JOURNAI
46
OF APPLIED METEOROLOGY
VOLUMAE 41
Economic Impacts of Climate Variability in South Africa and Development of
Resource Prediction Models
MARK R. JURY
Fnvztirooinental Studies Deportment, Uniiversity of Zulaund, KwvaDlangezwa, South At/ica
(Manuscript received 13 November 2000, in final form 28 June 2001)
ABSTRACT
An analysis of food and water supplies atnd economic growth in South Africa leads to the realization that
climate variability plays a major role. Sumtner rainfall in the period of 1980-99 is closely associated (variance
-= 48%) with year-to-year clan,ges in the grOss domlestic product (GDP). Given the strong links between cliniate
antd resources, statistical niodels are formulated to predict maize yield, river flows, and GDP directly. The most
influential predictor is cloud depth (outgoing lonigwave radiation) in the tropical Indian Ocean in the preceding
spring (Septemilber-November). Reduced monsoon convection is related to enilanced ratinfall over South Africa
in the following summer axid greater econonmc prosperity during the subsequent year. Methodologies are outliined
and risk-reduction strategies are reviewed. It is estimated that over US$1 billion could be saved annually
through uptake of tirtiely and reliable long-range forecasts.
17
Introduction
Anticipating the swings of climate is important for
sustainable economic. grow'th in these times of increasing pressure on the earth's limited resources. Recent
improvements in the global climate observing system
have been made with increasing in situ ocean data, satellite technology, and advances in climate modeling.
Institutional systems are being developed to exploit
these advances in support of sustainable developmienit,
through a reduction of risks associated with extreme
climate events. Climate information is now beinig applied directly or indirectly in many developed couuntries
to a wide range of activities, including agricultural practice, resouree mianagement, econotnic planninig, initernational relations, hydrology, and health.
The inmproved understanding of climate variability is
being exploited to provide seasonal climate forecasts
through regional outlook forums. These activities have
been instrumental in,clarifying the requirements of both
"producer" and "user" communities for adequate climate information and prediction services. Africa hosts
three such foruns, and climate scientists are actively
involved in providing guidance to local users (Mason
et al. 1999). This paper will cover one sucth effort to
prtovide meaningful input to the southern African regional climate outlook forum (SARCOF), initiated in
1995 to create consensus predictions. An initial survey
Corrvsponding authrr address: Mark R. Jury, Environmiental Stud-
ies Dept., U,niversity of Zululand, KwaDl)angezwa 3886, South Africa.
E-mail: [email protected]
2002 American Mrctorological Society
©M)
of mitigation strategies tailored to southern Africa is
reported in Natural Resources Itnstitute (1996). Harrison and Grahamn (2001) provide estimates of costbenefits attributable to long-range climi-ate pred-ictions.
The level of climate information uptake in southern
Africa is growing in parallel with experitnental riskhedging operations, thus inhibiting a stable quantitative assessment. Hence, this study is limited to analyzing climate impacts on resources and development
of predictive models.
a. Background
Southern Africa is dependent on subsistence agriculture to meet the basic needs of its population of more
than 100 millio.n. It is estimated that miore than one-half
of gross domestic product (GDP) and three-quarters of
all jobs are attributable to rain-fed agriculture (Hulme
1996). South Africa exports a considerable volume of
food stuffs and manufactured products to Africa, and
its economy is interdependent with the rest of the continent. The work presented here will outline how South
Africa's GDP responds to rainfall variability. The impacts of this variaibility on crop (maize) production, the
streamilow of major rivers, and overall economiiic
growth are assessed in the context of the El NOino-Southern Oscillation (ENSO). The potential for predictability
is explored, and statistical methodologies are provided.
Southerni Africa is particularly vulnerable to swinlgs inl
clinate owing to its location within the arid subtropics
(Fig. 1) and the potential impacts of a warmer and drier
future (Bohle et at. 1994).
JANUARY
2002
JURY
47
b
to
-
4
to
N
1rE
25 E
LongaLde
30E
35E
40'E
H[o. 1I.Jan rainfall over southern Africa, illustratinIg semiarid conditions (<4 mm day I at the seasonal
optimuum) to the south of 20WS. The maize-growing region and Orange River are located near the
"South Africa" tatxl (analyzed from CRU rainfall climatology via the IRMLDEO NWeb site).
The population of southern Africa is growing at a rate
of 3% per year. The population of South Africa is onehalf rural, and rates of urbanization outstrip population
growth, implying a continual shift of people away from
the land (Vogel 1994). The resulting decline in agriculture's economic conitribution could have a rnarked
influence on regional energy consunmption. Measures of
production such as GDP indicate that many countries
of southern Africa are poverty-stricken, with per capita
annual incomes of less than U.S.$1000 (W0orld Economic Forum 2000). In South Africa, the economy is
diversified as shown in Fig. 2. Manufacturing and milling contribute 'ibotut one-quarter of GDP; agriculture
contributes less than 5%; and commerce, communications, and trade contribuLte more than 40% (South African Reserve Bank 1999. personal communiication,
hereinafter SARB99). This economic diversification has
not reduced vulnerability to climate, as will be shown
later.
