3.2. Principal Component Analysis
The KMO Index estimated for the monthly streamflow series corresponded to 0.78 and 0.61 for the ARB and the PRB, respectively. According to Diaz and Morales [86
], a KMO measure above 0.60 is tolerable for factor analysis purposes and, in our case, indicated high multicollinearity among the monthly records of the studied basins flow-gauge stations and the feasibility of performing PCAs.
The PCA results indicate a single significant PC for each basin with eigenvalues above 1 (Table 3
). The selected PCs explained around 80% of the variance of the original data for ARB and PRB. The percentage of explained variance was acceptable given the number of flow-gauge stations in each basin and the complex nature of the studied variable. Likewise, the analysis of the PCs of the annual, maximum, and minimum streamflows indicated a single significant PC with eigenvalues above 1 for each case.
shows the time series of the first principal components (PC1s) for ARB and PRB. Positive (negative) values for the two PC1s, during the 1988–1989, 1998–1999, and 2010–2011 (1997–1998 and 2015–2016) periods coincided with the LN (EN) events recorded by Trenberth [87
] and Tedeschi et al. [32
]. These events have been related with an increase (reduction) in rainfall, soil moisture, and streamflows in Colombia [20
The above-mentioned events strongly impacted the Colombian territory. The 1997–1998 EN had a particularly strong effect on ecosystems, agriculture, energy production, and other sectors of economic development in Colombia (IDEAM, 2002). The 2010–2011 LN caused flooding, storms, and landslides, which mainly affected the Colombian Pacific region [62
]. During 1997–1998 EN, northern areas of South America suddenly received less rainfall, conversely to the increased rainfall recorded in the south-easterly part of the continent due to the presence of an anomalous Walker cell with its descending branch between 10° N and 10° S in the South American continent and its ascending branch over the eastern Pacific Ocean [33
]. The 2010–2011 LN effects were recorded in northern parts of South America and were established by anomalies induced by ENSO in the eastern Pacific, the Caribbean Sea, and the North Atlantic, and by the intrinsic variability of the SST anomalies in the Atlantic Ocean [35
3.3. Kendall’s Tau Correlation and Cross-Correlation with Climate Indices
The influence of large-scale climate phenomena on the variability of the ARB and PRB hydrology was studied using correlation analysis. Figure 4
shows the correlation coefficients between the PC1s of both studied basins and the selected climate indices in lag 0. The monthly correlations for ARB showed statistically significant positive (negative) relations mainly with precipitation (PRP), CJ, and SOI (ONI, MEI, SST1+2 and SST4) (Figure 4
a). For PRB (Figure 4
b), the highest positive (negative) correlations statistically significant were presented with PRP, CJ, and SOI (ONI, MEI, SST3.4, and SST4). These results indicate that a direct (inverse) relation exists between the atmospheric (oceanic) variables and the streamflows of the ARB and PRB.
These results are consistent with those obtained by Carvajal et al. [56
], Puertas and Carvajal [58
], Poveda et al. [20
], and Ávila et al. [57
], who reported a reduction (increase) in the streamflows associated with the positive (negative) values of the oceanic ENSO-related indices. Poveda et al. [19
] and Rueda and Poveda [89
] also reported that a direct teleconnection existed between the streamflows of the rivers in western Colombia and the SOI and CJ indices. These findings are coherent with those presented herein. According to Poveda and Mesa [63
], during EN events the CJ weakens strongly with the consequent reduction in rainfall while in LN events the CJ intensifies owing to the quantity of moisture that it transports which, in turn, interacts with easterly trade winds to result in high atmospheric instability and, consequently, in vast amounts of convection and rainfall.
The results in Table 4
show significant negative (positive) correlations between the minimum PRB streamflows and TNA (NAO). This result shows that the PRB streamflows can be influenced by factors other than ENSO. This finding is coherent with Poveda and Mesa [60
], who indicated a nonlinear dependence of Colombian hydroclimatology with NAO, and also with Carvajal et al. [56
], who reported negative correlations between TNA and the streamflows of rivers that lie south of the Colombian Pacific.
