INTERANNUAL VARIABILITY IN THE URUGUAY RIVER BASIN

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 23: 3 5 (23) Published online in Wiley InterScience ( DOI:.2/joc.853 INTERANNUAL VARIABILITY IN THE URUGUAY RIVER BASIN C. M. KREPPER, a N. O. GARCÍA b, * and P. D. JONES c a Centro de Recursos Naturales Renovables de la Zona Semiárida (CERZOS), Universidad Nacional del Sur, Bahía Blanca, Argentina b Facultad de Ingeniería y Ciencias Hídricas, Universidad Nacional del Litoral, Santa Fe, Argentina c Climatic Research Unit, University of East Anglia, Norwich NR4 7TJ, UK Received 3 June 2 Revised 22 July 22 Accepted 22 July 22 ABSTRACT Both (El Niño southern oscillation ENSO)-scale and near-decadal variations in precipitation and river flow have been noted in southeastern South America. Here we focus on the Uruguay river basin and its subcatchments. Both river flow and precipitation are analysed with singular spectrum analysis in an examination of how the signals of the subcatchments are related to the overall basin signal. The approximate 6 year signal and the 3.5 year ENSO signal are the two statistically significant peaks for the river flow. Both signals are present in all the available river flows, but only in the precipitation of the upper two-thirds of the basin. Precipitation in the lower part of the basin shows positive trends. The ENSO signal is strongly modulated at the secular time scale; we construct a simple sinusoidal model for this modulation. Copyright 23 Royal Meteorological Society. KEY WORDS: southeastern South America; climatic variability; precipitation; river discharge; singular spectrum analysis. INTRODUCTION The relationship between the climatic regimes over a river basin and its hydrologic response, through its streamflow, presents different degrees of complexity according to the physical characteristics of the basin. Chiew et al. (995) have mentioned that the variations in precipitation are amplified in streamflow, and that, in general, it is easier to detect a change in discharge than directly in the basic climatic variables (e.g. precipitation or temperature). Streamflow is a synthesis of precipitation, evapotranspiration and the rest of the hydrologic cycle components, together with possible anthropogenic influences. The Uruguay river is the second tributary in importance of the Río de la Plata basin with a length of approximately 6 km, with its own basin covering an area of around 365 km 2 (Figure ). The source of the river is located in Brazil, in Serra Geral at 28 S, where it takes the name Pelotas river at a height of approximately 8 m above sea level (Bischoff et al., 2). The whole upper part of the Uruguay river basin is located in Brazil, and downstream of the confluence with the Pepirí Guazú river it forms the border with Argentina, up to the confluence with the Cuareim at 3 5 S. From that point on the river forms the border between Argentina and Uruguay as far as the Río de la Plata. The main tributaries of the Uruguay river come from the east of the basin. The Ijuí, Ibicuí and Cuareim rivers have relatively small basins but they contribute considerable streamflow. For example, calculation of the specific discharge of the Ibicuí basin, for the period shows a value of 2. l km 2 s, whereas for the Cuareim it is 8.3 l km 2 s. The largest tributary of the Uruguay river is the Negro, with a mean annual discharge of 5 m 3 s and an approximate length of 56 km. This river basin is located almost completely in Uruguay (Figure ). Several studies performed on southeastern South America have used river flows as indicators of climatic variability from the interannual to * Correspondence to: N. O. García, Hydroclimatic Research Unit, Department of Hydrologic Sciences, National University of Littoral, CC 27, (3) Santa Fe, Argentina; nogarcia@fich.unl.edu.ar Copyright 23 Royal Meteorological Society