Agricultural production has exhibited increasing fluc-
tuations in recent decades. Following drought in 1992,
the GDP of Zambia, Zimbabwe, and South Africa declined 9%, 8%, and 3%. respectively (Benson and Clay
1994), contributing to rising uneniployment (>3 0 %) and
economic stagnation. Economic adjustmnents imposed by
external creditors reduced consurimptive demand, thus deflating national economies. Yet with increased political
stability cotnes interdependence and the potential for
overall growth in the region (Huline 1996).
b. El Nino impacts on regional cliimlate
An El Nifno produces warm, dry conditionis over
southern Africa (Ropelewski and Halpert 1987; Janowiak 1988; Jury et al. 1994; Hastenrath et al. 1995;
Shinoda and Kawnamura 1996: Rocha and Simmonds
1997) via westerly winds that invade the more productive eastern highlands of southern Africa, increasing
evaporative losses. Convergent subtropical jet streams
are generated in the upper level of the atmosphere in
48
JOURNAL OF APPLIED ME/'FOROLOGY
Agr,For,Fish (45/%)
GovService (15.0
Y
X-
VOLUME 4 1
(19.9%)
Socialtother (6.3%)
-- Mining (6.4%)
-Const,Elec,Water (6.6%)
Commerce (1 7.9%)
Trans,Comms (1 OaT)
Fit3. 2. Sectors Of the South African economy expressed as a percentage of the total GDP for 1999
(based on information supplied by SARB99).
response to the tropical warming associated with El
Niflo. The jet streams accelerate and swerve equatorward, producing subsidence over the interior (Mwafulirwa 1999). Rain-bearing troughs are carried eastward
rapidly, and wet spells become short lived and patchy.
Many of these attuospheric responses owe their origin
to changes in tropical sea surface temperatures (SST)
in the western Indian Ocean, which tend to warm in
sympathy with the eastern Pacific (Cadet 1985). Trogether these factors displace cloud bands to the adjacenit
oceans in the November-March suiimer season, when
the El Nifno is at peak intensity (Meehl 1988; Jury 1995).
El Nifno summers are clharacterized by two areas of
opposing weather centered oni Botswania (25°E) and
Mauritius (55°E). Warmer sea temperatures in the tropical Indian Ocean enhance local winds and clouds, causing a surplus of rainfall over Mauritius and the surrounding ocean. Some 3000 km away, the plateau of
southsern Africa becomes moisture deficient. The zonal
overturning circulation is often part of a stationary
"wave train" in the subtropical westerlies (KaTnay et
al. 1986). Satellite normalized difference vegetation index images for composite El Ninio years reveal that
vegetation south of 15°S is water stressed and is considerably "browner" than usual. Jury and Lyons (1994)
12
I-
I
Luj
--
--
_-
-
-
-
_
_
-
South Airica
>\
II
-
A
Zi o
C.
Ci~
found that surface dewpoint temperatures decline below
zero for extended periods during El Nifno. Heat waves
are produced when subsiding westerlies strengthen the
Botswana high pressure cell, blocking the input of noisture from the Tropics.
The El Niflo inffluence on southern Africa's summer
climate has been documented from historical observations (Vogel 1994). Since the 1950s, major El Nifios
have occured in 1964, 1966, 1973, 1983, 1992, and
1998. Dry summers occasionally are unrelated to El
Niuo, for example in 1968 and 1970. Drought over
southern Africa is characterized by rainfall of about 60%
of normal, temperatures of about 5°C above normnal, and
evaporative losses of tnore than 1(0 mm day I (Lema
and Seely 1994). The percentage of land in southern
Africa that can support crop production shrinks from
about 30% to 5% of the total---an area too small to
support a growing population. Maize yields decline below I Mg ha-', a fivefold reduction over the plateau.
Yields for crops such as sunflower, sugarcane, and so
on, show about a threefold decline. Annual agricultural
income fronm dry-land crops valued at about U.S.$1 billion in South Africa, declines to about 30% of average
(SARB99). The result for commercial farmers is a cashflow crisis and bankruptcy, followed by a vicious circle
of higher interest rates and shrinking profit margins.
Even miiore serious is the impact of a failed crop on
subsistence farming, which constitutes two-thirds of
all jobs in southern Africa (Hulme 1996). Food shortages contribute to a high risk of malnutrition and starvation, affecting more than 30% of the population in
southern Africa (Natural Resources Institute 1996).
Comparable impacts are found across developing regions, in which strategies for coping with El Niflo may
benefit from the application of improved predictions
(Nichols 1997).
5-
c. Clinmate impacts on the economy
4-
I96C
31
rT'r
1955
rr-r
4sl
1970
9
1975
19s0
-r
1905
T
1990
Tr-rT1
1 95
YEAM
Fc;. 3. Trends in the contribution of agriculture to the South
African econonry (analyzed from SARB99).
Although agriculture accounts for less than 5% of
economic activity in South Africa (Fig. 3), its indirect
infltience is estimated to be about 25%Yo (Lindesay 1990),
attributable to the need for manufacturing and service
JANUARY
2002
J JURY
support in commercial farming operations. Agriculture's
share in export trade averaged over southern Africa is
36% (Rwelamira and Kleynhans 1996). Among the 15
countries of southern Africa, food stuffs are the largest
single item of trade. The widespread nature of drought
limits the potential for some parts of southern Africa to
supply food to other parts during their time of need.
