1 Introduction

The streamflow is a relevant part of the hydrologic cycle since it relates to the amount of discharge in natural streams, allows for the transportation of mass and energy through watersheds, and maintains the mass balance of water in the natural system (Rice et al. 2015; Tamaddun et al. 2016). However, human-induced global warming is modifying the water cycle generating hydroclimatic variations that change the frequency and magnitude of hydrologic variables influencing the natural dynamics of climate variability. Climate variability is a natural geophysical phenomenon that modulates rainfall patterns (Mkuhlani et al. 2020); however, due to anthropogenic forcing, frequency and magnitude of extreme weather events, such as droughts and intense rainfall, have increased, significantly influencing the freshwater systems and their management (Oñate-Valdivieso et al. 2020; Kundzewicz et al. 2019) and affecting the capacity of the hydrographic basin to sustain the multifunctional water system that produces environmental services for society and ecosystem conservation (Sánchez Torres et al. 2022).

According to Liu and Alexander (2007), the hydroclimatic variability in a region can be linked with the ocean-atmospheric patterns that occur at great distances through teleconnections. That allows the atmosphere to act as a bridge between different parts of the ocean and enables the ocean to act like a tunnel linking different atmospheric regions. The atmospheric bridges connect two or more sites far away by atmospheric processes such as the Walker and Hadley circulation that can link the Pacific Ocean with other ocean basins (Wang 2019) due to the wind and ocean circulation, influencing the sea surface temperature (SST) variations in different oceans. For example, the anomalous warming or cooling of the Pacific and Atlantic oceans can alter the Walker circulation and create an inter-basin SST gradient variability that produces surface zonal wind anomalies over equatorial South America and some regions (Wang 2006, 2019; Cerón et al. 2021a).

The most significant teleconnection patterns, with hydroclimatological influence in South America (SA), are linked mainly with the SST in the Pacific and Atlantic Oceans (Zhang et al. 2022). However, the presence in SA of the most extended mountain range in the world, Los Andes, influences the hydroclimatology due to its long latitudinal extension and prominent topography that crosses Venezuela, Colombia, Ecuador, Peru, Bolivia, Chile, and Argentina. According to Arias et al. (2021), the hidroclimatology in the Andes is influenced by climatic oscillations such as El Niño-Southern Oscillation (ENSO), Maden Julian Oscillation (MJO), Atlantic SST variability, Pacific Decadal Oscillation (PDO), and the Atlantic Multidecadal Oscillation (AMO) and is also modulated by large-scale phenomena such as tropical easterly waves, atmospheric rivers, and low-level jets. However, ENSO is the dominant mode of interannual variability that may induce different responses in the Andean hydroclimate and the tropical North Atlantic influences, mainly in the northern Andes and the southern tropical Andes. ENSO can be classified as Canonical ENSO or Modoki ENSO, where the El Niño (La Niña) Canonical – ENC (LNC) register positive (negative) SST anomalies in the east of the Tropical Pacific Ocean, whereas that El Niño Modoki (ENM) is characterized by positive SST anomalies in the Central Pacific Ocean and bounded by negative SST anomalies in the eastern and western Tropical Pacific Ocean, the opposite behavior corresponds to the La Niña Modoki (LNM) (Ashok et al. 2007).

In particular, Colombia registers negative (positive) anomalies in rainfall and streamflows during the El Niño-EN (La Niña-LN) phase of ENSO. Their duration and magnitude vary across the main geographic regions of the country (Andes, Pacific, Caribbean, Orinoco, and Amazon) (Arias et al. 2015; Poveda and Mesa 1997; Waylen and Poveda 2002; Salas et al. 2020). According to Salas et al. (2020), during the LN phase, Colombia registers wetter than neutral periods in the Pacific, Andean, and Caribbean regions; and the opposite conditions occur during the EN phase. Meanwhile a warmer Tropical North Atlantic (TNA) is linked with increased precipitation in Colombia and a weaker Caribbean low-level jet (CLLJ), allowing atmospheric water vapor transport toward northern South America (Arias et al. 2015, 2020). On the other hand, low-frequency phenomena can also affect the hydroclimatology of the country. For instance, the cold (warm) phase of AMO strengthens EN (LN) events due to the presence of a negative inter-Pacific-Atlantic SSTs (Kayano and Capistrano 2014), and the cold (warm) phase of PDO favors (inhibits) the negative SST anomalies in the tropical Pacific (Kayano et al. 2019).

