Variability in the discharge of South American rivers and in climate / Variabilité des débits de rivières d'amérique du Sud et du climat

Transcription

1 Hydrological Sciences Journal ISSN: (Print) (Online) Journal homepage: Variability in the discharge of South American rivers and in climate / Variabilité des débits de rivières d'amérique du Sud et du climat To cite this article: (25) Variability in the discharge of South American rivers and in climate / Variabilité des débits de rivières d'amérique du Sud et du climat, Hydrological Sciences Journal, 5:3, -478, DOI: /hysj To link to this article: Published online: 15 Dec 29. Submit your article to this journal Article views: 52 View related articles Citing articles: 38 View citing articles Full Terms & Conditions of access and use can be found at Download by: [ ] Date: 19 December 217, At: 4:39

2 Hydrological Sciences Journal des Sciences Hydrologiques, 5(3) June Variability in the discharge of South American rivers and in climate NORBERTO O. GARCÍA 1 & CARLOS R. MECHOSO 2 1 Unidad de Investigaciones Hidroclimáticas, Facultad de Ingeniería y Ciencias Hídricas, Universidad Nacional del Litoral, CC 217 Ciudad Universitaria, 3 Santa Fe, Argentina nogarcia@fich.unl.edu.ar 2 Department of Atmospheric Sciences, University of California Los Angeles, 7127 Math Sciences Building, 45 Hilgard Avenue, Los Angeles, California , USA mechoso@atmos.ucla.edu Abstract Changes in trend and quasi-periodicities are sought in the time series of river discharges in all major South American basins. The relationship between trends and quasi-periodicities found and climate variations on interannual and longer time scales are discussed. Consideration of multiple rivers gives insight into the geographical extent of hydrological signals and climate impacts. It is found that the streamflow of all major rivers of South America has experienced an increased trend since the early 197s. It is suggested that this simultaneity may reflect the impact of a large-scale climate change. All the time series of river streamflows that were analysed show El Niño-like periodicities. Only for La Plata Basin do these explain a larger part of the total variance than the other quasi-periodicities. There are two other quasi-oscillations in the time series analysed: one of them with a longer period around 17 years and the other of about 9 years. Previous work has related these oscillations to sea-surface temperature anomalies in the Atlantic Ocean. Key words changes in trends; climate variability; ENSO impacts; periodicities Variabilité des débits de rivières d Amérique du Sud et du climat Résumé Des changements de tendance et des quasi-périodicités sont recherchés dans les séries temporelles de débits des principaux bassins d Amérique du Sud. La relation entre les tendances et les quasi-périodicités identifiées, et les variations climatiques interannuelles et à des échelles de temps plus longues, est discutée. La prise en considération de multiples rivières donne une meilleure vision de l extension géographique des signaux hydrologiques et des impacts du climat. Il apparaît que les débits de toutes les principales rivières d Amérique du Sud suivent une tendance croissante depuis le début des années 197. Nous suggérons que cette simultanéité reflète l impact d un changement climatique à grande échelle. Toutes les séries temporelles de débit qui ont été analysées montrent des périodicités semblables à celle d El Niño. Mais celles-ci n expliquent une part de la variance totale plus grande que ne le font les autres quasi-périodicités, que pour le basin de La Plata. Il existe deux autres quasi-oscillations dans les séries temporelles étudiées: l une avec une longue période autour de 17 ans et l autre avec une période de l ordre de 9 ans. Un précédent travail a relié ces oscillations à des anomalies de température de surface de la mer dans l Océan Atlantique. Mots clefs changements de tendances; variabilité climatique; impacts de l ENSO; périodicités INTRODUCTION Discharges of major rivers provide an integrated view of the hydrological properties of their drainage basins and, as such, contain valuable information on regional climate variability. This information is particularly valuable in basins poorly covered by climate monitoring stations (Hastenrath, 199) or with short climate records. However, the relationships between climate variability in a basin and the behaviour of its rivers are complex and eventually basin-dependent. Most studies on regional climate variability based on river behaviour represent the latter in terms of streamflow time series at one or more gauging stations. River streamflows are influenced by many factors inclu- Open for discussion until 1 December 25

