Introduction

The El Niño Southern Oscillation1 (ENSO) is a periodic sea surface temperature (SST) fluctuation across the equatorial Pacific Ocean. It is classified into a warm phase (positive or El Niño), a cold phase (negative or La Niña), and a neutral phase. The associated variations in SST have the potential to impact South American precipitation at regional scales: during the warm phase, the atmosphere above the sea is heated, allowing moisture-rich air to rise, be transported by synoptic-scale teleconnections and develop into rainfall over continental land masses; whereas during the cold phase, less water is transferred to the atmosphere, reducing precipitable water in the atmosphere and consequent transport to downwind regions. This large-scale oceanic sensible and latent heat exchange also influences circulation patterns, steering moisture to different target regions depending on the phase of ENSO.

Although ENSO is a large-scale climate phenomenon affecting all continents2,3,4,5,6,7,8,9, its influence in South America (SA) is notable. The ship captain D. Camillo N. Carrillo described in 1892 an ocean countercurrent on the Equator-Peru-Chile coast flowing from north to south, and whose existence has been known by other captains and local sailors for many years10. He noted that the waters of this current are likely warmer, given its direction and that would be the cause of abundant vegetation in the Patagonia: [Pudiera inclinarnos a suponer que la corriente antártica deja un gran seno entre ella y las costas occidentales de Patagonia, donde, acaso, las aguas cálidas de la contra-corriente del Niño, entibia el frío ambiente de aquellas regiones, favoreciendo su abundante vegetación.] [We could think that the Antarctic current leaves a large hollow between itself and the western coasts of Patagonia, where perhaps the warm waters of the El Niño countercurrent warm the cold environment of these regions, enhancing their abundant vegetation.]10. Today, we would further argue that this vegetation response likely also results from rainfall induced by the Pacific warmer waters of El Niño. ENSO is popularly described as a driver of heavy rainfalls during El Niño on the west coast of Ecuador, Peru, and Southeastern South America (SESA) and a driver of droughts during La Niña, while in Colombia and the Amazon droughts could occur during El Niño11,12,13,14,15,16. In Ecuador from 1970 to 2014, 50% of the registered meteorological droughts developed in a neutral ENSO period, 25% developed in El Niño periods, and 25% of drought events occurred during La Niña accompanied by a negative Pacific Decadal Oscillation (PDO)4. The frequency changes in extreme rainfall in SA produced by El Niño and La Niña episodes during neutral years seem roughly symmetric15. Notwithstanding such a general climatological understanding, however, other recent studies suggest that ENSO’s influence is not entirely understood11.

ENSO is also altered by the effects of different climate variability modes, like the PDO17, Madden-Julian Oscillation (MJO)18, the Atlantic Multidecadal Oscillation (AMO)19 and the interactions between different ocean basins, particularly the Pacific and Atlantic11,17,20. El Niño also varies according to its anomaly center, which ranges from the Equatorial Eastern Pacific (EP or canonical) to the Equatorial Central Pacific (CP or Modoki). CP and EP events can lead to different or even opposite effects due to the differences in how they connect through both tropical and extratropical pathways. Along the western coastal regions, flood conditions may not arise during a CP El Niño event; conditions may be drier than usual11,21.

The processes that lead to floods and droughts depend not only on concurrent climate and meteorological events but also on basin characteristics and watershed initial conditions, such as water levels and antecedent soil moisture22, which reflect an integration of prior sequences of local climate. Therefore, analyzing the regional influence of ENSO on flood and drought events through a combination of both precipitation and hydrological variables could enhance our understanding of its influence. Our study builds on prior work, which indicates that the climatological mean differences in average runoff between El Niño and La Niña years are approximately five times greater (relative to average flows) than in precipitation during these periods23. The impact of ENSO on streamflow can be more or less pronounced than rainfall due to concurrent effects on soil moisture and evapotranspiration, which can compound or compete with the ENSO-rainfall impact5. The first global modeling study on ENSO’s influence on flood duration and frequency24 revealed that the impact of ENSO on basin-scale flood duration was more pronounced than its effect on flood frequency. Predicting flood hazards during ENSO events is very complex, and the likelihood of increased or decreased flood risks varies worldwide25.