In southwestern Colombia (SC), the hydrological basins register a significant spatio-temporal variability of their hydroclimatology due to natural geographic split and the influence of the aforementioned phenomena. According to Canchala et al. (2022) the SC has two homogenous hydroclimatic regions: Andean Region (AR) and Pacific Region (PR). In AR, the rainfall is partly modulated by a west–east SST gradient between the eastern tropical Pacific and the TNA. In PR, the rainfall is modulated by the SST anomalies in the eastern Pacific mainly (Cerón et al. 2021a). In this region is located the Patía River Basin (PRB), in which, according to Canchala et al. (2020a, b), the SST in the central and western tropical Pacific Ocean influences the variability of monthly flows, which decreases (increases) when EN (LN) events are recorded approximately 0–5 (5–9) months before. Although the variability of monthly flows is known to be influenced by ENSO, there are knowledge gaps related to the seasonal streamflow variations, their relationship with the ocean-atmospheric patterns that occur in the Pacific and Atlantic oceans, and the temporal evolution of these patterns. Therefore, we seek to improve the predictability of seasonal flows and provide useful information for planning water resources in the study area.

In this sense, the main objective of this research is to study the Pacific and Atlantic sea surface temperature influences on seasonal streamflow variability in the PRB through the non-linear principal component analysis and quantify the relationships with ocean-atmospheric conditions using correlation and composite analysis. The study area and data are presented in Sect. 2, methodology in Sect. 3, results and discussion in Sect. 4, and finally, conclusions in Sect. 5.

2 Study area and data

2.1 Study area

The PRB is located in the SC, between 0° and 3° N and between 76° and 79° W; it rises in the Colombian massif on the western slopes of the Central Mountains range of the Colombian Andes, with a drainage area of 24,400 km2, an average annual flow of 1453 m3.s–1, and an altitude range of 0 – 5368 ma.s.l (Velásquez-Restrepo and Poveda 2019; Canchala et al. 2020a, b). Figure 1a shows the watershed that extends over the Andean region and the Pacific region of Nariño, and the location of the main streamflow gauge stations, highlighting that the Los Nortes (NOR) gauge station is the most downstream flow measurement station of the PRB which records up to this point a mean annual discharge 367.9 m3.s–1 (See Table 1).

Fig. 1
figure 1

Study Area. a) The geographic location of the Patía River Basin shows the streamflow stations (blue circles) and b) Monthly variability of streamflow of Los Nortes (NOR) gauge station over the 1984–2018 period 

Full size image
Table 1 Descriptive statistical analysis of the monthly streamflow data in the Patía river basin (1984–2018) 
Full size table

The PRB is a tropical river basin, which is strongly influenced by the Intertropical Convergence Zone (ITCZ) due to the precipitation pattern response to its seasonal movement (Syvitski et al. 2014). This Basin register an average annual precipitation of 1448 mm.year−1 and a precipitation bimodal regime with two dry periods (December-January–February – DJF and June-July–August – JJA) and two wet periods (March–April-May – MAM and September–October-November – SON), consistent with the precipitation regime found in the Andean Region of Southwestern Colombia (Canchala et al. 2022). As a consequence, a bimodal regime characterizes the annual streamflow cycle in the PRB with low discharges in January–February-March (JFM) and July–August-September (JAS) and high discharges in April–May-June (AMJ) and October–November-December (OND). Cross-correlations performed between the monthly precipitation and monthly streamflow in the PRB depicted significative correlations up to a delay of three months (Lag-0: 0.69; Lag-1: 0.71; Lag-2:0.51 and Lag-3:0.22) where the precipitation precede the streamflow. Figure 1b shows the bimodality described above in the NOR gauge station, with the lowest and highest mean monthly streamflow during JAS (201.41 m3.s–1) and OND (434.38 m3.s–1), respectively. The bimodal regime in the study area is due to the combination of the ITCZ, the atmospheric moisture advection by the Choco low-level jet (CJ) mainly, easterly waves, and land surface-atmospheric interactions (Urrea et al. 2019; Yepes et al. 2019; Arias et al. 2021; Canchala et al. 2022). The annual streamflow cycle in the PRB indicates the lowest (highest) discharge in the season JAS (OND); therefore, these seasonalities were selected for this study.

