1 Introduction

Rainfall variability governs spatial and temporal availability of water resource (Chowdhury and Beecham 2010) and might affect the critical socioeconomic sector by changing characteristics of water supply for human consumption and agricultural activities (Shahid 2010). In particular, regions whose economy depends on agricultural production demand a better knowledge of climate variations as part of the agronomic sector's strategic planning tools; this is the case for Southwestern Colombia.

Regarding large-scale phenomena, several studies evidence the influence of the El Niño–Southern Oscillation (ENSO) in the tropical Pacific on the interannual rainfall variations in many parts of the globe, including South America (Ropelewski and Halpert 1986; Kousky and Ropelewski 1989; Grimm et al. 2000; Zhou and Lau 2001; Kayano and Andreoli 2006; Kayano et al. 2009, 2013; Hill et al. 2011; Araújo et al. 2013; Tedeschi et al. 2015; Tedeschi and Collins 2016; Canchala et al. 2020a), and Australia and India (Felton et al. 2013; Girishkumar et al. 2015; Bhowmick et al. 2019). In South America, a reduced rainfall anomaly occurs in the northern and northeastern sectors and an increased rainfall anomaly prevails in the western equatorial coast and southern/southeastern areas during the warm ENSO (El Niño) phase, and reversed rainfall anomaly conditions occur during the cold ENSO (La Niña) phase (Pabón and Montealegre 2000; Poveda et al. 2001a, 2002; Córdoba-Machado et al. 2015a, b; Kayano et al. 2013; Tedeschi et al. 2013; Ávila et al. 2019; Kayano and Andreoli 2006; Hill et al. 2011; Arias et al. 2015; Andreoli et al. 2017). The oceanic ENSO signal is associated with a large-scale oscillation of sea level pressure with action centers over Indonesia and the southeastern Pacific (McPhaden et al. 2006; Rasmusson and Arkin 1985; Grimm et al. 1998, 2000; Wolter and Timlin 2011). ENSO's influence on the equatorial region occurs through variations in the Walker circulation and the local Hadley circulation (Kousky et al. 1984; Rasmusson and Mo 1993; Zhou and Lau 2001).

ENSO impacts in the Colombian rainfall, in general, have resulted in negative socioeconomic and environmental consequences (Poveda et al. 2001a, 2003; Hoyos et al. 2013a; Córdoba-Machado et al. 2016; Cerón et al. 2020d). During the 1997–1998 El Niño event, Colombia recorded a decrease in rainfall that affected the country's most important agricultural export (10% collapse in coffee production) (Poveda et al. 2001a), and nearly 12,000 forest fires were reported (Unidad Nacional para la Gestión del Riesgo de Desastres—UNGRD 2016). During the 2015–2016 El Niño event, one of the strongest on record (Jacox et al. 2016; Martínez et al. 2017), the rainfall deficit was 40% in the Andean region and 37% in the Caribbean, affecting over one million hectares of crops, mostly in the states of Atlántico, Córdoba, and Nariño (Southwestern Colombia) (UNGRD 2016). Furthermore, in endemic areas of Colombia, El Niño has also been associated with an increase of malaria cases (Poveda et al. 2001b), due to the rise of temperature and decrease of rainfall, which are favorable conditions o the formation of breeding sites, and therefore, the increment of vectors.

Conversely, during the La Niña events of the 2010–2012 period, extreme rainfall and flooding occurred in Colombia, causing disasters in the Magdalena and Cauca river valleys (Arias et al. 2015; Enciso et al. 2016), and in the Atrato river valley (Hoyos et al. 2013b). Overall, most parts of the Colombian territory recorded increased river discharges that affected approximately 9% of the total population and brought economic losses of around US $7.8 billion (Comisión Económica para América Latina y el Caribe (CEPAL) 2013; Hoyos et al. 2013b).