The cross-correlation analysis with a lag of up to 12 months between the PC1s of the ARB and PRB and the climate indices revealed a significant persistence up to a delay of 9 months (Figure 5
). Such persistence might be due to the linkages of the rainfall, temperature, evapotranspiration, soil moisture, and run-off in the hydrographic basins [18
]. This result is coherent with that obtained by Navarro et al. [22
], who reported persistence for up to 6, 7, 8, and 9 months for teleconnections between rainfall in stations in western Colombia and the oceanic ENSO-related indices. For the ARB, the highest correlation coefficients between streamflows and the climate indices for the following lags were 0 (SST1+2, SST3, PRP, and CJ), 1 (SST4, SST3.4, and ONI), 2 (MEI and SOI), and 5 (PDO) months. For PRB, the maximum correlation coefficient was recorded with a zero-lag for all the indices, except for PDO, with a 5-month lag. The cross-correlation in lag 0 estimated by Pearson was consistent with the relation shown through the Kendall correlation.
Simultaneous significant correlations were identified with atmospheric indices PRP and CJ (Figure 5
). These indices are related to one another because as the CJ reaches the western mountain range of the Colombian Andes, it rises due to the orographic effect and interacts with the trade winds facilitating the rise of a vast quantity of moisture. It generates an accelerated process of deep convection [63
], which causes positive rainfall and streamflow anomalies for the Pacific region. These results are also coherent with Serna et al. [53
], who reported that CJ is related to SST in the eastern TPO, since the winds associated with the CJ present more variability during canonical EN events, which reduce the SST and pressure gradients between the eastern TPO and western Colombia due to a more pronounced heating in the eastern TPO.
Unlike the ARB, the maximum correlations were obtained for the PRB with all the ENSO-related ocean indices synchronously (i.e., with lag 0). In spatial terms, this indicates that the influence of the ENSO phenomenon on the streamflows in the Colombian South Pacific was more immediate than the influence recorded on the streamflows in the Colombian North Pacific. Nonetheless, the ENSO phenomenon influence was stronger on the ARB than on the PRB because the correlation coefficients between the indices and streamflows of the ARB were higher than the coefficients recorded for the PRB. This result reveals that the influence of the oceanic climate indices is stronger on the streamflows of the basin closer to the TPO. This finding agrees with Poveda et al. [19
], who indicated that the relation between the SOI index and the monthly streamflows of 50 rivers in Colombia was stronger in the stations closer to the TPO.
Finally, statistically significant negative correlation coefficients between the PDO Index and the PC1s of both ARB and PRB were observed (Figure 4
), with maximum values for a 5-month lag (Figure 5
). According to Kayano and Andreoli [91
], PDO can influence the frequency of ENSO events depending on the phase. Thus, the negative correlations with PDO and ENSO indicated that the occurrence of higher (lower) streamflows is favored while negative LN/PDO (positive EN/PDO) events occur.
The correlation analyses show that the ARB and PRB streamflows relate negatively with the ENSO indices and positively with SOI (Table 4
). Previous studies have shown that under EN the tropical atmosphere features higher than normal SLP over Indonesia, below normal SLP over the central and eastern TPO, and an anomalous Walker circulation eastward displaced with anomalous upward movements over the eastern and central equatorial Pacific, where positive anomalies of SST prevail, and anomalous downward movements to the west and northern and northwestern South America [92
]. In association to this anomalous Walker cell, low-level easterlies are established in the extreme eastern equatorial Pacific and adjacent areas including the ARB and PRB. Thus, the CJ is weakened, reducing the transport of moisture from the Pacific to northwestern South America. It is worth noting that the CJ index and the ARB and PRB streamflows are positively correlated (Table 4
). Consequently, the rainfall and streamflow are reduced in the ARB and PRB. On the contrary, the LN presents the opposite anomaly patterns, as documented by Hoyos et al. [62
] and Arias et al. [35
Negative correlations between ARB and PRB with PDO indicate a positive teleconnection between ENSO and PDO. In this way, ENSO teleconnections for the streamflows and rainfall in the study basins are related to the PDO, which creates a background for these teleconnections that act constructively (destructively) when ENSO and PDO are in the same (opposite) phase, confirming the findings of Kayano and Andreoli [91
] for South America.