2.2 Data description

2.2.1 Streamflow Dataset

The monthly streamflow dataset between 1984 and 2018 of five flow-gauge stations of the PRB was obtained from the Instituto de Hidrología, Meteorología y Estudios Ambientales (IDEAM) of Colombia. Guachicono (GCH), Puente Juanambu (PTJ), Agroyaco (AGR), Puente Nuevo Patía (PNP), and Los Nortes (NOR) were chosen by the quality of the data, time series extension (35 years of data records), and representativeness of the domain of the study. Table 1 depicts each flow gauge station's main features and statistical details.

From Table 1, it is relevant to highlight that PNP (AGR) is a gauge station with a drainage area of 5,065 (3,992) km2 located in the northern (southern) zone of the upper basin of the Patía River with minimum streamflow records of 10.4 (21.2) m3.s−1. Although the drainage area of PNP is more extensive, this area registers a more substantial influence of the ENSO phenomenon than the south zone; therefore, when an El Niño phenomenon is recorded, rainfall and flows are strongly reduced and even more so when the El Niño phenomenon coincides with the season lowest rainfall which is recorded in July, August and September.

2.2.2 Climate indices

In this study, we selected thirteen climate indexes linked with the Pacific and the Atlantic oceans, which were: anomalies of SST in the regions 1 + 2, 3, and 4 (Trenberth 1997, 2001), Oceanic Niño Index (ONI) (Yu and Kim 2013), Multivariate Enso (MEI) (Wolter and Timlin 2011), Southern Oscillation Index (SOI) (Trenberth and Shea 1987), Trans Niño Index (TNI) (Trenberth and Stepaniak 2001), Niño Modoki Index (EMI) (Ashok et al. 2007), Pacific Decadal Oscillation (PDO) (Mantua et al. 1997), Caribbean SST Index (CAR), North Tropical Atlantic SST Index (NTA) and Tropical Northern Atlantic Index (TNA) (Enfield et al. 1999), and South Atlantic Tropical Index (SAT) (Chang et al. 1997). It is worth mentioning that SST3.4 Index wasn’t considered to avoid redundancy, due to that the ONI index is estimated considering the temperature of the Niño 3 + 4 region. The indices described are available online in the National Oceanic and Atmospheric Administration (NOAA) website (https://www.esrl.noaa.gov/psd/data/climateindices/list/).

2.2.3 Sea surface temperature information

Gridded (2° × 2°) monthly SST data were obtained from NOAA; the Extended Reconstructed SST V5 (ERSST) data set (NOAA/ERSST) (Huang et al. 2017) is available online at https://www.esrl.noaa.gov/psd/data/gridded/. That variable was obtained from 1984 to 2018 in the following domains: 60°N–60°S and global.

2.2.4 Climatological precipitation data

In the study area there are in situ precipitation measurement stations, however these are scarce and are not uniformly distributed. Therefore, the historical precipitation data used in this study were obtained from the ERA5 reanalysis dataset, produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) (Hersbach et al. 2020) and available online by the Copernicus Climate Change Service (C3S) at https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5. ERA5 data cover both land and sea areas at a horizontal resolution of 0.25° × 0.25°, providing high-quality climatological information on hourly or monthly average basis. This variable was obtained for the 1984–2018 period in the following domains: 20°N–10°S and 60°W–90°W. Its hourly output and the high horizontal resolution of 31 km provide significantly finer details of oceanic and atmospheric phenomena relative to previous global reanalysis (Hoffmann et al. 2019; Hersbach et al. 2020).

