The South American Monsoon System (SAMS) is characterised by intense convective activity and precipitation that peaks in tropical South America during the austral summer (December–February) (Horel et al., 1989; Lenters and Cook, 1997; Zhou and Lau, 1998; Marengo et al., 2001; Jones and Carvalho, 2002; Vera et al., 2006; Silva and Carvalho, 2007; Bombardi and Carvalho, 2009). Many studies have demonstrated that SAMS varies over broad ranges of temporal and spatial scales and important local and remote linkages controlling its variability have been identified (Liebmann et al., 1999; Nogues-Paegle et al., 2000; Fu et al., 2001; Zhou and Lau, 2001; Nogues-Paegle and Mo, 2002; Grimm, 2003, 2004; Carvalho et al., 2004; Liebmann et al., 2004; Grimm et al., 2007; Ma and Mechoso, 2007; Muza et al., 2009).
An important characteristic of SAMS is the timing of its onset and demise, which has drawn considerable interest. The pioneering works of Kousky (1988) and Horel et al. (1989), for instance, used 5-day averages (pentads) of outgoing longwave radiation as a proxy to describe the annual cycle of convection and determined that the mean onset of the wet season over the Amazon and associated Bolivian high circulation usually occurs within a single month. Typically, the onset and demise occur in mid-October and mid-April, respectively. Marengo et al. (2001) used rainfall station data from 1979 to 1996 and found the climatological onset to occur in mid-October and reach the core of the Amazon near the end of the year with large interannual variability. Moreover, they found that onset/demise dates in northern and central Amazon appear to be related to sea surface temperature (SST) anomalies in the tropical Atlantic and Pacific Oceans such that warm Pacific and cold Atlantic SST conditions result in a delayed onset and early withdrawal of SAMS. Other studies defined SAMS onsets and demises using different data sets as well as criteria based on atmospheric circulation characteristics (Liebmann and Marengo, 2001; Gan et al., 2006; Liebmann et al., 2007; Garcia and Kayano, 2009; Raia and Cavalcanti, 2008).
The variability of SAMS onset and demise on multi-annual to decadal time scales, however, is less known due in large part to limited observations in the Amazon basin and surrounding areas. The number of precipitation records over key areas of SAMS was significantly lower before the late 1970s and the network of stations was quite sparse (Liebmann and Allured, 2005). Daily precipitation records extending back to the 1940s are usually available only over the northeast, southeast and southern regions in Brazil, which precludes a detailed characterisation of long-term changes in SAMS.
This note describes results of an ongoing investigation to further understand decadal changes in SAMS. We employ here an alternative method to describe the onset and demise of SAMS. In a previous paper, Silva and Carvalho (2007) introduced the large-scale index for the South American Monsoon (LISAM; Silva and Carvalho, 2007). LISAM was developed to represent combined changes in precipitation and large-scale circulation associated with SAMS using satellite data and atmospheric fields from reanalysis (1979–2005). That study revealed that LISAM is a useful approach to represent intraseasonal-to-interannual variations in SAMS and the determination of SAMS onset/demise dates agreed well with other studies using different methods. The present study extended LISAM using all fields from National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis (1948–2008) and investigated possible changes in SAMS characteristics with emphasis on the climate transition of the 1970s (e.g. Graham, 1994).
The variability of SAMS was investigated using the NCEP/NCAR reanalysis (hereafter NNR) (Kalnay et al., 1996; Kistler et al., 2001). Pentads (5-day averages) of precipitation (P), zonal (U) and meridional (V) wind components, temperature (T) and specific humidity (Q) at 850-hPa were used from 1 January 1948 through 31 December 2008. To complement the analysis, pentads of precipitation from the Global Precipitation Climatology Precipitation (GPCP) (Xie et al., 2003) were used (1 January 1979–31 December 2007).
LISAM was employed to describe changes in SAMS characteristics. To derive LISAM, long term (1948–2008) means were first removed from the time series of P, U, V, Q and T. LISAM is then defined as the first time coefficient of a combined empirical orthogonal function (EOF) analysis of [P, U, V, Q and T]. LISAM was computed in two different ways: (1) all variables from NNR (1948–2008) and (2) P from GPCP and U, V, Q, T from NNR (1979–2008). Differences between the two computations during 1979–2008 are discussed in the next section. The inclusion of all variables in the LISAM computation results in the best representation of SAMS variability. It is appropriate to mention that some studies have cautioned in using P from NNR in the tropics, since it is not consistent with observed precipitation (e.g. Kinter et al., 2004). To examine this issue, we also computed LISAM only with [U, V, Q and T] and compared against LISAM with [P, U, V, Q and T] during 1948–2008. SAMS characteristics are insensitive to LISAM computed with all fields from NNR. Mean differences in onset, demise and duration between both computations are less than 1 pentad and correlations are above 0.90.