2 4 C. M. KREPPER, N. O. GARCÍA AND P. D. JONES Figure. The study area the secular scale (Aceituno, 998; Rickey et al., 989; Hastenrath, 99; Mechoso and Pérez Iribarren, 992; Marengo, 995; García and Vargas, 996, 998; Genta et al., 998; Robertson and Mechoso, 998). García and Vargas (998), in particular, analysed the Río de la Plata basin and found positive trends in the runoff since 97, and changes in the series around the years 97 8 and 943. Genta et al. (998) examined long discharge series of the major rivers in southeastern South America looking for long-term changes. Robertson and Mechoso (998) examined the power spectrum of the combined Uruguay, Negro and Parana Paraguay river discharges, with the purpose of determining the existence of oscillatory components and their possible relationship with sea surface temperature (SST) variations. These authors have found evidence of the existence of nonlinear trends in the combined river discharges and oscillatory components with periods of around 3.5 and 6 years. The different basins of the region have been affected by anthropogenic change, such as deforestation and changes in land use, which may have influenced the river flow characteristics, as was shown by García (2) for the Río de la Plata basin. The main purpose of this work consists of studying the interannual variability of the Uruguay river basin, particularly the quasi-oscillatory variability that is present in the series. It is also of interest to be able to determine the presence of possible interaction among the different signals, generally of nonlinear signals in the oscillatory components. A second objective consists of determining probable relations between the oscillatory modes of the river flow and those of precipitation on different subcatchments of the basin. A simple procedure, filtering the series by means of moving averages over different periods, is used to study the low-frequency variability in both the river flow and in the precipitation. A singular spectrum analysis (SSA) is used to isolate the interannual oscillation modes in the series and to reconstruct the oscillatory components. This technique has several advantages over other types of spectral analysis (Ghill and Mo, 99; Vautard et al., 992); in particular, the temporal empirical orthogonal function need not be sinusoidal, which implies that the method does not need to decompose a single nonlinear oscillation into a large number of sine waves. SSA assumes only stationarity; however, even this assumption seems to be weak in practice. In Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

3 RIVER FLOW VARIABILITY 5 fact, one of the remarkable results of Vautard and Ghill (989) and Ghill and Vautard (99) is that, although the assumption of stationarity is usually implicit in SSA, the analysis performs very well in the presence of long-term nonstationarities, like trends. 2. DATA AND METHODS OF ANALYSIS 2.. Data used Four gauging stations on the Uruguay river with at least 6 years of discharge data were used in the present study. The names and location of the stations, as well as the periods of the record and the annual mean discharge for a common period (93 92), are shown in Table I. Precipitation data for the catchment come from the.5.5 gridded data of New et al. (999, 2). These data cover the 9 95 period. This database uses, among others, data coming from the 92 stations collected in the Assessing the impact of future climatic change on water resources and hydrology of the Río de la Plata basin project (Jones et al., 999) Methodology SSA is a statistical method related to principal components analysis (PCA) but it is 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 =, 2,...,N),where t is the sampling interval, in terms of its lagged autocovariance structure (Vautard and Ghill, 989). The eigenvalue decomposition of the lagged autocovariance matrix C(M M), up to lag M t, produces temporal-empirical orthogonal functions T-EOF s (s =,...,M) and statistically independent temporalprincipal components T-PC s, with no presumption as to their functional form. M is the maximum number of lags and is also called the window length. Each T-PC s has a variance λ s [(eigenvalue) and represents a filtered version of the original series X i. The SSA develops a set of data adaptive filters, in such a way that the original series X i is decomposed in the following way: X(k) = A M [T-PC s (i)][t-eof s (j)] () s= i+j=k where X(k) isthe(i + j)th value of the time series. The index i denotes a moment in time, and the index j denotes a lag from time i. T-EOF s (j) isthejth element of the sth filter, and T-PC s (i) is the amplitude of the signal captured by the sth filter. The value of A is generally /M, except near the beginning and end of the time series (Vautard et al., 992; Plaut and Vautard, 994). In the SSA context, a quasi-oscillatory structure will be present in X(k) when the following criteria are met: (a) two consecutive eigenvalues, λ s and λ s+, are nearly equal (degenerate pair); (b) the two corresponding T-EOF s and T-EOF s+ are nearly periodic, with the same period, and in quadrature; (c) the associated T-PC s and T-PC s+ are also in quadrature. SSA has the interesting property that oscillations with small frequency variation (within some small frequency range) Table I. Names and location of the stations used Gauging station Latitude Longitude Data period Mean annual discharges (93 92) (m 3 s ) Itá S 52.3 W Santo Tomé S 56. W Paso de los Libres S 57.4 W Concordia 3.23 S 58.2 W Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