3 46 ding events external to the physical system. García (2), for example, argued that different sub-basins of La Plata Basin have been affected by anthropogenicallyinduced changes, such as deforestation and changes in land use consisting of substitution of coffee plantations for annual crops. Nevertheless, the association between river streamflows and regional climate can be such that the former amplify the latter (Chiew et al., 1995; Berbery & Barros, 23). This study examines the time series of streamflows for large rivers in all major catchments of South America: Amazon, Orinoco, Tocantins, San Francisco, Paraná, Paraguay, Uruguay and Negro. Hydrological fluxes in these catchments are comparable to those in other major rivers of the world (Sellers, 1965; Street-Perrot et al., 1983). The aim is to find changes in trend and quasi-periodicities in the time series, and to relate those changes to climate variations on interannual and longer time scales. Consideration of multiple rivers allows for an insight into the geographical extent of hydrological signals and climate impacts. The authors recognize that some of the basins selected for study are large enough to be influenced by more than one climate system. Maheu et al. (23), for example, demonstrated that interannual variations in both the South American monsoon system (SAMS) and the extratropical atmospheric circulation can affect the Paraná and Paraguay rivers in La Plata Basin. Ambiguities of this sort are more easily addressed in a multiple-river context. On interannual time scales, several studies have demonstrated that El Niño/ Southern Oscillation (ENSO) can impact the behaviour of South American rivers. Poveda & Mesa (1997) examined the links between ENSO and river discharges in Colombia, and reported that those links develop progressively later in the calendar year from east to west. They also found that impacts during the positive phase of the Southern Oscillation (cold events in the Equatorial Pacific referred to as La Niña) are more pronounced than those during the positive phase (warm events or El Niño). Poveda et al. (21) showed that ENSO impacts are comparatively stronger for streamflow than for precipitation, due to concomitant effects on soil moisture and evapotranspiration. Hastenrath (199) found anomalously high river discharges almost everywhere in northern South America (except in the Orinoco basin) during the high phase of the SO. Mechoso & Perez Iribarren (1992) demonstrated that streamflows of the Uruguay River at Salto Grande (31 9 S, 58 W) tend to be above average during November February in El Niño years, with values that are occasionally well above average. Streamflows of the same river have a clear tendency to be below average during cold ENSO events. Marengo (1995) found interannual variations that characterize the hydrology of tropical South America in association with the extreme phases of the SO. In particular, he detected the signature of El Niño events in the river flows of northwestern Peru, southern Brazil, and northern Argentina, as well as in the rainfall over Northeast Brazil, while only very strong events seem to affect the Negro River in the northwestern portion of the Amazon Basin. Robertson & Mechoso (1998) examined the power spectrum of the combined Uruguay, Negro and Paraná Paraguay river discharges and found evidence of nonlinear trends and of oscillatory components with periods of around 3.5 and 6 years. Bischoff et al. (2) inspected the relationships between the different phases of ENSO and the timing of strong floods in the Uruguay River basin. They demonstrated that flows during warm and cold events are significantly different. In general, the discharges of the Uruguay River tend to be

4 Variability in the discharge of South American rivers and in climate 461 above the mean from the southern spring through to the next March during the warm phase of ENSO, and the largest occur in association with El Niño events. Berri et al. (22) confirmed that river discharges in La Plata Basin during El Niño events are always larger than those during La Niña events. There are also reports on the signature of oceanic anomalies other than ENSO on river streamflows. Rickey et al. (1989) used river streamflow data as indicators of the interannual variability of the northern Amazonia climate and showed that the oscillations in river discharges predate significant human influences in the Amazon basin and reflect both remote and regional factors. Genta et al. (1998) demonstrated that the 3-year moving average of annual mean discharges in several rivers of South America tends to increase monotonically in time following an index representative of sea-surface temperature (SST) in the eastern Equatorial Pacific Ocean. Robertson & Mechoso (1998) found a near-decadal component in the annual streamflow of the Paraná and Paraguay rivers, which is strongest in the austral summer and such that high river discharge corresponds to anomalously cool SSTs over the tropical North Atlantic. Changes in trend and jumps in the multi-year means have also been found in the time series of streamflow in South American rivers. García & Vargas (1998) detected jumps in the time series of annual mean streamflows in La Plata Basin rivers in the late 191s and early 194s. They also found a significant and rapidly increasing trend in the annual mean runoff since the 197s. These features have counterparts in the regional climate. Minetti & Vargas (1983) and Vargas et al. (1995) identified an abrupt climate change signature in the time series of water vapour flux, temperature and precipitation series in southern South America around the late 195s early 196s (see also Castañeda & Barros, 1994; Barros & Doyle, 1996). This paper starts with a presentation of the river basins selected for study and a summary of major features in the distribution of precipitation over South America. Next the seasonal cycles of river discharges are described. The findings on trends and jumps, ENSO impacts, and quasi-periodicities in the time series of annual mean streamflows are presented next. The last section of the paper includes a summary of the results and a discussion on the possible associations with climate variability. SELECTED RIVER BASINS South America extends from Cape Gallinas at 12 3 N to Cape Hornos at 56 S, and from Cape Branco at 35º3 E to Cape Fariñas at 81º2 W. Its land mass, extending over 17 8 km 2 mostly in the Southern Hemisphere and representing 14% of the continental masses on Earth, is much wider in the tropics than in the high latitudes. The Andes mountains run along the entire western coast forming an effective meteorological barrier between the Pacific Ocean and the continent. All rivers selected for this study discharge into the Atlantic Ocean (Fig. l). The Amazon runs west to east across the continent from its sources in the Andes slightly south of the Equator to its mouth almost exactly at the Equator. The Negro, which runs northwest to southeast near the Equator, is one of the most important affluents of the Amazon. The Orinoco runs southwest to northeast entirely north of the Equator. The Tocantins and San Francisco run from south to north in eastern Brazil. The Paraguay runs north to south in central South America until its junction with the Paraná, which

5 462 Fig. 1 Map of studied South American river basins. runs northeast to southwest and joins the Uruguay to form the Plata River. The time series of monthly mean streamflows from one gauging station in each river were analysed, except for the Paraná river where two data sets were taken, one from a gauging station located upstream (Posadas station) and another located downstream (Corrientes station) of its confluence with the Paraguay River. Table 1 gives the geographical location of the stations selected and the periods covered by the data sets available, which are several decades long in all cases. The combination of the wide range of latitudes covered by the South American continent, the existence of major orographic features, and significant oceanic influences results in a great diversity and complexity of climate patterns. Of special relevance to river behaviour is the distribution of precipitation. The SAMS starts developing during the southern spring as the region of intense convection in northwestern South America shifts rapidly to the southern Amazon Basin and Brazilian highlands, where rainfall peaks in January February. The Bolivian high-nordeste trough system establishes at upper levels in the atmosphere as precipitation increases south of the Equator becoming most intense during the high summer. The South American Convergence Zone (SACZ), an important component of the SAMS, shifts east to west from December to January. The SAMS starts decaying in February

