Rainy-Season Duration Estimated from OLR versus Rain Gauge Data and the 2001
Drought in Southeast Brazil

SERGIO H. FRANCHITO AND V. BRAHMANANDA RAO

Centro de Previsão de Tempo e Estudos Climáticos, Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil

PAULO R. B. BARBIERI

Fundação Cearense de Meteorologia (FUNCEME), Fortaleza, Brazil

CLOVIS M. E. SANTO

Centro de Previsão de Tempo e Estudos Climáticos, Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil

(Manuscript received 12 February 2007, in final form 20 September 2007)

ABSTRACT

Large precipitation deficits observed during the 2001 austral summer over the southeast region of Brazil
contributed to the worsening of the energy crisis that was occurring in the country, with unprecedented
social and economic consequences. Reliable information on the beginning of the rainy season was essential
for the Brazilian government to manage the energy crisis. The purpose of this study is to determine the rainy
season in this region and to point out the risk of using outgoing longwave radiation (OLR) data to estimate
the beginning of it. The results show that when OLR data are used the beginning and the end dates of the
rainy season are wrongly anticipated and delayed, respectively. The present study aims to provide useful
information for the management of the impact of adverse climate conditions such as the one in 2001 by
basing the analysis on rainfall data instead of on OLR.

1. Introduction

The southeast region of Brazil (SEB) is the most
populous and plays an important role in the economy of
the country, being characterized by high industrial ac-
tivity, agricultural productivity, and hydroelectric gen-
eration. Although Ramage (1971) did not consider
South America (SA) as having a true monsoon regime,
some studies revealed that a large part of subtropical
SA experiences typical monsoon circulation (Zhou and
Lau 1998; Gan et al. 2004). Particularly, SEB shows
characteristics of a monsoon system, where the rainy
season is the austral summer and the dry season is the
austral winter.

During the 2001 austral summer the total rainfall was
much lower than the normal values in southeastern,

northeastern, and central Brazil. Large precipitation
deficits observed over these regions contributed to the
worsening of the energy crisis that was occurring in the
country. To control the crisis, the Brazilian government
imposed rationing on the consumption of power in the
beginning of the dry season (June 2001), which contin-
ued until March 2002. This led to an approximately
24% reduction in Brazilian consumption of energy, af-
fecting even other regions where there was no rationing
(Bardelin 2004). This energy crisis provoked significant
social and economic consequences.

To help the Brazilian government to manage the en-
ergy crisis, the Minister of Science and Technology cre-
ated a working group at Center of Weather Prediction
and Climate Studies (CPTEC) that was entrusted with
the task of analyzing the situation of the water level in
the power-plant reservoir and predicting the beginning
of the rainy season. To this working group was given
the responsibility of issuing bulletins in real time. Many
efforts were made to provide accurate information on
the beginning of rainy season, which was essential for
the Brazilian government to manage the energy crisis.

Corresponding author address: S. Franchito, Centro de Previsão
de Tempo e Estudos Climáticos, Instituto Nacional de Pesquisas
Espaciais (INPE), CP 515, São José dos Campos, SP 12245-970,
Brazil.
E-mail: fran@cptec.inpe.br