Recent studies have shown that close subregions in South America might be affected differently by the ENSO (Campozano et al. 2016; Morán-Tejeda et al. 2016; Canchala et al. 2020a). Morán-Tejeda et al. (2016) and Vicente-Serrano et al. (2017) showed a strong influence of Niño 1+2 on the magnitude and seasonal distribution of rainfall along the Ecuadorian coast and a more pronounced impact of Niño 3.4 in the western Andes mountain range. They also observed that El Niño and La Niña phases promote droughts in the Andean regions and over the western plains, respectively. Hurtado (2009) and Guzmán et al. (2014) identified substantial variability in rainfall within small distances in Colombia due to gradients induced by the topography, mainly in the Pacific western plains and the mountain slopes. In this respect, recent studies show that rainfall in Southwestern Colombia, an essentially rural region whose economy depends on agriculture, has a strong negative correlation with the ENSO phenomenon in the mountain region, while the western plains near the Pacific Ocean exhibit no significant correlation (Canchala et al. 2020b,c). Since the rainfall in these regions is fed by the regional moisture transport (Mesa-Sánchez and Peñaranda-Vélez 2015), the previously documented differences in rainfall associated with the same ENSO phase are further analyzed here in terms of the oceanic and atmospheric circulation features, which might drive the differences in this transport. Therefore, the present work focuses on the large-scale and regional climatic features related to the rainfall variations in Southwestern Colombia subregions through wavelet analysis and total and partial correlations. The data and methods are presented in Sect. 2; results and discussions in Sect. 3; and finally, conclusions in Sect. 4.

2 Data and Methods

2.1 Rainfall Data

Monthly rainfall time series at 44 stations in Southwestern Colombia, whose locations are illustrated in Fig. 1, have been provided by the Institute of Hydrology, Meteorology, and Environmental Studies (IDEAM) of Colombia. These time series were previously subjected to quality control procedures, and missing data were estimated with nonlinear principal component analysis (NLPCA) using an artificial neural network approach by Canchala et al. (2019). Canchala et al. (2020a) classified the rainfall stations into the Andean Region (AR) and Pacific Region (PR) according to the rainfall regionalization for Southwestern Colombia, using NLPCA and self-organizing map approaches.

Fig. 1
figure 1

Study area and location of the meteorological stations used in this study. Black triangles: Western Plains or Pacific Region (PR) stations. White triangles: Andean Region (AR) stations

Full size image

Southwestern Colombia occupies a geostrategic position because it is crossed by the Andes Mountain Range, near the Tropical Pacific Ocean, and has topography changes in small distances. Concerning the microclimate of Southwestern Colombia, 8% of its territory belongs to the Amazon rainforest region, one of the outstanding biodiversity reserves in the world, 52% corresponds to the Pacific plain with high rainfall and exceptional conditions in the diversity of communities and species, and the remaining 40% belongs to the Andean area where the moors and volcanoes stand out (Gobernacion de Nariño 2016). The AR typifies the mountain range in Southwestern Colombia and covers about 15,500 km2, while PR represents the western plains near the Pacific Ocean coast with 14,700 km2 (Hurtado 2009; Canchala et al. 2020b). Rainfall in the southern Colombian Andes (AR) presents values between 1000 and 2000 mm year–1 and an average monthly rainfall around 130 mm month–1, contrasting with the 3000–7000 mm year–1 and 350 mm month–1 in the western plains (PR) (Cerón et al. 2020b; Canchala et al. 2020b; Hurtado 2009; Rodríguez-Eraso et al. 2010).

Since the observed rainfall data are limited to Southwestern Colombia, for a more extensive view on the rainfall anomalies in northwestern South America, the gridded rainfall data from the Climate Hazards Group Infrared Precipitation with Stations (CHIRPS 2019; Funk et al. 2014, 2015), version v2.0, were also used. The CHIRPS data set has daily time steps from January 1981 to present, a high spatial resolution of 0.05°, and almost global coverage (50° S–50° N and 180° W–180° E), available online at https://www.chc.ucsb.edu/data/chirps/. Following a previous study that described the rainfall variability in the entire Colombia territory (Cerón et al. 2020c), the CHIRPS rainfall data at 0.25° × 0.25° of horizontal resolution were used. Therefore, monthly rainfall data for the same period as the observed data were obtained in the domain between 5° S–15° N and 65° W–85° W.