6 Variability in the discharge of South American rivers and in climate 463 Table 1 The nine gauging stations used in this study. Gauging station River Lat. Long. Period Obidos Amazon 1º57 S 55º 31 W Manaus Negro 3º8 S 6º5 W Ciudad Bolivar Orinoco 8º8 N 63º31 W Tucurui Tocantins 3º47 S 49º41 W Itaparica San Francisco 9º6 S 38º18 W Posadas Paraná 27º38 S 55º54 W Puerto Bermejo Paraguay 27º33 S 58º34 W Corrientes Paraná 27º59 S 58º48 W Paso de los Libres Uruguay 29º47 S 57º4 W and continues to weaken through southern autumn as convection gradually retreats northward toward the Equator. (See Nogues-Paegle et al., 23, for more details and appropriate references.) In the subtropics, precipitation is more uniform throughout the year as synoptic disturbances become active during the cold season. Annual mean rainfall in La Plata Basin tends to decrease both from north to south and from east to west. Rainfall is high in the upper reaches of the Paraguay and Paraná river basins. The northern part of the basin has a well-defined annual cycle with maximum precipitation during summer (December February). The central region (northeast Argentina/southern Brazil) has a more uniform seasonal distribution, with maxima during spring and autumn. Since the major rivers in the basin generally run from north to south, this rainfall regime contributes to a downstream decrease of the seasonal cycle amplitude. It is necessary to mention that the La Plata Basin, Tocantins River basin and San Francisco River basin have several dams with storages of more than one km 3 of volume. One should add to this the topographical features in order to verify that the effects of these works are non-existent on the monthly mean flows in high waters and are inside the measurement error in low waters, still with the retention that they could carry out; and they are totally null in the annual mean flows (García & Vargas, 1994). Therefore, with the purpose of analysing the hydrological variability, naturalized flows were used as suggested by García & Vargas (1996), who used corrections in each dam starting from daily values by means of the expression: Natural flow Discharge + Difference of volume As shown by García & Vargas (1996), the weak effect of the dams on the monthly and annual flows is due to the characteristics of the works of infrastructure, which turn the power stations into passing-stations, with little regulation capacity. This suggests that the mean values could only vary from one year to another as a function of the precipitation variability, the state of the vegetation, the land cover and use, etc. MEAN SEASONAL CYCLE OF STREAMFLOWS Figure 2 shows the mean annual cycle of the streamflow in the selected rivers as obtained by using the entire time series for each case. The seasonal cycle of the

7 464 Tocantins San Francisco Parana Paraguay Uruguay Amazon Orinoco Qmm (m3/s) Qmm (m3/s) 5 5 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Fig. 2 Seasonal cycle of the different rivers: right axis: Amazon and Orinoco rivers; left axis: Tocantins, San Francisco, Paraná, Paraguay and Uruguay rivers. Orinoco, in the Northern Hemisphere, shows a strong annual component with peak values in August September, when precipitation over South America is strongest north of the Equator. The seasonal cycle of the Amazon, whose catchment is mostly in the Southern Hemisphere covering western and central Equatorial South America, also has a strong annual component that peaks in May June, which is a few months after the peak precipitation associated with SAMS. The strong annual components of the Tocantins and San Francisco both peak earlier and closer in time to the mature phase of SAMS. The seasonal cycle of the Paraguay peaks in June July; this river runs through the Pantanal, one of the largest wetlands on Earth. The Uruguay has a weak seasonal cycle with lowest streamflows in summer from December to April, and largest values in August and October. The annual cycle of this river is primarily determined by precipitation associated with synoptic systems of extratropical nature (García & Vargas, 1996). Since the peaks in all cases are well within the calendar year, annual mean will be defined on the average for the January December period. TRENDS AND JUMPS IN ANNUAL MEAN STREAMFLOWS It is assumed in this study that monthly or annual river discharges are independent and normally distributed. The adequacy of these assumptions was verified by application of the method of Accumulated Periodgram (Anderson, 1977; García & Vargas, 1998). Table 2 presents the mean annual streamflows of the rivers selected for this study, together with the standard deviation, skewness, kurtosis, standard errors and specific discharge of the time series (Chou, 1974). All values shown correspond to the period , in which data for all rivers were obtained. The annual mean river discharges in Table 2 cover a broad range of values. The highest one corresponds to the Amazon, which is the largest river in the world. In terms of discharge, the Amazon is followed by the Orinoco. These two tropical rivers have approximately the same specific discharge. Next is the Paraná with different