4 6 C. M. KREPPER, N. O. GARCÍA AND P. D. JONES are identified and reconstructed as a single oscillation, rather than as several separate signals as in most other spectral analysis methods. However, SSA can only distinguish between neighbouring spectral peaks, with frequencies f k and f k+,if f k f k+ > /M. The choice of the window length M is a sensitive issue. The choice represents a compromise between information content (large M) and statistical confidence (small M) (Plaut and Vautard, 994). Within the context of this work, in order to avoid confusions with other definitions found in the literature, we can divide the broad interannual band into two. The interannual band (IAB) itself, with periods between and years, and the low-frequency band (LFB), with periods over years. 3. VARIABILITY OF THE URUGUAY RIVER FLOWS Monthly mean river flow series from Itá, Santo Tomé, Paso de los Libres and Concordia on the Uruguay river were analysed. The series corresponding to Itá was not used for the analysis of long-period fluctuations owing to its shorter length (62 years). To observe possible long-period changes in LFB, the annual average flows were examined over the different running average periods (centred on, 2 and 3 years). Figure 2 shows the results obtained for a smoothing of and 3 years, where the resulting flows have been normalized ((X X mean )/σ ) so as to make comparison easier. Similar behaviour shown by the Santo Tomé, Paso de los Libres and Concordia series can also be observed. Despite the river flow increasing about 59.8% (Table I) between Santo Tomé and Paso de los Libres, the smoothed flow behaviour does not show any differences. Figure 2(a) shows a period of absolute minima between 94 and 95, coinciding with several important droughts that developed in the river basin during 943, 944 and 945 (Genta et al., 998). Such an extreme drought period separates the Uruguay river flows into two contrasting periods. First, there is the period before 94, which is characterized by oscillations with periods over years; second, after 95 an appreciable positive trend occurs. The same type of behaviour is maintained in the smoothed flows series (Figure 2(b)). This shows that the characteristics of variability in the LFB are given by very long-period signals (T >3 years). The period after 95 shows a change in the trend around 97, as has been discussed by García and Vargas (998). When a spectral analysis (SA) of the mean annual river flow series (Santo Tomé, Paso de los Libres and Concordia) is performed (Figure 3), the only significant peaks at the 95% confidence level (WMO, 966) occur within the IAB with periods around 3 and 6 years, in agreement with those found by Robertson and Mechoso (998). The general spectral shape does not change from one station to another, only the relative magnitude of the peaks. In order to analyse the oscillatory structures within the IAB, an SSA was applied to the series of flows corresponding to Itá, Santo Tomé, Paso de los Libres and Concordia, using, in each case, all the data available. Table II summarizes the results obtained with the most suitable window length possible for each case. It is possible to appreciate the relative importance of each oscillatory mode, in accordance with the percentage of explained variance. The series of mean annual river flows at Itá presents only one oscillation mode, with a dominant period of around T 3 years, formed by the T-EOF 3 and T-EOF 4 pair; however, in the remaining series, an oscillation mode of around T 6 years, formed by T-EOF and T-EOF 2, and a secondary one with T 3 years (T-EOF 3 and T-EOF 4 ), are evident. For all these gauging stations, the oscillatory components with a period close to 6 years represent a quasiharmonic oscillation, whereas the components with T 3 years, especially downstream at Santo Tomé, show nonlinear interactions between frequencies of different ranges. In particular, Figure 4 shows the T-PC s for Paso de los Libres associated with such oscillation modes. Figure 4(a) shows the series reconstructed by T-PC 3 and T-PC 4, RC(3 4), computed by Equation () when s = 3 and 4, which accounts for 3.% of the variance and the reconstruction using the first four T-PC s, RC( 2 3 4) (Figure 4(b)). The oscillatory pairs with dominant periods of 3 and 6 years, together, explain 35.8% of the variance in the river flow series. Figure 4(a) gives evidence of an amplitude modulation of RC(3 4), which must be a reflection of what occurred in T-PC 3 and T-PC 4. The RC(3 4) presents a very distinct attenuation between 94 and 96, which would be associated with a modulation by a long wave belonging to the LFB. With the objective of Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

5 RIVER FLOW VARIABILITY Standardized values Concordia Paso de los Libres Santo Tomé 3 2 Standardized values Concordia Paso de los Libres Santo Tomé Figure 2. (a) The year centred moving averages of the streamflow series: Santo Tomé (mean = 256 m 3 s, σ = 29 m 3 s ); Paso de los Libres (mean = 4 m 3 s, σ = 499 m 3 s ; Concordia (mean = 4487 m 3 s, σ = 365 m 3 s ). (b) The 3 year centred moving averages of the streamflow series: Santo Tomé (mean = 245 m 3 s, σ = 67 m 3 s ); Paso de los Libres (mean = 393 m 3 s, σ = 34 m 3 s ; Concordia (mean = 4386 m 3 s, σ = 365 m 3 s ) fitting the T-PC 3 and T-PC 4 signals, the following simple nonlinear model is proposed: Y(t) = F(t){[sin(2πf t + φ) + C]}+B sin(2πf t + φ) (2) where /f corresponds to the dominant period of the oscillation mode and F (t) is another sine wave with frequency f, assuming that f << f. The best fitting obtained for the T-PC 3 of Paso de los Libres can be seen in Figure 5, and corresponds to an expression of the model in Equation (2) with the shape Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