ENSO has been attributed as the main cause of recent severe events in SA, such as the floods that occurred in Rio Grande do Sul, Brazil, in 202326 and 2024; the 2012 flood in the Amazon27; floods in SA in 198328; the severe drought in SA during 2019/202029,30; and droughts in the Amazon in 201631. However, instead of attributing an individual or series of events to a specific phase of climate variability, a more proactive inquiry is to assess whether a particular climate variability phase can alter the risk of experiencing floods by predisposing regions to higher or lower-than-average wetness conditions32. This approach shifts the focus from association to understanding and estimating the risk patterns associated with ENSO phases and can enhance the understanding of ENSO as a predictability factor.

To date, no studies have comprehensively analyzed long-term observed flow time series focusing on understanding the influence of the ENSO on droughts and floods in SA. Most existing studies primarily focus on precipitation and climate patterns11,14,15,16,17,18,19,33,34,35,36,37,38, and the few studies on streamflow are outdated, dating back to the early 2000s39 and conducted on a global scale23,40. The most recent works are analyzed by simulated time-series24,25,41,42, or at a basin scale43. Given the ENSO’s influence on South American climate, it is crucial to review our understanding of its influence on extreme hydrological events with the recent years of observed data and a continental approach. Such targeted research is essential for improving prediction models and implementing effective management strategies in response to these extreme conditions. Several regional and national institutions in SA integrate ENSO information into climate services to enhance resilience and risk management. The International Research Center on El Niño (CIIFEN) coordinates ENSO monitoring and forecasting across sectors. National Meteorological and Hydrological Services (NMHSs) such as the National Service of Meteorology and Hydrology of Peru (SENAMHI), the National Institute of Meteorology (INMET) and the Center for Weather Forecast and Climate Studies (CPTEC) in Brazil, and the Chilean Meteorological Directorate (DMC) incorporate ENSO signals into seasonal outlooks. The World Meteorological Organization’s (WMO) Regional Climate Outlook Forums (RCOFs) provide ENSO-informed regional forecasts, and the United Nations Office for Disaster Risk Reduction (UNDRR) promotes the use of ENSO insights in disaster preparedness efforts.

In this study, we identify patterns of ENSO’s impact on extreme flow in South American rivers by utilizing observed streamflow time series from gauge stations across the continent and the Oceanic Niño Index (ONI)44, which is an operational index used to quantify ENSO variability. The analysis is structured around three main research questions: First, is a particular ENSO phase more likely to record an extreme hydrological event (flood or drought)? This will be assessed by analyzing changes in the relative conditional likelihood of streamflow observations occurring in each extreme category—i.e., below the 10th percentile of the monthly minimum daily streamflow time series or above the 90th percentile of the monthly maximum daily streamflow time series—associated with each ENSO phase (determined by the ONI index value). Second, is there a distinct wet or dry streamflow influence on extreme flows across SA associated with different ENSO phases, as the strength of influence is expected to vary by region? This will be assessed by analyzing each phase’s relative conditional likelihood of streamflow occurring above or below the historical full-record median streamflow of each extreme category, i.e., below the 50th percentile for minimum or above the 50th percentile for maximum time series. Third, does ENSO influence the timing of floods and droughts in SA? This will be assessed by analyzing the month of occurrence of the events above and below extreme thresholds for each timeseries (90th percentile for maximum and 10th percentile for minimum) and for each ENSO phase. The time series are detrended and deseasonalized to better capture ENSO influence. To add context to the analysis of maximum flow extremes, we perform the same tests on precipitation and investigate the correspondence between these conditional changes in high extremes for both rainfall and streamflow events. This final analysis targets whether the land surface filtering of rainfall variations into hydrological outcomes either amplifies or ameliorates the ENSO influence.