3.4. Wavelet Analysis
Both time and frequency variations were analyzed for the monthly PC1 of the ARB streamflows obtained by the wavelet transform. The energy spectrum showed a dominant variability on the 2–3-year and 4–6-year time scales owing to the significant variations (at the 5% level) during the 1995–2000 and 1990–2003 periods (Figure 6
a). The energy spectrum of the PC for PRB showed a dominant and significant variability on the interannual 3–4-year and 4–6-year scales observed during the 1997–2003 and 1991–2009 periods (Figure 7
a). Strong variability was found on the 10–15-year decadal scale owing to the nonsignificant variation during the 1997–2005 period (Figure 7
a). The global wavelet spectrum (GWS) of the PC1s of both ARB and PRB are found in Figure 6
b and Figure 7
b, respectively. These figures indicate 4–6-year interannual peaks, which are significant at the 5% level for both basins. Nonsignificant peaks are seen on the 2–3- (3–4-) year interannual scale for ARB (PRB), and a peak on the 10–15-year decadal scale for PRB.
The scale-averaged wavelet power (SAP) scales were built for the 2–3- and 4–6-year scales for ARB, and for the 3–4-, 4–6-, and 10–15-year scales for PRB (Figure 6
c and Figure 7
c, respectively). The SAP was obtained from using Equation (24) described in Torrence and Compo [79
], which allows to examine the scale modulations in the same time series [83
]. In general, both basins show similar time fluctuations of the SAP (in variance units) series on the 4–6-year scale with significant values for ARB between 1990 and 2002 and for PRB between 1995 and 2004. For the ARB (PRB), the 2–3- (3–4-) year interannual scale showed significant values for the 1994–1999 (1997–2002) period. The SAP series for the 10–15-year scale for PRB presented almost significant values during the 1995–2010 period.
3.5. Wavelet Coherence and Phase Difference
The wavelet coherences between the PC1 for ARB and the ONI, MEI, SST1+2, and SST4 (SST3) indices led to similar relations, mainly on the 2–7-year time scale from 1990 to 2010 (1985 to 2017), with a phase difference from −160° to 180° (Figure 8
a and Table 5
). The −160° phase difference for the 2–7-year interannual scale indicates that dry (wet) conditions preceded the mature EN (LN) phase by approximately 1–5 months. The 180° phase difference means that the mature phase of the EN (LN) events occurred simultaneously with minimum (maximum) streamflows. These series also displayed considerable coherence on the 8–16-year decadal scale from 1995 to 2007 with a −160° phase difference, which indicates minimum (maximum) streamflows preceded the mature EN (LN) phase by approximately 5–9 months, except for SST3, for which no decadal coherence was observed (Figure 8
The relationship between El Niño 3.4 and PC1 for ARB was more reduced. Figure 8
c shows significant coherence at an interannual scale (3–6 years) from 1988 to 2010 with a phase difference from −160° to 180°, and for the decadal-scale (10–12 years) from 1997 to 2007 with a phase difference of −170°. The phase difference from −160° to −170° indicates that the maximum (minimum) streamflow anomalies precede the minimum (maximum) variations in SST3.4 by 2–4 and 3–4 months for the interannual and decadal scales, respectively.
Significant coherences were observed with SOI on the 2–7-year interannual scale from 1987 to 2013, with phase differences from 0° to 20° (Figure 8
d). A 20° phase difference indicates that the negative (positive) streamflow anomalies were followed by negative (positive) SOI values with a delay of approximately 1–5 months, whereas 0° suggests that both time series were simultaneously related. Coherence was observed on the 10–14-year decadal scale from 1995 to 2005 with a phase difference from 0° to 10°, indicating that a decrease (increase) in streamflow precedes a low (high) pressure in the eastern TPO by approximately 3–5 months.