8 Variability in the discharge of South American rivers and in climate 465 Table 2 Basic statistics properties of the time series, using the hydrological year in each case, for the common period River / Station Annual discharge (m 3 s -1 ) Area (km 2 ) SD (m 3 s -1 ) Skewness Kurtosis Standard error (m 3 s -1 ) Specific discharge (l s -1 km -2 ) Amazon / Obidos Orinoco / C. Bolivar Tocantins / Tucurui S. Francisco / Itaparica Paraná / Posadas Paraguay / P. Bermejo Paraná / Corrientes Uruguay / P. de los Libres Negro / Manaus * m m m - * Negro River has altitude in metres only. Table 3 Autocorrelation coefficients for each river. Lag (years) Orinoco Negro Tocantins San Francisco Paraná values in its upper and lower parts, the latter of which includes the Paraguay streamflow. The Tocantins is also a large river near the SAMS region. The other rivers have substantially lower discharges. Relative to these mean values, the standard deviations in Table 2 increase with increasing latitude of the catchment. In this regard, La Plata Basin rivers have a relatively high interannual variability. The San Francisco also has a high standard deviation, but the catchment of this river is relatively small. The generally small values of skewness in Table 2 imply that the probability distribution of annual discharges is fairly symmetric about their means. Skewness is very important for the Upper and Lower Paraná only. According to the values of kurtosis in Table 2, the distributions are fairly flat, except again for the Upper and Lower Paraná. The series randomness was analysed by means of autocorrelation coefficients (see Table 3) and Anderson s test (Anderson, 1977). To determine changes of trend in the time series, Mann s test (Sneyers, 1975; García & Vargas, 1998) was applied. Let x 1,..., x n be a time series of annual mean

9 466 streamflows, and R 1,..., R n the corresponding time series of streamflows ranks. The null hypothesis is tested that the observations are randomly ordered, against the alternative hypothesis that there is a time trend. First, the time series is computed: [ ( R j Ri )] s sng j 1,..., n (1) j i j where sng(ξ) ξ/ ξ if ξ, and if ξ. For normal distribution of mean zero, the variance of ranks is defined by: var(j) j(j 1) (2j + 5)/18 j 1,..., n (2) Therefore the following time series is defined: u(i) s ij /var(i) 1/2 j 1,..., n (3) Plots of u(i) facilitate the visualization of points of change in the mean. The test of accumulated deviations (Q) is also applied in order to detect jumps in the mean (Buishand, 1982). The Q test examines the ratio between the deviation from the mean and the square root of the variance. Let x be the mean of the time series x 1,, x n. The time series of deviations from the mean is given by: s j ( xk x) * s j 1,..., n (4) *, j k 1 and the ratio between these deviations and σ ( x) s / σ ** * i s i j The position of the maximum of s i ** i n i 1 1/ 2 2 x / n to yield: i i, 1,..., n (5) could be taken as an estimate of change points m. The length of the time series used in this paper only allows for detecting jumps in the period from the mid 196s to the mid 197s. Table 4 shows the timings of changes in trend and of jumps in the annual mean streamflow detected by application of Mann s test and the Q test. (Note that the Amazon is not included in this analysis due to the lack of data for the period ) Figure 3(a) (c) shows the curves u(i) obtained from equation (3) by using the complete time series for the Negro, Orinoco and Paraná rivers, respectively. Mann s test indicates a change in trend around the early 197s in all river streamflows, except in the San Francisco, where the change in trend takes place a few years later in The Q test detects a jump in the same period from the late 196s to the early 197s in all time series, except in that for the San Francisco. Here, as well as in the time series for the Uruguay, the jump is in the late 197s. ENSO IMPACTS An ENSO event, either warm or cold, unfolds during a two-year period with the mature phase around the end of the first calendar year. Several studies demonstrate that impacts of both warm and cold ENSO events on precipitation over South America are

10 Variability in the discharge of South American rivers and in climate 467 Table 4 Timing of the change of trend and jumps in the mean. River Station Change of trend (Mann s test) Negro Manaus Orinoco Ciudad Bolivar Tocantins Tucurui San Francisco Itaparica Paraná Posadas Paraguay Puerto Bermejo Paraná Corrientes Uruguay Paso de los Libres El Niño 3 Region Jump in the mean (accumulated deviation) more clear in the southern spring summer, i.e. from October to the following March (Ropelewski & Halpert, 1987, 1989; Pisciottano et al., 1994). In order to explore the event-dependence of ENSO impacts on river discharges separate plots were made of monthly-mean streamflows during years when warm and cold events started, according to the classification by Kiladis & Diaz (1989), as well as for the remaining or neutral years. (Kiladis & Diaz, 1989, define that warm events start in years when the SO index, SOI, changes sign from positive to negative, while cold events start in years when the SOI changes sign from positive to negative.) Figure 4(a) shows that mean Amazon streamflows during warm ENSO events are lower than during neutral years from February to September, while those during cold events are higher than during neutral years. The strongest signal is for cold events. Mean Orinoco streamflows (not shown) during warm ENSO events are slightly higher than during neutral years from July to August, while those for cold events and neutral years are practically identical in all months. None of these features, however, is statistically significant. Figure 4(b) shows that mean San Francisco streamflows during warm ENSO events are very similar to those in neutral years, and that mean streamflows during cold events are generally lower than those during neutral years, particularly from January to May. Again, none of these features is statistically significant. Statistically significant ENSO impacts are only detected in rivers of La Plata Basin. Figure 4(c) and (d) shows that mean Paraná (Corrientes) and Uruguay streamflows during warm ENSO events are always larger than during neutral years. This feature is statistically significant for the Paraná from December to February and in May June, and (marginally) statistically significant for the Uruguay from November to March. On the other hand, for both the Paraná and Paraguay, streamflows during cold events and neutral years are very similar.

11 468 (a) (b) U ( i ) U ( i ) Test of jump in the mean: accumulated deviations: 1969 Changes of trend Changes of trend i 1959 i Test of jump in the mean: acumulated deviations: Test of jump in the mean: acumulated deviations: U (i ) (c) Changes of trend i Fig. 3 Curve u(i) vs i for (a) the Negro River at Manaus, (b) the Orinoco River at Simón Bolivar, and (c) the Paraná River at Corrientes.

