GEOPHYSICAL RESEARCH LETTERS, VOL. 23, NO. 7, PAGES 729-732, APRIL 1, 1996
Satellite-based identification of linked vegetation index and
sea surface temperature anomaly areas from 1982-1990
for Africa, Australia and South America
R•ng• B. Myneni
Department of Geography,University of Maryland and BiosphericSciencesBranch,
NASA Goddard Space Flight Center, Greenbelt, Maryland
Sietse 0. Los
ScienceSystems Applications Inc. and Biospheric SciencesBranch,
NASA Goddard SpaceFlight Center, Greenbelt, Maryland
Compton J. Tucker
BiosphericSciencesBranch, NASA Goddard SpaceFlight Center, Greenbelt,Maryland
Abstract.
Meteorological
satellitedatafrom 1982to 1990 Data
wereusedto identify areasof significantassociationbetween
and
Methods
The National Ocearficand Atmospheric Administration's
tropicalPacificseasurfacetemperature(SST) andremotely
sensednormalizeddifferencevegetationindex (NDVI) ano- polar-orbiting meteorologicalsatellitesobtain daily global
malies,heretaken asa surrogatefor rainfall anomalies.During this period, large areas of arid and semi-arid Africa,
Australia and South America experiencedNDVI anomalies
directly correlated to tropical Pacific SST anomalies. The
results are limited by the relatively short time period of
analysis. However, they confirm the disruptive effects of
large-scaletropical Pacific SST variations on arid and semiarid continental rainfall patterns in Africa, Australia, and
South America, as reported previously.
Introduction
information from the advanced very high resolution radio-
meter (AVHRR) instrument. Data from the .AVHRR in-
strumentare usedto producea vegetation
•ndexdirectly
related to the density of chlorophyll contained in plants on
land [Sellers,1985]. The satellite-sensed
vegetationindexis
expressed
asa normalizeddifferenceof two channels[(Chan-
nel 2- Channel1) / (Channel2 + Channel1)], whereChannel I recordsreflected light from 0.55-0.70/•m and Channel
2 recordsreflectedlight from 0.72-1.1/•m. Combiningdata
from these adjacent spectral regionscompensatesfor differences in irradiance and provides an estimate of the intercepted fraction of the photosynthetically active radiation or
capacity[Asrar et al., 198•]. Thesedata are
Severalworkershave comparedsurfaceprecipitationpat- photosynthetic
strongly
related
to total primary production when summed
terns with E1 Nifio Southern Oscillation cycle sea surface
temperature(SST) anomalies,identifyingareaswhererain- or averagedoverthe growingseason[Funget al., 1987].
Daily daytime 4-kin AVHRR data were usedfrom NOAA-
fall anomalieswere correlatedwith tropical Pacific SST anomalies [McBride and Nicholls, 1983; Ropelewskiand Hal-
7 (1982 to 1984), NOAA-9 (1985 to 1988), and NOAA-11
pert, 1987; 1989]. Point rainfall frequentlydoesnot ade- (1989 to 1990) coveringAfrica, Australia,and SouthAmerquatelyrepresentareal rainfall, especiallyin semi-aridtrop- ica. The eruption of Mt. Pinatubo in mid-1991 injected a
ical regions where rainfall is often erratically distributed.
Thus accurateestimatesof the geographicalextent of SST
linked precipitation anomalieshas been hamperedin many
areas. We have used satellite-derived
normalized
3OO
SeaSurfaceTemperature(C) Anomal) (198•1990)
difference
vegetationindex (NDVI), whichseveralstudieshaveshown
to be highlycorrelatedwith arid and semi-addrainfall [Malo
and Nicholson, 1990; Nicholsonet al., 1990; Tucker et al.,
1991]. Our approachhas been to correlatetropical Pacific
2OO
SST anomalies
from the NINO3 region(5ø S to 5ø N latitude and 90ø W to 150ø W longitude)and NDVI anoma-
*'
000 A
Se
p188
Apnl190
lies, thus delineating anomalousrainfall areas through direct measurement of vegetation amount and condition in a
spatially continuousmanner for arid and semi-add Africa,
South America, and Australia.
