Increased Amazon Basin wet-season precipitation and river discharge since the early 1990s driven by tropical Pacific variability

The Amazon River is by far the Earth's largest river by flow, responsible for 15%–20% of the freshwater discharged into the oceans (Molinier et al 1996, Dai et al 2009). Its basin holds the largest planetary watershed and rainforest, which play key roles in the global climate system and carbon cycle and contain a wealth of biodiversity (Foley et al 2002, Malhi et al 2008). Recent extensive Amazon Basin dry-season droughts and fires have heightened concerns about climate change and deforestation (Nobre et al 2016, Marengo et al 2018). Additionally, starting in the early 1990s, the Amazon Basin experienced a dramatic increase in wet-season (December–May) precipitation, resulting in a strengthening of the hydrological cycle, heightened peak river discharge, and disruptive flood events (Gloor et al 2013, 2015, Marengo and Espinoza 2016, Barichivich et al 2018). The increased river discharge also reduced tropical Atlantic surface salinity (Gouveia et al 2019), with potential impacts on ocean circulation and large-scale climate (Jahfer et al 2017). Understanding the causes of this recent Amazon Basin hydrological intensification is important for assessing the uncertainties associated with climate change projections.

The Amazon Basin hydrological cycle intensification coincided with a pronounced climatic shift in the tropical Pacific. Over the 20-year period beginning in the early 1990s, the Pacific trade winds experienced a strengthening which was unprecedented within instrumental and reanalysis records (Balmaseda et al 2013, de Boisséson et al 2014). This was associated with sea surface temperature (SST) cooling in the equatorial central and eastern Pacific, and a reduced rate of global surface temperature increase compared to previous decades, sometimes referred to as the hiatus (Meehl et al 2011, Kosaka and Xie 2013). The trade wind and SST changes were connected to a strengthening of the Pacific Walker circulation and more frequent La Niña phases of the El Niño/Southern Oscillation (ENSO) (Dong and Lu 2013, Liu and Zhou 2017). On decadal timescales, these trends reflected a transition from the warm to cold phases of the Interdecadal Pacific Oscillation (IPO; Zhang et al 1997, Power et al 1999), with a larger amplitude than previous transitions (England et al 2014).

Observational analysis has connected the Amazon Basin wet-season precipitation increase from the early 1990s to the Pacific trade wind strengthening, SST cooling, and intensified Walker circulation (Barichivich et al 2018), building on well-established ENSO teleconnections in the tropics (Ropelewski and Halpert 1987, Dai and Wigley 2000). However, it is difficult using observations alone to separate the influence of Pacific variability on Amazon precipitation from that of greenhouse gases (GHGs) and other external climate forcing agents over the same period. As such, climate model sensitivity experiments can be useful to differentiate the contemporaneous impacts of different drivers (Hegerl and Zwiers 2011).

In this study, we use a set of companion large-ensemble climate model simulations with tropical Pacific surface wind stress nudging. The two experiments have identical external forcing agents but impose different wind stress regimes in the tropical Pacific, while allowing the climate system to evolve freely outside the tropical Pacific. In the first ensemble, W-CLIM, tropical Pacific wind stress is constrained to its long-term climatology, simulating the counterfactual trajectory of the climate system holding a fixed neutral IPO state. In the second ensemble, W-FULL, observed tropical Pacific wind stress anomalies are imposed, simulating the evolution of the climate system with realistic tropical Pacific decadal variability, including the trade wind acceleration from the early 1990s and IPO phase transition. By comparing the Amazon Basin hydroclimate response in W-CLIM and W-FULL, we can distinguish the contributions of the external forcings common to both simulations from those related to Pacific trade wind variability found only in W-FULL.

We examine a set of experiments with the IPSL-CM5A-LR global climate model (Dufresne et al 2013). Its atmospheric component is LMDZ5A (Hourdin et al 2013), with a resolution of 3.75° × 1.875° and 39 vertical levels. The land surface component is ORCHIDEE (Krinner et al 2005), with a similar horizontal resolution as the atmosphere, and the ocean component is NEMOv3.2 (Madec 2008). The experimental setup, further described in Gastineau et al (2019), consists of two 30-member ensembles, each initialized in 1979 from a spread of random phases of the IPO and the Atlantic Multidecadal Oscillation (AMO) (Kerr 2000) from the IPSL-CM5A-LR fully-coupled historical large ensemble (Frankignoul et al 2017). Both ensembles are forced by identical Fifth Coupled Model Intercomparison Project (CMIP5) historical and Representative Concentration Pathway 8.5 (RCP8.5) GHG and ozone emissions from 1979 to 2014 (Taylor et al 2012). Volcanic aerosols are included, but anthropogenic aerosols are fixed at 1940 values in both experiments.