6 8 C. M. KREPPER, N. O. GARCÍA AND P. D. JONES.2 Santo Tomé (99-997) Normalized spectral distribution function Frequency (cycles/year).4 Paso de los Libres (99-997) Normalized spectral distribution function Frequency (cycles/year).2 Concordia (99-997) Normalized spectral distribution function Frequency (cycles/year) Figure 3. Spectra of annual mean streamflow series Y(t) = A sin(2πf t + φ ) sin(2πf t + φ 2 ) + B sin(2πf t + φ 2 ) + C sin(2πf t + φ ) (3) where f =.2855 cycles/year (T = 3.48 years), f =.66 cycles/year (T = 5.5 years) φ =.9, φ 2 =.93, A = 395., B = 382. andc = The correlation between the model given by Equation (3) and T-PC 3 is.9. It is worth mentioning that the fitting to a model such as Equation (3) is extremely sensitive to the number of data values used, especially in Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

7 RIVER FLOW VARIABILITY 9 Table II. Results obtained with the most suitable window length possible for each case Gauging Lag window First oscillation pair Second oscillation pair station M (years) Eigenmodes Dominant period (approx.) T (years) Explained variance (%) Eigenmodes Dominant period (approx.) T (years) Explained variance (%) Itá 2 T-EOF 3, T-EOF Santo Tomé 3 T-EOF, T-EOF T-EOF 3, T-EOF Paso de Ios Libres 3 T-EOF, T-EOF T-EOF 3, T-EOF Concordia 3 T-EOF, T-EOF T-EOF 3, T-EOF (a) Annual mean streamflow (m3/s) (b) Annual mean streamflow (m3/s) Paso de los Libres Stream flow RC(3-4) Paso de los Libres Stream flow RC(-2-3-4) Figure 4. (a) Annual mean streamflow at Paso de los Libres and RC(3 4). (b) Annual mean streamflow at Paso de los Libres and RC( 2 3 4) the f determination. Thus, the objective of such fitting is not to determine the modulating signal frequency f accurately, but only to prove that the condition f << f is satisfied. Equation (3) can be written as the sum of different sine waves with frequencies f + f, f f, f and f. Consequently, f + f, f f (side bands) and f (carrying frequency) are so close that they can not be discriminated by a traditional SA (as shown in Figure 2), and the Uruguay river flow series are too short to resolve a modulating frequency as low as f. Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

8 C. M. KREPPER, N. O. GARCÍA AND P. D. JONES 6 Paso de los Libres T-PC3 Model Figure 5. Paso de los Libres streamflow: T-PC 3 and the best fitting A similar fitting was performed (using Equation (3)) for the T-PC 3 of Concordia (not shown here); some differences were obtained in the estimate of the modulating frequency (T 32 years), as a consequence of the different lengths of the river flow series at Paso de los Libres and Concordia. 4. VARIABILITY OF PRECIPITATION IN THE URUGUAY RIVER BASIN Annual precipitation in the Uruguay river basin shows a high degree of spatio-temporal variability. Figure 6 shows the mean annual precipitation distribution, calculated for the 9 95 period, over a wide region of southeastern South America, which includes the whole Uruguay river basin. The precipitation grid (29 grid points) allows us to calculate the basin mean series and for different subcatchments between the years 9 and 995. This information, together with the streamflow series, helps Figure 6. Annual mean precipitation Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