Adding to the problem is a 30% coefficient of variation
for summer rainfall, which plagues many of South Africa's food and water resources. These variations appear
to be growing over time, which suggests that the clirnate
of southern Africa is become more extrenme (Jury and
Majodina 1997).
Maize contributes a large slice of agricultural (GDP
in South Africa. The timing of rainfall during maize
growth in early February is critical to the food supply
of southern Africa (Jury et al. 1997). Other factors that
govern maize production include the amount of fertilizer
used and the loan interest rates, which are tied to inflation and macroeconomics (Lindesay 1990). Interest
rates in southern Africa are high by world standards.
Over the past 25 yi; South Africa's official rate of inflation has oscillated above 10%, with peaks in 1975,
1981, 1986, and 1991, following "good" rainfall years
when consumption increases. However a weak quasibiennial cycle in southern African rainfall (Mason and
Tyson 1992) means that a "negative cascade" often
occurs. After a good rainy season for which inflation
and interest rates are high, farmers must undertake crop
preparation and planting with the prospect of a dry year
looming.
Lindesay (1990) finds that the GDP in dry years is
83% of that in wet years. Rainfall departures of 10%
lead to GDP changes of about 1%. In the 1983 El Ninlo
year, the South African GDP dropped by about 7%
(VanZyl et al. 1988). In other countries north of South
Africa, agriculture makes a substantially larger contribution to GI)P, as follows: Angola 46%, Botswana 3%,
Lesotho 20%, Malawi 37%, Mozambique 61 %, Namnibia
11 %, Swaziland 24%, Tanzania 63%, Zambia 11 %, and
Zimbabwe 20% (Rwelamira and Kleynhans 1996). It
can therefore be expected that fluctuations in climate
will have a proportionately larger influence on economic
activity. Zhakata (1996) indicates how the extended
drought of 1982-87 adversely affected crop production
in Zimbabwe. Both the 1983 and 1984 harvests were
about 30% of the long-term mean for maize, sorghum,
and groundnuts. Ninety percent of the smaller inland
dams dried up, and food supplies dwindled in urban
centers. In the 1992 drought, 5.5 millioni communal
farmers requested drought relief assistance, constituting
75% of the rural population. Glantz et al. (1997) estimates that U.S.$60 million in foreign exchange losses
occurred in Zimbabwe alone following the 1992
drought. Vogel (1994) indicates that 22% of agricultural
debt can be traced directly to drought impacts. Multiyear
droughts can lead to severe water shortages, as pointed
out by Dilley (1996) in the case of the prolonged El
49
Nifio of the 1992-95 period. Vulnerability to climatic
fluctuations is considerably greater in countries directly
reliant on agricultural production to feed their people
(VanZyl 1993).
Water is not a guaranteed commodity in southern Africa, mainly because evaporative losses are more than 2
times rainfall across most of the region. Water resources
are depleted by prolonged dry spells; average runoff declines to about 1% of rainfall, and inflows to the Vaal
River catchment, near Johannesburg, decline sixfold on
average (Lindesay 1990). Limitations of water supply
affect domestic and industrial activity in many areas,
leading to quotas and consequent production losses across
diverse sectors of the economy. Dam levels across South
Africa cdrop to about 30% of capacity if drought is followed or preceded by a another dry year (Vogel 1994).
Historical data extending back to 1900 indlicate that
the most severe and widespread drought was 1983, followed by 1992, both in association with a negative
Southern Oscillation index (SOI). There conversely has
been an increase in short-lived flood events, such as
those associated with Tropical Cyclone Demoina in
1984, the September 1987 flood near Durban, those near
Johannesburg in 1988, snowstorms over Lesotho in
1996, and widespread floods across east Africa in November of 1997 and southern Africa in February of
2000. These short-term flood events have more immediate and localized impacts, making the infrastructure
more vulnerable to climate change. The current consensus of dynamical models on long-term trends is that
evaporative losses are likely to rise by about 20% over
the next century (Hulme 1996), turning southern Africa
into a water-scarce region. Schulze (1996) points out
that seasonal and interannual changes in rainfall are significantly greater than long-termn trends and decadal cycles; so efforts toward understanding and predicting hydrological impacts are critical to survival in the region.
Rainfall during a dry year may be 50%-80% of normal,
but runoff is typically 20%-50% of normal and is patchy
in its spatial distribution. The compounding effect on
resources is clear.
The forestry contribution to the economy of South
Africa is important (--5%). Drought slows the growth
rate of trees and contributes to fire hazard. Because climate fluctuations are integrated over a longer time period,
say 10 yr for the average conimercial pine or gum tree,
it is likely that multiyear drought and decadal cycles of
climate significantly impact this economic sector. In
southern Afiica, an 18.6-yr cycle of rainfall has been
noted in the past century (Mason and Tyson 1992). so
forestry production would have benefited in the wetter
1950s, 1970s, and 1990s, whereas slower growth and
losses from fire would have occurred in the 1940s, 1960s,
and 1980s. Fisheries are governed by shifts in ocean
currents and by overexploitation. ENSO phase causes
latitudinal shifts in upwelling and productivity in the Benguela Current along the west coast of South Africa. During El Nifio, warmer waters near Cape Town modify the
50
JOURNA}, OF APPLI ED METEOROLOGY
ecological environmiienit through the food web, and a consequent shift in fisheries species, habitat, and abundance
occurs. Fisheries management systems will be able to
respond to predicted clhanges, thus reducing consequetnces along the coast. Climate variability touches every facet
of economic activity in southern Africa, particularly for
the less-developed counitries dependent on subsistence
agriculture for food (Hulne 1996).
d. Resource prediction
Some of the year-to-year fluctuations in climate are
the result of random sequences of events. A region may
experience a dry spell simply because no storms happen
to pass that way for some time. Prediction for these
aspects of climate is unlikely. It is the climatic variations
that are part of coherent large-scale patterns that are
more predictable, especially where the "memory" of
the initial state of the ocean plays an imnportant role. If
atmospheric circulation patterns are coupled to SST,I
then the observation and prediction of marine conditions
can be used to forewarn of climatic consequences.