In both ensembles, wind stress (τ) is prescribed in the ocean component in the tropical Pacific from 20° S to 20° N with a buffer region from 15° to 25° latitude (outlined in figure 1(a)), while leaving the climate system unconstrained outside the tropical Pacific. Prescribing wind stress tightly constrains SST through its effects on zonal advection, thermocline depth, and eastern Pacific upwelling (Clarke 2008). It is somewhat similar to nudging SST (e.g. Kosaka and Xie 2013, McGregor et al 2014), but favors the dynamical consistency between the atmosphere and ocean, and reduces artificial heat fluxes imposed into the ocean (Watanabe et al 2014, Douville et al 2015). In the representative climatological ensemble, W-CLIM, the 1979–2014 climatological wind stress from the IPSL-CM5A-LR historical and RCP8.5 CMIP5 ensemble mean is prescribed, with additional model noise to deepen the mixed layer; this is further described in text S1 (which is available online at stacks.iop.org/ERL/16/034033/mmedia).

Zoom In Zoom Out Reset image size

In the second ensemble, W-FULL, daily wind stress anomalies from 1979 to 2014 from the ECMWF ERA-Interim Reanalysis (ERA-I; Dee et al 2011) are added to the model climatology and prescribed in the ocean model component. Figure 1(a) shows the 1992–2012 ERA-I mean τ trend over the nudging region during the Amazon Basin wet season (December–May). Figure 1(b) shows the corresponding zonal wind stress (τ_x ) anomaly over the Niño 4 region (160° E–150° W, 6° S–6° N) from ERA-I and applied in W-FULL, illustrating the significant (p< 0.05) strengthening trend from 1992 to 2012. An error in the treatment of the ERA-I τ anomalies induced τ amplitudes about 20% larger in W-FULL than ERA-I, though the overall interannual changes are very similar and the spatial patterns are identical (Gastineau et al 2020).

For precipitation, we examine the monthly 2.5° × 2.5° combined satellite-gauge Global Precipitation Climatology Project (GPCP), version 2.3 (Adler et al 2003, 2018). We take December–May as the Amazon Basin-wide wet season, which is associated with the South American monsoon system, while recognizing that the seasonal cycle varies regionally within the Basin (Marengo et al 2012, Wang et al 2018). IPSL-CM5A-LR generally reproduces the observed wet-season regional precipitation climatology (figures S1(a)–(c)). In common with many CMIP5 models, it suffers from a double intertropical convergence zone (ITCZ) in the eastern Pacific (Jones and Carvalho 2013, Li and Xie 2014). There is also excessive precipitation over the Andes relative to the Amazon Basin interior (Dufresne et al 2013). IPSL-CM5A-LR effectively simulates the observed seasonal cycle of Amazon region mean precipitation (figure S1(d)), though similar to many CMIP5 models, it underestimates total precipitation, particularly during the July–October dry season (Yin et al 2013).

To complement precipitation, we examine monthly Amazon River discharge (in m³ s^–1) from the farthest downstream long-term gauge at Óbidos (55.7° W, 1.9° S; indicated in figure 2(a)), which captures runoff from most of the Amazon Basin (Dai and Trenberth 2002). River discharge provides an integrated measure of Amazon Basin hydroclimate, though it does not allow us to differentiate between recirculated and non-recirculated precipitation (Gloor et al 2013, Dai 2016). For the model, we use the output of freshwater flux into the ocean from rivers (in kg m^–2 s^–1) multiplied by the respective gridcell area to obtain a monthly volume flux, which we integrate over the Amazon River outflow region (45°–51° W, 2° S–6° N; blue box in figures 2(b)–(c)).