LISAM only with [U, V, Q and T] was also computed with the European Centre for Medium Range Forecast Reanalysis (ERA40) (1958–2001) and compared against similar calculations using NNR. The correlations between dates of onset and demises were 0.68 and 0.86, respectively. The correlation between durations of SAMS was 0.81. This ensures consistency of results between different reanalysis products available. For completeness, this study employs LISAM with all fields and the spatial patterns of the first EOF are shown in Figure 1. The reader is referred to Silva and Carvalho (2007) for additional details on LISAM.
In order to investigate potential changes in SAMS characteristics, we also used monthly river flow from the Parana River basin from 1905 to 2006. The gauge site is located in Timbues, Argentina, (60.71°W and 32.67°S). The river flow data were collected and archived at NCAR and originates from various sources of streamflow data, including those collected by the United Nations Educational Scientific and Cultural Organisation, the United States Geological Survey, University of New Hampshire, (UNH), Russia's State Hydrological Institute, the Global Runoff Data Centre (GRDC) in Germany and many national archives such as those for South American nations (see Dai et al., 2009). While this station and its catchment base are situated south of the core of SAMS, this independent data record is used to check the consistency with the NNR results. Mean annual river flow was computed by averaging the monthly data from 1 July through 30 June for each year during 1905–2006.
The variability of SAMS was analysed in terms of the following properties: amplitude, onset and demise date and duration. To identify onsets and demises of SAMS, LISAM was smoothed 15 times with a 3 point moving average (equal weights) and the beginning and ending dates were identified when the index became positive and negative, respectively. Smoothing is necessary to avoid identification of false onset and demise periods, for example, when intraseasonal events occur before (after) the onset (demise) dates and the unsmoothed LISAM remains positive (negative) for a short time (e.g. ∼15–20 days). The selected number of passes was deemed appropriate and the results are not sensitive for a reasonable choice of additional passes (∼20 passes). Lastly, SAMS' duration is defined as the interval between onset and demise dates.
Figure 2 displays onset and demise dates of SAMS (bars). For completeness, LISAM computed with combination of GPCP and NNR is also shown (solid lines). Mean bias (GPCP-NNR combination minus all NNR) in onset dates is 1.8, whereas for demise dates is − 0.9 pentads. Correlations for onset and demise dates are 0.90 and 0.87, respectively. This indicates that the inclusion of P from NNR in the LISAM computation is not substantially large. More importantly, the trend in onset is toward earlier dates from 1948 to the early 1970s. Since 1972 they have generally remained earlier than pentad 60 (23–27 October). Likewise, progressively later demise dates were observed from 1948 to the late 1970s. Since then they have almost always been later than the 21–25 April pentad (23).
To determine the amplitude of SAMS during each season, the mean annual cycle of LISAM was calculated by averaging each of the 73 pentads per year over all years and then smoothing the resulting series with 20 passes of a 1-2-1 filter. The annual cycle was removed from the original (unsmoothed) LISAM and the amplitude of LISAM was defined as the integral of positive anomalies between onset and demise for each SAMS season. Figure 3 shows the amplitudes of LISAM during 1948–2008. The most obvious feature is the change in amplitudes in the early-to-mid 1970s, which does not coincide with the introduction of satellite data in NNR and suggests a possible modification in SAMS behaviour. Moreover, it is interesting to note that amplitudes during 1952–1954 are comparable to amplitudes after 1979, which gives confidence in using LISAM based on all fields from NNR to investigate the variability of SAMS.