-3 •
I s2 i 83 ! 84 I 85 1 86 I 8- I 88 I 89 ! •
Copyright1996by theAmericanGeophysical
Union.
I
Figure 1. SST anomaliesfrom the tropical Pacific NINO3
Papernumber96GL00266
0094- 8534/96/96GL-00266503.00
area of 5øS to 5øN latitude and 90øW to 150øW longitude
from 1982 to 1990 [Woodruffet al., 1993].
729
730
MYNENI
ET AL.: LINKED
VEGETATION
INDEX
AND SST ANOMALY
MAY B•
-
"FL
T•ELVE
Figure 2. Summaryof areaswith linkedtropicalPacific
MONTII
aPRIL
AREAS
03
NINO"
NI)VI
ANOMA[.•
-0.1
0.1
C(}MPOSIrE
SST and NDVI anomalies for Africa, Australia and South
America. The 1982-90time period wasdividedinto 4 ENSO
events,characterizedby a consecutive
12 monthperiodthat
includedthe peak SST anomaly and the southernhemispheresummerrainfallperiod(May 1982- April 1983;March 1985 - February 1986;March 1987- February 1988;May
1988- April 1989). For everyland pixel, correlations
were
performedbetweenNDVI and SST anomaliesfor eachof
MAY BB -
APRIL
B9
"LA N1NA'
the 4 events. Areas with 2 occurrences are labeled blue,
3 occurrencesare labeled pink, and 4 occurrencesare la-
beledyellow. An occurrence
is definedas (1) correlation
coefficients
greaterthan [0.5[(2) cumulative
twelvemonth
NDVI anomalygreaterthan0.3 or lesserthan-0.3, and(3)
surroundedon all sidesby eight similarlypixels. The con- Figure 4. Areas of linked tropical PacificSST and NDVI
tours indicateregionsfrom whichthe time seriesshownin anomalies for Australia and South America. The pixels identified are those with SST-NDVI anomaly correlation coefFigure 5 and the entriesin Table 1 werebased.
ficients
greaterthan 10.5[.Thin linesplottedon the South
Americanimageare rainfallcontoursof •750 mm/year;the
boldlinesindicate•1500 mm/year of rainfall. For Australia
significantquantity of aerosolsinto the ßstratosphere
which
thesecontoursindicate•300 and •500 mm/year of rainfall.
preventedus from using data for 1991 onwards. Data were
processedinto a maximum value NDVI compositeby month
which minimizes atmospheric effects, cloud contamination,
scanangle effects,and solar zenith angleeffects.Data were time serieswas determined,pixel by pixel, by evaluatingthe
numerically calibrated to a common referenceto permit id- crosscorrelationcoefficientusing the procedureCORREL
entification of NDVI anomaliesamong satellitesand have a outlinedin NumericalRecipes[Presset al., 1986]. This
spatialresolutionof •7.6km [Loset al., 199•].
method identifies land areas where NDVI
anomalies are sim-
Pacific sea surface temperature anomaliesfrom 5øS to ilar to SST anomalies, thus providing direct measurement
5øN latitude and 90øW to 150øW longitudewere obtained of location and areal extent of affected arid and semi-arid
from the ComprehensiveOcean- AtmosphereData Set pro- areas. Our techniqueis not accurate for areas receiving
gram[Woodruff
et al., 1993].MonthlymeanNDVI was morethanm1000
mm/yrprecipitation
astherelationship
computed
andanomalies
fromeachmonthdetermined.
The betweenNDVI andrainfalltendsto saturate.
similarity between the SST and NDVI
monthly anomaly
Est. NDVI anomaly = 0.028 -0.079'(SST
anomaly)
Est. NDVI Anomaly =
r^2
r2 = 0.66
- 0.053 + 0.102 {SST Anomaly)
= 0.63
0.3
ßwarm
ENSO'82-'83
l
ß
0.2
0.1
warm ENSO'87-'8Sl
0.0
•E
o
-0.1
0
'-
-•.2
ß
ß
warm ENSO'87-'88
A
cool ._NSO'88-'89
ß
ß
ß
-0.3
'
ß
•
"
i
i
1
Monthly
ea S face Temperature.