The demand for global long-range forecasts received
impetus from the 1983 El Nifio and its laobal consequences (>U.S.$1 0 billion damages; Glantz et al. 1997).
The ability to model atmospheric responses to ENSO
has grown such that ensemble numerical models are able
to make skillful projections of rainfall and temperature
in nmany parts of the world. However, the direct prediction of food and water resources involves complex
interactions, which can be parameterized_ albeit with
compounding errors. Another approach is the statistical
one, in which precursor "signals" are trained directly
onto historical records of resources (e.g., Cane et al.
1994). This method requires a time series of high quality, relatively free of nionclimatic factors. South Africa
is fortunate in this regard, with a long history of resource
data available. The empirical technique attempts to replicate fluctuations in resources at a lead-timne of 3-6
months over the past 25-40 yr using key indicators of
the tropical ocean climate system (Jury et al. 1999).
2. Data analysis and model development
In this section, a statistical methodology is outlined
to give insight to the predictability of resources influenced by climate. To formulate a statistical model, it is
necessary to have a specific target in miind. After a
decade of research, more than 20 target time series have
been generated based on user needs, analyses of climatic
data, and consideration of economic value. The targets
consist of summer rainfall and temperature, yields for
various crops based on quality-checked agricultural records, and naturalized catchment inflows to the major
water supply dams of South Africa. The area averaging
to state (provincial) level is guided by principal component analysis of gridded summer rainfall and tenmperature fields. For these clitnatic variables, dynamrical
VOL1JME, 4 1
model products are available. Reliable crop-yield data
were available at provincial (state) level for maize, sugar, sorghum, sunflower, and so on. Analysis in the period
of 1971-93 revealed increases due to improved farming
practice, so linear detrending was applied. For the hydrological targets, specific catchments with highest water volu-mes were identified, including Gariep, Vaal,
Pongola, and Hartebeespoort. The hydrological data derive from streamfiow gauges, supplemented by daily
catchnment rainfall to interpolate missing records (typically <5%), and reduce anthropogenic effects (of dams
built, etc.). For brevity, the analysis here is restricted to
the national maize yield and the flow of the Or-ange
River GDP data were obtained from the Statistical Division of the South African Reserve Bank. The GDP is
a useful measure of economic prosperity and depends
on productivity, money supply, and other factors. To
create a more stable, climate-oriented time series, yearon-year changes in the GDP (growth rate) are considered. H-lence, statistical models are trained directly onto
food, water, and financial resource data.
A pool of candidate predictors is assembled from a
subset of indices outlined in Jury et al. (1999). These
include time series such as the SOI and quasi-biennial
oscillation (QBO), SST in the tropical oceans, tropical
air pressure and wind components, and indices of monsoon convection [outgoing longwave radiation (OLR)]
averaged over areas of more than 100 x 100 (Table 1).
The total number of predictors is 18. Two seasons are
considered: July-September (JAS) and September-November (SON). The predictors have been identified on
the basis of principal component analysis of the field
parameters and by correlation and composite mapping
with respect to summer rainfall (Jury et al. 1999). They
reveal interactions betweeni the regional and tropical climate systems around Africa (Mason and Jury 1997).
NModels are formulated in stepwise fashion to a maximum of three predictors, and colinearity tests are applied
to screen out adjacent variables with variance 7-2 greater
than 8%. Although ongoing research seeks to consolidate predictors, the current formulation has a candidate
pool nearly equal to the number of years trained (1971-93, or 22). Thus a "penalty" is needed to deflate the
hindcast fit of the model, considering the candidate pool,
the number of predictors utilized, and any autocorrelation in the target series. In southern Africa, persistence
of rainfall from summer to summer is negligible owing
to an underlying quasi-biennial cycle (Mason and Tyson
1992). To evaluate the effect of candidate pool size, a
Monte Carlo test was conducted by scrambling the existing predictors into random sequences. The random
numbers then were applied to the target time series to
produce "optimiial" three-predictor algorithmis. The average hindcast fit was deemed to be artificial skill. The
r2 penalty. to accoLunt for a candidate pool of 18 with
three predictors retained is 12%. To achieve 98% significansce, the model rH fit should exceed 29%S, after
application of the penalty.
JANUARY
2002
J UR Y
Predictors.