ERA5 reanalysis dataset was chosen based on previous studies that indicated its better performance in the region like the one carried out by Avila-Diaz et al. (2023), who evaluated the performance of a subset of HighResMIP models, members of CMIP6, in the simulation of daily extreme weather events of temperature and precipitation over Latin America and the Caribbean region during 1981–2014. For this purpose, he compared three gridded datasets (ERA5, CHIRPS and GMFD). Its results indicated that, in general, the ERA5 dataset presents better results compared to GMFD and CHIRPS. Furthermore, recent studies have used this reanalysis for extreme event trends associated with climate variability and climate change (Cerón et al. 2021c; Yuan et al. 2021; Avila-Diaz et al. 2021), surface wind speed, variability in the South American region (Aguirre et al. 2021; Gil Ruiz et al. 2021), and land–atmosphere interactions (Baker et al. 2021).

3 Methodology

3.1 Non-linear principal component analysis

The Non-linear Principal Component Analysis (NLPCA) is a non-linear statistical method that generalizes the principal components by capturing non-linear forms and extracting nonlinear components with the least loss of information (Hsieh 2001). NLPCA uses the multi-layer perceptron of an auto-associative topology, known as a bottleneck, and the output of this method is a low-dimensional curvilinear coordinate system that tracks the relevant features of the data (Scholz 2012).

In recent years, NLPCA has been widely employed to study hydrology, climatology, meteorology, and oceanography (Hsieh and Tang 1998; Kenfack et al. 2017; Miró et al. 2017; Canchala et al. 2020a, b; Canchala et al. 2022). Its advantage with regard to the Principal Component Analysis (PCA) is the maximization of the explained variance (Hsieh 2001). That is particularly useful in the knowledge areas where non-linear processes are involved. In this study, the objective of NLPCA is to extract the non-linear components that show the main modes of variability of seasonal streamflow of PRB following the neural network model indicated in Fig. 2. To obtain the hierarchically ordered features, we use the Non-linear PCA toolbox available online at http://www.nlpca.org/matlab.html. More details about this method are available in Scholz (2002) and Hsieh (2001).

Fig. 2
figure 2

Diagram of Non-Linear Principal Component Analysis (NLPCA) with a [n-4–3-m-3–4-n] network topology where n and m are dimensions of x and NLPCS, respectively. The diagram depicts five hidden layers of neurons inserted between the input layer \({x}_{n}\) and the output layer \({\overline{x} }_{n}\). \(h(x)\) is the encoding layer, \({y}_{m}\) is the bottleneck layer and \(h(y)\) is the decoding layer

Full size image

Thus, we performed several experiments to choose the architecture with the highest percentage of explained variance. It is highlighted that, before applying NLPCA, we turned the flow values into anomalies by subtracting each month's mean, which allows for a better understanding of the magnitude of the variable over time.

3.2 Composite analysis

Composite analysis, often known as epoch analysis, is commonly employed to identify connections between climate phenomena and potential causative elements, requiring two steps: i) form sets of events and ii) calculate the composite mean values or calculate the composite anomalies (Li and Dolman 2023). Initially, we calculated the percentile of the seasonal NLPCs between 1984 and 2018 and selected events above (below) of the 75th (25th) quantile (set of events) which make up the positive (negative) events. Subsequently, monthly standardized anomalies of SST and precipitation for the same events were obtained for later use in the composite analyses. The estimation of the SST and precipitation anomalies was performed by subtracting the mean and dividing by the standard deviation using the same resolution and area with which the data were obtained. Finally, we got the seasonal composites of the precipitation and SST anomalies for JAS and OND; the composite refers to the arithmetic mean of the 3-month mean anomalies of the selected events. The composite analysis has been considered a relevant indicator of ocean-atmospheric conditions during specific climate states, which provides information on the connections pathways (Boschat et al. 2016; Cerón et al. 2021a, b, c). Furthermore, lagged composite analysis was performed between SST anomalies, precipitation anomalies, and streamflow when SST and precipitation are leading streamflow between 1 and 4 seasons. The acronym (JFM, AMJ, JAS y OND) indicates the season for lagged SST and precipitation. In the map, the term (–1) means that the seasonal SST and precipitation correspond to the preceding year of the seasonal streamflow. The statistical significance of the composites was evaluated using the Student’s t-test at a 95% confidence level (Wilks 2011), and the degrees of freedom were the number of negative and positive events for each period.