Anomaly (.C) @ (5N-5S by 90W-150W)
2
3
Monthly. Sea Surface-Temperature
Ancmaly (C) @ (5N-5S by 90W-150W)
Figure3. Linked
SSTanomalies
fromthetropical
Pacific
NINO3areaandNDVIanomalies
for(a)Northeastern
Brazil
and (b) EasternAfrica.
MYNENI ET AL.: LINKED VEGETATION INDEX AND SST ANOMALY AREAS
731
Table 1. Areas of Africa, Australia and South America where linked tropical Pacific SST and NDVI anomalieswere
determinedfrom 1982 to 1990. The areasin this table are the areaswheretime seriescorrelationcoefficients
are greater
than 10.5]andwherethe cumulative
twelvemonthNDVI anomaly
wasgreaterthan0.3 or lesserthan-0.3. The regions
are shownin Figure 2.
ENSO Cycle Sea Surface Warming
May 1982 to Apr 1983
Region
Northeastern Brazil
Southeastern S. America
Eastern Australia
Central Australia
Eastern Africa and Horn
Southeastern Africa
Central Southern Africa
Results
and
ENSO Cycle Sea Surface Cooling
Mar 1987 to Feb 1988
Mar 1985 to Feb 1986
May 1988 to Apr 1989
+ve NDVI
-ve NDVI
+ve NDVI
-ve NDVI
+ve NDVI
-ve NDVI
+ve NDVI
-ve NDVI
Anomalies
Anomalies
Anomalies
Anomalies
Anomalies
Anomalies
Anomalies
Anomalies
75,000
406,200
106,400
67,800
245,900
137,500
34,200
360,700
285,200
499,100
313,400
260,200
170,900
212,300
1,001,700
26,500
557,000
82,600
996,600
173,100
105,200
744,400
16,500
201,300
1,665,800
549,900
16,800
73,300
373,900
93,400
670,400
83,700
8,600
406,800
256,800
20,900
72,900
759,800
307,300
100,300
25,700
19,500
83,200
485,100
1,000
62,500
555,200
34,800
125,600
498,600
144,300
1,743,500
39,700
4,900
391,000
60,700
areas studied, was the least affected in terms of area of linked
Discussion
tropical Pacific SST and NDVI anomalies(Table 1). Our
Duringthe 1982to 1990time periodthe easternequato- analysis identified three areas of Africa which were most
rial Pacific experienced two warm and two cold sea surface
strongly associatedwith linked tropical Pacific SST and co-
temperatureepisodes(Figure 1). Of these,the 1982-1983
ENSO cycleSST warmingand the 1988-1989ENSO cycle
SST coolingwere more pronounced.Our analysisidenti-
(a)
fied sizeableareas of Africa, Australia, and South America
where precip.itationand henceour NDVI anomalieswere
SST
N
NORDES rE, BRAZIL
,
2.0
associated
with tropicalPacificSST anomalies
(Table1).
SouthAmerica and Australia weremoredirectly associated
with linked tropical Pacific SST and NDVI anomalies than
-2.(}
wasAfrica (Figure 2).
We confirmpreviousanalysesfor NortheasternBrazil wh-
ich haveassociateddroughtconditionswith tropicalPacific
SST warmingsand excessive
rainfall with tropical Pacific
SST coolings[Chu, 1991]. We foundareasrangingfrom
360,700
km2 to 1,001,700
km2 whereNDVI/precipitation
and SST anomalies were linked in northeastern Brazil. Fur-
182 I 831 84 I gSI 861 871•1891
thermore, we found an inverselinear relationshipbetween
the NINO3 index and Northeastern Brazil NDVI
•1
anomalies
(Figure 3). By contrast,southeastern
SouthAmericaexperiencedgreaterthan averagerainfallwith tropicalPacific
SST warmings;tropical Pacific SST coolingswere associated with both excessive
rainfallin March 1985to February
1986andseveredroughtin May 1988to April 1989(Table
1).