12 Variability in the discharge of South American rivers and in climate 469 (a) Qmm (m 3 s -1 ) Qmm (m 3 s -1 ) (c) OCT NOV DEC JAN FEB MA MAYJUN JUL APR AUG SEP OCT NOV DEC FEB JAN MARAPR MAY JUN JUL AUGSE WARM NEUTRAL COLD P OCT NO DEC JAN FEB MA APR MAY JUN JUL AU SEP OCT NOV DEC FEB MAR JAN APR MAY JUN JUL AUG SEP WARM NEUTRAL COLD Fig. 4 (a) Amazon river streamflows; (b) monthly mean streamflow of the San Francisco River; (c) monthly mean streamflow of the Paraná River; and (d) monthly mean streamflow of the Uruguay River, during different phases of ENSO. QUASI-PERIODICITIES IN THE TIME SERIES To analyse quasi-periodicities in the time series of river streamflows the method of singular spectrum analysis (SSA) was applied. SSA is a statistical method related to principal component analysis (PCA) but applied in the time domain. The objective is to describe the variability of a discrete and finite time series x i x(i t)(i 1, 2,..., N), where t is the sampling interval, in terms of its lagged autocovariance structure (Vautard & Ghil, 1989). The eigenvalue decomposition of the lagged autocovariance matrix C(M M) up to lag M t produces temporal-empirical orthogonal functions T-EOF, and statistically independent temporal-principal components T-PCs, with no presumption as to their functional form. M is the maximum number of lags and is referred to as the window length. Each T-PC has a variance λ s (eigenvalue) and represents a filtered version of the original series. Table 5 shows the quasi-periodicities detected by the SSA in monthly mean Niño3 SSTs using window lengths of M 1 years and M 2 years. The periods in the table were obtained by computing the power spectrum of the principal components for each oscillatory pair. The ENSO signature on SST appears in the oscillatory pairs (T-PC1, T-PC2) and (T-PC3, T-PC4), which have periods of about 3.5 and 4.5 years. A third (b) (d)

13 47 Table 5 Leading quasi-periodicities detected by the SSA method in the time series of Niño3 SST. Window length (years) Oscillatory pair (components) Trend or dominant period (year) 1 T-PC1 and T-PC T-PC3 and T-PC T-PC6 and T-PC T-PC1 and T-PC T-PC3 and T-PC Explained variance (%) oscillatory pair (T-PC6, T-PC7), with a quasi-biennial period of 2.5 years is obtained for M 1. The time series of streamflows for the two large tropical rivers show ENSO-like periodicities of about 3.5 years for the Amazon, and 4.5 years for the Orinoco. A quasi-biennial oscillation with a period of about 2.5 years is also apparent in both rivers. None of these quasi-oscillations correspond to the dominant modes of variability for both rivers, which is consistent with the small and not statistically significant differences between streamflows during ENSO events and neutral years shown in Fig. 3(a) and (b). The dominant modes have near-decadal or longer periods. Therefore, one can restrict consideration to the results obtained with M 2, noting the coincidence with the results obtained using the shorter window. For both the Amazon and Negro, the leading mode has a period of about 17 years and explains 24% of the variance. For the Orinoco, the leading mode has near-decadal period of about 8.5 years, and explains 19% of the variance (see Table 6(a)). To illustrate the SSA results, the first three oscillatory pairs detected in the time series for the Negro River and obtained by using M 2 are presented in Fig. 5(a) (c). Consistency in the phase lag between the two components of each pair can be noted. It is also apparent that the amplitude of the longest quasi-oscillation varies relatively little in time, while those of the quasi-biennial and ENSO-like ones are stronger in the first part of the record. The leading mode for the Negro River has a period of about 17 years (18.4 with M 1 and 17.9 with M 2) as in the Amazon River. The second leading mode has a dominant period of 1.3 (with M 2 only). There are two ENSOlike frequencies at 5 and 3.5 years, but only one of them (3.5 years) is clearly dominant after 195 (Fig. 6). The quasi-biennial oscillation has very small amplitudes from about 194 to 196. The time series of the Tocantins and San Francisco running south north in Northeast Brazil also show ENSO-like periodicities, which are not the leading modes of variability either (see Table 6(b)). There are also indications of a component with a quasi-biennial period. The leading mode of variability has a period of about 17 years and explains 31 and 33% of the variance for the Tocantins and San Francisco, respectively. It is noted that the ENSO-like period is longer for the Tocantins than the San Francisco (5.3 and years, respectively). This result is not entirely unexpected since, despite the proximity of their basins, the basic statistics of the rivers have substantial differences (see Table 2). The leading oscillatory mode of variability for the Paraná and Paraguay rivers in La Plata Basin has a quasi-decadal period (see Table 6(c)). This period is clearly seen in the time series of the mode s principal components (Fig. 7(a)).

14 Variability in the discharge of South American rivers and in climate 471 MANAUS SSA (M2) - T-PC1 & T-PC2 (24.2%) (a) Time (years) T-PC1 T-PC2 (b) (c) MANAUS SSA (M2) - T-PC3 & T-PC4 (18.9 %) Time (years) T-PC3 T-PC4 MANAUS SSA (M2) - T-PC8 & T-PC9 (8.8%) Time (years) T-PC8 T-PC9 Fig. 5 Negro River at Manaus: (a) T-PC1 and T-PC2; (b) T-PC3 and T-PC4; and (c) T-PC8 and T-PC9. Spectral analysis reveals, respectively, a leading period of around 17 years, a leading quasi-biennial period and leading ENSO-like periods of around 5 and 3.5 years.