Zoom In Zoom Out Reset image size

In observations, the Amazon River discharge climatology exhibits a relatively smooth seasonal cycle (figure S1(e)), gradually peaking in May and June. Óbidos discharge lags Basin-average precipitation by around 3 months due to the time for surface runoff to travel downstream (Dai and Trenberth 2002, Marengo 2005). The model seasonal discharge maximum arrives 1–2 months early, peaking in April, and lasts for a shorter period compared to observations, which reflects ORCHIDEE deficiencies over the Amazon region (Guimberteau et al 2012, 2014) and its floodplain storage not being activated in the CMIP5 experiments (Agnès Ducharne, personal communication). To allow for comparison between observations and simulations given the annual cycle phase differences, we track high-season discharge by taking the mean of the five largest-discharge months of each year (with the discharge year spanning from February to January). In addition to precipitation and discharge, we use the monthly 2° × 2° NOAA Extended Reconstructed SST dataset (ERSST), version 5 (Huang et al 2017), as well as monthly 2.5° × 2.5° atmospheric fields from ERA-I.

Our analysis focuses on the Amazon region wet season, for which the tropical Pacific has a relatively large influence on interannual variability (Yoon and Zeng 2010, Andreoli et al 2012), over the 21-year trend from 1992 to 2012, corresponding to the maximum Pacific trade wind strengthening (Balmaseda et al 2013, de Boisséson et al 2014). Trend significance is calculated based on a two-tailed Student's t-test of the slope, adjusting the standard error and degrees of freedom assuming a first-order autoregressive (AR1) noise model (Santer et al 2000). We also examine interannual correlations of linearly-detrended time series over the entire 1980–2014 period, with significance evaluated based on a two-tailed Student's t-test, adjusting the degrees of freedom assuming an AR1 noise model (Bretherton et al 1999). We describe trends and correlations as significant when p < 0.05. The W-CLIM and W-FULL ensemble means are analyzed in order to emphasize the robust signals in each experiment.

3.1. Precipitation and discharge trends

The 1992–2012 observed wet-season precipitation trend (figure 2(a)) is positive over the Amazon Basin and northern South America extending into the equatorial Atlantic. In the eastern Pacific, it is positive from 5° to 10° N and negative along the equator. W-FULL (figure 2(b)) well reproduces the spatial pattern and magnitude of the observed trend, though the precipitation increase in South America is shifted to the south, with some drying in the northern Amazon region. W-FULL also overestimates the positive trend in the equatorial Atlantic and the negative trend in the eastern equatorial Pacific. W-CLIM (figure 2(c)), which has much smaller amplitudes (note doubled color scale), shows wettening in the eastern equatorial Pacific contrasting with drying in the western Amazon region, and a weaker precipitation increase in the eastern Amazon region extending into the equatorial Atlantic.

Figure 2(d) shows the observed and simulated wet-season mean precipitation anomaly averaged over the Amazon region (75°–50° W, 12.5° S–5° N) and 1992–2012 linear trend. W-FULL reproduces the significant positive 1992–2012 trend found in observations (table 1). It also captures much of the observed interannual variability, with a significant correlation with GPCP (table 2). The mean 5-month high-season Amazon River discharge is shown in figure 2(e). Like precipitation, W-FULL captures the 1992–2012 positive trend and is significantly correlated with Óbidos observations. W-CLIM lacks a discernible trend in basin-wide precipitation or river discharge.

Table 1. Wet-season trends. 1992–2012 trends for observations and W-FULL and W-CLIM ensemble means, in decade^–1; mean slope and 95% confidence intervals. Top: December–May mean Amazon region (75°–50° W, 12.5° S–5° N) precipitation. Middle: mean high-season (5 largest months) Amazon River discharge. Bottom: December–May mean Pacific–Atlantic interbasin SST contrast (tropical central and eastern Pacific (180°–90° W, 10° S–10° N) minus tropical Atlantic (40° W–15° E, 10° S–10° N) SST).