The coherence analyses between the PC1 of the PRB and the indices gave similar results for the ONI and SST4, mainly on the 4–8-year time scale from 1987 to 2011, with a phase difference from −160° to 180°, establishing a phase difference around 2–5 months between minimum peak of streamflow and the maximum peak of the indices. Coherences were found on the 8–14-year decadal scale from 1994 to 2007, with a −160° phase difference, which indicates that the anomalies of the streamflow in PRB precede the variations in the SST4 by about 5–9 months (Figure 9
a). The 180° phase difference indicates simultaneous alterations.
b also shows significant relations on the 2–7-year interannual time scale from 1986 to 2013, with a phase difference from −150° to 180°. For this time scale, the phase difference of −150° generates a lag between the maximum peak of the indices and the minimum peak of the streamflow of 30° (2–7 months). In other words, the maximum (minimum) streamflow anomalies occur in the development phase of an LN (EN) event. Simultaneous variations were also observed between SST anomalies in the Niño 3.4 region and PRB streamflow on the 4–8-year time scale from 1986 to 2011 (Figure 9
c). The relation between SOI and the PRB streamflow can be seen in Figure 9
d. One link was found, mainly during the 1990–2010 period, with a phase difference from 0° to 20° on the 4–7-year scale indicating 3–5-month lags. These series also showed coherence on the 10–12-year time scale for the 1996–2004 period, with a 45° phase difference, that means lags of approximately 15–18 months (Figure 9
Therefore, the present work reveals that the SST anomalies in different ENSO regions cause distinct ARB and PRB streamflow responses, whose temporal persistence also varies. This could be due to the orographical diversity in West Colombia [26
], and also to the influence of different atmospheric factors, which indicate considerable complexity in the spatiotemporal structure of precipitation and other hydroclimate variables in the region [96
The coherence between the PC1 for ARB (PRB) and PRP was significant on the 2–8-year time scale from 1985 to 2011 with a phase difference from 0° to 10° (0° to 45°) (Figure 10
a,c). For this scale, the 10° (45°) difference indicated that the influence of PRP on streamflows was recorded with 1–3 (3–12) month lags, whereas 0° indicated that both time series were simultaneously related. On the 10–14 (8–14) year decadal scale, the indices presented −45° (0° and 10°) phase difference, which implies that the influence of PRP on streamflows of ARB (PRB) was reflected within a 15–21 (3–5) month time interval.
The CJ index and PC1 for ARB showed coherences on the 2–7-year interannual scale from 1988 to 2013 (Figure 10
b), with phase differences from 0 to −45°, meaning a delay of approximately 3–11 months. Significant coherences between the CJ and PC1 for PRB were observed on two interannual scales (Figure 10
d): (i) from 1987 to 2003 on the 2–3-year scale, with a −90° phase difference, which indicates a time of 6–9 months beforehand between the CJ index and the basin streamflows, and (ii) from 1990 to 2011 on the 4–6-year scale, with phase differences from 0° to −10°, indicating lags of 1–2 months between the CJ index and the PRB streamflows.
The influence that CJ has on both basins’ streamflow anomalies has been previously documented and is relevant for the vast quantities of moisture being transported toward the continent, which would explain why Colombia is considered one of the rainiest areas in the world [28
]. Poveda and Mesa [52
] reported an average moisture transport rate close to 3.78 × 106 kg·s−1
or 3774 m3
, which would directly contribute to the annual discharge of ≈5000 m3
for the ARB and San Juan River, and is one of the highest run-off rates worldwide. The wavelet coherences analyses with CJ index identified different responses of streamflows such that the anomalous moisture transport associated with CJ responds more immediately to the PRB streamflow anomalies (0–9 months) compared to ARB (3–11 months).
The coherence results obtained with PDO are depicted in Table 5
. It was not possible to define a significant coherence with the ARB streamflows, and the relation with PRB was observed during the 1987–2011 period on the 5–7-year scale, with a phase difference from −170° to −150° that generates a lag between the maximum of the PDO index and the minimum peak of streamflows by 2–7 months beforehand (figure not shown).