The length of SAMS (i.e. duration) (Figure 4) varies widely from 1948 to 2008, with a mean of 38 pentads (190 days). The previous results suggest that a change in SAMS characteristics may have occurred in the 1970s, which is consistent with the climate transition discussed in many studies (Kayano et al., 2009). To quantify this change, the statistical method discussed in Rodionov (2004) was employed to detect shifts (i.e. changepoints) in the mean duration of SAMS. The method is based on a sequential data processing technique which tests differences in the mean between two data segments of length L. Different values of L (from 7 to 12 years) were used to test the sensitivity of the method in finding robust changepoints. Using L = 10 and significance of 5% indicated three changepoints in SAMS duration: 1971–1972, 1981–1982 and 1997–1998. Mean durations during the corresponding periods were: 34 (1948–1972), 39 (1972–1982), 41.3 (1982–1998) and 38.5 pentads (1999–2008). More importantly, the changepoint during 1971–1972 is insensitive to a wide range of L, which implies an important long-term modification in SAMS duration in the early 1970s. It is also worth noting that the systematic bias in LISAM computed with all NNR fields and combination of GPCP-NNR does not explain the changepoint in 1971–72. In other words, even if one decreases SAMS duration by ∼3 pentads during 1979–2008 (Figure 4), it still does not affect the changepoint in the early 1970s.
To characterise changes in the large-scale atmospheric circulation associated with the climate transition of the 1970s, we examined vertically integrated zonal and meridional moisture fluxes (Mu, Mv) (surface to 300-hPa). Figure 5 shows (Mu, Mv) averaged during all SAMS seasons. It illustrates climatological features of equatorial easterlies in the Atlantic and Pacific Oceans, whereas easterly moisture fluxes in tropical South America are observed near the equator and northerly/northwesterly fluxes near Bolivia and Paraguay.
Changes in the mean state during SAMS onset months were assessed by computing differences in (Mu, Mv) during 1982–2008 minus 1948–71 (Figure 6). Local statistical significance was based on a t-test with effective degrees of freedom estimated according to (Wilks, 2005). To illustrate the differences between the two time periods, the evolution of the anomalies is shown during September, October and November separately. The results show larger eastward moisture fluxes over the equatorial eastern Pacific extending over tropical South America from 1982 to 2008 relative to 1948 to 1971. Two other features are worth noting over South America. First, anti-cyclonic anomalies develop over the northern Amazon in September, intensify in October and move eastward over the tropical eastern Atlantic in November. Second, southerly anomalies develop to the east of Bolivia in September, which appear to intensify and close into cyclonic anomalies in October. Lastly, westward moisture fluxes have increased during all 3 months along the west coast of Africa in the later period. The general patterns shown in Figure 6 suggest increased moisture fluxes in tropical South America during onset months after the 1970s climate transition.
Variations in moisture fluxes during SAMS demise months (Figure 7) also exhibit westerly anomalies over the equatorial eastern Pacific extending over tropical South America, although they are more intense and zonally oriented relative to the onset months. Cyclonic anomalies over central Brazil persist during March and April. There are also northerly northeasterly flux anomalies over the northern coast of Brazil, which persists during March–May.
Changes in moisture fluxes during the peak of the monsoon season (December–February) are displayed in Figure 8 and show weakening of the easterlies in the equatorial eastern Pacific. As previously noted in the onset and demise months, westerly moisture flux anomalies extend over tropical South America. It is interesting to note that as westerly flux anomalies enter the South American continent they are deflected slightly southward as they cross over the Andes Mountains and then contribute to the cyclonic anomaly over the southern Amazon and central Brazil. It is also worth noting that moisture flux anomalies over the northern Amazon in Brazil are relatively small compared to those in the southern region. There is an increase in northerly moisture fluxes over the northern parts of South America and strengthening of easterlies in the eastern tropical Atlantic. Statistically significant anomalies further suggest weakening of the anti-cyclonic circulation in the southern Atlantic (compare Figures 5 and 8) and strengthening and poleward displacement of the subtropical jet (∼35°–60°S).
To investigate the consistency of the results obtained with LISAM with an independent measure of climate change in South America, Figure 9 shows the annual flow at the Parana River from 1905 to 2006. The same test previously described indicated a statistically significant (5% level) changepoint in the series in 1972, which is largely consistent with the previous results shown here. Additional evidence for large-scale climate changes in South America has been reported in other studies. Piovano et al. (2002) analysed sediment records in Laguna Mar Chiquita, a highly variable closed saline lake located in the Pampean plains of central Argentina, and reported a significant increase in lake level also in 1972 (see their Figure 2).
4. Summary and discussion
We used LISAM to examine long-term changes in onset, demise and duration of SAMS during the period of available reanalysis. The results show that onset dates have changed in the early 1970s and tend now to occur earlier than pentad 60 (23–27 October). Likewise, progressive increases in demise dates were observed from 1948 to late 1970s and have, in general, occurred later than pentad 23 (21–25 April) after the mid-to-late 1970s. Consequently, SAMS duration determined with LISAM shows a changepoint in the summer of 1971–1972 such that the mean duration was 34 pentads during 1948–1972 and 39 pentads in 1972–1982. Modifications in the mean state associated with the 1970s climate transition were assessed with differences in vertically integrated moisture fluxes during 1982–2008 and 1948–1971.