2.2 Atmospheric and Sea Surface Temperature Information

Gridded (2° × 2°) monthly SST data were obtained from the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST V5 (ERSST) data set (NOAA/ERSST) (Huang et al. 2017). SST data were obtained in the area bounded at 120° E, Greenwich longitude, 50° N and 50° S. Data on zonal (u) and meridional (v) winds and specific humidity (q) at surface pressure levels from surface to 300 hPa were obtained from the European Centre for Medium-Range Weather Forecasts Interim Reanalysis (ECMWF/ERA-I 2019) (Dee et al. 2011). The atmospheric variables are in 1° horizontal resolution. All data sets were obtained for the 1983–2016 period.

The zonal and meridional winds and specific humidity data were used to calculate the vertically integrated moisture flux (VIMF) from surface to 300 hPa, and its divergence (kg m−1 s−1) following the equations shown in Peixoto and Oort (1992). These calculations were done in the domain between 15° S–25° N and 65° W–110° W. Additionally, the Oceanic Niño Index (ONI) was obtained from the NOAA Climate Prediction Center, available online at https://origin.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ONI_v5.php.

2.3 Methods

Taking into account the previous classification of rainfall in Southwestern Colombia into the two regions AR and PR (Guzmán et al. 2014; Canchala et al. 2020c), rainfall indices were calculated for each region. The AR and PR include, respectively the 33 and 11 gauge stations illustrated in Fig. 1. Rainfall indices were constructed separately for each region. First, the individual months of all years were area-averaged, such that a monthly rainfall time series was obtained. Then, monthly standardized anomalies were calculated considering the corresponding monthly means and standard deviations of the 1983–2016 period. These time series are the AR and PR rainfall indices.

For the gridded variables (SST, components of VIMF, divergence of VIMF, and CHIRPS rainfall), the time series of monthly standardized anomalies at each grid point was obtained using the monthly means and the standard deviations in the 1983–2016 period.

Following the computational procedures of the wavelet analysis reported by Torrence and Compo (1998) and Grinsted et al. (2004), the Morlet wavelet was used for the time–frequency analysis of the AR and PR indices to determine the dominant oscillations and their time variations (Torrence and Compo 1998). Morlet's wavelet is a complex exponential modulated by a Gaussian:

$$\psi \left(t\right)={e}^{i{\omega }_{0}\eta }{e}^{-\frac{{\eta }^{2}}{2}}$$
(1)

where \(\eta =t/s\), \(t\) is the time, \(s\) is the wavelet scale according to time \((s=2/\partial t)\), and \({\omega }_{0}\) is nondimensional frequency (Torrence and Compo 1998). In order to have a comparable wavelet transform over many scales, the wavelet function is normalized in each scale. For a given scale \(s\), the global wavelet power (GWP) is the mean time (Torrence and Compo 1998):

$${\overline{{W}^{2}}}= \frac{1}{N} \sum_{n=0}^{N-1}{\left|{W}_{n}\left(s\right)\right|}^{2}$$
(2)

The wavelet analysis was also used here as a band-pass filter reconstructing the time series as the sum of the real part of the wavelet transform over an interval of scales \({j}_{1}\) and \({j}_{2}\) (Torrence and Compo 1998):

$${x}_{n}^{\prime}= \frac{\delta j\delta {t}^{1/2}}{{C}_{\delta }{\psi }_{0}\left(0\right)} \sum_{j={j}_{1}}^{{j}_{2}}\frac{\mathfrak{R}\left\{{W}_{n}\left({s}_{j}\right)\right\}}{{s}_{j}^{1/2}}$$
(3)

where \({\psi }_{0}\left(0\right)\) is a factor that removes the energy scaling, \({s}_{j}^{1/2}\) converts the wavelet transform into an energy density, \({C}_{\delta }\) is a constant (for Morlet, \({C}_{\delta }=0.776\)).

Here, this filter was used to isolate the interannual timescale fluctuations in the variables, rainfall indices, and ONI. The interannual band is specified in the results.