TABLE L
Acronym
_ P-arameter
Pressure dipole
30-hPa zonal wind
SST
SST
SSTI'
Meridional wind
Zonal wind
200-hiPa zonal wind
SST
SST
SST
Zonal wind
Meridional wind
Cloud depth
Meridional wind
Air pressure
Zonal wind
Air pressure
Sol
QBO
Nino3
naST
scaST
eaV
att]
atU2
socnS
ciST
wiST
wiu
wiv
OLR
niV
seiP
citJ
eiP
_
_
Region
_
West Pacific
Upper Tropics
Central Pacific
North Atlantic
Southeast Atlantic
East Atlantic
Central Atlantic
Central Atlantic
Southern Ocean
Cetttral Indian
West Indian
West Indian
West Indian
Central Indian
North Indian
Southeast Indian
East Indian
East Indian
3. Results
_~~~~~~~~~ego Central lat, long
5°S. I80°W
2°S, 40 1100E
0°, l20°W
12ON, 25-W
10°S, 50E
5°S, 5QE
2°S. 3( 0'W
2°S, 20QW
45°S, 20°E
2°S, 62 0E
20S, 47 0E
2°S, 47°E
2°S, 47°E
2°S. 62°E
12°N, 65 0E
12'S, 77°E
2°S, 904E
2°S,
90°E
0
0
2 S, 90 E
rying capacity of this semiarid environment is nearing
a population threshold. This increasing vulnerability
arises despite the influences of internal political and
external market forces.
In this section, statistical models are developed to
predict the interannual variability. The first target for
consideration is the detrended maize yield of South Africa. Much of the maize is grown in an east-west belt
on the high plateau just to the south of Johannesburg.
An adjusted fit of 38% is obtained by two predictors:
OLR in the central Indian Ocean and the stratospheric
QBO. Convection (Ol.R) is known to be influenced by
ENSO through changes in the underlying SST (Cadet
1985) and monsoon circulation systems (Hastenrath
The GDP response to rainfall over South Africa in
the preceding sumnmzer is illustrated in Fig. 4. No clear
relationship is evident in the l960s and early ]970s,
when the growth rate remiained above 3% and the contribution of both gold mining and agriculture was greater. After the mid-1970s, summier rainfall and the GDP
fluctuate together. Forty-eight percent of variance in the
GDP growth rate may be attributable to rainfall in the
second half of the record. It is not clear why the economy has become more vulnerable to climate in recent
decades, given the levels of diversification. The withdrawal of government subsidies and high interest rates
are contributing factors. It is also possible that the car-
-4
78
2
51
2~'
c1
t 980
Year
-
Gcowtvh
-
Rain
IFIG. 4. Comparison of GDP annual growthl rate and Nov-Mar rainfall over South Africa in the
preceding sutnmer. The cross-correlation rF value is 3% up to 1979 but 48%e thereafter.
VOLLIUME41
JOURNAL OF APPLIED METEOROLOGY
52
,
-
2.5
,,
--
rr
, -,-
'-
-rT
TABLE
S
_
1.4
9
v
0
-
8.8
V~11-it
N
la
l
iTJ!~
0l;
-'.E
i
-1.5
1974
3
SA.maize
Org.river
SA.GDP
over the western Indian Ocean and eastern Africa in the
1978
1982
1986
1982
1988
1998
1994
S
2
3
a
r-
I
5.
ci
;[F45
'U
S.
co
-2
1974
1978
-
197i
i
S?4
].978
1982
1978
1974
1978
1982
7
197 8
99 1988
-
1994
1.5
S
4.'
0.5
M
-e.5
CO
-1.8
-1
.I.
_;
._L_..L_._L
-2.E
Algorithmi
+0.66(oOl.R) - 0.34(oQBO)
I O.39(aeaV) - 0.70(oseiP) - 0.45(osocnS)
4-I->64(oOLR) + 0.53(awiST)
spring are consistent with wet conditions over
I,_,1_91I 7 .I....................
1 austral
southern Africa in the following suminer, leading to
L,
5
1978
0
8U
Target
ern Africa (Goddard and Graham 1999). Reduced clouds
(5)
-2.
2. Multivariate models, where prefix o denotes SON andl a
denotes JAS.
j
988
L990
1994
1888
1998
1994
YEARS
FIG. 5. Comparison of model-predicted amd observed (dots) national maize yield (top), Orange River flow (mi.ddle), and GDP growth
rate (bottom), based on algorithms given in Table 2 for a lead-time
of 6-9 mnonths. Error bars are provided. The vertical axis refers to
standardized departures.
2000) A zonal overturning Walker cell is prominent
during the SON season. with vertical mnotions over east
Africa and Indonesia and zonal flow in between. As the
ITCZ shifts south of the equator under the inifluence of
thie northeast monsoon, a dipole rainfall pattern is observed, with contrasting regimies over eastern and south-
increased maize yields. The stratospheric QBO acts to
reinforce the zonal overtuirning in an unknowwn manner
and favors increased maize yields during west phase.
From this result, over one-third of the variability of
maize yield can be predicted at a lead time of 6 months
(e.g., November forecast for April harvest). The model
predicts within the error bar 17 times in 22 yr (Fig. 5),
and may be considered to be sufficiently reliable to adjust farming effort via exposure to credit.
inflow
The second target considered is the catchment
3
to the Gariep Dam (volume about 10"l i ) on the
Orange River to the west of Lesotho (30°S, 25°E). This
reservoir is the largest water resource in South Africa,
and its iniflow is accumulated each year with a 1--2month lag on rainfall. The model fitting the time series
has three predictors: meridional wind in the tropical east
Atlantic, pressure in the southeast Indian Ocean, and
SST south of Africa. The southeast Indian Ocean pressure has the strongest influence anid is closely related
to ENSO (correlation coefficient r =- 0.83 with Nifio3 SST for N = 22 cases). When pressure is low (La
0
Nif.a), river flow is high. The lower pressure near 90 E
heralds an alternating pattern of higher pressure at 60°E
and lower pressure at 201E (Rocha and Simmonds
1997), which contributes to southern African rainfall.