This study demonstrates a clear and spatially consistent influence of ENSO phases on extreme streamflow events across SA, based on 45 years of streamflow observations from 788 gauges. El Niño events are linked to a heightened likelihood of floods in the La Plata Basin (up to 160%) and droughts in the Amazon Basin (up to 160%), while La Niña tends to increase flood risk in the northern Amazon (over 120%) and Southeastern Brazilian Coast (over 80%) and drought risk in the La Plata Basin (over 80%). ENSO phases also shape wet and dry streamflow tendencies, with stronger effects on extreme events than on above- or below-median flows and influence the timing of floods and droughts in distinct regional patterns. Maximum precipitation and streamflow signals are generally aligned, but flood flows exhibit stronger amplification than high precipitation, suggesting a cumulative climate effect on extreme events.

Results

High flows

Figure 1a shows the ENSO phase that caused the highest change in the likelihood of extremely high streamflow—defined as monthly maximum daily streamflow values above the 90th percentile—occurring in any month of the year. To isolate the influence of ENSO phases, the baseline likelihood of such events (10%) was subtracted from the conditional likelihoods. The results were then normalized by dividing by 10%, expressing the change as a proportion of the baseline likelihood.

Fig. 1: Influence of ENSO phases on high flows.
figure 1

ENSO phase with the highest historical relative added likelihood of registering a maximum monthly daily streamflow. a above the 90th percentile within any month of the year, and b above the 50th percentile within any month of the year. Color intensity reflects the magnitude of added likelihood, which means the baseline historical likelihood is subtracted (10% in (a) and 50% in (b)) from the conditional ENSO phase likelihood. Triangles represent gauge stations with maximum time series with low significance (p value > 0.05 on the Kruskal-Wallis Test). Marker size varies with the density of stations for better visualization.

Full size image

In the Amazon Basin, floods were over 120% more likely to occur during La Niña, as depicted by the red dots concentrated on the North half of the basin. Similarly, around the Paraíba do Sul River on the Southeast Brazilian Coast, La Niña emerged as the phase that increased the likelihood of floods over 80%. The El Niño phase consistently increased the likelihood of floods in the La Plata Basin, as depicted by the blue dots, with a likelihood increase of more than 120% along the Paraná River and certain tributaries of the Uruguay River. In the central portion of SA, in the transition between the blue and red areas, the prevalence of the neutral phase underscored the independence of flood risk from classic El Niño or La Niña impacts.

Figure 1b shows the likelihood of maximum monthly daily streamflow above the median (50th percentile) during different phases of the ENSO. Here, we refer to volumes above the median, including those above the 90% threshold, and the baseline or random occurrence likelihood of 50% was subtracted to isolate the added influence of ENSO phases on these hydrological extremes. The spatial patterns visualized here are very similar to those in Fig. 1a, with most of the difference relying on the magnitude of the relative added likelihood, which is smaller. In the Branco River and tributaries, the analysis indicates that La Niña conditions were most conducive to increased streamflow. The La Plata Basin exhibited a higher incidence of above the median flows during El Niño phases, adding over 40% to the likelihood of these events. Supplementary Fig. S1 presents the results of Fig. 1, separated into two periods of the year. See methodology for months separation in Section “ONI Index”.

Figure 2 illustrates the median month of occurrence for the monthly maximum daily streamflows exceeding their 90th percentile for each ENSO phase. Since the seasonal component was removed from the data, the timing of floods shown on Fig. 2 does not represent the traditional period of wet conditions. Instead, the maps show when the anomalies usually happened in each ENSO phase. Visually, we can see that El Niño tended to anticipate flood anomalies in the central portion of Brazil (from April up to January), and delayed floods in the upper Amazon and South of Brazil. La Niña tended to delay flood anomalies in the upper portion of Southeastern SA, and switch from winter to summer the floods on the east coast.