MAY 2008 F R A N C H I T O E T A L . 1493

DOI: 10.1175/2007JAMC1717.1

© 2008 American Meteorological Society

JAMC1717



Several criteria were used to estimate the beginning of
the rainy season. Some studies used satellite-based out-
going longwave radiation (OLR) data to infer the pre-
cipitation. Although most of the studies that have been
based on OLR anomalies are on the subject of intrasea-
sonal variability of precipitation in SA (Nogues-Paegle
and Mo 1997; Paegle et al. 2000; Carvalho et al. 2004;
Gonzalez et al. 2006; and many others), there are a few
investigations using OLR to identify and predict the
onset of the rainy season in SA, mainly the onset in the
Amazon basin (Kousky 1988; Horel et al. 1989;
Marengo et al. 2001b). Other criteria to determine the
beginning of the rainy season are based on winds (Gan
et al. 2004, 2006) and rainfall data from rain gauge sta-
tions (Rao and Hada 1990; Cavalcanti et al. 2001). For
the particular case of the severe drought of 2001,
Marengo et al. (2001a) used OLR data to suggest that
the rainy season over the basins generating electric en-
ergy in SEB begins between the end of September and
beginning of October. Cavalcanti et al. (2001), using
daily data of precipitation, showed that there is large
variability in the beginning of the rainy season in SEB.
However, on average it begins between the end of Oc-
tober and the beginning of November. So the two meth-
ods used to determine the onset of the rainy season
showed different results. Since OLR is a measure of
high cloud, the criterion commonly used to determine
the rainy season with OLR data—values of OLR lower
than 240 W m2—indicates the presence of deep clouds
that only indicate the probability of precipitation. How-
ever, nonprecipitating high clouds like cirrus also show
OLR values lower than 240 W m�2, and consequently
the method may lead to wrong estimations of rainfall.
Thus, a method to determine the rainy season based on
rain gauges seems to be a more reliable indicator of
precipitation characteristics.

The purpose of the present study is to determine the
beginning of the rainy season in SEB based on rainfall
data and to point out the risk of using OLR data to
estimate it. This study aims to provide information for
the management of the impact of adverse climate con-
ditions such as the severe drought of 2001, which is a
long-range objective of CPTEC.

2. Data and methodology

Rainfall data of 189 rain gauge stations for the period
1981–96 were obtained from Agencia Nacional de En-
ergia Eletrica (ANEEL) and Sistema Integrado de
Gerenciamento de Recursos Hidricos do Estado de Sao
Paulo (SIGRH). Figure 1 shows the location of these
stations. OLR data for the same period were obtained
from NCEP–NCAR reanalysis (Kalnay et al. 1996).

Pentad (5-day averaged) rainfall and OLR data were
calculated.

To determine the rainy season using precipitation
data for each rain gauge station (i ) for each pentad ( j )
for each year (k), the value of the accumulated precipi-
tation at each 5 days [ p(i, j, k)] was compared with
[P5(i)], which represents the 5-day accumulated precipi-
tation if the annual rainfall was uniformly distributed
during the whole year. Starting from 1 January, a ratio
r(i, j, k) � p(i, j, k)/P5(i) was calculated, where P5(i) �
P(i)/73, and P(i) is the mean annual rainfall. As a cri-
terion for the beginning (end) of the rainy season, it was
assumed that at least 50% of the values of r (i, j, k) are
higher (lower) than 1 (threshold value) at four consecu-
tive pentads. Although this criterion might seem arbi-
trary, it was based on some important characteristics of
the premonsoon and monsoon precipitation regimes. In
the premonsoon period, rainfall may occur during some
consecutive days (less than 10 days). These episodes of
intermittent rainfall are not related to the beginning of
the rainy season. The monsoon period (rainy season) is
characterized by regular (not intermittent) rainfall,
which occurs during at least 10 days continuously. How-
ever, after around 30 rainy days, periods with lack of
precipitation may be related to the 30–60 day oscilla-
tion (Jones and Carvalho 2002). After these breaks,
rainy days are again observed. Thus to characterize the
rainy season we considered that the rainfall must occur
continuously for a period of at least 10 days but less
than 30 days. So we assumed that a criterion of the
beginning of the rainy season is that at least one-half of

FIG. 1. Station location map. The schematic map of the entire
country of Brazil including the location of the area of study is
shown in the upper left of the figure. The letters SP, MG, RJ, and
ES correspond respectively to the states of São Paulo, Minas
Gerais, Rio de Janeiro, and Espirito Santo.

1494 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 47



the rain gauge stations (50%) in the region show rain-
fall higher than the threshold value at four consecutive
pentads (20 days).

To determine the rainy season using OLR data, the
method used by Kousky (1988) was adopted. The cri-
teria for the beginning (end) of rainy season were: (i) a
pentad with OLR mean value lower (higher) than 240
W m�2, (ii) at least 10 of the 12 earlier pentads with
mean values of OLR higher (lower) than 240 W m�2,
and (iii) at least 10 of the 12 subsequent pentads with
mean values lower (higher) than 240 W m�2.