The adjustedc model fit for the Orange River flow is high
at 59%, but the hindcast tercile "hit" rate is 15 of 22
and so offers a two-out-of-three reliability consistent
with other prediction models.
The final target considered here is the GDP growtl
rate. The model uses predictors in the Jully-Novemiiber
period in the preceding year: a lead time of 6-9 months.
The cloud depth (OLR) in the central Indian Ocean is
the main determinant (same as for maize yield; see Tabie
2); SST in the west Indian Ocean provides a secondary
input. With application oL the penalty for artificial skill,
the adjusted hindcast fit is 36%°.The model suggests
that, when clouds are reduced in austral spring (SON)
0
0
in the area of I 0N-E5°S and 45 -80 E, wet weather
leading to
sumimier,
during
develops over South Africa
an increase in GDP the following year. Hence, improved
knowledge of ocean- -atmosphere coupling in the Indian
Ocean will underpin long-range forecasts of climaterelated economic activity in southern Africa. With a
1999 GDP valued at U.S.$168 billion and variations in
growth rate of t2% (e.g., U,S.$7 billion), a significanit
JANUARY
2002
J UR Y
level of risk could be "hedged" through prediction and
mitigation (Harrison and Graham 2001).
Validation tests have been conducted oni the empirical
formulations, as outlined in Jury et al. (1999), using independent periods for development and testing and using
jacknife validation within the record. Given the low degree of persistence in the target data, it is found that both
hindcast fit and tercile hit rate are reasonable measures
of skill, consistent with more sophisticated performance
checks. The inherent 2-5-yr variability also means that
a 25-yr history will capture a sufficient number of interannual events to lend confidence in predictive applications. The statistical models are currently implemented,
and results could be found online at the time of writing
(ht t p: / / weather. i a rica. cco m /f ore casts!
cip_seasonaLoutlook.html). This site was receiving over
1(000 hits per day during the peak season (October). Most
of the users are from South Africa and are divided equally
among financial analysts, commercial farmers, and other
resource managers. According to response surveys conducted via e-mail, users prefer deterministic percentages
to be listed together with error bars.
A validation of sunmmer rainfall forecasts over eastern
South Africa from predictive models based on similar
techniques reveals hits in 1992, 1993, 1998, and 2000;
near misses (e.g., within the error bar) in 1994, 1996,
1997, and 1999; and a complete miss (e.g., outside the
tercile category) in 1995. In 4 yr out of 9 (44%), users
were adequately forewarned and could gain a cost advantage. In another 4 yr, ambiguous forecasts suppressed potential benefits. The misleading forecast in
1995 may have incurred losses among users. Potential
gains or losses affect many kinds of users, from institutions (banks, government departments) to donor agencies (financial and resource intervention) and individuals (commercial farmers, subsistence communities).
The cost-benefit of a forecast hinges on its uptake (Harrison and Graham 2001) and hence the level of confidence in its use. There is considerable scope for hedging
losses, as outlined below.
Mitigationi strategies
A reliable long-range forecast will have no value unless it is disseminated, taken up, and acted on. It may
be relatively easy to develop predictive tools in southern
Africa because of the known ENSO influence. Yet dissemninationi becomes difficult in some instances, particularly to small farmers without access to external media.
A number of channels of commiunication need to be
explored. SARCOF, which includes the region's weather
services and other agencies, is the main institution coordinating and disseminating long-range forecasts. The
main response of users is to modify farming effort
through exposure to borrowed finances, amounting to
U.S.$l billion per year. When a drought is forecast,
farmers (and their bankers) in the maize belt of South
Africa react by reducing debt to manageable levels,
53
planting less acreage and utilizing drought-resistant varieties, purchasing less fertilizer, and deferring outlays
on capital equipment (Natural Resources Institute 1996).
Many subsistence farmers continue with "business as
usual" in hopes of random rainfall on their plot, irrespective of the large-scale climate anomalies and forecasts thereof. Increasing sophistication with regard to
forecasting the shift of seasonal rains (e.g., early in
"dry" years, late in "wet") enables farmers to adapt
planting dates accordingly. Another potentially useful
mitigation strategy would involve varying credit exposure east-west across the background gradients in
rainfall. In this scheme, the level of financing made
available for farming operations would shift east in dry
years and west in wet years.
There is a system of rural extension officers who journey across the countryside every spring (October) to
inform subsistence farners and the rural public on the
prospects for the coming season. This community approach to information transfer has worked very well in
recent years. In other regions, early-warming systems
are in place and involve in situ and satellite monitoring
of resource status, analysis of household economic levels, nutritional analysis at rural schools, and administrative mechanisms to ensure that supplies are available.
It is important that partnerships are forged among government departments, researchers, nongovernmental aid
organizations, and commercial and industrial sectors of
the economy, to make the most of the shifting and variable climate of southem Africa (Thomson et al. 1998).