Furthermore, the relationships between the unfiltered AR, PR, and ONI indices were assessed from wavelet coherence analysis by calculating their cross-wavelet spectrum and the wavelet coherency phase difference according to Torrence and Webster (1999). The wavelet transform coherence (WTC) measures the coherence of two time series, which can be interpreted as the square of the correlation coefficient. According to Torrence and Webster (1999), given the two time series, X(t) and Y(t), with their respective wavelet transforms, \({W}^{X}\left(t,s\right)\) and \({W}^{Y}\left(t,s\right)\), the WTC of the two time series is

$${R}_{n}^{2}\left(s\right)=\frac{{\left|S\left({s}^{-1}{W}^{XY}\left(t,s\right)\right)\right|}^{2}}{S\left({s}^{-1}{\left|{W}^{X}\left(t,s\right)\right|}^{2}\right) \dot S\left({s}^{-1}{\left|{W}^{Y}\left(t,s\right)\right|}^{2}\right)}$$
(4)

where \(S\) is a smoothing operator, \({W}^{XY}\left(t,s\right)={W}^{X}\left(t,s\right){W}^{Y*}\left(t,s\right)\) is the cross-wavelet spectrum, and \({R}_{n}^{2}\) can range from 0 to 1, with 0 indicating no correlation between the two series, and 1, a perfect correlation. Monte Carlo methods were used to estimate the statistical significance of the wavelet analysis and wavelet coherence (Torrence and Webster 1999; Grinsted et al. 2004).

Maps of total linear correlations between the filtered atmospheric and oceanic variables, and the filtered AR and PR indices were obtained for the period 1983–2016. The statistical significance of the correlation was established with the Student t test at a 95% confidence level, using 34 degrees of freedom, which correspond to the number of years in the rainfall time series. Partial correlation maps were also obtained. According to Panofsky and Brier (1968), the partial correlation between two variables, \(A\) and \(B\), while excluding the influences of a third independent variable, \(C\), is defined as

$${r}_{AB\cdot C}= \frac{\left({r}_{AB}- {r}_{AC} \times {r}_{BC}\right)}{\sqrt{1- {r}_{AC}^{2}}\sqrt{1- {r}_{BC}^{2}}}$$
(5)

where \({r}_{AB}\), \({r}_{AC}\), and \({r}_{BC}\) are the total linear correlations between \(A\) and \(B\), between \(A\) and \(C\), and between \(B\) and \(C\), respectively. The partial correlation maps between the filtered AR or PR indices and the filtered atmospheric and oceanic data excluding the influence of the ONI were calculated. For conciseness, the reference to these partial correlations in the results will mention only the rainfall index (AR or PR) and involved atmospheric or oceanic variables. Here, the statistical significance uses the two-tailed Student t test at a 95% confidence level.

3 Results and Discussion

3.1 Dominant Modes of Variability

Figure 2a presents the wavelet power spectrum (WPS) for the AR index in the time-period plot which shows the largest power at interannual timescale (3–6 year) during the 1995–2002 and 2008–2015 periods, and at decadal timescale (8–15 year) during the 1997–2016 period. Consistently, the GWP presents significant peaks in the 3–6- and 8–15-year time scales (Fig. 2b). Figure 2c shows the filtered series for the 3–6-year and 8–15-year time scales. The filtered time series for the 3–6-year time scale clearly indicates a minimum during 1997–1998 and a maximum during 2010–2011, which represent the AR rainfall response to the strong El Niño and strong La Niña that occurred in these years, respectively. Thus, these strong ENSO events were responsible for the largest interannual power of the AR index during the 1995–2002 and 2008–2015 periods. Furthermore, filtered time series for the 8–15-year time scale shows conspicuous decadal oscillations during the 1995–2015 period, which overlaps the period with the largest decadal power of the AR index.

Fig. 2
figure 2

a Local wavelet power spectrum (WPS) of the continuous wavelet transform (CWT) of the Andean Region (AR) index; b global wavelet power (GWP) using variance units; and c filtered series. The region where edge effects are significant is on the U-shaped curve in a. The dotted curve in a, b is the 95% confidence level

Full size image

Otherwise, WPS and GWP of PR evidence the concentrated power on the interannual band of the 3–6-year time scale, during the 1998–2010 period (Fig. 3a, b). Figure 3c shows the filtered series for the 3–6-year time scale, which shows that most minima during the 1998–2010 period are not related to the El Niño occurrences, and only three maxima coincide with La Niña of 1998–1999, 2000–2001, and 2010–2011. Thus, this result indicates a weak relation of the PR index interannual variations with the ENSO.