The number of the pentads and the respective calen-
dar days relate as follows: pentad 1 corresponds to 1–5
January, pentad 2 to 6–10 January, . . . , pentad 73 to
27–31 December.

3. Results

The prediction of the onset, intensity, and withdrawal
of the monsoon constitutes a set of complex problems
on which many researchers are working (see Sperber
and Yasunari 2006). While the prediction of the arrival
of the monsoon is a long-range goal, the present study
has a more modest and reasonable goal of determining
when the monsoon arrives, that is, to nowcast rather
than to forecast. So we use two different methods to
determine the onset of the monsoon and then compare
the results. To evaluate which of the two methods bet-
ter represents the onset and end dates of the rainy sea-
son, we must first identify these dates. In this paper we
use the definition of the rainy season given by Gan et al.
(2006). They determined the rainy season by consider-
ing that the onset (end) is defined when the precipita-
tion is more (less) than 4 mm day�1 for at least six of
the subsequent eight pentads. Gan et al. (2006) used
this definition to analyze the interannual variability of
the rainy season associated with the monsoon over SA
for an 18-yr period of precipitation time series (1979/
80–1996/97). Their results indicated that the earliest
and latest onset dates occurred in 13–17 September and
12–16 November, respectively. The mean onset and the
standard deviation for the onset were 13–17 October
and 15 days, respectively (see the two first columns of
their Table 1).

First, the results using precipitation data from
ANEEL and SIGRH are presented. As seen in Fig. 2a,
the SEB shows distinct rainy and dry seasons. Starting
from pentad 53 (18–22 September) the rainfall begins
to increase. The change from the dry to the wet season
occurs during pentad 58 (13–17 October). According to
the criteria described in section 2, the beginning of the
rainy season occurs during pentad 59 (18–22 October)
and the end during pentad 18 (27–31 March). The

change from the wet to the dry season occurs around
midautumn (pentad 19, 1–5 April). The period with
highest number of meteorological stations with r � 1
occurs from the beginning of December (pentad 68, 2–6
December) to the end of January (pentad 5, 21–25
January; Fig. 2b). During the rainy season, the percent-
age of meteorological stations with r � 1 is higher than
50% in all of the pentads.

The results shown above indicate that the onset date
of the rainy season obtained using the method based on
rainfall is in agreement with the mean onset date and
the variability of the onset dates for the monsoon in SA
obtained by Gan et al. (2006). However, the end of the
rainy season occurs 3 pentads (15 days) earlier than the
mean demise date given in that study.

Figure 3 shows the distribution of OLR for the be-
ginning and end of the rainy season in SEB. By mid-
September (pentad 52, 13–17 September), a deep con-
vection band, which develops over Amazonia (Fig. 3a),
advances southeastward, arriving in SEB by the end of
September and beginning of October (pentad 55, 28
September–2 October; Fig. 3b). By mid-October (pen-
tad 59, 18–22 October), a large cloudiness band with
OLR values of 240 W m�2 is found over the entire SEB
(Fig. 3c). By mid-April (pentad 21, 11–15 April) the
cloudiness band starts to move toward the equator, and

FIG. 2. (a) Values of r(i, j, k) at each pentad, and (b) the per-
centage of meteorological stations with r(i, j, k) � 1 at each pen-
tad. The dates correspond to the center of the pentads.

MAY 2008 F R A N C H I T O E T A L . 1495



in the beginning of May (pentad 25, 1–5 May) the end
of the rainy season in SEB occurs (Figs. 3d–f). These
results are in general agreement with earlier studies
using OLR data, such as Kousky (1988) and Marengo
et al. (2001a), who found that the beginning of the rainy
season occurred in the beginning of October and be-
tween the end of September and beginning of October,
respectively.