4 Results and discussion

4.1 Non-linear principal component analysis

The main modes of streamflow variability for PRB were estimated for the seasons JAS and OND using PCA and NLPCA, however for JAS (OND), NLPCA lead to more variance being explained by its first mode 77,31% (89.15%), compared to PCA results where the first mode explained 47,57% (88.31%), in this sense we decided to use NLPCA. The results indicated one significant NLPC for each season, where the explained variance for JAS and OND was 77.31%, and 89.15%, respectively. We assessed several topologies (See Table 2) to define the best (the highest explained variance). We found a common [5–5-5–1-5–5-5] network topology for both seasons. The components followed a hierarchical order, with a hyperparameter (weight decay coefficient) of 0.01 and 5000 iterations. The weight decay coefficient is a gradual reduction in learning rate over time to accelerate learning which was carried out iteratively.

Table 2 Explained variance of the NLPCs in each season. The best Artificial Neural Network (ANN) topologies are in bold
Full size table

Figure 3 shows the time series of the NLPCs for each season and the ONI index. The main negative (positive) values of the NLPCs in the season OND matching in the years 1987, 1991–1992, 1998, 2002–2003, 2015–2016 (1984, 1999–2000, 2008, and 2011), which are linked with the EN (LN) events of the ENSO phenomenon as observed through the ONI index, which is a measure of the ENSO: values of + 0.5 or higher indicate EN, and values of –0.5 or lower indicate LN. The relationship between the streamflows during OND with the ONI index is confirmed by Pearson correlations that show negative correlations of –0.57 (Fig. 3b). It is relevant to highlight that EN events that occurred during 1991–1992, 1997–1998, and 2015–2016 left significant socio-economic impacts on Colombia. The first event led Colombia to register an energy crisis, the second notably affected the country's main crops, and the third was the strongest on record, causing a rainfall deficit in the Andean region with adverse effects on the agriculture and livestock in Southwestern Colombia (Poveda et al. 2001 and Martínez et al. 2017).

Fig. 3
figure 3

Nonlinear principal components for seasons JAS and OND (1984–2018) for the PRB and ONI index (red line). a) NLPC – JAS and b) NLPC – OND

Full size image

Furthermore, the positive values of NLPCs coincide with LN events triggered in Colombia emergencies linked with slow floods, flash floods, and landslides, resulting in human and material losses (Hoyos et al. 2013; Arias et al. 2015). According to IDEAM (2019), during LN 2010–2011, the main rivers in Colombia, which cross the country from south to north, maintained critical levels exceeding overflow levels for numerous towns located along the riverside zone, which affected infrastructure and caused many victims due to destruction. In addition, the heavy rains caused numerous sudden floods in mountain rivers and tributaries to the upper basins of the main rivers, located in the southwestern and central areas of the country. This event affected 11.3% of the country's population leaving 5.2 million affected people in 1,027 municipalities (87%), including the Capital District, and the economic losses exceeded 4,870 million dollars (Sedano-Cruz et al. 2013). On the other hand, Fig. 3a shows the time series of the NLPC of JAS, which does not indicate a significant synchronous relationship with the ONI index (r = –0.14), showing that the ENSO phenomenon does not instantly influence the flows in this season.

4.2 Links between main modes of variability of the seasonal streamflow and climate indices

Once the main modes of variability of each season have been established, the degree of the relationships between NLPCs of each seasonal flow and the climate indices were estimated using Pearson’s correlation. Figure 4 shows the cross-correlations analysis between NLPCs of the JAS and OND and the climate indices influencing each season with a lag of up to 12 months. The cross-correlation for NLPC-JAS depicted statistically significant negative correlations with CAR, NTA, and TNA, where the influence of CAR remains until lag 9. In addition, an inverse influence is observed with SST4 and ONI that persists to lag 12 and 6, respectively. These results establish that the variability of flows in JAS is influenced mainly by ocean-atmospheric phenomena in the Caribbean Sea and the Atlantic Ocean. Otherwise, the cross-correlations for NLPC-OND showed significant negative (positive) correlations (ordered from greatest to least) with SST4, MEI, ONI, SST3, EMI, PDO, CAR, and SST1 + 2 (SOI and TNI) indicating inverse (direct) relation with ocean-atmospheric indices linked to ENSO phenomenon mainly. Furthermore, it is observed that the highest correlations are recorded synchronously (i.e., lag 0) and that the persistence of the relationship is statistically significant for up six months (lag 6) in most indices.