Fig. 3
figure 3

a Local wavelet power spectrum (WPS) of the continuous wavelet transform (CWT) of the Pacific Region (PR) index; b global wavelet power (GWP) using variance units; and c filtered series. The display is the same as in Fig. 2

Full size image

3.2 Wavelet Coherence and Phase Difference

The wavelet coherence and phase difference between the ONI index and the AR and PR indices present contrasting differences (Fig. 4). The AR shows significant coherence on the interannual to decadal scale from 1990 to 2014, and phase differences varying from −135° to 180° (Fig. 4a). For the 3–6-year scale, the phase difference of −135° generates a lag between the maximum peak of the AR index and the minimum peak of the ONI of 45° (4–9 months); that is, the maximum rainfall in the AR anomalies occur during the development phase of a La Niña event, and minimum rainfall during the developmental phase of an El Niño event. The phase difference of 180° (out of phase) indicates that the maximum wet conditions coincide with the mature stage of the La Niña event in the AR; opposite conditions are noticed during El Niño. Conversely, the PR index only shows significant coherence in the 1.5–3-year scale during the 1998–2002 period with a phase difference of 45° (Fig. 4b); in other words, negative ONI precedes negative rainfall anomalies in the PR, and vice versa, with a time lag of about 2–5 months.

Fig. 4
figure 4

The wavelet coherence and phase differences between a AR and ONI indices, and b PR and ONI indices. The closed contours represent the squared wavelet coherence. The region where the edge effects are significant is under the U-shaped curve (the cone of influence). The significant cone of influence (COI) is delimited by a thin line (C). The arrows indicate the phase differences as follows: in phase (0°), pointing out to the right; out of phase (180°), pointing out to the left; the first time series leading the second by 90°, pointing downwards; and the first time series lagging the second by 90°, pointing upwards

Full size image

The results above indicate that the ENSO has strong influences on the AR rainfall, and weak linkages with the PR rainfall. These findings are consistent with the previous results by Córdoba-Machado et al. (2015a), who reported inverse relationships of the ENSO and AR rainfall, with El Niño related to dryness and La Niña with wetness, and no significant effects on the rainfall in the western plains in Southwestern Colombia. They suggested that the complex topography alters the ENSO effects in Southwestern Colombia. Similar results have been reported in Ecuador, a neighboring country with the same topographic conditions (Campozano et al. 2016; Morán-Tejeda et al. 2016; Vicente-Serrano et al. 2017). These studies showed rainfall increase in the Andean region of Ecuador linked to negative Niño 3.4 SST index, and dry conditions in the west sector of the country were associated with negative Niño 1+2 SST index, whereas a reduction in rainfall in the Andean Region is associated with a positive Niño 3.4 SST index, and wet conditions in the western sector are linked to positive Niño 1+2 SST index.

Now, considering that the regional indices present a 3–6-year interannual variability, the strong influence of the ENSO on the AR, and its weak influence on PR rainfall variability, the next section evaluates the relationships of these indices with the oceanic and atmospheric variables at a 3–6-year scale.

3.3 Relationship Between Andean Rainfall and Regional Climate

Figure 5 shows the correlation maps between the AR index and the SST. The negative correlations between the AR index and the SST in the tropical Pacific along the western coast of the American continent, extending westward along the equator, with the largest values between 120° W, 150° W, 15° N, and 15° S, are surrounded by positive correlations in the western tropical and subtropical Pacific (Fig. 5a). Positive correlations in the equatorial Atlantic, surrounded by negative correlations centered at 25° N (tropical North Atlantic—TNA) and in the Southwestern Atlantic between 20° S–40° S were also observed. Thus, dry conditions in the AR are related explicitly to El Niño events, and wet conditions to La Niña events, with the largest SST anomalies in the central Pacific region. This result is consistent with many studies regarding the impact of the ENSO on Colombian rainfall, mainly in the Colombian Andes (Arias et al. 2015; Hoyos et al. 2013a, b; Navarro-Monterroza et al. 2019; Canchala et al. 2020b; Tootle et al. 2008; Poveda et al. 2011).