15 Water Level (m) Time (years) Period: 3.6 (M1) Period: 5.1 (M2) Fig. 6 Reconstructed time series for the two ENSO-like periodicities: 3.5 years (darker), 5 years (lighter) for the Negro River at Manaus ( ) Time (years) T-PC2 T-PC Time (years) T-PC5 T-PC6 Fig. 7 Time series of (a) T-PC2 and T-PC3 and (b) T-PC5 and T-PC6 for the Paraná River at Corrientes.

16 Variability in the discharge of South American rivers and in climate 473 Table 6 Single spectrum analysis results. Basin Window length Oscillatory pair (years) (components) (a) Amazon and Orinoco time series: Amazon 1 T-PC1 and T-PC2 T-PC3 and T-PC4 T-PC8 and T-PC9 2 T-PC1 and T-PC2 T-PC3 and T-PC4 T-PC8 and T-PC9 Negro 1 T-PC1 and T-PC2 T-PC3 and T-PC4 T-PC7 T-PC8 and T-PC9 2 T-PC1 and T-PC2 T-PC3 and T-PC4 T-PC5 and T-PC6 T-PC7 T-PC8 and T-PC9 Orinoco 1 T-PC1 and T-PC2 T-PC5 and T-PC6 T-PC7 and T-PC8 2 T-PC1 T-PC2 and T-PC3 T-PC5 and T-PC6 T-PC7 and T-PC8 (b) Tocantins and San Francisco time series: Tocantins 1 T-PC5 and T-PC6 T-PC8 and T-PC9 2 T-PC1 and T-PC2 T-PC3 and T-PC4 T-PC9 and T-PC1 San 1 T-PC1 and T-PC2 Francisco T-PC5 and T-PC6 2 T-PC1 and T-PC2 T-PC4 and T-PC5 T-PC8 and T-PC9 (c) Paraguay and Paraná time series: Paraná 1 T-PC1 (Posadas) T-PC2 and T-PC3 T-PC5 and T-PC6 2 T-PC1 T-PC2 and T-PC3 T-PC4 and T-PC5 Paraguay 1 T-PC1 T-PC2 and T-PC3 2 T-PC1 T-PC3 and T-PC4 T-PC8 and T-PC9 Paraná (Corrientes) (d) Uruguay time series: Uruguay 1 T-PC1 T-PC2 and T-PC3 T-PC5 and T-PC6 T-PC8 and T-PC9 2 T-PC1 T-PC2 and T-PC3 T-PC5 and T-PC6 T-PC8 and T-PC9 1 T-PC1 and T-PC2 T-PC3 and T-PC4 T-PC5 and T-PC6 2 T-PC1 and T-PC2 T-PC3 and T-PC4 T-PC6 and T-PC7 T-PC14 and T-PC15 Trend or dominant period (year) Trend Trend Trend Trend Trend Trend 9.3 Trend Trend Trend Variance fraction (%)

17 474 According to Table 6(c), the streamflows of the two rivers show the ENSO signature. In this regard, note the similarity in the behaviour of the quasi-oscillation with a dominant period of about 3.5 years for the Negro and the Paraná in Figs 6 and 7(b), respectively, particularly the large amplitudes towards the end of the period in both cases. The discharges of the Paraná (Corrientes) also show the hint of a quasi-biennial period. The Uruguay River is also in La Plata Basin, but runs entirely outside of the tropics and extratropical influences are comparatively more important than for the other cases analysed. The Uruguay is the only river in the selected set whose variability is dominated by ENSO-like time scales (see Table 6(d)). Nevertheless, the results also show quasi-decadal and quasi-biennial periodicities in the streamflows of this river. SUMMARY AND DISCUSSION Seasonal cycle The results verify the decrease in amplitude of the seasonal cycle with increasing latitude (see also Maheu et al., 23). Tropical rivers receive the largest seasonal contribution in the warm season from precipitation associated with SAMS; extratropical rivers receive comparable contributions from warm season disturbances and cold season synoptic systems. Trends and jumps It was found that the streamflow of all major rivers of South America has experienced an increased trend since the early 197s. The simultaneity of events across the continent suggests that the mechanism of generation has a large spatial scale. Genta et al. (1998) proposed an association between the increasing trend in streamflows and the warming experienced by the tropical Pacific Ocean around the mid 197s. Giese et al. (22) argue that this climate change originates with subsurface temperature anomalies in the subtropical Pacific in the late 196s that become visible at the surface in the eastern Equatorial Pacific several years later. Several studies have identified a change in the behaviour of moisture advection, temperature and precipitation in southern South America around the late 195s early 196s (Minetti & Vargas 1983; Vargas et al., 1995). The increased precipitation appears to be associated with a reduction of the meridional gradient of surface temperature in northern Argentina (Castañeda & Barros, 1994). Consistently, García & Vargas (1998) detected increases in the time series of annual mean streamflows in La Plata Basin rivers since the 197s. It has been suggested that this regional feature has been at least partially due to anthropogenically-induced regional changes in deforestation and changes in land use consisting of substitution of coffee plantations for annual crops (García et al., 24). Nevertheless, the approximate simultaneity in the dates of change in trend for all rivers examined indicates that the relative extent to which anthropogenic activities and climate variations affect La Plata Basin rivers requires further investigation.

