectrum of the NAOI, the largest negative contribution to the correlation is observed between 3.79 years and 4.3 years, with a distinct peak at 3.79 years. Those frequencies at 3.8 years and 2.3 years are observed in the S O I. spectrum (Nicholson and Entekhabi, 1986) and similar frequencies are observed in the Nino4 spectrum (Fig.5.9). Squared coherence between Nmo4 and SEAR (Fig.5.11) shows very large coherence at many periodicities including squared coherences greater than 0.7 between 11.4 to 13.0 years, at 5.7 years and between 2.45 to 2.6 years. Meaningful shared periodicities can be observed by comparing the dominant spectral powers of the cospectrum and the squared coherence at the same frequencies. The large positive spectral power in the cospectrum at about 2.3 years is discarded due to the lack of a large squared coherence at that frequency. The large negative spectral powers between 3.9 to 4.1 years and also at 7.5 to 8.3 years are also discarded for the same reason. The only large spectral peak that is retained on the basis o f having a large squared coherence is the periodicity between 11.3 and 13.0 years. However, examination of the individual spectra (Figs.5.7 and 5.9) shows that this frequency is not statistically significant at the 95% level in either spectrum. In summary, comparison of the significant spectral powers of the individual spectra do not show any spectral peaks in common. The cospectrum between the NAOI and Nîno4 indices (Fig.5.12) shows prominent spectral contributions to the small negative correlation between them (r=-0.19, significant at 112 0.9 0.8 (D O c 0.7 (ü 03 0.6 .C O o 0.5 "D 03 0.4 o 3 cr 0.3 00 0.2 0.1 0.0 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.11 Squared coherence between Nino4 and SEAR, 1899-1989. Periodicity is on the x-axis in years, and squared coherence is measured on the y-axis. 113 0 .1 5 0.10 0.05 03 5 O û _ o 0.00 o 0) Q. -0 .0 5 00 - 0.10 -0 .1 5 15.1667 7.5833 5.0556 3.7917 3.0333 2.5278 2.1667 Period Figure 5.12 Cospectrum between the NAOI and Nino4, I899-I989. Periodicity is on the x-axis in years, and spectral variance is measured on the y- axis. 114 the 90% confidence level) occur at low fi’equencies, with negative peaks observed at 13.0 years, 5.7 years and at 4.55 years. A distinctive large positive contribution to the covariance is observed at 3.79 years and is by far the single largest positive contribution to the overall negative correlation. Some smaller positive spectral contributions are observed in the high fi-equency end of the spectrum and particularly in the biennial frequencies. The large positive contribution at 3.79 years is also observed in the cospectrum between the Nino4 index and SEAR (Fig.5.10), this time as a large negative peak near 3.79 years and in the spectrum of the S.O.I. (Nicholson and Entekhabi, 1986) which suggests a possible quasi-periodic interrelationship between the NAOI, ENSO and SEAR at this frequency. Rogers (1984) performed cospectrum analysis between the NAOI and normalized winter SLP anomalies at Darwin between 1900-1983 and shows that the largest power occurs at 5.7 years. The squared coherence spectrum between NAOI and Nino4 (Fig.5.13) has relatively large values occurring at 13.0, 5.67, between 5.75 and 5.67 years, and at 2.27 years. These findings indicate that the relatively large spectral peak at 5.7 years in the cospectrum has relatively large amounts of variance and spectral peaks at those frequencies, and are thus possibly statistically meaningful periodicities. Cospectral analysis of the negatively correlated SEAR and NAOI (r=-0.48, significant at the 99% confidence level) time series (Fig.5.14) depicts relatively large peaks between 7.58 and 9.5 years; smaller peaks are observed at 5.6 years and 3.8 years. The peak between 7.58 and 9.5 years in the cospectrum is evident in the individual spectra of the NAO (Fig.2.6) and SEAR (Fig.5.7). Meaningful spectral relationships are again determined using the squared coherence in relation to the cospectrum between the indices. The squared coherence 115 0.9 0.8 0.7 (D ü C D 0.6 (UL_ s z o 0.5 O x> 0.4 i_OJ o =3 0.3 CT GO 0.2 0.1 0.0 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.13 Squared coherence spectrum between the NAOI and Nino4. Periodicity is on the x-axis in years, and squared coherence is measured on the y-axis. 116 0.1 0.0 - 0.1 (ü u c o - 0.2 a > - 0 .3 "5 Dü - 0 .4 C l 00 - 0 .5 - 0.6 -0 .7 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.14 Cospectrum between the NAOI and SEAR, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y- axis. 117 0.9 0.8 Q) 0.7 2 0.6 (D O 0.5 0.4 3 0.3 cr ^ 0.2 0.1 0.0 2.275 2-0222 6.0667 4.55 18.2 Period measured on the y-axis. 118 spectrum (Fig.5.15) between the NAOI and SEAR shows relatively large squared coherence occurring at 13.0 years, at 7.58 years, between 2.67 to 2.75 years and at 2.3 years. The peaks at 3.79 years, 4.55 years, 5.68 years, 7.58 years, 13.0 years and at 22.75 years in the cospectrum have relatively large coherence. Examination of the statistically significant periods in the individual spectra (Figs.2.6 and 5.7) show that only the meaningful periodicity at 7.58 years is statistically significant in both spectra. The squared coherence spectrum (Fig.5.15) shows that only the periodicity at 7.58 and at low fi-equencies beyond 15 years have large coherence. Neither the statistically significant peaks at 3.25 years observed in the spectrum of SEAR (Fig.5.7) nor the peak at 2.67 years in the spectrum of the NAOI (Fig.2.6) have large squared coherence, and therefore share little common variance. Cospectrum analysis between SAFR and the NAOI shows (Fig.5.16) large positive contributions to their covariance at almost every periodicity. The largest spectral peak occurs at the extremely low frequencies (beyond 90 years), at 7.58 years, 5.68 years, 4.55 years and between 2.67 to 2.75 years. The squared coherence between SAFR and the NAOI (Fig.5.17) has many large peaks with very large magnitude across the spectrum. Especially large peaks occur at infinity and 91, 13, 7.58 and between 5.68 and 4.33 years, as well as at other high frequencies. Inspection of the SAFR and NAOI spectra (Figs. 5.8 and 2.6) indicates that only the spectral power at 2.67 years is statistically significant in both o f the individual spectra in addition to having a large squared coherence at this periodicity. Cospectrum analysis between SAFR and the Nifio4 SST index depicts (Fig.5.18) relatively large contributions to the correlation (r = -0.295, significant at the 99% confidence level) at all frequencies with the exception of between 2.8 - 4.0 years, where relatively large 119 0.3 L_ 5 O 0.2 û_ o u (D 0.1 enCL 0.0 - 0.1 18.2 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.16 Cospectrum between SAFR and the NAOI, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y- axis. 120 0.9 0.8 D U 0.7 C (U L_ 0.6 d) _c o 0.5 o x> d) 0.4 o 3 cr 0.3 00 0.2 0.1 0.0 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5.17 Squared coherence spectrum between SAFR and the NAOI, 1899- 1989. Periodicity is on the x-axis in years, and squared coherence is measured on the y-axis. 121 0.2 0.1 0) ? O û_ o 0.0 o (U Cl en - 0.1 - 0.2 18.2 9.1 6.0667 4.55 3.64 3.0333 2.6 2.275 2.0222 Period Figure 5,18 Cospectrum analysis between SAFR and Nino4, 1899-1989. Periodicity is on the x-axis in years, and spectral variance is measured on the y-axis. 122 signs of the cospectra between rainfall and the S.O I. are negative over SEA and positive over South Africa (Nicholson and Entekhabi, 1986) in keeping with the sign of the correlation coefficient between the Nino4 index and rainfall (Tables.5.2 and 5.4). Cospectral analysis of the negatively correlated SEAR and NAOI time series (Fig.5.14) depicts spectral peaks at 7.58 and 9.5 years which are not found in many Africa rainfall spectra (Nicholson and Entekhabi, 1986). However a peak at 7.0 years is observed in the rainfall spectra of many northern African stations which appears to be distinct from the peak found between 5-6 years (Nicholson and Entekhabi, 1986). The NAOI and SEAR are strongly related at 7.58 years indicating that variability at this frequency produces much of the large negative correlation between the indices. 123 0.9 0.8 (U o c 0:7 (U ^a,_ 0.6 ! g Oj X> V 0 .4 o & G.3 (/) 0.2 0.1 0.0 Z.0222 18.2 9.1 6.0667 Period C/117R and Nino4, 1899-1989. measured on the y-axis. 124 NAOI Nino4 SEAR SAFR NAOI 1.0 Nino4 -0.19 1.0 SEAR -0.48 O il 1.0 0 0 SAFR 0.43 -0.30 -0.10 1.0 Table 5.4 Correlation coefTicients between atmospheric circulation and precipitation indices. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined. 125 5.5 Regression Analysis between the NAOI and Climatic fields over Africa. The anomalous circulation associated with the NAOI is described using OLS regression analysis. The standardized slope coeflBcients are mapped and are interpreted as the standard deviation change in the dependent variable (Y) caused by a 1.0 standard deviation increase in the independent variable (XJ and are thus dimensionless ratios. SLP coefBcients associated with a 1.0 standard deviation increase in the NAOI reveals the NAO pattern in the Northern Atlantic (Fig. 5.20). The signs of the regression coefBcients are the same as those of pressure anomalies during the positive NAO phase. The NAO-related SLP departures observed over Africa are dominated by large positive coefBcients centered over the Atlantic Ocean, associated with the Atlantic subtropical high, extending far inland and dominating northern Africa and much of the Mediterranean basin. Eastern Africa has a higher than normal pressure along its eastern coastline from Ethiopia to South Africa but the coefficients are only statistically significant at the 95% level. These positive pressure coefBcients extend far inland into central equatorial Afiica and cover the region encompassed by the regional precipitation anomalies. A second area with large positive coefBcients can be observed over southeast South Africa which extends into the southwestern Indian Ocean. When the NAO is positive (negative) the SLP over SEA is higher (lower) than normal and rainfall is lower (higher) than normal consistent with the correlations in Table 5.2. Streamlines indicating anomalous 1000 hPa flow associated with the NAO (Fig.5.21) indicate the flow regime that is imposed upon the climatology (Fig.4.3). Analysis o f these streamlines must take into account the feet that they represent changes in flow resulting from a 1.0 standard deviation NAOI increase and are thus imposed upon the climatological flow. 126 SON SON SON SON 40N 40N 20N 20S 40S SOW 40W 20W CM 20E 40E 60E Figure 5.20 SLP standardized regression coefficients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 127 The climatological low pressure center over eastern Angola intensifies to the north and east of its climatologcal location (20TE, 2Cf S) while divergence and enhanced easterly flow occurs over the SEA during the positive phase of the NAO. Simultaneously, divergence and strong southerly flow occurs along the coastline of Mozambique and Kenya and over Tanzania. The strong southeasterly streamlines over most of eastern Afiica to the north of 20° S indicates that the deep penetration of the moist Zaire air masses associated with increased rainfall over SEA does not occur, nor does a northeasterly monsoon flow from the Indian Ocean. Easterly anomalies over Zaire reduce the strength of the westerly flow during the positive NAO phase. During the negative NAO phase the opposite occurs and the moist westerlies are enhanced over SEA (see next section); the result of these NAO-related circulation differences are the significant negative rainfall-NAO correlations (Tables 5.2 and 5.4). The 1000 hPa u-wind component coefBcients associated with the NAO (Fig.5.22) show statistical significance over most of the North Atlantic Ocean, over most of Northern Europe, along the Eastern coast of the United States to South America. Statistically significant negative coefficients are observed over much of Northwestern Afiica from the Mediterranean to the Gulf of Guinea. The significant negative (easterly) coefficients are observed far to the south of the Equator over Eastern Afiica, over the same regions as the significant positive SLP anomalies, and cover much of western South Afiica and the southwestern Indian Ocean. Significant easterly flow is observed over central SEA stretching from Zaire to the Tanzanian coastline, as rainfall over eastern Afiica is strongly associated with westerly flow o f moist Zairean air consistent with the significant negative correlation coefficients between the NAOI and rainfall observed (Fig.5.1) over eastern Africa. 128 aoN '-■f}— BON 40N iJ 40S^ 60W 40W SOW CM 20E 40E 60E Figure 5.21 1000 hPa streamlines associated with the NAOI, 1958-1995. Light and dark shading represents the statistical significance of the average u- and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 129 SON SON SON -j SON 40N 40N 20N :# ^ 3 ïv rf'.