	Observations	W-FULL	W-CLIM
December–May Amazon region precipitation	0.54 ± 0.41 mm d–1 (GPCP)	0.52 ± 0.28 mm d–1	–0.00 ± 0.06 mm d–1
5-month max Amazon River discharge	15.2 ± 13.0 × 103 m3 s–1 (Óbidos)	34.4 ± 18.3 × 103 m3 s–1	–1.3 ± 4.5 × 103 m3 s–1
December–May Pac–Atl $\Delta$SST	–0.55 ± 0.37 °C (ERSST)	–0.75 ± 0.45 °C	0.03 ± 0.025 °C

Table 2. Wet-season interannual correlations. 1980–2014 detrended correlations of observations with W-FULL and W-CLIM ensemble means. Asterisks indicate statistically significant correlations at the 95% confidence level. Top: December–May mean Amazon region (75°–50° W, 12.5° S–5° N) precipitation. Middle: mean high-season (5 largest months) Amazon River discharge. Bottom: December–May mean Pacific–Atlantic interbasin SST contrast (tropical central and eastern Pacific (180°–90° W, 10° S–10° N) minus tropical Atlantic (40° W–15° E, 10° S–10° N) SST).

	W-FULL	W-CLIM
December–May Amazon region precipitation (GPCP)	0.72*	0.32
5-month max Amazon River discharge (Óbidos)	0.57*	0.41*
December–May Pac–Atl $\Delta$ SST (ERSST)	0.90*	0.05

3.2. SST and atmospheric circulation

The 1992–2012 trends in relative SST and low-level (850-hPa) wind are shown in figures 3(a)–(c). Relative SST (referred to as SST*), the deviation from the tropical mean (20° S–20° N) SST, captures the spatial gradients of SST and low-level moisture which are important for tropical convection (Vecchi and Soden 2007, Ma and Xie 2013, Khodri et al 2017); absolute SST is shown in figures S2(a)–(c). ERSST (figure 3(a)) features SST* cooling in the tropical central and eastern Pacific resembling the pattern associated with the transition from positive to negative IPO phases, with equatorial SST* cooling extending into both hemispheres along the eastern boundary, and SST* warming in the tropical Atlantic. W-FULL (figure 3(b)) shows a similar pattern, though with larger-amplitude equatorial Pacific SST* cooling and without extending toward the mid-latitudes. These differences may reflect the larger-amplitude τ anomalies applied (Gastineau et al 2020) or model cold tongue SST and double ITCZ biases (Bellenger et al 2014, Li and Xie 2014). In contrast, W-CLIM (figure 3(c)) shows equatorial Pacific SST* warming, as expected from global warming processes found in coupled models (Dinezio et al 2009). ERA-I and W-FULL feature increased 850-hPa flow into the Amazon region from the tropical North Atlantic, one of the key pathways for Amazon climatological moisture convergence (Grimm 2003, Arraut et al 2012).

Zoom In Zoom Out Reset image size

Figure 3(d) presents variations of the Pacific–Atlantic interbasin SST contrast (Pac–Atl $\Delta$SST), which reflects the strength of the Walker circulation (Wang 2006, McGregor et al 2014, Chikamoto et al 2015) and has been linked to wet-season Amazon region precipitation (Gloor et al 2015, Barichivich et al 2018). Pac–Atl $\Delta$SST is calculated as the difference between the tropical central and eastern Pacific (180°–90° W, 10° S–10° N) and the tropical Atlantic (40° W–15° E, 10° S–10° N; boxes in figures 3(a)–(c)). We find that Pac–Atl $\Delta$SST is significantly negatively correlated with Amazon Basin wet-season precipitation and high-season river discharge in observations and the simulations (table S1). Consistent with the precipitation and discharge trends, W-FULL captures the significant 1992–2012 observed negative Pac–Atl $\Delta$SST trend (table 1). The discrepancy between tropical North Atlantic SST* warming in ERSST compared to tropical South Atlantic SST* warming in W-FULL is not reflected in Pac–Atl $\triangle$SST. In contrast, the W-CLIM Pac–Atl $\Delta$SST trend is an order of magnitude smaller and of opposite sign.

In order to more clearly identify the zonal circulation changes, we examine the trends in pressure vertical velocity ($\omega$) and the zonal component of the divergent wind (u_div) averaged over the maximum precipitation trend signal over the Amazon latitudes (0–10° S; figure 4). ERA-I and W-FULL feature increased ascent over the Amazon region and weakened ascent over the central equatorial Pacific (figures 4(a) and (b)), depicting strengthening of the climatological Walker circulation branch between the Pacific and South America (Dong and Lu 2013, Liu and Zhou 2017). In contrast, W-CLIM features a small weakening of the ascent over the western Amazon region (figure 4(c)), consistent with the weakening of the Walker circulation expected from global warming (Vecchi and Soden 2007). The meridional circulation, depicted through $\omega$ and the meridional component of the divergent wind (v_div) zonally averaged from 75° to 50° W (figure S3), also reveals enhanced Amazon uplift in ERA-I and W-FULL, though displaced to the south in W-FULL.