A number of studies have discussed a significant shift in the global climate in the mid-to-late 1970s (Zhang et al., 1998; Deser et al., 2004; Deser and Phillips, 2006). Some of the features of this climate transition are an intensification of the Aleutian low pressure system and changes in SST patterns related to the Pacific Decadal Oscillation (Hare and Mantua, 2000). Furthermore, there is strong evidence that the climate transition of the 1970s was associated with SST changes over the Pacific and Indian Oceans and it is widely agreed that the Indian Ocean and western Pacific have substantially warmed since 1977 (e.g. Deser et al., 2004).
The impact of the 1970s climate transition in South America has been addressed in previous works (Marengo, 2004; Agosta and Compagnucci, 2008 and references therein). Marengo (2004), for instance, developed a detailed analysis of precipitation and circulation changes in the Amazon and showed an overall negative trend in rainfall over the entire Amazon basin during 1950–1998. However, that study also pointed out large regional differences in the sign and magnitude of trends. Based on the available rainfall gauges, Marengo (2004) developed indices for northern and southern Amazon regions. While northern Amazon received less precipitation after 1975–1976, southern Amazon exhibited a positive trend in rainfall after the 1970s climate transition. More recently, Marengo (2009) extended the analysis of river flow and precipitation showed regional differences between northern and southern Amazonian regions such that low-frequency variations are not in phase. Interannual changes appear strong in the northern Amazon, whereas decadal and multi-decadal variations are more evident in the southern Amazon.
The increase in moisture fluxes over South America after 1971–72 shown here is consistent with other studies (Curtis and Hastenrath, 1999; Marengo, 2004). In particular, the pattern of moisture flux anomalies displayed in Figure 7 indicates favouring conditions for enhanced precipitation over southern Amazon and central Brazil. This is in agreement with the positive trend in rainfall over southern Amazon after 1975–1976 reported in Marengo (2004).
Likewise, enhanced meridional moisture flux anomalies (Figure 7) over tropical South America after the summer of 1971–1972 may represent important clues about changes in the dates of onset, demise and duration reported here. This is substantiated by the study of Wang and Fu (2002), who developed an index based on the meridional component of the wind and showed that a northerly regime is associated with precipitation shifts toward the Amazon basin and subtropical South America.
It is also appropriate to recall that SAMS varies on many different spatial and time scales. On intraseasonal time scales, Carvalho et al. (2002) showed observational evidence that active and break periods in SAMS are associated with coherent structures in circulation and convective activity. Low-level westerly (easterly) anomalies correspond to enhanced (suppression) convection and precipitation over central Brazil and suppression (enhanced) over the northern parts of South America. Carvalho et al. (2010) demonstrated that enhanced amplitudes of SAMS on intraseasonal timescales as determined with LISAM are associated with northwesterly moisture flux anomalies over the Amazon and central Brazil. In the context of intraseasonal anomalies, some studies found a significant increase in the activity of the Madden-Julian Oscillation (MJO) after the mid-1970s (Jones and Carvalho, 2006; Pohl and Matthews, 2007; Jones, 2009; Jones and Carvalho, 2009). The MJO is the most fundamental mode of tropical intraseasonal variations with a key role in SAMS variability.
The results presented here emphasised SAMS onset, demise and duration. A change in its behaviour occurred close to the 1971–1972 season. This is relatively earlier than the mid-to-late 1970s climate transition reported in several studies, although Baines and Folland (2007) discussed a number of important changes in the global atmospheric circulation and climate centred on the late 1960s. The processes that control the onset and demise of SAMS remain elusive. An important issue that remains to be determined is the physical mechanism in which changes in the mean state and tropical intraseasonal variations interacted over SAMS before and after the 1970s climate transition. This issue is being further investigated and results will be presented elsewhere.
This research was funded by NOAA Office of Global Programs (NOAA/NA05OAR4311129, NA07OAR4310211 and NA08OAR4310698). We thank the financial support from the European Community's Seventh Framework Programme (FP7/2007-2013) under Grant Agreement N°212492 (CLARIS LPB). NCEP/NCAR Reanalysis data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.cdc.noaa.gov.