- V^r-J 20S n&y 40S <3 )^'t:< 3 B £ £ ^ SOW 40W 20W Figure 5.22 1000 hPa u-wind standardized regression coefficients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test 130 The 1000 hPa v-wind coefficients associated with a 1.0 standard deviation increase in the NAO (Fig.5.23) show significant northward component to the flow over the central North Atlantic Ocean. Significant northerly flow (negative coefficients) is observed over much of northwestern and northeastern Afiica with isolated pockets of positive coefficients observed over northcentrai Afiica. A small southerly flow is located just south of the equator including SEA, with a small region of significant coefficients observed immediately west of SEA Strong and significant southerly flow is observed between the eastern coast of Afiica and Madagascar. Significant northerly flow is observed over western South Afirica and due south of the continent. Climatological streamlines at the 850-hPa level (Fig.4.12) shows strong convergence just north of the equator in the vicinity of Lake Victoria and over the climatological surface low over eastern Angola. Streamlines associated with the anomalous NAO circulation at the 850-hPa level (Fig.5.24) show southeasterly flow over much of SEA during the positive phase of the NAO, in contrast to the northeasterlies in the climatology (Fig.4.12). The small high over Zaire (at 20°E, 10°S) enhances easterly flow south of the equator during the positive phase of the NAO and is imposed upon the mean westerly climatological flow (Fig.4.12). The convergent flow into the cyclone further south over eastern Angola is displaced slightly east and associated with enhanced northeast flow over south-central Afirica. The statistical significance of the flow anomalies associated with the NAO can be ascertained by examination of the u- and v-wind coefficients at 850 hPa. The u-wind coefficients (Fig.5.25) show an extremely strong easterly flow (negative coefficients) over much of North Afiica fi’om the Mediterranean Basin south over northwestern Afiica to the 131 SON SON SON ■— — SON 40N H 40N 20N 20S 40S 40S SOW 40W 20W GM 20E 40E SOE Figure 5.23 1000 hPa v-wind standardized regression coefficients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 132 :r-3 8 0 N 60N 40N SOW 40W 20W CM 20E 40E 60E Figure 5.24 850 hPa Streamlines associated with the NAOI, 1958-1995. Light and dark shading represents the statistical significance of the average u-and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 133 SON BON 60N BON 40N ' 40N 20N ^ 20S 40S 40S 60W 40W 20W GM 20E 40E 60E Figure 5.25 850 hPa u-wind standardized regression coefTicients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 134 Gulf of Guinea. This is comparable to the easterly flow anomalies observed over the same areas at 1000 hPa (Fig.5.22). Significant easterly flow is observed to the south of the Equator, extending fi'om the Congo, through Zaire to SEA. A small westerly flow (positive coeflBcients) anomaly is observed over Angola and covers some of the South Atlantic Ocean while others stretch from the central Indian Ocean along the coast of eastern Afiica to cover much of eastern South Afiica. A large and significant easterly anomaly is observed to the southeast of South Afiica with westerly anomalies observed just to the southwest. These u- wind coeflBcients at 850 hPa are very similar to those observed at 1000 hPa but have larger magnitudes and greater statistical significance possibly related to a decrease in the data variance away fi'om the surface of the Earth. The small area with westerly flow to the west of Madagascar slightly inhibits the penetration of moist Indian Ocean air masses over eastern Afiica which are strongly associated with increased rainfall over Zimbabwe during summer (DIFM) (Jury, 1996). The v-wind coefficients associated with the NAO at 850 hPa (Fig.5.26) show very similar patterns to those observed at 1000 hPa (Fig.5.23). An extensive positive anomaly is located over the central North Atlantic Ocean and Northern Europe and Scandinavia. The Mediterranean Basin, much of southern and eastern Europe and the Middle East are covered by anomalous northerly flow. Northwestern and northeastern Afiica are covered by significant northerly wind anomalies. Significant southerly flow is observed between the coast of eastern Afiica and Madagascar, with significant northerly flow over southern Namibia and northwestern South Afiica. 135 BON SON 60N ir—------60N 40N 5 = 40N 20N 20N EQ ------80S 20S 40S 4 0S 60W 40W 20W GM 20E 40E 60E Figure 5.26 850 hPa v-wind standardized regression coefficients associated with the NAOI, 1958-1995. Negative isollnes are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 136 aoN aONiœ 60W 40W 20Tf Figure 5.27 300 hPa Streamlines associated with the NAOI, 1958-1995. Light and dark shading represents the statistical significance of the average u- and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 137 Streamlines at 300 hPa (Fig.5.27) show that southern Afiica is dominated by an anomalous high pressure system related to strong northerly flow over the west coast and strong southeasteriy flow over SEA. The anticyclone is an enhancement of the mean 300 hPa high in the climatology (Fig.4.13) during positive NAO cases. The imposition of the upper level high on the climatological mean flow at 300 hPa ÇFig.4.I3) results in relatively cold air between SEA and Madagascar in addition to relatively warm air over, and to the west of, the west coast of Afiica. The dynamics associated with this configuration results in upper level divergence over and to the east of SEA and upper level convergence over the west coast of Afiica and to the east of the trough. If this system conserves mass in a similar manner to mid level easterly waves (Hastenrath, 1985) then upper level divergence over SEA will be associated with low- and mid-level convergence resulting in uplift, while divergence associated with subsidence occurs in the low- and mid-levels between SEA and Madagascar. This situation appears to enhance (reduce) uplift (subsidence) associated with increased (decreased) rainfall over SEA during the positive (negative) NAO phase, while the correlation coefficients between the NAOI and SEAR (Fig.5.1) are negative. Anomalous u-wind component associated with the NAO at 300 hPa (Fig.5.28) has a strong westerly component to the flow over all of southern Afiica south of 15°S (the axis of the anticyclonic center in Fig.5.27). A strong easterly component to the flow dominates immediately to the north and extends to between 10°-15“ N and northeast towards the Persian Gulf as well as extending south of the equator to 10°-15°S. North of the equator is an extremely strong and highly significant westerly wind anomaly which extends fi'om the eastern coast of South America over Afiica and into central Asia, which possibly represents a 138 SON SON SON SON 40N 40N 20N ^ 20N EQ 20S 20S 7 ^ ^ 40S 40S SOW 40W 20W GM 20E 40E SOE Figure 5.28 300 hPa u-wind standardized regression coefficients associated with the NAOI, 1958-1995 Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 139 strengthening and shift to the southeast of the subtropical Jetstream during the positive phase of the NAO. The distribution of statistically significant v-wind coefficients at 300 hPa (Fig.5.29) are similar to those observed at lower levels of the atmosphere and show extensive southerly flow over the central North Atlantic, northerly flow over Iberia, Algeria and Morocco. Central Europe and parts of the Middle East are under a northerly flow centered over the Black Sea and the Caucasus. Northcentrai Afiica is covered by a large positive flow extending south into the tropics. Southeastern Afiica is covered by a strong southerly flow, while western Afiica exhibits strong northerly flow in keeping with the enhanced anticyclone at 300 hPa over Angola in the positive NAO phase (Fig.5.27) The 850 hPa specific humidity anomalies associated with the NAO (Fig.5.30) form a distinct pattern over the North Atlantic Ocean closely related to the surface and near-surface circulation. Northwestern Afiica and much of the Mediterranean Basin are covered by significant negative humidity coefficients, small areas with significant positive humidity coefficients are observed over Egypt and Sudan. A region along the Gulf of Guinea extending to central Afiica exhibits a significant negative response to the positive NAO phase. No coherent humidity response is observed over Afiica south of the equator. Significant positive humidity coefficients are observed over large areas of the southern Atlantic Ocean between 20°S and 50°S. Specific humidity coefficients at 300 hPa (Fig.5.31) are somewhat different from those observed closer to the surface. The North Atlantic Ocean is covered by a series of zonally oriented humidity coefficients of alternating sign stretching from Greenland to the equator. 140 80N 80N 60N 60N 40K 40N 20N 20S 40S SOW 40'ff 20¥ GM 20E 40E 60E Figure 5.29 300 hPa v-wind standardized regression coefTicients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 141 BON BON BON BON 40N 40N 20N 20N EQ 20S 20S 40S 40S SOW 40W 20W GM 20E 40E SOE Figure 5.30 850 hPa Specific Humidity standardized regression coefficients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test 142 BON BON BON 60N 40N 40N 20N 20S 40S 40S 60W 40W 20W GM 20E 40E 60E Figure 5.31 300 hPa Specific Humidity standardized regression coefficients associated with the NAOI, 1958-1995. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 143 Negative coefiScients extend in a coherent manner from relatively small significant areas over the eastern Pacific Ocean across Central America and the Caribbean to cover the entire Mediterranean Basin and much of the Caucasus region of central Eurasia. Large significant positive humidity coefiScients cover much of northern South America and extend to the northeast across the tropical Atlantic, becoming a thin ribbon over northern Afiica. Isolated areas of negative coefficients are observed over SEA. The relationship between the NAOI and precipitable water (PWAT) over SEA (Fig.5.32) is weakly negative, with the exception of significant negative coefiScients centered on Lake Malawi, indicating a decrease in precipitable water associated with the positive NAO phase. A large significant negative PWAT relationship is observed from the Gulf of Guinea through the Sahel to Ethiopia. Northern Afiica (including Morocco), the Mediterranean Basin and southern Europe show the expected large negative PWAT relationship to the NAO associated with the strong northerly flow around the enhanced subtropical high, while Northern Europe is positively correlated due to the strong westerlies These Northern Hemisphere relationships correspond to those found by Hurrell (1995). The outgoing longwave radiation at the top of the atmosphere (OLR) serves as a proxy for tropical convective cloud heights. Negative OLR anomalies, such as those over South America (Fig.5.33), are associated with increased convective cloudiness since convective cloud top temperatures are low. The large positive OLR anomaly implied by the positive regression coefiScients centered over Lake Malawi represent a decrease in convective cloudiness during the positive NAO phase. A similar relationship appears over the Sahel region, Saudi Arabia and over southern Europe. It must be noted that the OLR data are 144 SON SON SON SON 40N 40N 20N 40S SOW 40W 20W GM 20E 40E SOE Figure 5.32 PWAT standardized regression coefficients associated with the NAOI, 1958-1995. Negative isollnes are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test 145 BON SON BON BON 40N 40N 20N 20N EQ 20S 20S 40S 40S 60W 40W 20W GM 20E 40E BOE Figure 5.33 OLR standardized regression coefficients associated with the NAOI, 1958-1995. Negative Isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 146 calculated within the reanalysis model and are consequently highly dependent on the model’s physics and must be treated with caution (Kalnay et al., 1996). The results over Africa nevertheless agree in spatial pattern with those of specific humidities and precipitable water. Significantly decreased convective cloud heights over SEA contrasts to the significant increases in convective cloud heights over South America, suggestive of an overturning Walker-type circulation between these regions associated with the phase of the NAO. In summary, decreased precipitation observed over SEA during positive (negative) NAO phases (Fig.5.1) is related to reduced (enhanced) penetration of moist and nearly unstable Zaire air masses strongly associated with rainfall over eastern Afiica. An easterly (westerly) propagating wave observed extending from southern to equatorial Afiica enhances convergence (subsidence) at 