The results presented above show that when OLR
data are used the beginning (end) of the rainy season
occurs four (seven) pentads earlier (later) than when
rainfall data are used. The lag is larger for the end of
rainy season. This may be due to the movement of deep
convection toward the equator that is slower than its
progress southeastward (Kousky 1988; Marengo et al.
2001b). Gan et al. (2004) suggest that the slower move-

FIG. 3. Distribution of OLR values for: (a) pentad 52 (13–17 Sep), (b) pentad 55 (28 Sep–2 Oct, the
beginning of the rainy season), (c) pentad 59 (18–22 Oct), (d) pentad 21 (11–15 Apr), (e) pentad 25 (1–5
May, the end of the rainy season), and (f) pentad 29 (21–25 May).

1496 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 47

Fig 3 live 4/C



ment of deep convection equatorward is perhaps due to
the storage of soil moisture during the rainy season. In
any case, results of the present study show that there is
a substantial difference in the determination of the
rainy season depending on whether OLR or rainfall
data are used. This discrepancy must be taken into ac-
count whenever an accurate determination of rainy sea-
son is crucial to establish strategies to deal with adverse
climate changes such as the 2001 severe drought.

The discussion above considered the mean start and
end of the rainy season when precipitation and OLR
data are used. The same criteria (section 2) were ap-
plied to years of El Niño and La Niña events. During
the period 1981–96, eight El Niño and four La Niña
episodes occurred (additional information is available
online at http://www.nws.noaa.gov). The El Niño and
La Niña episodes and the pentad numbers correspond-
ing to the beginning and end of the rainy season are
given in Fig. 4. During El Niño events, the maximum
delay of the beginning of the rainy season (pentad 69,
7–11 December) occurred in the 1991/92 and 1993/94
episodes, while in the 1982/83 episode the rainy season
extended up to the end of April (pentad 23, 21–25
April). The shortest (17 pentads) and longest (40 pen-
tads) rainy seasons occurred respectively in the 1991/92
and 1982/83 episodes. In the case of the La Niña events,
the earlier end of the rainy season occurred in the 1983/
1984 episode (pentad 6, 26–30 January).

When the determination of the beginning, end, and
duration of the rainy season using OLR data is com-
pared with the values obtained using precipitation data,
large differences can be noted, as seen in Fig. 4. For
example, in the longest rainy season observed during El
Niño events, which occurred in 1982/83, the OLR
method delays the end of the rainy season by 11 pen-
tads. This wrongly overestimates by 13 pentads the du-
ration of the rainy season. The beginning and the end of
the shortest rainy season observed in the 1991/92 El
Niño event were respectively anticipated and delayed
by 16 pentads, thereby causing an erroneous increase of
32 pentads in the duration of the rainy season with the
use of OLR data. During years of La Niña events, large
differences between the two methods are also seen. In
particular, in the 1983/1984 episode there was an erro-
neous delay of the end of the rainy season (pentad 29,
21–25 May) with OLR data when compared with the
result when precipitation data were used (pentad 6, 25–
29 January). This discrepancy caused an apparent in-
crease of 26 pentads in the duration of the rainy season.
One should, however, be careful in analyzing the results
of the differences between El Niño and La Niña events
because of the fact that the number of events is not
large in the time series used in the present study. The

conclusions should be reconsidered with further studies
using a longer time series.

The results of the present work show that there is an
overestimation of the rainy season using OLR data.
The anticipation of the onset of rainy season using
OLR data is in agreement with the results of other
studies on the energy crisis in SEB (see Marengo et al.
2001a and Cavalcanti et al. 2001). We tested the same
criteria based on rainfall (as described in section 2) for
the meteorological stations located in the region of the
basins responsible for the generation of electric energy
in SEB and found that the beginning of the rainy season
in SEB occurs at the end of November. The definition
we used to determine the onset of the rainy season is a
rigorous criterion in which the rainfall amount was
enough for raising the low water levels in the reservoirs

FIG. 4. Beginning, ending, and duration of rainy season during
El Niño and La Niña events using precipitation (P) and OLR
data. Also shown are the mean climatological data (Mean).
An asterisk indicates the shortest rainy season and the earlier
beginning or end of rainy season; a double asterisk corresponds to
the longest rainy season and the maximum delay in the beginning
or end of rainy season.