18 Variability in the discharge of South American rivers and in climate 475 ENSO impacts To study the influence ENSO on the South American rivers, the time series of streamflows was divided into three categories: warm events, cold events and neutral years. The tropical rivers in central and western South America (Amazon and Orinoco) show a tendency towards slightly increased streamflows during cold events and a very small impact of warm events. The tropical rivers in northeastern South America (Tocantins and San Francisco) show the opposite behaviour with decreased streamflows during cold events and practically no differences for the warm events. None of the differences is statistically significant. La Plata Basin rivers (Paraná and Uruguay), on the other hand, show significant streamflow increases during warm events and small differences during cold events. These results are broadly consistent with the current view of the region s climate. Over tropical South America during El Niño events, rising motion tends to be weaker than average and rainfall over the eastern Amazon and northeast Brazil tends to be below average, with broadly the reverse situation during La Niña (e.g. Marengo & Hastenrath, 1993). One would expect, therefore, a tendency towards stronger rainfall and higher streamflows during cold events and the opposite during warm events. In general, however, ENSO only explains a small fraction of the interannual variance of rainfall in Amazonia (e.g. Marengo et al., 1993, 21; Rao et al., 1996; Dettinger et al., 2). Consequently, the results on streamflows have small statistical significance. The Tocantins and San Francisco rivers are in the northeastern part of South America, where the regional climate is also influenced by SST and ITCZ anomalies in the Atlantic. La Plata Basin s climate is significantly affected by ENSO through atmospheric teleconnections (Cazes et al., 23). Quasi-oscillations All the analysed time series of river streamflows show El Niño-like periodicities. Only for rivers in La Plata Basin do these quasi-periodicities explain a larger part of the total variance than elsewhere. There are two other quasi-oscillations, and these are the most important contributors to variability in all tropical rivers. The quasi-oscillation with the longer period around 17 years is dominant for the Amazon, Tocantins and San Francisco. The other quasi-oscillation has a quasi-decadal period of about 9 years. These rank among the first three SSA modes for all rivers in this study. Robertson & Mechoso (2) associated the 17-year cycle to changes in the intensity of the SAMS. (Independent confirmation of this cycle was found in the behaviour of SSTs over the southwest Atlantic between 2 and 3 S.) Precipitation on the Tocantins and San Francisco basins is directly associated to the South Atlantic Convergence Zone (SACZ), one of the major components of SAMS. Figure 8 shows the reconstructed time series for the 17-year quasi-periodicity in the Tocantins and San Francisco rivers. This figure can be contrasted with Figure 1 in Robertson & Mechoso (2), which is for the difference between the January March means of the Paraná/Paraguay and Uruguay/Negro [here Negro refers to the river in southeastern South America] streamflows as well as those for sea-surface temperatures (averaged

19 Time (years) San Francisco Tocantins Fig. 8 Reconstructed time series for the quasi-oscillation with a dominant period of about 17 years: San Francisco River (darker) and Tocantins (lighter). over the southwest Atlantic) and an index representative of the SACZ intensity. There is an approximate phase coherence between all reconstructed time series. This coincidence suggests that the 17-year quasi-periodicity is generated by interdecadal variability of the SACZ. Robertson & Mechoso (1998) associated the quasi-decadal component to SST anomalies over the northern Tropical Atlantic. In their scenario, these SST anomalies in the Tropical Atlantic are linked to the location of the Atlantic ITCZ and moisture flux from the ocean to the continent. This is transported southward by the low-level jet along the eastern slopes of the Andes to converge in the Plata Basin. Therefore, the Negro on the path of the low-level jet and the Plata Basin rivers near the region in which moisture transported converges, are expected to show this oscillation. Acknowledgements The authors are grateful to the anonymous reviewers for their helpful comments on the earlier version of this paper. This research was supported at the Universidad Nacional del Litoral, Argentina by Program C.A.I.+D. 2 no. 28, Project 198, and at the University of California Los Angeles by NOAA Grant NA6GP511. REFERENCES Anderson, O. D. (1977) Time Series Analysis and Forecasting. Butterworths, UK. Berbery, E. H. & Barros, V. R. (23) The hydrologic cycle of the La Plata Basin in South America. J Hydromet. 3(6), Berri, G. J., Ghietto, M. A. & García, N. O. (22) The influence of ENSO in the flows of the Upper Paraná River of South America over the past 1 years. J Hydromet. 2(1), Bischoff, S. A., García, N. O., Vargas, W. M., Jones, P. D. & Conway, D. (2) Climatic variability and Uruguay river flows. Water Int. 25(3), Buishand, T. A. (1982) Some methods for testing the homogeneity of rainfall records. J. Hydrol. 58,