300 hPa (Fig.5.27) over SE A This configuration of upper level dynamics acts to enhance (reduce) precipitation over SEA during positive (negative) NAO extremes. In light of the cluster of stations in SEA having statistically significant negative precipitation correlations to the NAOI, the contribution o f the dynamics at 300 hPa to reduced precipitation receipt is considered secondary to the near surface advection of Zaire air masses over SEA 147 5.6 Regressions of the Southeast African Rainfall Index (SEAR). At best, with correlations of about -0.7 over the period 1958-1989 (Table 5.2), SEAR only has a coefiBcient of variation (r^ with the NAO o f about 50%. As such the variability of the rainfall index is examined separately in relation to the atmospheric circulation. The SLP response to precipitation variability over SEA (Tig.5.34) exhibits a dipole in the North Atlantic almost identical to the anomalous circulation associated with the NAO but with opposite sign to the NAO (Fig.5.20). The statistical significance of this pattern is comparable to that of Fig.5.20 and is in keeping with negative relationship between SEA precipitation index and the NAOI. The SLP field over much of southern Afiica is only slightly negative and is not statistically significant, which is unexpected as increased precipitation receipt might be expected to be associated with sizable circulation variations and a significant decrease in SLP over this region. The 1000 hPa geopotential height coefficients associated with precipitation over SEA (not shown) closely resemble the SLP coefficients (Fig.5.34). The 1000 hPa streamlines related to precipitation variability over SEA (Fig.5.35) show significant flow over the North Atlantic Ocean dominated by a severely weakened Icelandic Low and weaker than usual Azores High, comparable to the negative phase of the NAOI. The pattern of air motion in Fig.5.35 is in many regards the inverse of that in Fig.5.21. The westerly flow over the North Atlantic is reduced by anomalous easterly coefficients and the northeast trades are also weakened by the imposition of a strong anomalous southwesterly flow over the tropical Atlantic Ocean that is statistically significant over the western coastline of northwest Afiica from the Mediterranean Basin to the Gulf of Guinea. Flow over SEA is principally fiom the north and northwest which is associated with 148 SON SON SON 40N 6 40N SON EQ 2QS SOS 40S 40S SOW 40W SOW GM 20E 40E 60E Figure 5.34 SLP standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 149 BON SOW 40W 20W GM 20E 40E 60E Figure 5.35 1000 hPa Streamlines associated with SEAR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients' t-scores at the 95% and 99% confidence levels respectively. 150 advection of moist Zairean air masses over the region. This flow appears related to enhanced convergence into anomalous cyclonic circulations over the Southern Indian Ocean east of Madagascar in addition to a similar feature over southern Mozambique. A smaller region of convergence is located over eastern South Africa and is associated with the convergence of mid-latitude air masses jfrom the South Atlantic and the southern Indian Oceans. The intensified climatologjcal low over eastern Angola is associated with the enhanced advection of mid-latitude air masses fi'om. the South Atlantic Ocean over southern Afiica and the advection of tropical Atlantic air masses firom the northeast. Statistically significant 1000 hPa u-wind coefficients associated with SEAR (Fig.5.36) are reduced somewhat fi’om the levels of significance in Fig.5.22. The anomalous westerly flow over Greenland, easterly flow into Northern Europe and Scandinavia, and westerly flow over the North Atlantic south o f 40°N, all exhibit some significance. Strong westerly flow is observed over much of eastern Afiica and most of SEA with statistical significance observed over the eastern part of SEA centered on the Tanzanian coast (at 10°S, 40°E). This westerly flow over eastern Afiica extends to the east and significant coefficients are observed to the west of India and significant easterly coefficients over Bangladesh and to the southeast of India. The 850 hPa streamlines associated with an increase of 1.0 a in SEAR (Fig.5.37) show strong westerly flow anomalies across the tropical Atlantic Ocean on both sides of the equator and southwesterly flow is observed from the Gulf of Guinea over the Sahel and north to the Mediterranean Basin. Much of southern Afiica is covered by anomalous flow from the northwest during wet summers over SEA and converges with southerly flow into a cyclonic 151 8 0 N SON SON SON 40N - 40N 'M 20N 20N 20S 20S '4 0 S 40S SOW 40W 20W GM 20E 40E SOE Figure 5.36 1000 hPa u-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 152 BON BON 40N 40S BOW 40W 20W GM 20E 40E 60E Figure 5.37 850 hPa Streamlines associated with SEAR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 153 center over northeastern South Afiica and southern Swaziland. Southerly flow is prevalent over Namibia and western and central South Africa. A cyclonic anomaly to the east of Madagascar (centered at 55°E, 20“S) contrasts to the anticyclonic flow anomaly over this location associated with the circulation diflference between wet and dry South African summers (Jury and Pathack, 1993; their Fig.4b). Southerly and southeasterly flow is observed over the central Indian Ocean and converges to the west of India with westerly flow from Africa. 850 hPa u-wind component coefiBcients associated with a 1.0 a increase in SEAR (Fig.5.38) depict westerly flow prevailing over most ofNorth Africa and southern Africa to the North of 16°S, to be significant in several areas. Westerly flow between the Horn of Africa and India is significant, as are two areas south of the equator. Streamlines at 300 hPa (Fig.5.39) again shows strong and significant circulation anomalies across the Atlantic Ocean from western Greenland south to the tropical Atlantic and equatorial Afiica although statistical significance is not as widespread as in response to the NAG (Fig.5.27). The anomalous circulation over the North Atlantic Ocean is similar to that at 1000 hPa (Fig.5.21) with the exception of the anomalous upper level high over the western tropical Atlantic (at 20°N, 50°W) which induces significant southwesterly flow anomalies to the northeast of the Caribbean and easterlies over the central tropical Atlantic. Flow about the weakened Azores High interacts with a second high occurring over northwestern Africa to induce the significant anomalous southwesterly flow over Morocco and Iberia. The interaction between this high in northwest Africa and the large low centered over the Red Sea induces significant northeasterlies over central North Africa and strong northeriy flow over much of central north and equatorial Africa. The Red Sea low produces 154 80N aoN 60N 60N 40N 40N 20N 20N EQ 20S 20S 40S 40S 60W 40W 20W GM 20E 40E 60E Figure 5.38 850 hPa u-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 155 BON BON S SON "7—-r 40N 20N © 40S 60W 40V 20V GM 20E 40E 60E Figure 5.39 300 hPa Streamlines associated with SEAR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 156 significant anomalous southwesterlies over Lake Victoria along the northern extent of SEA and along the coast of eastern Afiica. Westerlies predominate Afiica south o f the equator. An anomalous low centered to the southeast of South Afiica represents a cyclonic vortex in the higher latitude westerlies inducing southwesterly and southerly flow over western and central Southern Africa, and establishes a northerly component to the flow over the LMR, Zimbabwe and Mozambique. Differences between Figs.5.27 and 5.39 over southern Afiica are striking. The 300 hPa anticyclone representative of the positive NAO phase (Fig.5.27) is replaced by cyclonically based westerly flow representative of the wet periods in SEA (Fig.5.39). The 300 hPa u-wind component coefiBcients associated with wet summers over SEA (Fig.5.40) form a sequence of zonally oriented anomalies of alternating sign appearing in high latitudes over Greenland and extending southward to equatorial Afiica as in Fig.5.28. The westerly wind maximum in the core of the subtropical Jetstream in the Northern Hemisphere located at 30° N over North Afiica is enhanced, with an increase in the westerly winds over the western Mediterranean basin and northwestern Afiica during anomalously wet periods in southeastern Africa. A large significant positive anomaly straddles the equator over much of eastern Africa, associated with a significant weakening of the climatological near-equatorial easterlies over Afiica (Fig.4.13). The weakening of the climatological easterlies is due to the enhanced u-component of flow associated with two anomalous cyclones at 20°N, 40°E and 35°S, 35°E in Fig.5.39. As in Fig.5.28, Fig.5.40 shows the strong significance of mean easterly wind strength variations around equatorial Afiica in the extremes of both the NAO and SEAR precipitation. Comparison between the u-wind coefficients associated with the 157 8 0 N r-T3 SON 60K 60N 40N — 40N 20K : 40S 40S sow 40W 20W GM 20E 40E 60E Figure 5.40 300 hPa u-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 158 NAO and SEAR (Figs.5.28 and 5.40) shows great consistency in the inverse relationship between the two fields across the globe. The 300 hPa v-wind coefficients at associated with wet summers over SEA (Fig.5.41) are useful in isolating the strong divergent nature of the flow over eastern Afiica south of the equator, and apparent in the streamlines (TFig.5.39). A region with significant northerly flow is observed over Mozambique in contrast to significant southerly flow over equatorial Kenya and Somalia. During the positive (dry LMR) NAO phase (Fig.5.29) the divergent flow in the v-component winds is apparent, with the flow over eastern Afiica dominated by southerly winds fi'om the Indian Ocean. The PWAT anomalies associated with increased rainfall over SEA (Fig.5.42) are very similar to those associated with the NAO in the Northern Hemisphere (Fig.5.32), but with opposite signs. An extensive positive PWAT anomaly covers much of maritime northwestern and northern Afiica and extends almost fi'om the equator over west Afiica as far north as the Mediterranean Basin. Eastern and central north Afiica have a non-significant negative relationship to rainfall over SEA A large and significant positive PWAT anomaly is observed over the eastern coastline o f Afirica, extending fi'om Eritrea and Ethiopia to the north to a maximum over SEA and extending far to the east into the tropical Indian Ocean. The PWAT coefficients associated with increased rainfall over SEA can be explained by their relationship to the streamlines at 1000 hPa (Fig.5.35), 850 hPa (Fig.5.37) and 300 hPa (Fig.5.39). The anomalous westerly flow into tropical Afiica south of the equator fi'om the Atlantic to Zaire is very strongly related to the significant positive PWAT anomaly and increased precipitation over Eastern Afiica and SEA. In keeping with this relationship 159 8 0 K BON 60N BON 40N 40N 20N / 20S 40S 40S BOff 40W 20W GM 20E 40E 60E Figure 5.41 300 hPa v-wind standardized regression coefficients associated with SEAR, 1958-1989. Negative isoiines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 160 SON SON 30fj SON 40N -X 40N 20N : SOS ^ —; 40S SOW 40W SOW GM SOE 40E SOE Figure 5.42 PWAT standardized regression coefTicients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 161 easterly flow and negative PWAT anomalies are observed during winters with below normal precipitation over the LMR. The specific humidity coefficients at various levels of the atmosphere are similar to those exhibited by the PWAT distribution. Specific humidity coefficients at 1000 hPa (Fig.5.43) are large and positive off the west Afiican coast in the North Atlantic and around SEA. In the North Atlantic this reflects the weakened northerly (increased southerly) flow over the Canary Current and western North Afiica as the subtropical high weakens. The same is true over SEA except that the source is enhanced westerlies over the South Atlantic across Zaire. At 850 hPa positive humidity coefficients are observed fi'om the Caribbean to North Afiica (Fig.5.44), where the coefficients become statistically significant, the eastern Mediterranean Basin and the Middle East. These anomalies are zonally oriented consistent with the strong 850 hPa u-wind anomalies over the same region (Fig.5.38). The 850 hPa specific humidity coefficients exhibit even larger areal significance over SEA, accompanying the strong westerly flow into the region at that level. OLR anomalies associated with increased precipitation over SEA (Fig.5.45) shows SEA encompassed by an extensive statistically significant negative anomaly indicating increased convective clouds over the region during with wet SEA summers. This is consistent