Fig. 4
figure 4

Cross-correlograms between climate indices and NLPCs of JAS, and OND. Significant correlation at the 0.05 level is shown in bold (|r|> 0.331)

Full size image

Notably, the streamflows during JAS are related to the SST over the Caribbean region and TNA. Usually, a warmer TNA and CAR are related to the increase in precipitation in northern South America (Arias et al. 2015, 2020) and most of the Colombian Andean Region (Cerón et al. 2021b). Although this is the behavior in most of the year, the results show that the streamflow variability in JAS is not directly linked to a Pacific forcing but to an inverse relationship between the SST in CAR and TNA. This inverse relationship during JAS (season of streamflow reduction) is due to the occurrence of two atmospheric physical mechanisms that coincide in this season: i) the movement of the ITCZ, which during the northern hemisphere summer (JJA) is located to the north of the equatorial line, bringing the moisture towards latitudes greater than 8°N, therefore decreasing rainfall and streamflows in the PRB (Guzmán et al. 2014; Estupiñan 2016) ii) during JJA, the easterly trade wind continues to flow westward to the Caribbean Sea, forming the CLLJ, which intensifies with wind speed larger than 13 m/s, that in turn influences the cooling of the sea surface and results in SST anomalies in the tropical Pacific Ocean (Wang 2007).

Likewise, the results indicate that the streamflow variability during OND is mainly influenced by indices linked to the ENSO phenomenon such as ONI, SOI, MEI, and SST in regions Niño1 + 2, Niño3 and Niño 4, consistent with Poveda et al. (2002); Rojo-Hernández and Carvajal-Serna (2010); Tootle et al. (2008); Cerón et al. (2020); Córdoba-Machado et al. (2015); Cai et al. (2020), among others, who have found that ENSO phenomenon influence the variability of flows and rainfall in Colombia. Particularly, in this research it is highlighted the significative correlations found during OND, when the ITCZ moves towards the south of the equatorial line, increasing the precipitations and streamflows in the PRB (Guzmán et al. 2014; Estupiñan 2016). In addition, during OND the CJ carries moisture toward the continent due to its strengthening with speeds that fluctuate between 5 and 7 m.s−1 (Serna et al. 2018; Cerón et al. 2020), which increases the streamflows in PRB.

The results shown in Fig. 4, also indicate that the relationship between ENSO indices and PRB streamflows during OND is inverse: warm (cold) ENSO events are associated with the decreased (increased) streamflow. Those results are consistent with Poveda et al. (2011), Bedoya-Soto et al. (2018), Canchala et al. (2020a, b), and Cerón et al. (2021b) who have reported impacts of ENSO on the Colombian Andes highlighting that El Niño (La Niña) are related with the decrease (increase) of precipitation, runoff, and soil moisture.

The lag of correlations might be due to the linkages of the rainfall, temperature, evapotranspiration, soil moisture, and runoff in the tropical hydrographic basins (Poveda and Mesa 1997; Poveda et al. 2001). For his part, Salas et al. (2020) reported that the positive (negative) hydrological anomalies (precipitation and streamflow) over the Pacific and the Andean regions of Colombia exhibit generalized synchronization with the ENSO signal during LN (EN), mainly in the Central Pacific (Niño 3.4 and Niño 4 regions). Furthermore, the results showed an inverse relationship between seasonal streamflow and the EMI index that captures the SST zonal gradient in the eastern and western tropical Pacific. This relationship also was evidenced by Córdoba-Machado et al. (2015) in gauge stations located in the department of Nariño, which were correlated with EN (LN) Modoki linked with the decrease (increase) of the rainfall and, therefore of the streamflows. On the other hand, the correlations between streamflow and TNI index (SST gradient between Niño 1 + 2 and Niño 4) indicated that a positive (negative) gradient influences the increase (decrease) of the streamflow in PRB. Regarding this, Tedeschi et al. (2013) indicate that the diversity of ENSOs such as Canonical, and Modoki, among others also implies a wide range of atmospheric teleconnections worldwide.