EasternAustraliawasfoundto experience
negativeNDVI
anomaliesfrom both tropicalPacificSST warmingsandcoolings;the largestmagnitudeNDVI anomalywasindicative
ibl..SST
SOUTttEA
At'-RICA
o
•ø'ø
I
-2.0
-•1--'
of a severedrought in 1982-1983associatedwith the 1982-
1983tropicalPacificSST warming.We estimatean area of
1,743,500
km• wasstrongly
associated
withthe 1982-1983
ENSO cycleSST warmingevent;by comparison,
the ENSO
cycleSSTcoolingin 1985-1986
and1988-89affected744,400
ioo
¾NDVI
(times 25)
.,,•, .•...k
L
_o
---•o•
i s2 i 83 I 84 • $5 i s6 I 87 i gs I 80 i 9o i
km:•and549,900
km:•,respectively.
Thesituation
forcen- Figure 5. Time seriesplots for tropical Pacific SST and
tral Australiais lessclear;a majornegativeNDVI and precipitationshortfallanomalyoccurredwith the minor ENSO
cycle SST coolingin 1985-1986,while other ENSO cycle
SST anomalyeventshad positiveand negativeNDVI and
hencerainfall anomalies[Allan, 1991]. In addition,Australian workers[Foranand Pearce,1989]havereporteda
NDVI anomalies from 1982-1990. Also plotted at the bot-
tom of each figure is the rainfall anomaly [Dai and Fung,
1993].The NDVI time seriesis the averagefrom pixelswith
two or more occurrencesof significant associationbetween
SST and NDVI anomalies(crosscorrelationcoefficients>
10.sI).Thepixelsarefroman areawithintheregions
shown
in Figure2: (a) NortheasternBrazil (2.6øS- 12øSx 36.4ø
highly anomalousgreeningof central Australia associated W- 42.6øW) and (b) Southeastern
Africa (Zimbabweand
Mozambique:20.0øS- 25.0øN x 29.2øE- 33.8ø E). The
withthe 1988-1989
ENSOcycleSSTcooling(Figure4).
Africa,'situatedfurthestfrom the tropicalPacificof the rainfall anomaly value is the average for the box.
732
MYNENI
ET AL.: LINKED
VEGETATION
incident NDVI anomalies: parts of Eastern Africa extending
toward the Horn of Africa; Southeastern Africa; and Central
Southern
Africa.
Eastern Africa and the Horn of Africa showeda tendency
INDEX
AND SST ANOMALY
AREAS
mating absorbed photosynthetic radiation and leaf area index
from spectral reflectance in wheat, Agron. J., 76, 300-306,
1984.
Cane, M. A., G. Eshel, and R. W. Bushland,ForecastingZimbab-
ings, which was more pronouncedwith the strong ENSO
cycle SST warming of 1982-1983. SoutheasternAfrica experienceddrought conditionsin 1982-1983associatedwith
the ENSO cycle SST warming of that time and precipitation excessassociatedwith the minor ENSO cycle SST cool-
wean maize yield using eastern equatorial Pacific sea surface
temperature, Nature, 370, 204-205, 1994.
Chu, P., Brazil's climate anomalies and ENSO, in Teleconnections Linking Worldwide Climate Anomalies, edited by M. H.
Glantz, R. W. Katz, and N. Nicholls, pp. 43-72, Cambridge
Univ. Press, New York, 1991.
Dai, A., and I. Y. Fung, Can climate variability contribute to the
"missing" CO2 sink?, Global Biogeochem. Cycles, 7, 599-606,
ingof 1985-1986;
in bothcases
m400,000
km:•of areawas
Foran, B., and G. Pearce, The Use of NOAA satellites and the
toward wetter conditionswith tropicalPacificSST warm-
identifiedin our analysisfor Southeastern
Africa(Figure5).
The influenceof eastern equatorial Pacific SST on vegeta-
1993.
green vegetation index to assessthe 1988/89 summer growing season in central australia, CSIRO Rangelands Research
Report, Alice Springs, Australia, 1989.
tion in this regionis so strongthat Cane et al. have recently proposedforecastingZimbabweanmaize yieldsusing Fung, I. Y., C. J. Tucker, and K. C. Prentice, Application of
advanced very high resolution radiometer vegetation index to
NINO3 SST anomaly[Cane et al., 199•]. CentralSouthern
study of atmosphere-biosphereexchangeof CO2, J. Geophys.