with the 300 hPa divergence observed over the region (Fig.5.39) which is dynamically associated with low level convergence and uplift during wet SEAR summers. The results have shown that SLP varies significantly over southern Afiica in conjunction with the NAO but SLP coefficients are statistically insignificant over SEA 162 SON SON SON SON 40N 40N 20N 20S 40S SOW 40W 20W Figure 5.43 1000 hPa Specific Humidity standardized regression coefficients associated with SEIAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 163 80N SON SON V e 60N 40N ^ 40N 20N^ 20S r 40S 60W 40W 20W GM 20E 40E 60E Figure 5.44 850 hPa Specific Humidity standardized regression coefficients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 164 SON 80N 60N 60N 40N ~Z¥eJ 40N 20N 20N EQ 20S 20S 40S 40S Figure 5.45 OLR standardized regression coefHcients associated with SEAR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 165 (Fig.5.34) in conjunction with SEAR. It is also apparent from the SLP patterns that SLP is higher (lower) over land than over the South Atlantic Ocean causing easterly (westerly) flow towards (from) the Ocean during dry (wet) SEA summers, comparable to positive (negative) NAO phases. Enhanced low level convergence into the intensified low pressure system over Angola just southwest of the climatological low is associated with the enhanced advection of moist Zaire air over SEA and a decrease in the advection of dry monsoonal air from the northeast and air masses from the Indian Ocean over SEA. The westerly flow of moist Zaire air is especially associated with precipitation over near-equatorial Africa due to near saturation as it is gradually uplifted, possibly aided by a potential destabilization of a neutral or weakly stable westerly flow near the equator by the vertical component of the Coriolis parameter. Divergence over SEA evident at 300 hPa (Fig.5.39) dynamically induces near surface convergence and uplift near the surface. This rising motion dynamically induced over SEA is associated with clouds and precipitation consistent with wet DJF periods over SEA. In addition to these dynamic causes ft)r the correlations between the circulation anomalies and SEAR, a large positive PWAT anomaly encompasses SEA, which is related to large specific humidity anomalies throughout the atmosphere (Fig.5.43 and Fig.5.44). OLR from the top of the atmosphere (Fig.5.44), negatively correlated to convective clouds, is significantly negative over SEA implying a significant increase in convective cloud heights. 166 5.7 Regressions of the Southern African Rainfall Index (SAFR) Explanation of the significant positive correlations between rainfall over SAF and the NAO is examined in this section using the SAFR index in a regression analysis to describe the anomalous circulation associated with rainfall at those stations significantly correlated to the NAO (shown in Fig.5.3). SLP coefficients associated with a 1.0 standard deviation increase in SAFR (Fig.5.46) depict a large and significant signal over the North Atlantic Ocean that is only somewhat similar to the NAO pattern. Negative SLP coefficients are observed over most o f the South Atlantic with statistically significant coefficients isolated in a small area. Negative coefficients extend into Afiica south of 10°N with the exception of over eastern South Afiica where a small area of positive coefficients is observed. In that area, a series of waves is observed in the SLP coefficients extending fi'om the western Indian Ocean across South Afiica and into the South Atlantic Ocean. This corresponds to a pattern with a ridge located over the western Indian Ocean to the east of Madagascar (at 20°S, 60°E), a trough over Madagascar, a ridge over much of South Afiica, a trough over the west coast of South Africa and a small ridge over the eastern South Atlantic. SLP anomalies create an easterly flow over South Afiica during wet years with a southeast flow of moist Indian Ocean air masses into the interior over the elevated eastern escarpment. 1000 hPa streamlines associated with wet SAF summers (Fig.5.47) depicts aspects of this wave train consistent with the SLP coefficients. The statistical significance of the circulation features associated with wet South Afiican summers can be ascertained fi'om the u- and v-wind components. As a reminder that the NAO is related to SAFR the v-wind component at 1000 hPa (Fig.5.48) depicts a cluster 167 SON SON 60N -J 60N 40N 40N 20N 20S 40S 40S SOW 40W 20W GU 20E 40E 60E Figure 5.46 SLP standardized regression coefficients associated with SAFR, 1958-1989. Negative isoUnes are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 168 80N 60N 40N 60W 40W 20W GM 20E 40E 60E Figure 5.47 1000 hPa Streamlines associated with SAFR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 169 BON BON BON BON 40N 40N 20N : g 20S 40S sow 40W 20W eu 20E 40E SOE Figure 5.48 1000 hPa v-wind standardized regression coefficients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 170 of significant coefBcients over the North Atlantic Ocean representing the anomalous flow about the Icelandic Low and Azores High aflfecting the v-wind over Greenland northwestern Afiica and over eastern North America and over much ofNorthem Europe, Scandinavia, over the Baltic Sea and Northern Russia. Alternating areas of southerly and northerly flow occur around SAF with significant southerly flow occurring over eastern South Afiica and extends to the southeast, an area with significant northerly flow is observed over Madagascar. The large v-wind components observed extending fi'om the western Indian Ocean over South Afiica to the eastern South Atlantic are consistent with the wavetrain observed in the SLP field (Fig.5.46) and in the streamlines (Fig.5.47). The u-wind component at 1000 hPa (Fig.5.49) has extensive and statistically significant coefiBcients over the North Atlantic and relatively small magnitude coefiBcients over the Southern Hemisphere. The intensified Icelandic Low induces significant easterly flow to its west over Greenland, and intensified westerly flow across the North Atlantic Ocean to its south. Zonal flow anomalies in the vicinity of SAF have some areas with statistical significance, including significant easterly flow extending fi'om Botswana to over the Ocean south of the continent to the east of SAFR. The significant coefiBcients from Botswana to the south of the continent are also an aspect of the wavetrain observed in the SLP field (Fig.5.46) and in the streamlines (Fig.5.48). Inflow into the low over Botswana at 25°E, 15°S associated with the southeasterly flow over southeastern South Afiica is evident in the 850 hPa streamlines associated with wet summers over the region (Fig.5.50). At this level the wavetrain is again evident in the streamlines and produces northerly flow over Madagascar, southeasterly flow over eastern 171 80 N BON SON SON 40N --a ^ 40N 20N : 20N EQ 20S -— 20S 40S 40S SOW 40W 20W GM 20E 40E SOE Figure 5.49 1000 hPa u-wind standardized regression coefficients associated with SAFR, 1958-1989. Negative isoiines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 172 SON 40?== 60W 40W 20W GM 2CE 40E 60E Figure 5.50 850 hPa Streamlines associated with SAFR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and v-wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 173 South Africa, easterly flow over much of SAP and northerly flow over the eastern South Atlantic Ocean. The dynamics associated with the easterly short wave (northerly flow associated with convergence and southerly flow with divergence) produces convergence at this level over Madagascar, divergence over the eastern escarpment of South Africa and over central South Africa, and the strong northerly flow over the eastern South Atlantic induces convergence. The dynamically induced convergence o f the relatively cold and moist Indian Ocean air masses over the eastern escarpment of South Africa is likely to produce rain over this region after uplift. Statistically signifrcant v-wind coefiBcients at 850 hPa (Fig.5.51) are observed over southeastern South Africa and are again coupled with a significant northerly flow to the east over Madagascar. 850 hPa u-wind coefiBcients associated with wet summers in SAF (Fig.5.52) are very similar to the distribution observed at 1000 hPa (Fig.5.49). The 300 hPa streamlines associated with wet summers in SAF (Fig.5.53) depict strong convergent southeasterly flow over the western Indian Ocean between about 18“S and 22°S from about 60°E to 40°E and strong northeasterly flow occurs over eastern SAF. At lower latitudes strong easterly flow represents an enhancement of the climatological equatorial easterlies (Fig.4.13), but at higher latitudes it represents a weakening of the climatological westerlies over SAF consistent with the results of Van Herdeen et al. (1988). The easterly propagating wavetrain is easily distinguished in the curvature of the streamlines and dynamically induces divergence over Madagascar and convergence over eastern South Africa at this level during wet SAF summers. Divergence over Madagascar coupled with dynamically induced surface and near surface convergence, and 300 hPa convergence with low level 174 80N SON SON - SON • / / ,^ 3 \ 3 40N 40N 20N m H#a a ' D 20S 40S SOW 40W Figure 5.51 850 hPa v-wind standardized regression coefficients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 175 8 0 N BON 60N - f c BON 40N -■, 40N 20N ^ 20M EQ 20S "• 20S 40S 40S 60W 40W 20W GM 20E 40E 60E Figure 5.52 850 hPa u-wind standardized regression coefficients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 176 80N |80K o sow 40E 60E Figure 5.53 300 hPa Streamlines associated with SAFR, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 177 divergence over South Africa, possibly accounts for the OLR dipole between these regions described by Jury and Pathack (1993). The 300 hPa v-wind components (Fig.5.54) are northerly over SAF but are not significant. Signifrcant anomalous u-wind components at 300 hPa (Fig.5.55) cross large expanses of both hemispheres to 20°S and represent a series of zonally oriented anomalies of alternating sign. The u-component of the Southern Hemisphere’s circulation at 300 hPa associated with wet SAF summers (Fig.5.55) shows highly significant easterly flow anomalies across much of the tropics. PWAT coefiBcients associated with wet South African summers (Fig.5.56) are more pronounced over the North Atlantic than in the vicinity of South Africa. A small area of significant coefficients is observed over Lake Victoria and is associated with the entrainment of dry desert air from the northwest observed in the anomalous near surface circulation (Figs.5.47 and 5.50). An area with statistically signifrcant PWAT coefficients encompasses southern Madagascar where flow is easterly from the Indian Ocean onto South Afiica in the near surface layers of the atmosphere (Fig. 5.47 and Fig.5.50). Small positive coefficients are observed along the west coast of South Africa extending to the north of the equator associated with the enhanced flow from the northeast over this area. Specific humidity coefficients at 1000 hPa and 850 hPa (not shown) associated with wet summers in South Africa have a very similar distribution to the PWAT coefficients (Fig.5.56) including an area of significance south of Madagascar. OLR coefficients associated with wet South African summers (Fig.5.57) are characterized by small patches of significant variations scattered over the globe. South Africa 178 80K TqBON SON 1— SON 40N m - r— 40N 20N 2 A - 20N - EQ 20S 20S 40S 40S SOW 40W 20W GM 20E 40E 60E Figure 5.54 300 hPa v-wind standardized regression coeffîcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 179 SON SON SON SON 40N 40N 20N 20N EQ 20S — 20S 40S 40S SOW 20E 40E Figure 5.55 300 hPa u-wind standardized regression coefficients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 180 SON SON 60N SON 40N ^ 20N ^ r^3g— 20N —J————— ———— — — - EQ 20S — ^ 20S 40S 40S SOW 40W 20W GM 20E 40E 60E Figure 5.56 PWAT standardized regression coefficients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 181 80N BON 60N BON 40K 40N 20N 20N EQ 20S 20S 40S 40S 60W 40W 20W GM 20E 40E 60E Figure 5.57 OLR standardized regression coeffîcients associated with SAFR, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 182 is covered by positive coefficients with small magnitude indicating a slight decrease in convective cloud heights over the region during wet summers. This implies that the increased precipitation over South Afiica is non-convective in nature and is possibly related to enhanced orographic rainfall over the eastern escarpment of South Afiica. Standardized rainfall receipt at individual stations in southeastern Afiica having significant correlations to the NAO are used here to create a regional rainfall index (Fig.5.5). At aU levels in the atmosphere the response to SAFR is linked