Fig. 2: Timing of high flow events by ENSO phase.
figure 2

Median month of occurrence for monthly maximum daily streamflows exceeding the 90th percentile, shown separately for a El Niño, b Neutral and c La Niña. Marker size varies with station density to enhance visualization.

Full size image

Figure 3a shows the relative added likelihood of maximum monthly daily rainfall above the 90th percentile during different phases of the ENSO, and Fig. 3b shows the correspondence rate between the extreme high rainfall and floods. See Section “Correspondence of extreme precipitation and streamflow” for correspondence methodology. Figure 3c shows the relative added likelihood of maximum monthly daily rainfall above the median during different phases of the ENSO. Similarly to Figs. 1a, 3a shows that El Niño had a higher likelihood of registering extreme precipitation in the La Plata Basin, while La Niña had a higher likelihood in the Amazon and Southeast Coast. However, the magnitude of the ENSO’s influence on precipitation was visually smaller than on streamflow.

Fig. 3: Influence of ENSO phases on high precipitation and flood correspondence.
figure 3

a ENSO phase with the higher relative added likelihood of registering a maximum monthly daily precipitation above the 90th percentile within any month of the year. b Correspondence between extremely high monthly daily precipitation and floods (Map of Fig. 1). c ENSO phase with the higher relative added likelihood of registering a maximum monthly daily precipitation above the 50th percentile within any month of the year. Color intensity reflects the magnitude of relative added likelihood, which means the baseline historical likelihood is subtracted ((a) 10% and (b) 50%) from the conditional ENSO phase likelihood. Triangles represent gauge stations with low significance (p value > 0.05 on the Kruskal-Wallis Test); Marker size varies with the density of stations for better visualization.

Full size image

The correspondence rate—i.e., the percentage of extreme events that appear linked between maximum precipitation and streamflow events—is up to 50% at several stations in southern Brazil. The correspondence rate was higher in regions with the lowest predictability time (i.e., the response time of a catchment; see the predictability map in the Supplementary Information Fig. S2). That might be because the shorter the response time of the watershed, the less chance there is for other factors to influence streamflow, and thus, the more direct the relationship between precipitation and streamflow.

Low flows

Figure 4a shows the ENSO’s phase with the highest relative added historical likelihood of registering a minimum monthly daily streamflow below the 10th percentile within any month of the year. For clarity, we refer to this dataset range as extreme low flows or droughts. To isolate the influence of ENSO phases, the baseline likelihood of such events (10%) was subtracted from the conditional likelihoods. The results were then normalized by dividing by 10%, expressing the change as a proportion of the baseline likelihood.

Fig. 4: Influence of ENSO phases on low flows.
figure 4

ENSO phase with the higher relative added likelihood of registering a minimum monthly daily streamflow a below the 10th percentile within any month of the year, and b below the 50th percentile within any month of the year. Color strength reflects the magnitude of added likelihood, which means the baseline historical likelihood is subtracted (10% in (a) and 50% in (b)) from the conditional ENSO phase likelihood. Triangles represent gauge stations with maximum time series with low significance (p value > 0.05 on the Kruskal-Wallis Test). Marker size varies with the density of stations for better visualization.

Full size image

Droughts were over 120% more likely to occur in the Amazon Basin during the El Niño phase, and over 80% more likely on the Southeast Coast of Brazil. In contrast, in the Paraná basin, La Niña was the ENSO phase with the highest relative added likelihood of registering droughts, being over 80% more likely in some of the Paraná River tributaries.