4.3 Composite analysis

Table 3 records the events of the NLPCs above (below) of the 75th (25th) quantile. From now on, we refer as positive (negative) events for the streamflows > 75th (< 25th) in the seasons JAS and OND.

Table 3 Years with positive and negative streamflow events in JAS and OND
Full size table

Figure 5a–d (Fig. 5e–h) shows the evolving SST anomaly patterns preceding the positive (negative) streamflow events registered in JAS. Consistent with the positive events, i.e., increases in flows above the 75th quantile in JAS, negative anomalies in the SST were noted in the central and eastern equatorial Pacific and in the TNA band between 0° and 30°N, including the Caribbean Sea and the Gulf of Mexico. In contrast, during negative events (flows under 25th quantile in JAS) (Fig. 5e–h) significant positive anomalies were observed in the Caribbean Sea, the Gulf of Mexico, and the North Atlantic south of 50° N from OND (–1) to JAS. However, the area and magnitudes of these anomalies were broader and higher in JAS than in the previous seasons. Furthermore, it is highlighted that no significant SST positive anomalies were recorded in the tropical Pacific Ocean during negative streamflow events.

Fig. 5
figure 5

Three-month composites of the SST anomalies during streamflow positive and negative events in JAS. The symbol σ refers to SST anomalies

Full size image

This result is also consistent with Pearson's correlation (See Fig. 4). The streamflow variability in PRB is complex and not uniform as it depends on the season analyzed, the results indicate that when positive flow events were recorded in JAS, the SST anomalies were cold in the central and eastern equatorial Pacific and the TNA band. In contrast, when negative flow events were recorded in JAS the SST anomalies were warm in the Caribbean Sea, the Gulf of Mexico, and the North Atlantic.

Likewise, the sequence of precipitation anomaly composites that results in positive streamflow events in the PRB during JAS shows positive precipitation anomalies mainly in JFM over the Andean and Caribbean regions of Colombia (Fig. 6a–d). On his part, when negative streamflow events are registered during JAS, negative precipitation anomalies occur in greater proportion in JAS mainly in the Amazon, Orinoco, and Andean regions of Colombia (Fig. 6e–h), where the rainfall decreased from JFM and negative anomalies intensified in JAS.

Fig. 6
figure 6

Three-month composites of the PRP anomalies of the positive (a–d) and negative (e–h) streamflow events in the Patía River Basin during JAS. Shading interval: 0.2 SD. Continuous contours indicate significant values at the 95% confidence level using the Student’s t-test. The symbol σ refers to precipitation anomalies

Full size image

Figure 7a–d (Fig. 7e–h) shows the evolving SST anomaly patterns preceding the streamflow positive (negatives) events registered in the PRB during OND. The negative anomalies of the SST were reported in the tropical Pacific along the western coast of the American continent, extending westward along the equator with the highest anomalies between 160°E, 100°W, 20°N, and 20°S, and were surrounded by anomalous warming in the western tropical and subtropical Pacific (Fig. 7d). Therefore, the increase in flows (positive events) during OND is linked mainly to LN events, with the largest negative SST anomalies in the central Pacific region. Instead, the decrease in flows (negative events) in OND is linked with the warming of the central pacific (EN event) (Fig. 7h).

Fig. 7
figure 7

Three-month composites of the SST anomalies during streamflow positive and negative events in OND. The symbol σ refers to SST anomalies

Full size image

Figure 7d (7 h) shows that the negative (positive) SST anomalies in the tropical Pacific are maximum in magnitude and extension in OND, a quarter in which the positive (negative) precipitation anomalies are maximum in magnitude over the PRB and cover the Andean, Pacific, and Caribbean regions (See Fig. 8d and 8h). Therefore, when SST anomalies in the tropical Pacific are positive (negative), precipitation anomalies over Colombia are negative (positive) (see Fig. 8).