Fig. 5
figure 5

Maps of the a total and b partial correlations for the period 1983–2016 on the 3–6-year scale between the Andean Region (AR) index and the sea surface temperature (SST). The grey shaded interval is every 0.2. Dark (light) shaded areas encompass positive (negative) values. Absolute correlations of 0.32 are significant at a 95% confidence level using the Student t test for correlations. The continuous (dashed) line indicates significant positive (negative) correlations

Full size image

Moreover, Fig. 5 highlights the significant role of the SST gradient between the equatorial Pacific and Atlantic basins in rainfall patterns in northern South America through an east–west circulation in the equatorial latitudes (Nogués-Paegle et al. 2002; Wang 2006; Kayano et al. 2011). Under La Niña, this SST gradient connects the two basins through an anomalous Walker circulation with downward motions in the tropical Pacific and upward motions over the northern and northwestern South America and the equatorial Atlantic. Thus, the atmospheric circulation and the convergence of humidity in northwestern South America would intensify, increasing rainfall in the Colombian Andes (AR).

The correlation pattern for the VIMF divergence shows positive values confined to the eastern Pacific band along 5° N and 3° S (Fig. 6a), with the largest magnitudes near the Ecuadorian coast and Southwestern Colombia between 77° W and 90° W, surrounded by negative correlations to the north and south of this oceanic band and over the Caribbean Sea; on the continental zone, positive correlations are observed in the western plains of Ecuador and Southwestern Colombia, and negative ones in a southwest–northeast-oriented band, extending from Ecuador, most of the AR, and reaching northeastern Venezuela. Thus, these correlation patterns indicate that positive rainfall anomalies over the AR are related to increased moisture convergence there (negative correlation) as a result of enhanced southeastward moisture transport from the equatorial coasts and the Colombian Pacific Ocean. Firstly, moisture is transported from the eastern Pacific northeastward along the extreme northwestern South America from where the flow turns southeastward, causing moisture increase in a southwest–northeast-oriented band that extends from Ecuador to northeastern Venezuela. The involved low-level circulation has been associated with the Choco Jet core (5° N–80° W), which exerts a strong influence on the Colombian hydroclimatology, mostly in western and central parts of the territory (Poveda and Mesa 2000; Poveda et al. 2014; Arias et al. 2015; Yepes et al. 2019; Cerón et al. 2020a). At the same time, west–east moisture transport over the Caribbean region indicates a weaker Caribbean Low-Level Jet (Wang 2007; Wang et al. 2007), strengthening the moisture transport from the Pacific to this region (Fig. 6a).

Fig. 6
figure 6

Maps of a total correlation and b partial correlations for the 1983–2016 period on scales of 3–6 years, between the AR index versus vertically integrated moisture flux (VIMF) and its divergence. The display is the same as in Fig. 5. Arrows show the correlations of moisture transport anomalies with AR

Full size image

According to this circulation pattern, the largest positive rainfall anomalies would be over the AR, Caribbean, and the northern Pacific regions of Colombia, and near-normal or negative rainfall anomalies over the PR and Ecuador. This rainfall anomaly pattern is confirmed in the correlation map between CHIRPS and the AR index presented in Fig. 7a. A closer review of the correlation pattern between AR and CHIRPS in Southwestern Colombia shows significant positive correlations in the AR, and nonsignificant correlations in the PR (Fig. 7b).

Fig. 7
figure 7

Maps of the a, b total correlation and c, d partial correlation for the period 1983–2016 on the 3–6-year scale between the AR index and the CHIRPS precipitation (PRP) anomalies. The display is the same as in Fig. 5