9 RIVER FLOW VARIABILITY (a) Standardized values IA-basin IB-basin II-basin IA-basin (Mean = 733. mm, σ = 64. mm); IB-basin (Mean = 8. mm, σ = 4.2 mm); II-basin (Mean = 65.4 mm, σ = 93.6 mm). (b) Standardized values IA-basin IB-basin II-basin IA-basin (Mean = mm, σ = 24.4 mm); IB-basin (Mean = mm, σ = 4.5 mm); II-basin ( Mean = 6.8 mm, σ = 4.5 mm). Figure 7. (a) The year and (b) 3 year moving averages of spatial annual mean precipitation for the upper basin to determine the contribution of the different drainage areas throughout a common period. The whole Uruguay river basin was divided into different drainage areas or sub-basins (Figure ): IA basin (9.4% of the whole area and 25 grid points) upstream of Itá; IB-basin (3.9% and 8 grid points) formed by the drainage area between Itá and Santo Tomé; II basin, whose main tributary is the Ibicuí river (9.4% and 25 grid points) between Santo Tomé and Paso de los Libres; III basin with the Cuareim as its main tributary (7.% and 22 grid points) between Paso de los Libres and Concordia; IV basin, with the Negro river as its main tributary (3.2% and 39 grid points) downstream of Concordia. With the help of Table I, and taking the mean annual flow at Concordia as reference, during the period it is possible to estimate the relative contributions of each drainage area. The IA-basin provides 22.2% of the mean flow at Concordia, the IB provides 32.7%, the II provides 32.8% and the III provides 2.3%. Then, the IB and II basins have approximately equal contributions, providing, as a total, 65.5% of the mean discharge that reaches Concordia. At Itá, the specific discharge (associated with the IA basin) is 22.9 l km 2 s, which represents the maximum of the Río de la Plata basin, whereas the value at Santo Tomé (associated with the IB basin) is 2.5 l km 2 s. These values coincide with the maximum of average annual precipitation totals, shown in Figure 6. In order to investigate low-frequency variability in precipitation, average annual values for each sub-basin were determined. These Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

10 2 C. M. KREPPER, N. O. GARCÍA AND P. D. JONES (a) Standardized values III-basin IV-basin III-basin (Mean = mm, σ = 96.2 mm); IV-basin (Mean = 98. mm, σ = 87.3 mm). (b) Standardized values III-basin (Mean = mm, σ = 96.2 mm); IV-basin (Mean = 98. mm, σ = 87.3 mm). III-basin IV-basin Figure 8. (a) The year and (b) 3 year moving averages of spatial annual mean precipitation for the lower basin were similarly filtered by means of centred moving averages over, 2 and 3 years. Drainage areas IA, IB and II (Figure 7) show a similar behaviour in LFB to that of the river flows (Figure 2). However, the drainage areas located to the south of Paso de los Libres, III and IV, present a completely different behaviour, consisting of strong and sustained positive trends (Figure 8) with average slopes or 3.4 and 3. mm year respectively. This is in agreement with what was found by Krepper and Sequiera (998) when the low-frequency variability in the precipitation of southeastern South America was studied. With the purpose of identifying oscillatory modes within the IFB, the mean annual precipitation over each of the drainage areas was analysed using SSA (Table III). An oscillation mode with a dominant period of around 3 years appears in the precipitation series of the upper part of the Uruguay river basin (IA, IB and II basins). The oscillatory modes of the IB and II drainage areas present the characteristics of a modulation over a wave with a period of approximately 3 years, as a result of a modulating wave within the LFB. Figure 9 shows the mean annual precipitation in the IB and II basins and the reconstructed series using the components corresponding to the 3 year oscillation mode (RC( 2) and RC(3 4) respectively). The 3 year oscillation mode disappears from the precipitation to the south of Paso de los Libres. It is worth mentioning that, even with a window length M = 2 for SSA, the III and IV basins show positive trends in their first temporal Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

11 RIVER FLOW VARIABILITY 3 Table III. Oscillation modes within the IFB the mean annual precipitation over each of the drainage areas Basin Lag window First oscillation pair Second oscillation pair M (years) Eigenmodes Dominant period (approx.) T (years) Explained variance (%) Eigenmodes Dominant period (approx.) T (years) Explained variance (%) IA 2 T-EOF,T-EOF IB 2 T-EOF,T-EOF T-EOF 3,T-EOF II (Ibicui) 3 T-EOF,T-EOF T-EOF 3,T-EOF III (Cuareim) 2 T-EOF,T-EOF Negro 2 (a) (mm) (b) (mm) Precipitation RC(-2) Precipitation RC(5-6) Figure 9. (a) IB basin: annual mean precipitation and RC( 2). (b) II basin: annual mean precipitation and RC(5 6) components (T-PC ) that were pointed out in the analysis with moving averages. On the other hand, only the III basin shows an oscillation mode formed by the pair (T-EOF 2,T-EOF 3 ) with a dominant period of about 5 years, which explains 6.% of the variance. The oscillatory pair with a period of about 6 years, which was manifested in all the river flow series except for that of Itá, only appears to be reflected in the IB and II basins mean precipitation. 5. SUMMARY AND DISCUSSION The study of river flow as an indicator of climatic variability, from the interannual to the secular scale, implies the determination of different climatic signals introduced by precipitation over the different subcatchments. At Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