MAY 2008 F R A N C H I T O E T A L . 1497



during adverse climate conditions such as the drought
of 2001. Another test of the method based on rainfall
data was made, this time considering all the available
meteorological stations of São Paulo State (SP in Fig.
1). Based on the same considerations about the pre-
monsoon and monsoon precipitation regimes men-
tioned in section 2, it was assumed as a criterion for the
beginning of the rainy season that r(i, j, k) was higher
than 1 for at least three consecutive pentads or for two
pentads followed by three pentads among four consecu-
tive pentads. This criterion allowed an earlier onset of
the rainy season when compared with that assumed in
section 2, namely, r(i, j, k) higher than 1 for four con-
secutive pentads. This new definition is sensitive to the
necessities of many sectors of society, such as agricul-
ture, human activities, and even the electric sector (in
situations different from that in the severe drought of
2001, when a heavy precipitation in the rainy season
was needed to raise the low levels of water in the res-
ervoirs). Despite this relaxation of the criterion, the
onset of the rainy season still occurred between the end
of October and beginning of November. Thus, even
when the onset of the rainy season is defined less strin-
gently, the onset as measured by rainfall data occurs
later than it does when measured by OLR data.

The overestimation of the rainy season using OLR
data may be attributed to fact that OLR is a measure of
high cloud, but not necessarily of rainfall. The criterion
described in section 2—values of OLR lower than 240
W m�2—indicates the presence of deep clouds that
show only the probability of precipitation. Since non-
precipitating high clouds like cirrus also show OLR val-
ues lower than 240 W m�2, the method may overesti-
mate the precipitation.

Although in the present study the validity of the rain
gauge network data for estimation of precipitation is
taken as granted, the method based on pentad rainfall
data, described in section 2, contains some limitations.
Since the method is based on precipitation time series
data, the estimation of the rainy season assumes that
the rainfall behavior is similar to that of the past. De-
spite this limitation, the onset date of the rainy season
obtained using this method is in agreement with the
mean onset date and the variability of the onset dates
for the monsoon in SA (Gan et al. 2006). Another
method to determine the onset of the rainy season,
which uses climate model’s predictions, is based on
winds. Since the skill of climatic models to predict the
wind components is better than to predict precipitation,
monsoon indices based on wind changes may be used to
estimate the rainy season in the future. This method has
been applied with some success to the west-central re-
gion of Brazil (Gan et al. 2004, 2006). However, there is

a need for similar tests for the various regions of the
country.

4. Conclusions

In this study, the rainy season in SEB was determined
and the risk of using OLR data for this purpose was
demonstrated. In addition, the changes in the beginning
and end of the rainy season during El Niño–La Niña
events as derived from OLR data were also examined.
The study aimed to provide information for the man-
agement of the impact of adverse climatic conditions
such as the severe drought in 2001. Pentad rainfall data
from 189 rain gauge stations and OLR data from
NCEP–NCAR reanalysis were used. The results
showed that when precipitation data were used, the
beginning and the end of the rainy season in SEB oc-
curred respectively during pentad 59 (18–22 October)
and pentad 18 (27–31 March). When OLR data were
used, the beginning and the end of the rainy season
were wrongly anticipated (four pentads earlier) and de-
layed (seven pentads later), respectively. When OLR
data were used for the determination of the rainy sea-
son during El Niño and La Niña events, large differ-
ences in relation to the precipitation data were also
noted. In the case of El Niño episodes, the beginning
and the end of the rainy season were wrongly delayed
and anticipated, respectively, up to 16 pentads while in
the case of La Niña episodes, the duration of the rainy
season was erroneously increased up to 26 pentads.

Thus, the wrong inference of the extent of rainy sea-
son with OLR leads to the conclusion that these data
should be used carefully, especially in the case of ad-
verse climatic conditions such as the severe drought of
2001, when accurate information about the beginning
of the rainy season was essential for the Brazilian gov-
ernment to manage the energy crisis. Since the date of
the onset of the rainy season obtained with precipita-
tion data showed a good agreement with that obtained
for the monsoon in South America, the criterion used
here may be useful to determine the rainy season in
SEB. The present study has attempted to provide
meaningful and reliable information for decision mak-
ers.