to large circulation anomalies over the Northern Hemisphere, including strong westerlies over the North Atlantic Ocean associated with wet SAF summers (Figs.5.47, 5.50 and 5.53), reflecting the large positive correlation (r =0.44) between SAFR and the NAOI. An easterly propagating wavetrain is observed in the streamlines (Figs.5.47, 5.50 and 5.53), and u- and v-wind components at all levels (Figs.5.48, 5.49, 5.51, 5.52, 5.54, 5.55). The easterly wave induces convergence over Madagascar, and the eastem South Atlantic Ocean and divergence over SAF at 1000 hPa and 850 hPa (Figs.5.47 and 5.50) inconsistent with increased rainfall receipt over SAF. The wave induces divergence at 300 hPa over Madagascar, and the eastem South Atlantic Ocean and convergence over SAF consistent with the conservation of mass within the atmospheric column. The strong southeasterly flow at 1000 hPa and 850 hPa (Figs.5.47 and 5.50) advects relatively moist and cold air masses off the Indian Ocean over the elevated eastem escarpment of eastem South Africa. This uplift produces enhanced orographic rainfall and potentially destabilizes the conditionally unstable air masses present in the atmosphere over SAF. Convective cloud heights (Fig.5.57) are not statistically significant over SAF implying that enhanced convective rainfall is not associated with wet summers over SAF. 183 The dynamics associated with the wave are consistent with the OLR dipole observed between Madagascar and South Africa in composite differences between wet and dry South African summers from 1975-1984 (Jury and Pathack, 1993) thought to represent a branch of the Walker circulation between the two regions. Composite differences of the circulation at 200 hPa between the same wet and dry South African summers depicts southeasterly flow from the western Indian Ocean over SAF. However no other similarities between this study and Jury and Pathack's (1993) composites are evident at this level or at 850 hPa. 184 5.8 Correlation analysis between Global Precipitation and El Nino - Southern Oscillation during the Boreal Winter. The relationship between ENSO and precipitation across the globe and over Africa south of the equator is now described. Correlation coefiBcients are calculated to quantify the ENS0-precipitation relationship during boreal winter (DJF). The standardized Nino4 SST index is used an indicator of the phase and intensity of EN (Trenberth, 1996) and seasonal rain&U from meteorological stations across the globe is used in this analysis. The statistically significant correlation coefiBcients (Pig.5.58) indicate very strong and coherent unlagged relationships between regional precipitation patterns across the globe and ENSO during boreal winter. The spatial coherence of the correlation coefiBcients during DJF (Fig.5.58) compares well to the regional precipitation-ENSO relationships described in Ropelewski and Halpert (1996). Northern North America has a strong negative response to ENSO from British Colombia through Alberta, Saskatchewan and Manitoba to the Great Lakes and extending to the south, with the exception o f a few isolated stations. In particular, the western Canadian coast is observed to have a large cluster of stations exhibiting significant negative relationship to ENSO. Stations as far to the south as Mexico depict a strong positive relationship to ENSO. Distinct bands of stations with positive correlations are picked out across North America, the southern portion o f the eastem seaboard o f the United States is highlighted as is a band covering much of the southwest and west of the United States and much of Mexico. An isolated negative response to ENSO appears in the Caribbean. Significant lag zero negative correlations are observed in Central America over much of the coastal area of northern South America. The negative correlations extend along the 185 coast o f northeast o f Brazil to about 8° S, south of which a region of coherent positive correlations is observed extending along the coast and somewhat inland. Significant negative correlations are also observed over southern South America between about 30° S to 40° S with isolated negative correlations evident poleward. The Hawaiian Islands have significant negative correlations but central and equatorial Pacific stations have positive correlations to Nino4 as the ocean is warming this area. Areas west of 170°W, away fi'om the equator, including Australia and Indonesia exhibit a significant negative response to the ENSO signal indicative of the droughts accompanying El Nino. Positive correlations cover much of Eurasia from Iberia through Siberia to coastal south China, with isolated negative correlations over northern Europe and Scandinavia and in the vicinity of Japan. Northern Afiica and some of the Middle East have isolated stations with negative correlations to ENSO; this is especially evident in the vicinity of the Gulf o f Guinea and further inland in sub-Saharan Afiica. The largest clusters of significant correlations to ENSO include a cluster of significant positive coefficients straddling the equator in eastem Afiica and a cluster of significant coefficients over Southern Africa. The negative correlations over southern Afiica occur along the western shore of Lake Malawi, to the southwest in Botswana, Zimbabwe and Zambia and also along the coast of Namibia and over much of South Afiica. 186 BON BON 20N 00 < 1 20S 20S BOS 60S 180W BOW 20W BOB lOOE 140E 160E Figure 5.58 Significant correlations between Nino4 and global precipitation during DJF; + and - refer to the sign of the relationship; large signs represent stations significant at the 99% confidence level, small signs refer to stations significant at the 95% confidence level. Stations not significantly correlated are not shown. 5.9 Regressions of the Nino4 SST Index Description and analysis of the rainfall response over Southern Africa to ENSO forcing is examined by regression of the standardized Nino4 SST index (XJ onto standardized NCEP/NCAR reanalysis data (Y). Global circulation anomalies associated with ENSO are described and the anomalous circulation in the vicinity of the African continent is related to the observed precipitation correlations. The 1000 hPa geopotential height coefficients (Fig.5.59) depict negative coefficients stretching northeast from Central America parallel to the North American coastline. The southern and equatorial Atlantic is covered by a large area of significant positive height anomalies between 15° N and about 30°S which extends across equatorial Afiica to the Indian Ocean and into the Indian Ocean. The significant positive coefficients become smaller in magnitude and statistically insignificant over most of Africa south of 15° S, with the exception of Madagascar which is covered by significant positive heights. Negative coefficients are observed to the south of the African continent, extending from the South Atlantic to the Indian Ocean to the south of 30°S. The significant positive height coefficients observed along the western coast of South Africa appear to correspond to the height anomalies observed during individual wet DJFM months over South Africa (VanHeerden eta l, 1988). This height distribution weakens (strengthens) the climatological westerly flow over South Africa during individual wet (dry) DJFM months. Streamlines at 1000 hPa (Fig.5.60) show the anomalous circulation associated with a 1.0 standard deviation increase in the Nino4 SST index. Southeastern Afiica is dominated by anomalous northerly flow south of about 10° S related to outflow from the high pressure system over Madagascar. This outflow also causes anomalous southeasterly flow over 188 La Nina Weak La Nina and El Nino Winters El Nino Winters Winters r -0.23 -0.59 -0.55 Observations 20 46 25 per category Table 5.5 Correlation coefficients between SEAR and NAOI stratified according to the Nino4 SST index. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined. The number of winters per category of Nino4 are also given. 189 La Nina Weak La Nina and El Nino Years El Nino Years Years r 0.41774 0.46585 0.31836 Observations 20 46 25 per category Table 5.6 Correlation coefficients between SAFR and NAOI stratified according to the Nino4 SST index. Correlations significant at the 95% level are in bold and correlations significant at the 99% confidence level are indicated by bold and underlined. The number of winters per category of Nino4 are also given. 190 coastal eastem Africa where the flow parallels the coast and extends as far North as Eritrea and Ethiopia. The area of cyclonic convergence over Angola (Fig.4.3) is missing but the anomalous northeriy flow (Fig.5.60) over that region represents an enhancement of the mean flow (Fig.4.3). Surface divergence is observed in the vicinity of the significant positive correlations over a large region of eastern Africa just to the south of the equator and over Kenya to the north of the equator. Strong southwesterly flow over western North Africa extends far into central North Afiica and represents a weakening of the climatological mean trade winds (Fig.4.3), as occurred over the tropical North Atlantic Ocean. The 850 hPa streamlines associated with the Nino4 SST index (Fig.5.61) show strong northwesterly flow across Namibia and Botswana, extending to the Highveld region of northeastern South Afiica. The offshore flow is observed far to the south of Madagascar and recurves in a wavelike manner becoming southwesterly. The wave propagates through the climatological westerlies to the east consistent with the geopotential height field at 1000 hPa (Fig.5.59). The dynamics associated with this westerly wave over Namibia, Botswana and South Afiica (Fig.4.15) produce divergence at 850 hPa associated with a strong northerly component to the flow, and convergence over Madagascar associated with a strong southerly flow component. SEA is dominated by air masses from the Indian Ocean associated with the significant positive correlations in the vicinity of the equator (Fig.5.58), while SAF is dominated by an enhanced westerly component to the flow associated with the significant negative correlations observed over SAF. The 850 hPa u-wind coefiBcients associated with Nino4 (Fig.5.62) are consistent with the wave observed in the streamlines at this level over SAF and depict statistically significant 191 80N 0ON 60N 60N 40N — 40N 20N 20N EQ 20S --- 20S 40S 40S SOW 40W 20W GM 20E 40E 60E Figure 5.59 1000 hPa Geopotentlal Height standardized regression coefficients associated with Nino4,1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 192 SON BON BOW ^ •• 40W 20W GM 20E 40E 60E Figure 5.60 1000 hPa Streamlines associated with Nino4, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 193 BON BON 60N> 40N 20N 20S 40S 60W 40W 2GW GM 20E 40E 60E Figure 5.61 850 hPa Streamlines associated with Nlno4, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 194 80N BON 60N — %: BON 40N — 40N 20N 2 20S 40S 60W 40W 20W GM 20E 40E 60E Figure 5.62 850 hPa u-wind standardized regression coefficients associated with Nino4,1958-1989. Negative isolines are broken; the isopleth interval Is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 195 coeflScients over SAF and the South Atlantic. Significant westerly flow is also observed over SAF associated with the strong northwesterly flow anomaly observed over this region (Fig.5.62). No statistically significant coefiBcients are observed over SEA, but statistically insignificant coefiBcients indicate easterly winds over most o f SEA, with the exception of a relatively small area o f westerly flow associated with Nino4 over the equatorial area having significant positive precipitation correlations to Nino4 (Fig.5.58). The 850 hPa v-wind coefiBcients associated with Pacific warm events (Fig.5.63) depicts significant southerly flow occurring between the South Atlantic high and the west coast of Angola. Significant northerly flow is observed fi'om Angola to northeast South Afiica over the area o f strong northwesterly flow in the streamlines. Southerly flow occurs in the vicinity of Madagascar consistent with the streamlines (Fig.5.60). Upper level flow associated with Nino4 is described by streamlines at 300 hPa (Fig.5.64). Strong westerly flow anomalies occur over Afiica south of the equator, with northwesterly flow over the lowest latitudes where easterlies should prevail (Fig.4.13). Strong southwesterly flow occurs over southern Africa, enhancing the climatological westerlies and associated with decreased geopotential heights (Fig.5.59). This agrees with strengthening (weakening) of the westerlies found in association with dry (wet) summers in South Afiica (VanHeerden et. al., 1988) and the general prevalence of westerly flow across the tropics associated with ENSO. The anomalous westerly flow over equatorial Afiica and the Indian Ocean is due to the disappearance or weakening of the mean anticyclone at 20°S, 20“E (Fig.4.13) and disappearance of the mean Somali anticyclone at IO“N, 50“E during warm episodes. Strong westerly flow is observed fi'om South America across the equatorial 196 BON BON BON 3 ^ BON 40N 40N 20N 20N O EQ 20S 40S 40S 60W 40W 20W GU 20E 40E 60E Figure 5.63 850 hPa v-wind standardized regression coefficients associated with Nino4,1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 197 i80K 60K kON 40S' 6 0 ¥ 40W 20W GM 20E 40E 60E Figure 5.64 300 hPa Streamlines associated with Nino4, 1958-1989. Light and dark shading represents the statistical significance of the average u- and V - wind coefficients’ t-scores at the 95% and 99% confidence levels respectively. 