Water resource managers act on long-range forecasts
in a more conservative manner, mainly because of the
longer lead tines required in hydrological planning.
Dam inflows and volumetric levels can be monitored,
and shortfalls can be reduced through the imposition of
quotas on domestic and industrial users. Yet, inefficient
agricultural irrigation schemes near perennial rivers can
evaporate large volumes of water, particularly during
early summer (November-December). W'ater managers
over the plateau of South Africa usually contain dam
outflows except when levels are full and short-term
flood warnings are issued. During drought, mitigating
strategies are limited in the hydrological sector and involve supply quotas and incremental-use cost structures.
Rural communities without formal water supplies rely
on local access to streamflow. This dependence could
explain the increasing sensitivity of South Africa's GDP
to rainfall.
4. Summary
A variety of economic outputs and resources in South
Africa fluctuate according to climate. This study demonstrates that a reasonable percentage of this variability
is predictable up to 6 months in advance. Statistical
models, based on indicators of tropical ocean climate
and ENSO, have been developed to anticipate shifts in
economic prosperity, crop yield, and river flow in South
54
JOURNAL, OF APPLIED METEOROLOGY
Atrica. The most influential predictor for nmaize yield
and GDP is the cloud depth in the central Indian Ocean.
Air pressure in the southeast Iindian Ocean is useful with
respect to river flow. Both predictors are sensitive to the
regiotnal uptake of ENSO. Risk reduction based on projected earninigs is an essential tool in financial management, which may be applied to forecast food and water
supplies. The econonmic analysis suggests that swings
of climate are manifested in GDP flucttuations on the
order of U.S.$5 billion. With an ability to predict and
to mitigate 20% of this variability through strategic
plannitng, more than U.S.$1 billion couild be saved annually in southern Africa, according to local user surveys (Harrison and Graham 2001). A cost-benefit ratio
in excess of 20 is estimated, based on a governtnent
agency being employed to produce comnputer-intensive,
numerical model forecasts of rainfall and so on. The
cost-benefit ratio may be enhanced using statistical
miodels trained directly on resources, as outlined here.
Comiparison of forecast products based on these two
distinict techniques helps forums such as SARCOF gain
an "ensenmble" view of the expected climate. The result
is belter managetnent of essential resources through
lonig-range planning inf-used with relevant information.
Acknowledgments. Dr. S. J. Mason of the Scripps Institution for Oceanography at the University of California, San I)iego, is thanked for providing Monte Carlo
tests to assess artificial skill for various candidate predictor pool sizes. This study was supported by the National Research Fottndation and the Water Research
Coniniission of South Africa.
REFERENCES
Benson, C.. ind E. Clay, 1994: The impact of drought on sub-Sahatran
African economies: A preliminary examination. Overseas Development Instinste Rep. 77, 60 pp.
Bohle. H. Cy., 'r. E. Downing, and M. J. Watts, 1994: Climate
change and social vulnerability. Global Environ. Change, 4,
37-48.
Cadet, D., 1985: 'lThe Southe-rn Oscillation over the Indian Occea. .1.
Climatol., 5, 189-21t0.
Cane, M., A. C. Eshel, and( R. W. Bucklatnd., 1994: Forecasting Ziinbabwean maize yield using eastern equatorial Pacific sea surface
temperatures. Nature, 370, 204-205.
Dilley, M., 1996: Usefulness of climate forecastinig information for
drought mitigation and response fromn a donor's standpoint. Proc.
Workshop on Reducing Clirnate-RelatedVulrzerabilitv in Sour/zero Africa, Victoria Falls, Zimbabwe 141-148.
Glantz, M., M. Betsill, and K. Crandall, 1997: Food security in southern Africa: Assessing the use and value of ENSO information.
NCAR NOAA Tech. Rep., 142 pp.
Goddard, L., and N. E. Grahtami 1999: 'hle importance of the tndian
Ocean for siniulating rainfall anomailies over eastern and southern Africa i. Geopshys. Res., 104, 19 t99-19 116.
flarrison, M. S. J., and N. E. Grahani, 20111: Forecast quality, forecast
applications and forecast value: Cases from southerni African
seasonal forecasts. WMO Bull., in press.
Hastenrath, S., 2000: Zonal circulations over the equatorial Indian
Ocean. J. Climate, 13, 2746--2756.
VOLUME 41
, L. Greischar, and J. vazi Heerden. 1995: Prediction of summler
rainfall over South Africa. J. Climate, 8, 1511-1518.
Hulme, M., El.,1996: Climate change in southern Africa: An exploration of some potential impacts and implications for the SADC
region. WWF Publ., 1(14 pp.
Janowiak, J. E. 198S: An investigation of interannual rainfall variability in Africa. J. C'iinate, 1, 240-255.
Jury, M. R., 1995: A review of research on ocean-atmosphere inteTactions and South African climate variability. South. A,fr. J.
Sci.. 91, 289 -294.
and S. W. Lvons, 1994: Contrasting synoptic weather events
over southern Africa iu the dry sunmner of 1983. Sou(th. Aft.
Geogr. J., 76, 6--110.
.and M. Majodina, 1997: A preliminary climatology of southern
Africa extreine weather: 1973-1992. Iheor.Appl. Climjiatol., 56,
103 112.