Figure 4b shows the likelihood of minimum monthly daily streamflow below the 50th percentile during different phases of the ENSO. Here, we refer to volumes below the median, including those below the drought threshold of 10%. The baseline or random occurrence likelihood of 50% was subtracted to isolate the change in the influence of ENSO phases on these hydrological extremes. The spatial patterns visualized here differ from those in Fig. 4a. The El Niño influence area is smaller and weaker, remaining only in the northern portion of SA and along the eastern coast, increasing the chance of below-median minimum flows by up to 80%. During La Niña the changes of low flows increased homogenously in the SESA region, being 40% more likely to occur. Figure 4b also shows a region of neutral phase prevalence, right above the SESA region. Supplementary Fig. S3 presents the results of Fig. 4, separated into two periods of the year. See methodology for months separation in Section “ONI Index”.

Figure 5 illustrates the median month of occurrence for the monthly minimum daily streamflows exceeding their 10th percentile for each ENSO phase. Since the seasonal component was removed from the data, the timing of droughts shown in Fig. 5 does not represent the traditional period of dry conditions, but rather when anomalies typically occurred in each ENSO phase. Visually, we can see that both El Niño and La Niña seem to delay drought anomalies in the SESA and anticipate droughts in the central portion of Brazil.

Fig. 5: Timing of low flow events by ENSO phase.
figure 5

Median month of occurrence for the monthly minimum daily streamflows below their 10th percentile, shown separately for a El Niño, b Neutral, and c La Niña. Marker size varies with the density of stations for better visualization.

Full size image

Discussion

This work investigated the influence of the ENSO phase on extreme streamflow events using a 45-year observed streamflow time series (1979–2023) from 788 gauges in SA, where the ENSO phase is defined using the ONI index. Our focus is limited by the areas where data are available, which does not exclude the ENSO influence on different regions of the continent. The time series were detrended and deseasonalized to better capture the influence of ENSO. Our research aimed to answer three key questions, discussed below.

  1. i)

    Does a particular ENSO phase have a higher likelihood of recording an extreme hydrological event?

Floods

The results indicate that during the El Niño phase, the La Plata Basin predominantly experiences more frequent floods, with flood chances increasing up to 160%. Similarly, La Niña increases the likelihood of flooding across the northern Amazon Basin and the Southeastern Brazilian Coast. For instance, during La Niña conditions, the Branco River exhibits over 120% increase in flood likelihood.

Droughts

Conversely, La Niña is associated with a higher likelihood of droughts in the La Plata Basin, being over 80% more likely, while the El Niño phase tends to increase drought conditions in the Amazon basin up to 160%.

  1. i)

    Is there a distinct wet or dry streamflow signal on extreme flows associated with different ENSO phases?

Wet streamflow tendencies

La Niña increases wet conditions in the North half of the Amazon basin and the upper east coast of SA by over 80%. El Niño enhances high flow probabilities in the La Plata Basin up to 40%. The more substantial impact on floods than above the median might indicate that ENSO’s influence is more pronounced on extreme wet hydrological events than on general wet hydrological conditions.

Dry streamflow tendencies

El Niño generally promotes drier conditions in the northern Amazon, increasing the likelihood of below-median streamflow up to 80%. Conversely, La Niña triggers lower flow conditions in SESA of up to 40%, highlighting its role in potentially exacerbating drought conditions in these regions.

  1. i)

    Does ENSO influence the timing of floods and droughts in SA?

Flood timing

El Niño tends to anticipate flood anomalies in the central portion of Brazil (from April up to January), and delays in the upper Amazon and South of Brazil. La Niña tends to delay flood anomalies in the upper portion of SESA, and switch from winter to summer the floods on the east coast.

Drought timing

Both El Niño and La Niña seem to delay drought anomalies in SESA and anticipate droughts in the central portion of Brazil.

In the flood context, results indicate a correspondence between precipitation and streamflow ENSO signals, which is intuitive and agrees with studies that have identified streamflow changes in response to precipitation changes23. A stronger signal in streamflow than precipitation manifests for extremely high precipitation and floods, likely because ENSO also affects temperature, not just precipitation, influencing streamflow through both parameters5.