198 Atlantic Ocean to West Africa related to flow about an anticyclone over the tropical South Atlantic (10°-30°S, 20°-40“W) and a strong cyclonic center over the coast o f West Africa (centered on 15“N, 20“W). Strong northwesterly flow from the central South Atlantic to the east of the anticyclone over the tropical South Atlantic produces a ridge of warm air over the eastern South Atlantic. The southwesterly flow over the southern part of South Africa builds a small trough of cooler air over South Africa and over SAP in particular. Northwesterly flow from the continental interior of southern Africa over SAP as far as Madagascar produces a ridge over this region. The juxtaposition of ridge, trough, ridge over mid-latitude southern Africa produces a wave-like flow at 300 hPa associated with Nino4. The westerly wave observed over Namibia, Botswana and South Africa at 850 hPa associated with Ninc4 (Pig.5.61) is oriented differently at 300 hPa (Pig.5.64) where flow over Namibia, Botswana and South Africa, in particular, is primarily from the southwest which dynamically produces convergence at this level. Between South Africa and Madagascar the streamlines recurve in a wavelike manner and dynamically induces divergent flow extending to the southeast of Madagascar. The upper level dynamics result in convergent flow above SAP where low level divergence is observed (Fig.5.61. Upper level support (Pig.5.64) reinforces the low level divergence observed over of Namibia, Botswana and South Africa (Pig.5.61) associated with Nino4. This is wholly consistent with the negative correlations observed across this region between the Nino4 SST index and DJP precipitation. Analysis of the u-component at 850 hPa (Fig.5.65) shows that zonal wind variations over all but southernmost Africa, south of the equator, are not significant. Extensive areas of statistically significant cross-equatorial westerly flow is observed from South America, 199 over the tropical Atlantic Ocean to the Arabian Peninsula, and over the Indian Ocean between 20“S to 20°N. Meridional flow coeflScients associated with Nino4 (Fig.5.66) depict a series of coeflScients with alternating sign extending fi'om the South Atlantic across much of southern Afiica to the Indian Ocean to the east of Madagascar consistent with a wave-like ridge-trough pattern over this region. Significant specific humidity anomalies at 1000 hPa associated with Nino4 (Fig.5.67) are observed throughout the tropics, and also in the mid-latitudes of the Northern Hemisphere. Positive coeflScients extend across Central America and parts of South America into the equatorial Atlantic Ocean, extending northeast to the western coast of North Africa and Iberia. A large area o f significant positive 1000 hPa specific humidity coefficients is observed over the equatorial Atlantic Ocean between 10° N and 15° S. A significant increase in specific humidity is observed over equatorial Afiica to the south cf the equator including negative différences to the south (over Angola, Namibia and South Afiica). An extensive area having large humidity coefficients is observed to the southwest of South Afiica in the region of strong northwesterly flow at this level (Fig.5.60) which advects relatively warm and moist air masses into the temperate regions to the south consistent with an enhanced subtropical high over the South Atlantic. The enhanced flow of moist Indian Ocean air masses into SEA at 1000 hPa (Fig.5.60) produces a negative PWAT anomaly over central Afiica fi’om just north of the equator to 10°S (Fig.5.68). With this exception statistically significant PWAT coefficients are not observed over Afiica south of the equator. A significant negative PWAT center is observed over the western South Atlantic (centered at 20°S, 35°W) to the west of the near surface 200 SON 80N SON 3 ^ 60N 40N — 40N 20N 20S 40S 60W 40W 2 0 ¥ GM 20E 40E 60E Figure 5.65 300 hPa u-wind standardized regression coefficients associated with Nino4,1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a Mest. 201 8 0 K BON 60N V ------BON 40N - 40N 20N 20N % EQ 20S z - z ^ 20S 40S 40S BOW 40W 20W CM 20E 40E 60E Figure 5.66 300 hPa v-wind standardized regression coefHcients associated with Nino4,1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 202 SON SON SON -3-- SON 40N 40N 20N 20N ? EQ 20S 20S 40S 40S SOW 40W 20W GM 20E 40E 60E Figure 5.67 1000 hPa Specific Humidity standardized regression coefficients associated with N ino4,1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 203 anticyclone where northwesterly flow predominates. A significant positive PWAT anomaly is observed to the southwest of this in a region dominated by near surface southeasterly flow (Figs.5.60 and 5.61). A large significant positive PWAT anomaly extends northeast fi'om the coast of South America into the central Atlantic Ocean between the equator and 20°N in a region characterized by divergent flow at 1000 hPa (Fig.5.60). Significant negative PWAT coefficients are observed parallel to the coast over the Gulf of Guinea and extend into equatorial Zaire. PWAT coefficients over the North Atlantic Ocean are very similar to those associated with the NAO (Fig.5.32). OLR coefficients associated with the Nino4 SST index (Fig.5.69) have limited significance over equatorial and southern Afiica with the exception o f immediately south of Madagascar and over an area of eastern South Afiica, both of which have significantly decreased convective cloud heights (positive coefficients). However a statistically insignificant increase in convective cloud heights (negative coefficients) is observed over and to the west of SEA where rainfall have significant positive correlations with the Nino SST index (Fig.5.58). The 300 hPa humidity coefficients associated with Nino4 depict much of the tropical and subtropical South Atlantic dominated by statistically significant positive coefficients which extend to the west coast o f South Afiica (Fig.5.60). A significant positive humidity anomaly is observed over southwest Afiica covering northern Angola and southwestern Zaire which extends into southcentral Zaire just west of SEA consistent with the OLR coefficients for this region (Fig.5.69). Statistical removal of S.O.I. variability from SST data was shown to have an effect on South Afiican rainfall-SST correlations (Walker, 1990). Walker performed a composite 204 60N BON 60N BON 40N 40N 20N 20N EQ 20S — 20S > V J i\ ; I / ' " 40S 40S 60W 40W 20W GM 20E 40E 60E Figure 5.68 PWAT standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test 205 BON 80N o ’ BON BON 40N 40N SON SON EQ EQ SOS SOS 40S 40S BOW 40W GMSOW 40ESOE 60E Figure 5.69 OLR standardized regression coefficients associated with Nino4, 1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 206 analysis and selected wet and dry South African summers having a relatively small value of the S.0.1. so that ENSO variability could be removed from the SSTs. EQs results (see section 2.3) show quite different relationships between the circulations during wet and dry South African summers before, and after, removal of ENSO variability. A comparable analysis is performed here to examine the effect of the removal of the ENSO signal on the magnitude of the NAOI-rainfall correlation over SEA The standardized NAOI and SEAR were stratified according to the phase and magnitude of the Nino4 SST index and correlations between the indices were performed for each category. Winters having Nino4 values ^ -0.5 (La Nina winters), between -0.49 and 0.49 (intermediate winters), and k 0.5 (El Nino winters) were stratified and correlations between SEAR and the NAOI were carried out for each category of the Nino4 SST index (Table 5.5). The statistically significant decrease (increase) in rainfall over SEA during positive (negative) NAO events is enhanced during years in which the ENSO signal is near zero and strongly positive (Table 5.5). This analysis was repeated with the NAOI correlated to SAFR for positive, negative and intermediate Nino4 cases (Table 5.6) showing a statistically significant correlation between SAFR and the NAOI during intermediate Nino4 cases, and an almost significant correlation is observed during La Nina events. Increased (decreased) rainfall over SAP during positive (negative) NAO events is enhanced during years in which the Nino4 signal is near zero and during La Nina events (Table 5.6). Both of these analyses imply that the phase and magnitude of ENSO modulates the strength of the correlation between SEAR, SAFR and the NAOI. In summary, the precipitation-Nino4 relationship over SAF can be explained in terms of circulation. The relationship between wet (dry) South African summers and anomalous 207 80N SON 60N : 60N 40N # 40N 20N 20N EQ 20S 20S 40S 40S SOW 40W 20W GM 20E 40E 60E Figure 5.70 300 hPa Specific Humidity standardized regression coefficients associated with Nino4,1958-1989. Negative isolines are broken; the isopleth interval is 0.1 (std/std); dark and light shading represents statistical significance at the 99% and 95% confidence level using a t-test. 208 easterly (westerly) flow (Van Heerden et aL, 1988; Jury and Pathack, 1993; Jury, 1996) is evident in the streamlines (Fig.5.60 and Fig.5.61) and 1000 hPa geopotential height coeflScients (Fig.5.59) associated with Nino4, consistent with the large negative correlations observed between Nino4 and summer rainfall. Enhanced westerly flow evident throughout the atmosphere (Fig.5.47 and Fig.5.51) associated with a 1.0 standard deviation increase in Nino4 overlies the significant negative precipitation-Nino4 correlations. Increased geopotential heights (Fig.5.59) are observed over South Africa with decreased heights to the south causing enhanced westerly flow associated with ENSO. These findings are very similar to what Van Heerden et al. (1988) describe. 209 5.10 Eastern African Hydrological response to Precipitation Variability Variability in the levels of the great lakes of SEA are known to be related to the variability of precipitation over their respective catchment areas (Grove, 1996). The lake levels are related to input of water through runoff and inflow from small rivers and streams and loss of water through evaporation (Grove, 1996). Human factors also can play a role in determining the levels of lakes through damming water and the extraction of water for irrigation. However the role of humans is not thought to play a role in the levels of the Great Lakes of Eastern Africa (Grove, 1996) and these lake levels are primarily determined by the interannual variability of precipitation and runoff over the catchment region. A dramatic increase in the level of Lake Victoria (Fig. 5.71), which is located in the far north of the LMR, occurred after the heavy rains of late 1961 (Rodhe and Viiji, 1976; Grove, 1996). The early 1960s are associated with a change from a wet to a dry epoch in the climate regime over South Africa (Tyson, ,1986; Kruger, 1999). This shift from a wet to a dry rainfall epoch over South Afiica and a dramatic increase in the level o f Lake Victoria, associated with a positive rainfall departures over eastern Africa, is consistent with the correlation coefficients between SEAR, SAFR and the NAOI. A gradual but steady decrease in the level of Lake Victoria is observed after the peak in 1964 to the late 1980s. The correlation between the level of Lake Victoria and SEAR is 0.23 (Table 5.7) over the 91 year period from 1899 to 1989 and is statistically significant at the 95% level. The correlation between the level of Lake Victoria and the NAOI for the same period is -0.32 (Table 5.7), significant at the 99% level. Grove (1996) suggests that Lake Victoria’s level had a quasi periodicity of about 11-years before this dramatic rise in levels in the early 1960s which is 210 substantiated by spectral analysis (performed during the research but not shown here) where relatively large variance occurs at the 11 year period which is not statistically significant. Cospectral analysis between the NAOI and the level of Lake Victoria (also performed during the research but not shown here) shows a small power at about 7 years and strong power at extremely low fi-equencies. The covariance between the standardized level of Lake Victoria and the NAOI (Fig.5.72) can be characterized by two distinct periods. The period between 1899 and 1962 has a negative relationship to the NAOI, with a correlation coefficient of -0.34. The period between 1963 to 1989 features a series of extremely large negative extremes, the largest of which occurs in 1963, which combine to produce a correlation of -0.52. The covariance between the NAOI and SEAR (Fig.S.4) also shows a change in their relationship in the early 1960s, suggesting the hydroclimatatological regime of eastern Afiica underwent a change or extremely large perturbation in the early 1960s as is suggested by Fig.5.71 and in the literature (Flohn, 1987; Shukla, 1987; Grove, 1996). In his analysis of the lake level record in eastern Afiica, Grove (1996) does not suggest that human influence is an important factor in the lake level record, but it is a possibility. Flohn (1987) describes an extremely large precipitation anomaly stretching fi'om over 30° E (over Eastern Afiica) almost to the Indian subcontinent (70° E) during the heavy rains of 1961/2. Shukla (1987) describes precipitation receipt greater than 130% o f the climatological average over the northwest and a decrease to the northeast of the Indian subcontinent during this period. The covariance between SEAR and the level of Lake Victoria (not shown) is almost the same relationship but with opposite sign and will not be commented upon. 211 - 1 -2 - 3 1890 1900 1910 1920 1930 1940 1950 1980 1970 1980 1990 Year Figure 5.71 Standardized level of Lake Victoria, 1899-1989. 