Fig. 8
figure 8

Three-month composites of the PRP anomalies of the positive (a–d) and negative (e–h) streamflow events in the Patía River Basin during OND. Shading interval: 0.2 SD. The symbol σ refers to precipitation anomalies

Full size image

These results are consistent with Grimm and Tedeschi, (2009); Kayano et al. (2013); Cai et al. (2020), and De Souza et al. (2021). They indicate that ENSO is the main phenomenon that modulates the interannual precipitation in SA, where EN (LN) influence the decrease (increase) in precipitation, runoff and soil moisture in north SA (Poveda et al. 2011; Bedoya-Soto et al. 2018). Furthermore, the same relationship between ENSO and the Colombian Andes was identified by Poveda et al. (2011) and Cai et al. (2020). They established that during LN (EN), colder (hotter) SST are related to surface pressure gradients between the coast of Peru and Ecuador and the Colombian Pacific that induces a stronger (weaker) atmospheric moisture transport to the region (Hoyos et al. 2018; Morales et al. 2021).

The results also coincide with Rasmusson and Carpenter (1982) and Larkin and Harrison (2002), who found that SST anomalies in the tropical Pacific linked with ENSO are maximum during late spring-early austral summer due to the ENSO generating changes in the upper-level air divergence and Walker circulation in the tropics. Therefore, the anomalous cells of Walker and Hadley circulation strongly influence the variability of precipitation in large areas of SA (Grimm and Silva Dias 1995).

5 Conclusions

In this study, we evaluate the Pacific and Atlantic sea surface temperature influence on seasonal streamflow variability in the PRB using the NLPCA, Pearson's correlation, and composite analysis. The main findings are summarized as follows:

  1. 1.

    The streamflow of the PRB registers a bimodal regime with low (high) discharges in JAS (OND), where the NLPCA showed only one main mode of variability for each season with an explained variance of 77.31% (89.15%), respectively. For the OND season, the streamflow variability is highly associated with the ENSO phenomenon, while that for JAS season, the streamflow association with ENSO is not as high as with CAR, NTA and TNA.

  2. 2.

    The cross-correlation analysis between the main modes of variability of OND (JAS) and climate indices depicted significant correlations mainly with ONI, SST3, SST4, MEI, EMI, PDO, SOI, and TNI (CAR, TNA, and NTA), climate indices linked with the ocean-atmospheric process that occurs in the Pacific (Atlantic) ocean. In general, the maximum correlations are registered until a lag of three months, and the influence over the streamflow variability by teleconnections remains until a lag of six months in its great majority.

  3. 3.

    The composite analysis for the seasons with minimum (maximum) discharges corresponding to JAS (OND) spatially confirmed their greater relationship with the SST in the Atlantic (Pacific) Ocean. The positive (negative) streamflow events during JAS that indicate the increase (decrease) of streamflow in PRB are related to negative (positive) anomalies in the TNA band, including the Caribbean Sea and the Gulf of Mexico. When the positive and negative events during JAS occur, precipitation increases and decreases, respectively, more intensely in the Colombia's Amazon, Orinoco, and Andean regions. Likewise, the positive (negative) events during OND are related to negative (positive) anomalies of SST corresponding to LN (EN) events. When the LN (EN) occur, the negative (positive) precipitation anomalies intensify over the Colombia's Andean, Pacific, and Caribbean regions.

The impacts of Tropical Pacific and Atlantic SST on streamflow in PRB indicate the importance of studying their influence on the seasonal scale, mainly in the most critical seasons such as OND and JAS. Those impacts could be addressed for prevention and risk management purposes, as well as to provide adequate planning and management of water resources in the study area. Furthermore, this study provides an understanding of the Pacific and Atlantic SST influence on seasonal streamflow variability in the PRB, which is a helpful information as an initial step to build forecasting models to estimate the seasonal streamflow, using climate indices as exogenous variables.