20 Variability in the discharge of South American rivers and in climate 477 Castañeda, M. E. & Barros, V. R. (1994) Las tendencias de la precipitación en el Cono Sur de América al este de los Andes. Meteorológica 19, 23 32, CAM, Buenos Aires, Argentina. Cazes, G., Robertson, A. W. & Mechoso, C. R. (23): Seasonal dependence of ENSO teleconnections over South America and relationships with precipitation in Uruguay. J. Climate 16, Chiew, F., Whetton, P. & McMahon, T. (1995) Detection of climate changes in recorded runoff volumes in south-east Australian rivers. Int. J. Climatol. 13, Chou, Y.-L. (1974) Análisis Estadístico. Editorial Interamericana, Mexico. Dettinger, M. D., Cayan, D. R. & Redmond, K. T. (2) United States streamflow probabilities based on forecasted La Niña, Winter Spring 2. Experimental Long-Lead Forecast Bulletin 8(4), García, N. O. (2) Anthropic impacts over hydrology of Rio de la Plata basin. Extended abstract in: Preprints 15th Conf. on Hydrology (AMS Annual Meeting, Long Beach, California, 9 14 January), American Meteorological Society. García, N. O. & Vargas, W. M. (1994) Análisis de la variabilidad climática en la cuenca del Rio de la Plata a través de sus caudales (abstract extendido). In: Anales del VIII Congreso Brasileiro de Meteorologia II Congreso Latino-Americano e Ibérico de Meteorología (18 25 Octubre, Belo Horizonte, Brasil), García, N. O. & Vargas, W. M. (1996) The spatial variability of runoff and precipitation in the Rio de la Plata basin. Hydrol. Sci. J. 41(3), García, N. O. & Vargas, W. M. (1998) The temporal climatic variability in the Rio de la Plata basin displayed by the river discharges. Climatic Change 38, García, N. O., Vargas, W. M., Venencio, M. V. & Valdes, J. B. (24) Assessment of anthropogenic impacts in La Plata River Basin. Climatic Change (submitted). Genta, J. L., Pérez Iribarren, G. & Mechoso, C. R. (1998) A recent increasing trend in the streamflow of rivers in Southeastern South America. J. Climate 11, Hastenrath, S. (199) Diagnostic and prediction of anomalous river discharges in northern South America. J. Climate 3, Kiladis, G. N. & Diaz, H. F. (1989) Global climatic anomalies associated with extremes in the Southern Oscillation. J. Climate 2, Maheu, C., Cazenave A. & Mechoso, C. R. (23) Water level fluctuations in the Plata Basin (South America) from Topex/Poseidon Satellite Altimetry. Geophys. Res. Lett. 3, 43 1/43 4. Marengo, J. A. (1995) Variations and change in South America streamflow. Climatic Change 31, Marengo, J. A & Hastenrath, S. (1993) Case studies of extreme climatic events in the Amazon basin. J. Climate 6, Marengo, J. A., Druyan, L. M. & Hastenrath, S. (1993) Observational and modelling studies of Amazonia interannual climate variability. Climatic Change 23, Marengo, J. A., Liebmann, B., Kousky, V., Filizola, N. & Wainer, I. (21) Onset and end of the rainy season in the Brazilian Amazon Basin. J. Climate 14, Mechoso, C. R. & Pérez Iribarren, G. (1992) Streamflow in Southeastern America and the Southern Oscillation. J. Climate 5, Minetti, J. & Vargas, W. M. (1983) El enfriamiento de la década de 195 en la República Argentina. Meteorológica 14, Nogues-Paegle, Mechoso, C. R., Fu, R., Berbery, E. H., Chao, W. C., Chen, T. C., Cook, K., Diaz, A. F., Enfield, D., Ferreira, R., Grimm, A. M., Kousky, V., Liebmann, B., Marengo, J. A., Mo, K., Neelin, J. D., Paegle, J., Robertson, A. W., Seth, A., Vera, C. S. & Zhou, J. (22) Progress in Pan American CLIVAR research: understanding the South American Monsoon. Meteorologica 27, Pisciottano, G., Díaz, A., Cazes, G. & Mechoso, C. R. (1994) El Niño-Southern Oscillation impact on rainfall in Uruguay. J. Climate 7, Poveda, G. & Mesa, O. J. (1997) Feedbacks between hydrological processes in tropical South America and large-scale Ocean Atmospheric phenomena. J. Climate 1, Poveda, G., Rave Herrera, C. C. & Mantilla, R. I. (21) Tendencias en la distribución de probabilidades de lluvias y caudales en Antioquía. Meteorología Colombiana 3, Rao, V. B., Cavalcanti, I. F. A. & Hada, K. (1996) Annual variation of rainfall over Brazil and water vapor characteristics over South America. J. Geophys. Res. 11, Rickey, J. E., Nobre, C. & Deser, C. (1989) Amazon river discharges and climate variability: Science 246, Robertson, A. W. & Mechoso, C. R. (1998) Interannual and decadal cycles in river flows of Southeastern South America. J. Climate 11, Robertson, A. W. & Mechoso, C. R. (2) Interannual and interdecadal variability of the South Atlantic Convergence Zone. J. Climate 11, Ropelewski, C. F. & Halpert, M. S. (1987) Global and regional scale precipitation pattern associated with the El Niño/Southern Oscillation. Monthly Weather Rev. 115, Ropelewski, C. F. & Halpert, M. S. (1989) Precipitation pattern associated with the high index phase of the Southern Oscillation. J. Climate 2, Sellers, W. D. (1965) Physical Climatology. The University of Chicago Press, Chicago, USA. Sneyers, R. (1975) Sur l analyse statistique des séries d observations. Tech. Note no. 143, World Meteorological Organization, Geneva, Switzerland. Street-Perrot, A., Beran, M. & Ratcliffe, R. (1983) Variations in the Global Water Budget. D. Reidel, Dordrecht, The Netherlands.

21 478 Vargas, W. M., Minetti, J. L. & Poblete, A. G. (1995) Statistical study of climatic jump in the regional zonal circulation over south America. J. Met. Soc. Japan 13, 1 8. Vautard, R. & Ghill, M. (1989) Singular spectrum analysis in nonlinear dynamics with applications to paleoclimatic time series. Physica D 35, Received 6 May 24; accepted 5 March 25