212 - 1 -2 - 3 - 4 -5 -6 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 Year Figure 5.72 Covariance between level of Lake Victoria and the NAOI, 1899-1989. 213 SEAR NAOI Lake Victoria Lake Malawi Lake Tanganyika SEAR 1.00 NAOI -0.48 1.00 1.00 Lake Victoria 0.23 -0.32 000100 1 Lake Malawi 0.32 -0.27 0.64 1.00 Lake Tanganyika -0.41 -0.50 0.91 0.43 1.00 Table 5.7. Correlations coefficients between East African Lake Levels, the NAOI and the African precipitation index, 1899-1989. Correlations significant at the 95% level are in hold and correlations significant at the 99% confidence level are indicated by bold and underlined. Data for Lakes Malawi and Tanganyika are only available for the period 1948-1978 inclusive. 214 The levels of Lakes Tanganyika and Malawi are also significantly related to precipitation variability over SEA (Table 5.7), with correlations of 0.41 and 0.33 respectively. These smaller lakes to the south of Lake Victoria also show the large rise in levels in the early 1960s. They are also both significantly correlated to the NAOI with correlations of r=-0.50 and r=-0.27 respectively (Table 5.7). Strong statistically significant regional covariability between the lake levels is observed, as each lake is positively correlated to the others (Table 5.7). This implies that the levels of each of these great lakes of Eastern Africa are driven by similar precipitation and climate anomalies. The hydroclimatological impact of the Afiican rainfall index and the NAOI is evaluated with respect to the levels of Lake Victoria, Lake Malawi and Lake Tanganyika. A strong positive correlation between rainfall and the lake levels and a strong negative relationship to the NAO are observed (Table 5.7). The levels o f the individual lakes are strongly positively correlated implying that they possibly share common hydroclimatological forcing mechanisms. 215 CHAPTER 6 CONCLUSIONS AND FUTURE RESEARCH The aim of this study was to describe and analyze associations between large scale atmospheric teleconnections and regional scale precipitation variability over south and southeastern Africa. The nature of the teleconnection relationships were outlined using correlation analysis, indicating the magnitude and direction of the relationship between both NAO and ENSO and African rainfall receipt during the boreal winter. Regional precipitation indices were constructed to summarize and average the precipitation receipt at stations significantly correlated to the NAOI or to ENSO (the Nino4 SST index). The resulting teleconnection-precipitation relationships are then examined using regression analysis to depict circulation anomalies statistically associated with regional precipitation indices. Univariate and bivariate spectrum analysis was used to describe the quasi-periodic variability and covariabilily between regional precipitation indices and the atmospheric teleconnections. This study used the NCEP/NCAR reanalysis dataset, having data for the mandatory atmospheric levels in addition to the surface. Precipitation data are obtained from the GHCN dataset o f7533 global meteorological stations with precipitation records for at least 10 years. 216 Results presented in Chapter 5 depict large scale circulation anomalies associated with the interannual variability of precipitation over Afiica south o f the equator. The tuning of the seasonality of the rainfell maximum over regions of southern and southeastern Afiica is found to be important in determining the magnitude and statistical significance of the correlations between the NAOI, the Nino4 SST index, and DJF precipitation. The large negative relationship between SEAR-NAO is strongly related to reduced (enhanced) penetration of moist Zaire air masses strongly associated with precipitation over SEA during the positive (negative) NAO phase. An easterly propagating wave pattern dynamically inducing uplift (subsidence) over SEA, and subsidence (rising air) to the west is also observed during the positive (negative) NAO phase but is thought to be secondary in importance to the near surfece flow of Zaire air. This anomalous atmospheric circulation is related to a large increase (decrease) in precipitable water over SEA and unusually low (high) OLR, inversely related to convective cloud heights, during the negative (positive) NAO phase. Positive correlations between precipitation over South Afiica and the NAO are associated with inflow and convergence of air fi'om both the Atlantic and Indian Oceans into a surface low pressure system located over the Highveld rather than easterly flow from the Indian Ocean, which is the usual source of moisture for the summer rainfall region. Streamlines at the 1000 hPa level during DJF periods with above normal precipitation over SEA show weakened trade winds over both the North and South Atlantic, and over the southwestern Indian Ocean. Enhanced convergence at 1000 hPa and 850 hPa into the intensified low pressure system over Angola, just southwest o f its climatological mean position and two smaller near surface cyclonic systems over Southern Afiica, is associated 217 with the enhanced advection of moist Zaire air over SEA and a decrease in the advection of dry monsoonal air from the northeast and air masses from the Indian Ocean over SEA. The westerly flow of moist Zaire air is especially associated with precipitation over SEA due to near saturation after uplift, and is possibly due to a destabilization of a neutral westerly flow near the equator by the vertical component of the Coriolis parameter. The circulation anomalies associated with precipitation variability over SAF is explained primarily with reference to near surface circulation anomalies. The strong relationship between wet (dry) SAF summers and easterly (westerly) flow (Van Heerden et a l, 1988; Jury and Pathack, 1993; Jury, 1996) is evident in the streamlines associated with ENSO and the strong negative correlations observed between ENSO and summer rainfall. The enhanced westerly flow evident throughout the atmosphere associated with a 1.0 standard deviation increase in the Nino4 SST index overlie the significant negative precipitation-ENSO correlations. Increased geopotential heights are observed over South Afiica with decreased heights to the south causing the enhanced westerly flow associated with ENSO. A westerly (easterly) propagating wave is observed at 850 hPa and in the upper level circulations inducing divergence (convergence) over SAF at 850 hPa supported by 300 hPa divergence (convergence) during El Nino (La Nifia) events consistent with the significant negative precipitation relationship. Geopotential height coefiBcients throughout the atmosphere increase (decrease) across this region and surface pressures are also observed to increase (decrease) during El Nino (La Nina) events consistent with the direction of the wave. The circulation anomalies relationship between SAFR and the NAOI depicts enhanced near-surface advection of moist and relatively cold Indian Ocean air masses from southeast 218 over the eastern escarpment of South Africa producing orographic uplift and possibly destabilizing any conditionally unstable air masses. However the dynamics associated with the easterly propagating wave over SAF produces divergence at both 1000 and 850 hPa and convergence at 300 hPa, inducing subsidence throughout the atmosphere and acts against precipitation production. The circulation anomalies associated with positive precipitation receipt at the 11 stations in SAF significantly correlated to the NAOI therefore do not appear to be very useful in explaining the mechanisms through which precipitation is generated over SAF. The relationship between SAFR and the NAOI is modulated by the phase and intensity of ENSO, with large correlations occurring during years with intermediate and strong positive Nino4 values. The magnitude and significance of the correlation is reduced during La Nina years (Tables 5.5 and 5.6). Spectral and cospectral analysis is used to quantify the dominant periodicities and their interrelationships between the NAOI, ENSO and SEAR. The spectra and cospectra show large spectral powers between variables representing the phase and intensity of all three phenomena at a series of periods, but a periodicity of 3.79 years is consistently evident in the cospectra of the three atmospheric phenomena and possibly represents the periodicity at which the three are most strongly interrelated. The NAOI and SEAR appear to have large covariance concentrated at periodicities of 7 years, corresponding to the strongest quasi- periodic signal found in the spectrum of the NAOI. The hydroclimatological impact of the Afiican rainfell index and the NAOI is evaluated with respect to the levels of Lake Victoria, Lake Malawi and Lake Tanganyika. Strong positive correlations between SEAR and the lake 219 levels and a strong negative relationship between the lake levels and the NAO are observed. The levels of the individual lakes are strongly positively correlated (Table 5.7) implying that they possibly share common hydroclimatological forcing mechanisms. During the course of this study the lack of understanding o f the climatological and meteorological situation of southern and southeastern Africa became apparent. Associations between African precipitation receipt and circulation features were not clearly explained and the various known associations could not be frtted into a consistent theoretical framework. Now that precipitation variability over SEA and SAF have been strongly linked to the NAO and ENSO identification of actual causal mechanisms by which precipitation variability over Africa south of the equator is related to interhemispheric climate fluctuations becomes an important issue. An ideal course of research of future research should include investigations using a general circulation models; including experiments to examine the influence of tropical SST variability on precipitation receipt over these regions through modulation of the location or intensity of the large scale features o f the Hadley circulation which dominate the climates of these areas. Future research suggested by this study includes analysis of the relation between precipitation variability and location and intensity of convergence zones over southern and southeastern Africa using daily surface and upper level data. Examination of precipitation variability in relation to monsoonal variability over Africa south of the equator is also interesting but, again, precipitation and surface wind data are necessary on daily timescales, and over relatively small spatial scales. The relationship between topography and precipitation variability over this part of the world is likely to be substantial and could be 220 examined using a general circulation model with and without topography to determine the orographic influence on rainfall variability. Atmospheric circulation mechanisms producing the large coherent clusters of significant NAOI-precipitation correlations across the globe also require analysis. 