Note that in regions where surface rain gauge sta-
tions are not available, the only available precipitation
data may be those inferred from satellite data (by OLR
and other means). However, in view of our results,
great care should be exercised in inferring quantitative
rainfall.

Acknowledgments. Thanks are given to Dr. Carlos A.
Nobre for the efficient organization of the energy crisis

1498 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 47



group at CPTEC. Thanks are also given to the anony-
mous reviewers for useful suggestions.

REFERENCES

Bardelin, C. E. A., 2004: The effects of rationing electric power in
Brazil in 2001 and 2002 with emphasis on the electric energy
consumption (in Portuguese). M.S. dissertation, Polytechnic
School, University of São Paulo, 113 pp.

Carvalho, L. M. V., C. Jones, and B. Liebmann, 2004: The South
Atlantic convergence zone: Intensity, form, persistence, and
relationships with intraseasonal to interannual activity and
extreme rainfall. J. Climate, 17, 88–108.

Cavalcanti, I. F. A., C. A. Souza, V. E. Kousky, and E. B. M.
Barbosa, cited 2001: Precipitation anomalies in the southeast
region of Brazil (in Portuguese). [Available online at http://
www.cptec.inpe.br/energia/saiba/portalpop2.shtml.]

Gan, M. A., V. E. Kousky, and C. F. Ropelewski, 2004: The South
America monsoon circulation and its relationships to rainfall
of over west-central Brazil. J. Climate, 17, 47–66.

——, V. B. Rao, and M. C. L. Moscati, 2006: South American
monsoon indices. Atmos. Sci. Lett., 6, 219–223.

Gonzalez, P. L. M., C. S. Vera, and B. Liebmann, 2006: Charac-
teristics of the intraseasonal variability of precipitation over
eastern Argentina. Proc. Eighth Int. Conf. on Southern Hemi-
sphere Meteorology and Oceanography (ICSHMO), Foz do
Iguacu, Brazil, Instituto Nacional de Pesquisas Espaciais,
1063–1088.

Horel, J., A. Hahmann, and J. Geisler, 1989: An investigation of
the annual cycle of convective activity over the tropical
Americas. J. Climate, 2, 1388–1403.

Jones, C., and L. M. V. Carvalho, 2002: Active and break phases

in the South American monsoon system. J. Climate, 15, 905–
914.

Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Re-
analysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.

Kousky, V. E., 1988: Pentad outgoing longwave radiation clima-
tology for the South American sector. Rev. Bras. Meteor., 3,
217–231.

Marengo, J. A., L. M. Alves, C. A. C. Castro, and D. Mendes,
cited 2001a: Information on the beginning of the rainy season
in the southeast, west-central and south of northeast regions,
which are affected by the energy crisis (in Portuguese).
[Available online at http://www.cptec.inpe.br/energia/saiba/
iniciochuvas.shtml.]

——, B. Liebmann, V. E. Kousky, N. Filizola, and I. Wainer,
2001b: Onset and end of the rainy season in the Brazilian
Amazon Basin. J. Climate, 14, 833–852.

Nogues-Paegle, J., and K. C. Mo, 1997: Alternating wet and dry
conditions over South America during summer. Mon. Wea.
Rev., 125, 279–291.

——, L. A. Byerle, and K. C. Mo, 2000: Intraseasonal modulation
of South American summer precipitation. Mon. Wea. Rev.,
128, 837–850.

Ramage, C. S., 1971: Monsoon Meteorology. Academic Press, 296
pp.

Rao, V. B., and K. Hada, 1990: Characteristics of rainfall over
Brazil: Annual variations and connections with Southern Os-
cillation. Theor. Appl. Climatol., 42, 81–91.

Sperber, K. R., and T. Yasunari, 2006: Workshop on monsoon
climate systems: Toward better prediction of the monsoon.
Bull. Amer. Meteor. Soc., 87, 1399–1403.

Zhou, J., and K.-M. Lau, 1998: Does a monsoon climate exist over
South America? J. Climate, 11, 1020–1040.

MAY 2008 F R A N C H I T O E T A L . 1499