12 4 C. M. KREPPER, N. O. GARCÍA AND P. D. JONES the same time, the intensity of the imprinted signals on the river flows will depend on the relative importance of the different sub-basin contributions. In the previous sections, an analysis of the interannual variability in the river flows and in the mean annual precipitation in several sub-basins of the Uruguay river has been undertaken. Our objective was the determination of oscillatory structures within the IAB as well as understanding the low-frequency behaviour (LFB). In Section 3 we presented evidence that the Uruguay river, measured at different locations, presented homogeneous behaviour in the LFB and that it reflects the precipitation variability in the sub-basins IA, IB and II. Discharge does not show what occurred in the III basin, because its contribution is only 2.3% of the flow in Concordia. The influence of the IV basin cannot be estimated, as no river flow measurements downstream of Concordia have been recorded. The flows within the IAB present only two interannual oscillation modes (with dominant periods of around 6 and 3.5 years), which, for Santo Tomé, Paso de los Libres and Concordia, explain jointly about one-third of the original series variability. These interannual oscillation modes are a reflection of the precipitation behaviour (Section 4) in some regions of the basin, especially to the north of Paso de los Libres. The oscillatory pair present in the Uruguay river basin with a period of around 6 years represents an oscillatory, almost harmonic structure, and is formed by the first two components of an SSA, applied to each one of the series downstream of Itá. Only precipitation averages of the IB and II basins, which span only a third of the whole Uruguay river basin, show a similar oscillatory component. There is no evidence of interactions with other frequencies. The oscillation mode with T 3 years that is recorded in all the flow series acquires importance within the IAB, due to its association with El Niño southern oscillation (ENSO) events in the eastern Pacific, as mentioned by Robertson and Mechoso (988). However, that oscillation mode does not correspond to a pure harmonic (Figure 4), as it presents clear evidence of nonlinear interactions between different frequency band waves. For Paso de los Libres and the remaining locations to the south, the filtered signal can be viewed as the modulation of a carrying wave of T 3 years, by a secular wave in the LFB, as shown in Figure 5. Such a modulation introduces a strong attenuation of the RC(3 4) (Figure 4) in the 94 to 96 period, in addition to a greater amplification of the oscillations after 96 with respect to those prior to 94. These characteristics must be taken into account when comparing such an oscillation with the Pacific SST anomalies. For example, in RC(3 4) for the Paso de los Libres flows, there is no year whose amplitude exceeds one standard deviation from the mean between 93 and 966; in the period, however, 24% of the cases exceed the limit of one standard deviation and the remaining 76% occur after 967. However, according to Quinn (992), during the period seven ENSO events occurred (one very strong, two strong and four moderate), without an apparent response in the T 3 years oscillation. We can speculate about the real reason why only the T 3 years mode is modulated by a secular wave and not the one that has a T 6 years as the dominant period, which is also within the ENSO-cycle variability range. Rasmuson et al. (995) show that the ENSO-cycle variability, which is concentrated on 2 to 6 year periods, should be seen as being superposed on variations that include periods greater than 3 years, whereas the secular variation in equatorial Pacific SST is similar to that in globally average SST. In other words, there is no evidence that the modulating wave corresponds to a similar wave in the ENSO behaviour. The oscillation mode in the flows, with T 3 years, is a clear reflection of what occurs with mean precipitation only over sub-basins IB and II (Figure 8). It is in the precipitation of the upper part (IA and IB basins) and middle part (II basin) of the Uruguay river where the major annual precipitation amounts are produced and the interannual oscillation modes are generated. The drainage area formed by the IB and II basins (which represent 33.3% of the total area), in particular, provides the major contribution to the Uruguay river flow (65.4% of the flow measured at Concordia) and becomes the main generator of the characteristics of oscillation modes (in the IAB) which come to Concordia and which represent more than a third of the total of the interannual variability. To the south of Paso de los Libres the mean precipitation does not show any of the oscillation modes mentioned (T 3 and 6 years). Only over the III basin is an oscillatory pair with T 5 years evident. The positive trends of precipitation (Figure 8) in the lower part of the basin (III and IV basins) are reflected in the first components of the SSA. Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)