221 BIBLIOGRAPHY African Data Dissemination Service, 1999; African reference map [WWW], http://edcintl.cr.usgs.gov/adds/cl/doc/maps/cl.gif. (Accessed 8/6/1999). Alexander, M.A. and J.D. Scott, 1997: Surface flux variability over the North Pacifrc and North Atlantic Oceans, J. Climate, 10, 2963-2978. Bamston, A. and R.E. Livezey, 1987: Classification, seasonality and persistence of low- jfrequency atmospheric circulation patterns. Mow. Wea. Rev., 115, 1083-1126. Burt, I.E. and G.M. Barber, 1996: Elementary Statistics for Geographers, Guildford Press, New York, 599 pp. Bromwich, D.H., F.M.Robasky, R.A_Keen, and J.F.BoIzan, 1993: Modeled variations of precipitation over the Greenland ice sheet, J. Climate, 6, 1253-1268. Chen, W.Y., 1982: Assessment o f the Southern Oscillation sea-level pressure indices, Mon. Wea. Rev., 110, 800-807. Cayan, D.R, 1992: Latent and sensible heat flux anomalies over the Northern Oceans: the connection to monthly atmospheric conditions, J. Climate, 5, 354-369. Dai, A, I. Y Fung, and A.D. Del Genio, 1997: Surface observed global land precipitation variations during 1900-88, J. Climate, 10, 2943-2962. 222 Davies, T.D., Vincent, C.E. and A.K.C. Beresford, 1985: July-August rainfall in West- Central Kenya, J. Climatol, 5, 17-33. Deser, C. and J.M. Wallace, 1987: El Nino events and their relation to the Southern Oscillation, Science, 228, 1085-1087. Deser, C. and M.L.Blackmon, 1993: Surface climate variations over the North Atlantic Ocean during winter: 1900-1989,/. Climate, 6, 1743-1753. Flohn, H., 1987; East Afiican rains o f 1961/2 and the abrupt change o f the White Nile discharge, Paleoecology of Africa, 18, 3-18. Griffiths, J.F., 1972 a: Eastern Afiica, Ch.9 in World Survey of Climatology, Vol. 10. Climates of Africa, Griffiths, J.F. (Ed.), Elsevier, New York, 604 pp. Griffiths, J.F., 1972 b: Mozambique, Ch. 12 in World Survey of Climatology, Vol. 10. Climates of Africa, Griffiths, J.F. (Ed.) Elsevier, New York, 604 pp. Grotzner, A., M. Latif and T.P. Barnett, 1998: A decadal climate cycle in the North Atlantic Ocean as simulated by the ECHO coupled model, J. Climate, II, 831-847. Carelton, A.M., 1988: Meridional transport of sensible heat in winters marked by extremes of the North Atlantic Oscillation, 1948/8-1979/80, J. Climate, I, 212-223. Grove, AT., 1996: Afiican river discharges and lake levels in the twentieth century. In T.C. Johnson and E.O. Odada (Eds.),The Limnology, Climatology and Paleoclimatology o f the East African Lakes, Gordon and Breach, Australia, 664 pp. Harrison, M.S.J., 1984: The annual rainfall cycle over the central interior of South Afiica, South African Geographical Journal, 66, 47-64. Hastenrath, S., 1985: Climate and Circulation o f the Tropics, D.Reidel, Dordrecht, Holland, 455 pp. 223 Holton, JJEL,1991: An Introduction to Dyncanic Meteorology, 3"* Ed., Academic Press, San Diego, 474 pp. Hurrell, J.W.,1995: Decadal trends in the North Atlantic Oscillation; regional temperatures and precipitation. Science, 269, 676-679. Hurrell, J.W., 1996: Influence of variations in extratropical wintertime teleconnections on Northern Hemisphere temperature, Geophys. Res. Lett., 23, 665-668. Hurrell, J.W and H. Van Loon, 1997: Decadal variations in climate associated with the North Atlantic Oscillation, Climatic Change, 36, 301-326. Jury, M.R., 1996: Regional teleconnection patterns associated with summer rainfall over South Afiica, Namibia and Zimbabwe, Int. J. ClimatoL, 16, 135-153. Jury, M.R, 1995; A review of research on ocean - atmosphere interactions and South African climate variability, S. Afr. J. Set, 91, 289-294. Jury, M.R and B.M.R Pathack, 1993: Composite climatic patterns associated with extreme modes of summer rainfell over Southern Africa: 1975-1984, Theor. Appl. ClimatoL, 47, 137-145. Jury, M R , C. McQueen and K. Levey, 1994: SOI and QBO signals in the African region, Theor. Appl. ClimatoL, 50, 103-115. Jones, PD. and M. Hulme, 1996: Calculating regional climatic time series for temperature and precipitation: methods and illustrations, IntJ. ClimatoL, 16, 361-377. Kalnay, E. M Kanimatsu, R Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Wollen, Y. Zhu, M. Chelliah, W. Ebisuazaki, W. Higgins, J. Janowiak, K.C. Mo, C. Roplewski, J. Wang, A. Leetmaa, R Reynolds, R. Jenne and D. Joseph, 1996, The NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteor. Soc., 77, 437- 471 plus CD-ROM. 224 Kapala, A., H. Machel and H. Flohn, 1998: Behavior of the centers of action above the Atlantic since 1881. Part H: Associations with regional climate anomalies. Int. J. ClimatoL. 18, 23-36. Kaplan, A., M.A. Cane, Y. Kushnir, A.C. Clement, M B. Blumenthal, and B. Rajagopalan, 1998: Analyses of global sea surface temperature, 1856-1991, J. Geophys. Res., 103, 18,567-18,589. Kelly, P.M, CM. Goodess, B.S.G. Cherry, 1987: The interpretation of the Icelandic sea ice record,/. Geophys. Res., 92C, 10,835-10,843. Koslowski, G. and P. Loewe, 1994: The western Baltic ice season in terms of a mass-related severity index: 1879-1992. Part 1: Temporal variability and association with the North Atlantic Oscillation, Tellus, 46A, 66-74. Kruger, A.C, 1999: The influence of the decadal-scale variability of summer rainfall on the impact of El Nino and La Nina events in South Africa, Int. J. ClimatoL, 19, 59-68. Kushnir, Y. and J.M. Wallace, 1989: Low frequency variability in the Northern Hemisphere winter: Geographical distribution, structure and time-scale, J. Atmos. Sci., 46, 3122- 3142. Kushnir, Y , 1994: Interdecadal variations in North Atlantic Ocean sea surface temperature and associated atmospheric conditions, J. Climate, 7, 141-157. Kutzbach, I E, 1970: Large-scale features of monthly mean Northern Hemisphere anomaly maps of sea-level pressure, Mon. Wea. Rev., 98, 708-716. Kushnir, Y., V I. Cardone, I.G. Greenwood and M.A. Cane, 1997: The recent increase in North Atlantic wave heights, J. Climate, 10, 2107-2113. Labitzke, K. and H van Loon, 1989: Association between the 11-year solar cycle, the QBO, and the atmosphere, m . Aspects of the association, J. Climate, 2, 554-565. 225 Lamb, P L and R.A. Peppier, 1987; North Atlantic Oscillation: concept and an application. Bull. Amer. Meteor. Sac., 68, 1218-1225. Lindesay, LA., 1988: South African rainfall, the Southern Oscillation and a Southern Hemisphere semi-annual cycle, J. ClimatoL, 8, 17-30. Machel, H., A. Kapala and H. Flohn, 1998: Behavior of the centers of action above the Atlantic since 1881. Part I: Characteristics of seasonal and interannual variability. Int. J. ClimatoL, 18, 1-22. Mason, S.L, 1990: Temporal variability of sea surface temperatures around Southern Africa: a possible forcing mechanism for the 18-year rainfall oscillation?, S. Afr. J. Sci., 86, 243-252. Mason, S.L, 1995: Sea-surface temperature - South African rainfall associations, 1910-1989, Int. J. ClimatoL, 15, 119-135. Mason, S.L, 1998: Seasonal forecasting of South African rainfall using a non-linear discriminant analysis model. Int. J. ClimatoL, IS, 147-164. Mason, S.L and P.D. Tyson, 1992: The modulation of sea surface temperature and rainfall associations over Southern Africa with solar activity and the quasi-biennial oscillation, J. Geophys. Res., 97, 5847- 5856. McClendon, M.J, 1994: Multiple Regression and Causal Analysis, F.E. Peacock, Itasca, 111., 347 pp. Meehl, G.A., 1978, Tropical teleconnections to the seesaw in winter temperatures between Greenland and Northern Europe, NCAR Cooperative Thesis No. 51, and INSTAAR Occasional Paper No. 28, Institute of Arctic and Alpine Research, University of Colorado. Meehl, G.A and H. van Loon, 1979: The seesaw in winter temperatures between Greenland and Northern Europe. Part IH: Teleconnections with lower latitudes, Mon. Wea. Rev., 107, 1095-1106. 226 Mitchell, JJM., B. Dzerdzeevskii, H. Flohn, WJL. Hofineyr, HJL Lamb, K.N. Rao, C.C. Wallen, 1966: Climatic Change. WMO Tech. Note No. 79, WMO No. 195TP100, 79pp. Moses, T., G.N. Kiladis, H.F. Diaz and R.G. Barry, 1987: Characteristics and frequency of reversals in mean sea level pressure in the North Atlantic sector and their relationship to long term temperature trends, J. ClimatoL, 7, 13-30. Nicholson, S.E. and D. Entekhabi, 1986: The quasi-periodic behavior of rainfall variability in Africa and its relationship to the Southern Oscillation, Arch. Met. Geoph. BiocL, Sen A 34, 311-348. Nicholson, S.E. and D. Entekhabi, 1987: Rainfall variability in Equatorial and Southern Africa: relationships with Sea Surface temperatures along the Southwestern coast of Africa, J. Clim. Appl. Meteor., 26, 561-578. Nicholson, S.E. 1993: An overview of African rainfall fluctuations of the last decade, J. Climate, 6, 1463-1466. Nicholson, S.E., 1996: A review of climate dynamics and climate variability in Eastern Africa, pp 25-56 in T.C. Johnson and E.G. Odada (Eds.),The Limnology, Climatology and Paleoclimatology of the East African Lakes, Gordon and Breach, Australia, 664 pp. Ogallo, L., 1979: Rainfall variability in Africa, Mon. Wea. Rev., 107, 1133-1139. Ogallo, L.J., 1988: Relationships between seasonal rainfall in East Africa and the Southern Oscillation, J. ClimatoL, 8, 31-43. Phillips, T., 1995: NCEP AMIP Model Description [FTP], ftp://wesley.wwb.noaa.gov/pub/ reanal/ random notes/model. (Accessed 6/11/1999) Preston-Whyte, R.A. and P D.Tyson, 1988: The Atmosphere and Weather of Southern Africa, Oxford University Press, Cape Town, 374 pp. 227 Rodhe, H. and H. Viiji, 1976: Trends and periodicities in East African rainfall data, Mon. Wea. Rev., 104,307-315. Rogers, J.C., 1984: The association between the North Atlantic Oscillation and the Southern Oscillation in the northern hemisphere, Mon. Wea. Rev., 112, 1999-2015. Rogers, J.C., 1990: Patterns of low-frequency monthly sea level pressure variability (1899- 1986) and associated wave cyclone frequencies, J. Climate, 3, 1364-1379. Rogers, J.C., 1997: North Atlantic storm track variability and its association to the North Atlantic Oscillation and climate variability of Northern Europe, J. Climate, 10, 1635- 1647. Rogers, J.C. and H.van Loon, 1979: The seesaw in winter temperatures between Greenland and Northern Europe. Part H: Some oceanic and atmospheric effects in the middle and high latitudes. Mow. Wea. Rev., 107, 509-519. Ropelewski, C.F. and M.S. Halpert, 1996: Quantifying Southern Oscillation-précipitation relationships, Climate, 9, 1043-1059. Serreze, M.C., F.Carse, R.G. Barry and J.C. Rogers, 1997: Icelandic low cyclone activity: chmatological features, linkages to the NAO, and relationships with recent changes in the Northern Hemisphere circulation, J. Climate, 10, 453-464. Shabbar, A., K. Higuchi, W. Skinner, and J.L. Knox, 1997: The association between the BWA index and winter surface temperature variability over eastern Canada and West Greenland, 7nr. J.ClimatoL, 17, 1195-1210. Shukla, J., 1987: Interannual variability of monsoons. In Fein, J.S. and P.L. Stephens (Eds.) Monsoons, Wiley, New York, 619 pp. Thompson, B.W., 1975: Africa: The Climatic Background, Oxford University Press, Ibadan, Nigeria, 60 pp. 228 Torrance, JJD., 1972: Malawi, Rhodesia and Zimbabwe, Ch. 13 inWorld Survey of Climatology, Vol. 10. Climates o f Africa, GrifiBths, JJF. (Ed.). Elsevier, New York, 604 pp. Trenberth, K.E and D.J. Shea, 1987: On the evolution of the Southern Oscillation, Mon. Wea Rev.. 115, 3078-3096. Trenberth, KJE., 1996: El Nino-Southem Oscillation, pp 145-173 in Climate Change: Developing Southern Hemisphere Perspectives, T.W. Giambelluca and A. Henderson-Sellers (Eds ), Wiley. Tyson, P JD., 1986: Climatic Change and Variability in Southern Africa, Oxford University Press, Cape Town, 220 pp. Schulze, G.C. 1972: South Africa, in Ed. JJ. GrifiBths,World Survey o f Climatology, Vol. 10, Climates of Africa, Elsevier, Amsterdam, 501-586. Van Heerden, J., D.E. Terreblanche and G.C. Schulze, 1988: The Southern Oscillation and South Africa summer rainfall, J. Climatol., 8, 577-597. van Loon, H. and J.C. Rogers, 1978: The seesaw in winter temperatures between Greenland and Northern Europe. Part I: General description, Mon. Wea. Rev., 106, 296-310. Vo se, R.S, R.L. Schmoyer, P.M. Steurer, T.C. Peterson, R. Heim, T.R. Karl and J.K. Eischeid, 1992: The global historical climatology network: long-term monthly temperature, precipitation, sea-level pressure and station pressure data. ORNL/CDIAC-53, NDP-041. Carbon Dioxide Information Analysis Center, Oak Ridge National laboratory. Oak Ridge, Termessee. 315 pp. Walker, N.D., 1990: Links between South African Summer rainfall and temperature variability of the Agulhas and Benguela current systems, J. Geophys. Res., 95 C3, 3297-3319. 229 Wallace, J.M. and PJ. Hobbs, 1977:Atmospheric Science : An Introductory Survey. Academic Press New York, 467 pp. Wallace, J.M. and D.S. Gutzler, 1981: Teleconnections in the geopotential height field during the Northern Hemisphere winter, Mon. Wea. Rev., 109, 784-812. Wallace, J.M., C. Smith and Y, Zhang, 1990: Spatial patterns of atmosphere-ocean interaction in the Northern Winter, J. Climate, 3, 990-998. Wallace, J.M., Y^Zhang, J.A.Renwick, 1995: Dynamic contribution to hemispheric mean temperature trends. Science, 270, 780-783. Warner, R.M., 1998: Spectral Analysis of Time Series Data, Guilford Press, New York, 190pp. 230