Africa was most strongly associatedwith tropical Pacific
Res., 9œ, 2999-3015, 1987.
SST coolings,but in an unpredictablefashion. ENSO cy- Los, S. O., C. O. Justice, and C. J. Tucker, A global I X 1
NDVI
data set for climate studies derived from the GIMMS
cle SST coolingsin the 1980'swerefound to be associated
with a positive NDVI or precipitation anomaly of 759,800
continental NDVI data, lnt. J. Remote Sens., !5, 3493-3518,
1994.
km2, whilethe ENSOcycleSSTcooling
of 1985-1986
was
Malo, A. R., and S. I3. Nicholson,A study of rainfall and vegetation dynamicsin the African SahelusingNormalized Differassociatedwith a negativeNDVI or precipitationanomalies
ence Vegetation Index, J. Arid Environ., 19, 1-24, 1990.
covering
406,800
km2 (Table1).
McBride, J. L., and N. Nicholls, Seasonal relationships between
Our analysisconfirmsthe disruptiveeffectsof large-scale
Australian rainfall and the Southern Oscillation, Mon. Wea.
SST anomalieson regional-scaleprecipitationpatternsand
Rev., 111, 1998-2004, 1983.
thereforevegetationin arid and semi-aridAfrica, Australia, Nicholson, S. E., M. L. Davenport, and A. R. Ma!o, A comand South America for the time period of 1982-1990. While
some areas such as Northeastern Brazil were consistently
parision of the vegetation response to rainfall in the Sahel
and East Africa, using Normalized DifferenceVegetation Index from NOAA AVHRR, Clim. Change, I7, 209-241, 1990.
dryerduringtropicalPacificSST warmingsandconsistently Philander, S.G. H., Unusual conditions in the tropical Atlantic
wetterduringSST coolings,no comparable
pattern waseviocean in 1984, Nature, $œœ,236-238, 1986.
dent from the other areasidentified by our analysisin Africa,
Australia, and South America (Table 1). This must not
be construed as true of all warm or cold events -- 9 years
Press, W. H., B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, Numerical Recipes,818pp, Cambridge University Press,
Cambridge, 1986.
Ropelewski, C. F., and M. S. Halpert, Global and regional scale
is obviouslytoo short a time period to estabisha pattern
precipitation patterns associatedwith the El Nifio/Southern
with statistical certainity. However, it is clear that tropiOscillation, Mon. Wea. Rev., 115, 1606-1626, 1987.
cal Pacific SST anomaliesperturb continentalprecipitation Ropelewski,C. F., and M. S. Halpert, Precipitationpatternsassociated with the high index phase of the Southern Oscillation,
patterns and thereforevegetationgrowthoverlarge spatial
J. Climate, œ,268-284, 1989.
Sellers, P. J., Canopy reflectance, photosynthesisand transpiration, lnt. J. Remote Sens., 6, 1335-1372, 1985.
Acknowledgements. The authorsthankA. Dai andI. Fung Tucker, C. J., H. E. Dregne, and W. W. Newcomb, Expansion and
for the rainfall anomaly data set and P. Schopffor the sea surcontraction of the Sahara desert from 1980 to 1990, Science,
scales.
face temperaturedata. The commentsof two anonymousreœ55, 299-301, 1991.
viewersare greatlyappreciated.This work has beensupported Woodruff, S. D. et al. ComprehensiveOcean-AtmosphereData
by NASA'sMissionto PlanetEarth research
program.
Set (COADS) Releasela: 1980-1992, NOAA, Washington,
D.C., 1993.
References
R. B. Myneni, S. O. Los, C. J. Tucker, Biospheric Sciences
Allan, R. J., Australasia,in Teleconnections
Linking Worldwide Branch, Mail Code 923, NASA Goddard Space Flight Center,
ClimateAnomalies,editedby M. H. Glantz,It. W. Katz, and Greenbelt, MD 20771.
N. Nicholls,pp. 73-120,CambridgeUniv. Press,New York,
1991.
Asrar, G., M. Fuchs,E. T. Kanemasu,and J. L. Hatfield,Esti-
(ReceivedJuly 10, 1995;revisedJanuary 10, 1996;
January16, 1996.)