13 RIVER FLOW VARIABILITY 5 ACKNOWLEDGEMENTS These research activities were supported by Universidad Nacional del Litoral Argentina (project 2/G22 of program C.A.I. + D. 96) and Consejo Nacional de Investigaciones Científicas y Técnicas Argentina (project PIP 464/98). REFERENCES Aceituno P On the functioning of the southern oscillation in the South America sector. Monthly Weather Review 6: Bischoff SA, García NO, Vargas WM, Jones PD, Conway D. 2. Climatic variability and Uruguay river flows. Water International 25(3): Chiew F, Whetton P, McMahon T Detection of climate changes in recorded runoff volumes in south-east Australian Rivers. International Journal of Climatology 3: García NO. 2. Anthropic impacts over hydrology of Rio de la Plata basin. In Preprints 5th Conference on Hydrology, AMS Annual Meeting, Long Beach, CA, 9 4 January; García NO, Vargas WM The spatial variability of runoff and precipitation in the Rio de la Plata basin. Hydrological Science Journal 4: García NO, Vargas WM The temporal climatic variability in the Rio de la Plata basin displayed by the river discharges. Climatic Change 38: Genta JL, Pérez, Iribarren G, Mechoso CR A recent increasing trend in the streamflow of rivers in southeastern South America. Journal of Climate : Ghill M, Mo K. 99. Intraseasonal oscillation in the global atmosphere. Part I: Northern Hemisphere and Tropics. Journal of Atmospheric Sciences 48: Ghill M, Vautard R. 99. Interdecadal oscillations and the warming trend in global temperature time series. Nature 35: Hastenrath S. 99. Diagnostic and prediction of anomalous river discharges in northern South America. Journal of Climate 3: Jones PD, García NO, Vargas WM, Conway D Assessing the impact of future climatic change on the water resources and the hydrology of the Río de la Plata Basin. Final report of work undertaken for the Commission of European Communities under contract ARG/B7-3/94/25. University of East Anglia Universidad Nacional del Litoral Universidad de Buenos Aires, Norwich Santa Fe Buenos Aires. Krepper CM, Sequeira ME Low frequency variability of rainfall in southeastern South America. Theoretical and Applied Climatology 6: Marengo JA Variations and change in South America streamflow. Climatic Change 3: Mechoso CR, Pérez, Iribarren G Streamflow in southeastern America and the southern oscillation. Journal of Climate 5: New MG, Hulme M, Jones P Representing twentieth-century space time climate variability. Part I: development of 96 9 mean monthly terrestrial climatology. Journal of Climate 2: New MG, Hulme M, Jones P. 2. Representing twentieth-century space time climate variability. Part II: development of 9 96 monthly grids of terrestrial surface climate. Journal of Climate 3: Plaut G, Vautard R Spells of low-frequency oscillation and weather regimes in the Northern Hemisphere. Journal of the Atmospheric Sciences 5: Quinn WH A study of southern oscillation related climatic activity for AD incorporating Nile River flood data. In El Niño: Historical and Paleoclimatic Aspects of the Southern Oscillation, Díaz HF, Markgraf V (eds). Cambridge University Press: Rasmusson EM, Wang X, Ropelewski CF Secular variability of the ENSO cycle. In Natural Climatic Variability on Decade-to- Century Time Scales. National Research Council, Academy Press: Washington DC. Rickey JE, Nobre C, Deser C Amazon River discharges and climate variability: Science 246: 3. Robertson AW, Mechoso CR Interannual and decadal cycles in river flows of southeastern South America. Journal of Climate : Vautard R, Ghill M Singular spectrum analysis in nonlinear dynamics with applications to paleoclimatic time series. Physica D 35: Vautard R, Yiou P, Ghill M Singular spectrum analysis: a toolkit for short, noisy chaotic signals. Physica D 58: WMO Climatic change. Technical Note 79. World Meteorological Organization, Geneva, Switzerland. Copyright 23 Royal Meteorological Society Int. J. Climatol. 23: 3 5 (23)