The results also show that droughts and floods do not occur exclusively in any ENSO phase; ENSO seems to be only one of the various factors influencing and triggering floods and droughts, and likely acts as an amplifying factor for meteorological systems driving hydrological extremes. Nevertheless, a clear and consistent spatial pattern of ENSO influence on streamflow was identified, illustrated in Fig. 6. Both wet and dry signals were observed during the same ENSO phase, highlighting a dual behavior. This finding suggests that hydropower plant systems could compensate for deficiencies in other HPP operations. It was also observed that ENSO has a more pronounced effect on extreme streamflow levels than on events above the timeseries median.

Fig. 6
figure 6

Overview of identified patterns of ENSO’s influence on extreme streamflow in South America.

Full size image

While we cannot directly compare our results to those of other studies due to methodological differences, our results show tendencies and spatial patterns similar to those of other studies based on long-term simulations25,42. Streamflow observations are often considered more reliable, yet the agreement between these observations and our analysis underscores the quality of such simulation studies.

Methods

Streamflow data

For this study, we downloaded and quality-controlled an extensive dataset of daily streamflow data across SA, focusing on records from the National Water Agency (ANA, Brazil), and the Institute of Hydrology, Meteorology, and Environmental Studies (IDEAM, Colombia), from 1979 to 2023. The dataset consists of time series from multiple gauging stations, with each station representing a distinct spatial location and containing repeated daily measurements over time—thus characterizing the data as spatiotemporal. To ensure data quality and relevance standards, we implemented a filtering process that includes:

  • Setting clearly erroneous zero value measurements and negative values to missing (NaN or ‘not a number’) to account for non-recorded data, given that none of these stations are in intermittent rivers.

  • Excluding streamflow measurements exceeding 900 mm/day and measurements more than 100 times the previous measurement as likely inaccuracies.

  • Retaining only stations with at least 40 years of data and less than 10% missing values.

  • Excluding stations without drainage area information.

This approach yielded 788 viable gauge stations in SA. See Figure S4 in the Supplementary Information for stations’ location and SA elevation maps and rivers. We proceeded with the analysis with maximum and minimum monthly daily streamflow in millimeters per day, meaning the streamflow was normalized by drainage area. We divided the monthly values by their ENSO phase and performed the Kruskal-Wallis Test to assess significance (p value < 0.05). The results and discussions focus on significant stations, although the low-significant ones are still indicated on the maps with triangle symbols.

Precipitation data

We used the MSWEP v2.8 precipitation database (1979–2023) to verify the linkage between extremely high precipitation and flood flows. We calculated the daily average precipitation for each gauge station drainage area and extracted the maximum monthly daily precipitation. We used the MGB–SA45 basin delimitations for delineating the catchment boundaries of each gauge station, enabling the comparison of these precipitation extremes with normalized flood flows.

Trend and seasonality removal

Given the strong seasonality and long-term trends present in South American hydrological data22,46,47,48,49,50,51, we performed a Seasonal-Trend decomposition using LOESS (STL) on each monthly time series. This process separated each series into trend, seasonal, and remainder components. We used the remainder component—containing the detrended and deseasonalized signal—for subsequent analysis. To ensure its stationarity, we applied the Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test, conducted in its standard two-sided form with a p-value threshold of 0.05 for significance (see Supplementary Fig. S5). Furthermore, Fig. S6 and Figure S7 on Supplementary Information show the maps of mean and standard deviation of the final data groups used in this study.