Full size image

The partial correlation maps between the AR index and the SST, VIMF, and CHIRPS precipitation while removing the effect of the ONI were also calculated. The partial correlation map for the SST shows significant positive values in most of the TNA, with the highest values centered between the equator and 10° N, and a dipole in the eastern Pacific with significant positive correlations extending westward from 20° N to 5° S along the western coast of Central America and Colombia to 120° W, and negative values extending westward from 10° S to 30° S along the Chilean and Peruvian coasts to 130° W (Fig. 5b). Thus, wet conditions in the AR without ONI influence are related to the anomalous warming in most of the TNA, the Caribbean Sea, and the eastern Pacific near the Colombian coast and the cooling of the Chilean and Peruvian coasts. The partial correlation for the AR and VIMF divergence shows a divergent southeasterly flow in the southern hemisphere along the Peruvian coast that shifts to a southwesterly flow as it crosses the equator. It increases the moisture convergence over the eastern Pacific north of the equator, the Colombian and equatorial Andes, and the Caribbean Sea. Therefore, positive partial correlations for CHIRPS rainfall are observed over western and northern Colombia, with the highest values over the AR and Ecuador, which decrease towards the western plains (Fig. 7c, d).

A noteworthy feature is the strong influence of the tropical Pacific SST anomalies on the rainfall over the Andean side of Colombia and Ecuador, and the weak influence of these SST anomalies on the rainfall in the western plains (see Figs. 5a, 7a, b). Consistently, Vicente-Serrano et al. (2017) found a strong influence of SST anomalies in the central Pacific on the rainfall in the Andes mountain range of Ecuador, which is a region bordering the AR, contrasting with the weak influence of these SST anomalies in the rainfall in the western plains bordering the PR. Also, when the ONI influence is removed, the positive rainfall correlations are more extensive in the western plains (Fig. 7c, d), likely due to the TNA SST action. These results support that rainfall in the tropical Americas is partly modulated by the SST anomalies of the tropical Pacific and TNA, and possibly through the SST gradient between these two oceanic sectors (Giannini et al. 2001; Gouirand et al. 2012, 2014).

3.4 Relationship Between Pacific Rainfall and Regional Climate

Total and partial correlation maps between the PR index and SST presented in Fig. 8, in general, show similar patterns with greater and better characterized partial correlations, pointing out a weak influence of the central Pacific. Positive significant partial correlations and nonsignificant total correlations in the eastern Pacific extend westward from the Colombian and Central American coasts to 120° W, and negative ones in a nearly zonal band between 20° S and 30° S near the Chilean coast extend westward up to 125° W, forming a dipolar correlation pattern in the eastern Pacific with the asymmetry axis about 10° S (Fig. 8b). Also, the partial correlation patterns in the Pacific sector for the PR and AR are quite similar (Figs. 5b, 8b, respectively). Moreover, the total and partial correlations maps for the PR in the Atlantic show very similar features with significant positive correlations in the Caribbean Sea and North Atlantic west of 60° W and along the equatorial Atlantic, with the highest values near the African coast, and negative correlations in the Southwestern Atlantic (20° S–40° S) near the South American coast and the TNA, but not significant in this latter region.

Fig. 8
figure 8

Maps of the a total and b partial correlations for the period 1983–2016 on the 3–6-year scale between the Pacific Region (PR) index and the sea surface temperature (SST). The display is the same as in Fig. 5

Full size image

Consistently, the PR index and the divergence of VIMF are significantly and negatively related through the total and partial correlations along the western plains of Ecuador and Southwestern Colombia, and significantly and positive correlated in the northern coast of western Colombia (Fig. 9a, b). Accordingly, the moisture divergence decreases and the rainfall increases over most of the western plains, and the opposite occurs in the northern coast of western Colombia. In the case of the PR correlations, the winds in the eastern Pacific between 110° W and 80° W and in the 5° N–15° S band cause a strong moisture divergence to the south of 5° S and convergence to the north of this latitude, which intensify moisture transport to the western plains. This moisture transport is significant and more pronounced when the effect of ONI is eliminated (Fig. 9b). Thus, the total and partial correlations between PR and CHIRPS show similar patterns with significant positive values decreasing eastward from most of Southwestern Colombia and Ecuador into some parts of the Colombian Andes mountain range (Fig. 10a, b). Meanwhile, the partial correlations show higher values on the western plains and are more extensive on the Colombian Andes mountain range (Fig. 10c, d). The VIMF circulation pattern shown in Fig. 9b was also observed in the correlation map between the 925-hPa winds and the PR index (figure not shown).