-, C. McQueen, and K. Levey, 1994: SOI and QBO signals in
the Atrican region. Theor. Appl. Cli/satol., 50, 103-115.
S. Weeks, and M. P. Gondwe, 1997: Satellite-observed vegetation as an indicator of climate variability over southern. Africa.
South. Afr. J. Sei., 93, 34-38.
li. M. Mulenga, and S. J. Mason, 1999: Explora.tory long-range
models to estimate stiminer cliniate variahility over southern
Africa. . Climate, 12, 1892-1899.
Kalnay, E., M. Kingtse, and J. Paegle, 1986: Large-amplitude, shortscale stationary Rossby waves in the Southern Hemisphere: Observations and mechanistic experiments to determine their origin. J. Ataios. Sei., 43, 252-275.
Lema, A., and lM. Seely, 1994: 'T'he climate factor: State of the Enviromttent in Souther/n Afric, Southern African Research and
Documentation Centre, Peirtise Press, 87-104.
Lindesay, C. J., 1990: Economic and socio-political impacts of rainfall variations. M.B.A. thesis, UJniversity of Witwatersranid, Johannesburg, South Africa, 181 pp.
Mason. S. J., and P. D. Tyson, 1992: The modulation of sea surface
teiiiperature and rainfall associations over southerii Africa with
solar activity amid the quasi-biennial oscillation. J. Geophys. RAes.,
97, 5847-5856.
and M. R. Jury, 1997: Climatic vmriability and chanige over
southern Africa: A reflection on underlying processes. Progr.
----.
Phyvs. Geogr., 21, 23-50.
I,. Goddard, N. E. Graham, E. Yulaeva, L. Sun, and R Arkin,
1999: rhe IRi seasonal cliniate prediction system and the 1997/
9S El Nifio event. Buill. Amer. Meteor, Soc., 80, 1853-1873.
Meeil, 'G. A., 19S8: Tropical-tmidlatitude interactions in the Indian
and Pacific sectors of the Southern Heniisphere. Mon. Wea. Rev.,
116, 472-484.
Mwafulirwa, N. D.. 1999: Climate variability and pre(lictability in
tropical southern Africa with a foccus on dry spells over Malawi.
M.S. thesis, Geography Deparutnet, University of Zululand, 187
PP.
Nattual Resources Institute, 1996: Drought risk management in southern Africa. Overseas Developmeiit Adnmin. Rep., 43 pp.
Nichols, N., 1997: What are the potential contribtutionis ot El NifioSoutherni Oscillation research to early watriiing of potential acute
food-deficit situations? internet J. Afi. Studies, 2, 59-65.
Rocha, A., and 1. Siminonds, 1997: Interanlnual variability of southern
Africa summer rainfall, 1: Relationships with air-sea interaction
processes. Ifo. J. Climnatol., 17, 235-265.
Ropelewski, C. F., and NI. S. Halpert, 1987: Global- and regionalscale precipitation patterns associated with the El Nino/Soutliern
Oscillation. Mon.
lWea. Rev., 115, 1606-1626.
Rweilamira, J. K., and T. E. Kleynhans, 1996: SADC agricultur.l
potential assessment--country profiles. Development Bank of
South Africa Rep., 124 pp.
Schulze, R. E., 1996: impacts of global climate change and variability
in a hydrologically vulnerable region: Challeitges to hydiolo-
JANIJARY
2002
JU RY
gists. Proc. Workslhop on Reducing Climnate Related Vulnerability in Southern Africa, Victoria 11alls, Zimbabwe, 85- 1 11
Shinoda, M., and R. Kawamura, 1996: Relationships between rainfall
over semi-arid southern Africa and geopotential heights and sea
surface temperatures. J. Meteor. Soc. Japan, 74, 21-36.
Thomson, A., P. Jenden. and E. Clay. 1998: Information, risk and
disaster preparedness: Responses in Southern Africa. Southern
Africa Regional Climate Outlook Forum Rep., 1-1().
VanZyl, J., 1993: Drought and the economy. InidicatorSA, 10, 4751.
55
, il. J. G. Nel, and J. A. Groenewald, 1988: Agriculture's contribution to the South African economy. Agrekon, 27, 10-16.
Vogel, C. H., 1994: T'he impact of extreme drought events. Human
Sciences Research Council Rep., Pretoria, South Africa, 36 pp.
World Economic Forumi, 2000: The Africa Comnjpetitiveness Report
2000. Oxford University Press, 292 pp.
Zhakiata, NV.,1996: Impacts of climate variability and forecasting on
agriculture (Zirmbabwe experience). Proc. Workshop on Reducing Climate-Related lnuiezerability in Southern Africa, Victoria
Falls, Zimbabwe, 131-139.
COPYRIGHT INFORMATION
TITLE: Economic impacts of climate variability in South Africa
and development of resource prediction models
SOURCE: Journal of Applied Meteorology 41 no1 Ja 2002
WN: 0200102564004
The magazine publisher is the copyright holder of this article and it
is reproduced with permission. Further reproduction of this article in
violation of the copyright is prohibited. To contact the publisher:
http://www.ametsoc.org/AMS/..ISN:0894-8763..SCJ:y..FTP:y
Copyright 1982-2002 The H.W. Wilson Company.
All rights reserved.