ONI Index

We used the Oceanic Niño Index (ONI) from the National Oceanic and Atmospheric Administration (NOAA, 2024) to classify the monthly streamflow time series into the three ENSO categories: El Niño, La Niña, and neutral. The index calculates the deviation from the norm, or anomaly, in average surface water temperatures over three months in the east-central tropical Pacific, close to the International Dateline. An ONI value of +0.5 or above signifies the occurrence of El Niño conditions, while a value of −0.5 or below denotes La Niña conditions. For an event to be officially recognized as a complete El Niño or La Niña episode, the thresholds must be surpassed continuously across at least five successive overlapping three-month periods. The ONI data is available from 1950 to the present every month. Since the data is presented as an average over three months, we used the last month of this average as the reference month for our analysis, as this offers better operational usability. During the analysis period (1979–2023), 139 months of El Niño, 154 months of La Niña, and 247 neutral months occurred. Figure 7 illustrates the distribution of monthly phases throughout the year. ENSO is more active from October to March, which led to the division of the data into two groups for seasonal decomposition of the results: October to March (ONDJFM) and April to September (AMJJAS).

Fig. 7: Monthly distribution of ENSO phases and ONI values.
figure 7

Monthly ENSO phase frequency (left) and Oceanic Niño Index (ONI) distribution (right). Bars in the left panel represent the number of El Niño (red), Neutral (gray), and La Niña (blue) events per month. The right panel shows box plots of ONI values.

Full size image

Historical likelihood estimation

We investigated ENSO’s influence by analyzing changes in the likelihood of streamflow observations occurring in dry (low) and wet (high) extreme categories, as defined by the 10 and 50th, and 50 and 90th percentiles of historical streamflow, respectively. Each monthly streamflow was paired with a corresponding ONI Index and ENSO phase. The conditional likelihood of streamflow occurrence was subtracted from the threshold likelihood to determine the altered likelihood. For instance, if streamflow corresponding to warm ENSO (El Niño) occurred at a 15% likelihood above the 90th percentile of historical flow, then the baseline likelihood of 10% was subtracted, and the altered likelihood is 5%. The altered likelihood was calculated for each ENSO phase and for all the gauge stations, and the maps show the higher added likelihood. For better comparison between thresholds, the highest added likelihood was normalized in relation to the thresholds. Figure 8 illustrates the methodology using the Alegrete station as an example. The El Niño phase presented the highest shifted (increased) likelihood of registering a maximum monthly daily streamflow above the 90th percentile (Fig. 8a). Thus, in the spatial map of Fig. 1a, the point representing this station is blue. Figure 8b shows that the La Niña phase presented the highest shifted (increased) likelihood of registering a maximum monthly daily streamflow below the 10th percentile; thus, in the spatial map of Fig. 4a, the point representing this station is red.

Fig. 8: Methodology visualization.
figure 8

a Maximum monthly daily streamflow above the 90th percentile, b minimum monthly daily streamflow below the 10th percentile. The density curves illustrate the distribution of each ENSO phase streamflow measurement, while the boxplot displays the thresholds for the entire dataset.

Full size image

For floods, we performed the same tests on precipitation and investigated the correspondence between extreme rainfall and streamflow events. Precipitation analysis was not performed for droughts because the correspondence between droughts and low precipitation can span many months (droughts can be multi-year events), preventing the analysis on the monthly timescale of the ONI Index. The maps show the ENSO’s phase with the highest shifted relative likelihood of exceeding one of the thresholds.

Correspondence of extreme precipitation and streamflow

To analyze the correspondence of ENSO impacts on flow versus precipitation, we recognize that there is a lag between precipitation falling on a catchment or river basin and the associated streamflow at the basin outlet, known as a basin’s ‘time of concentration’ or response time. This period reflects the time it takes for all the water that has entered the basin to pass through the system. Here, we apply the concept that if the time between a precipitation extreme event and a streamflow extreme event is less than or equal to the basin’s response time, the events are associated. In this work, the basin’s response time is defined as the null rainfall predictability (days) of the river52. Predictability was estimated as the number of days where a perfect model streamflow simulation with null rainfall maintains a KGE value higher than 0.9552, although we recognize that other metrics and approaches to estimating response time could be applied (such as the e-folding limit of correlation, which is often used in statistical analyses for calculating sampling degrees of freedom for timeseries displaying auto-correlation).