Fig. 9
figure 9

Maps of a total correlation and b partial correlations for the 1983–2016 period on scales of 3–6 years, between the PR index versus VIMF and its divergence. The display is the same as in Fig. 5. Arrows show the correlations of moisture transport anomalies with the PR

Full size image
Fig. 10
figure 10

Maps of the a, b total correlation and c, d partial correlation for the period 1983–2016 on the 3–6-year scale between the PR index and the CHIRPS precipitation (PRP) anomalies. The display is the same as in Fig. 5

Full size image

In the context of the low-level wind anomaly patterns in the eastern Pacific, previous studies indicate that moderate El Niño events feature a strong SST anomaly asymmetry with cooling and anomalous southeasterlies to the south and warming and anomalous southwesterlies to the north of the equator, and in association, intensifies deep convection only north of the equator (Dewitte and Takahashi 2019; Peng et al. 2020). Besides, the authors suggest consistent relationships between the wind velocity and SST variations, such that to the south of the equator, the strengthening of the wind velocity coincides with decreases in the SST, while to the north of the equator, weakening of the wind velocity concurs with increases in the SST. Thus, the wind plays a crucial role in the SST asymmetry outside the equatorial region. The results here in terms of the VIMF and SST correlations with the PR index in the eastern Pacific shown in Figs. 7b and 8b have some correspondences with these previous findings, but with the asymmetry axes here at 10° S for the SST and at 5° S for the VIMF. Specifically, the anomalous southerlies drive anomalous westward Ekman currents south of the equator, transporting cold water out of the southeastern Pacific and decreasing the SST in that oceanic region, as observed in Dewitte and Takahashi (2019) and Peng et al. (2020). It is worth recalling that the results here refer to correlations without the ENSO influence (Fig. 8b).

4 Conclusions

Here, the rainfall variability in Southwestern Colombia was analyzed by distinguishing the Andean Region (AR) and Pacific Region (PR) and in the context of a complex influence of the ENSO. Both regions show a dominant scale of the rainfall variability at the interannual (3–6-year) time scale. Nevertheless, the wavelet coherence analysis shows that the rainfall variability in the AR is strongly connected to the ENSO, whereas the PR is weakly connected. These linkages for the 3–6-year time scale are such that the maximum rainfall anomalies occur up to 4–9 months in advance and simultaneously with the minimum SST indices. Meanwhile, minimum rainfall anomalies occur with the same time leads but with maximum SST indices.

In contrast, significant coherences occurs in a short period (1998–2002) on the 1.5–3-year time scale, with negative ONI preceding negative rainfall anomalies in the PR and vice versa, by approximately 2–5 months. These differential relations between ONI and AR and PR rainfall become more evident in the correlation analyses. Indeed, the SST anomalies in the central and eastern Pacific regions have a strong impact on the AR rainfall, and a weak impact on the PR rainfall.

The total correlation analyses for the AR indicate rainfall increases in the center, west, and north of the Colombian territory, except in the western plains (PR) during La Niña events, and opposite effects occur during El Niño events. The partial correlations show that a north–south anomalous SST dipole in the eastern Pacific nearly symmetric around 10° S with a warming in the northern sector and a cooling in the southern together with a warming in the TNA and the Caribbean Sea cause rainfall increase in most of the Colombian Andean Region, and to a lesser extent, on the western plains. For the PR, the total and partial correlation maps of all variables show quite similar features, but with larger magnitudes and more extensive patterns for the partial correlations. Anomalous SST and low-level winds in the eastern Pacific register a substantial influence over the PR rainfall such that an SST anomaly dipole with asymmetry axis at 10° S and positive anomalies to the north and negative anomalies to the south, in association with an anomalous VIMF dipole with asymmetry axis at 5° S with southeasterlies to the south and southwesterlies to the north, acts to enhance rainfall in the PR. The most outstanding result is that the positive correlations for the PR cover the western plains of Colombia and Ecuador and some parts of the Colombian Andes mountain range, with higher partial correlations on the western plains, which are also more extensive in the Colombian Andes mountain range. The results here might be useful for future modeling studies of the Southwestern Colombian climate variations.