Zoom In Zoom Out Reset image size

3.3. Moisture convergence

We next examine moisture convergence, which is the predominant contributor to Amazon Basin wet-season precipitation and the leading driver of its interannual variability (Angelini et al 2011, Satyamurty et al 2013b, Drumond et al 2014). ERA-I and the simulations feature climatological wet-season moisture convergence over the Amazon region (table S2(a) and figure S4). Consistent with the precipitation and discharge increases, ERA-I and W-FULL have significant positive moisture convergence trends over the Amazon region from 1992 to 2012 (table S2(b) and figure S5). Following Satyamurty et al (2013a, 2013b), we also integrate the moisture fluxes across the four Amazon region boundaries. Increased ERA-I moisture convergence is mostly due to northeasterly moisture advection from the tropical North Atlantic, while W-FULL simulates increased westerly moisture flux from the equatorial Pacific.

We additionally use the linear approximation of Huang (2014) to separate the moisture convergence components related to dynamics and thermodynamics in the 1992–2012 Amazon region trend:

$$\begin{equation*}\Delta P\sim\Delta \omega \cdot \bar q + \bar \omega \cdot \Delta q,\end{equation*}$$

where P is precipitation; q denotes 925-hPa specific humidity; and $\omega$ is 500-hPa pressure velocity. The dynamic component ($\Delta \omega \cdot \bar q$) accounts for nearly all of the Amazon region moisture convergence increase in ERA-I and W-FULL, indicating that atmospheric circulation changes are predominant for the precipitation trend (table S3). In W-CLIM, which has little overall moisture convergence change, the increased thermodynamic component ($\bar \omega \cdot \Delta q$) opposes the reduced dynamic component related to the tropical circulation weakening.

Using a companion set of climate model experiments with identical external forcing agents but differently constrained tropical Pacific wind stress, we attribute the 1992–2012 increases in wet-season Amazon Basin precipitation and high-season river discharge to the contemporaneous Pacific trade wind strengthening. In simulations with observed tropical Pacific wind stress anomalies (W-FULL), the imposed Pacific trades cause equatorial Pacific SST cooling, sharpening the Pacific–Atlantic zonal SST contrast (Pac–Atl $\Delta$SST) and strengthening the Pacific-South America branch of the Walker circulation. The enhanced central and eastern Pacific subsidence and compensating convergence over the Amazon climatological ascending region lead to increased Amazon Basin wet-season precipitation and high-season river discharge (Gloor et al 2015, Barichivich et al 2018).

In contrast, the companion experiment with climatological tropical Pacific wind stress simulating a neutral IPO state (W-CLIM) does not produce significant changes in Amazon Basin wet-season precipitation or high-season river discharge. W-CLIM does have a larger 1992–2012 trend in absolute tropical Atlantic SST than W-FULL (figure S2(d)), though not in relative Pac–Atl $\Delta$SST (figure 3(d)) which is important for Amazon Basin wet-season convection. While atmosphere-only experiments have shown a strong influence of Atlantic SST warming alone on the Amazon region wet-season precipitation increase from 1979 to 2015 (Wang et al 2018), these simulations are not directly comparable to our approach since they impose climatological SST outside the perturbed SST basin, whereas we allow ocean-atmosphere feedbacks through interactive air-sea coupling and include SST evolution from external forcings.

The W-CLIM setup assumes that tropical Pacific wind stress did not respond to external forcing during the observed strengthening period. This assumption has been challenged by findings that anthropogenic Asian sulfate aerosols contributed to the increased Pacific trade winds (Smith et al 2016, Takahashi and Watanabe 2016), though the linkage has been questioned (Oudar et al 2018). There is also evidence that the Pacific trade wind increase and Walker circulation strengthening were partly driven by teleconnections from Atlantic SST warming (McGregor et al 2014, Chikamoto et al 2015).

Interestingly, we find that W-CLIM high-season Amazon River discharge is significantly correlated with observations, though explaining only about half the variance as W-FULL (table 2). Examination of the Amazon River discharge time series reveals that much of the skill comes from decreases in 1983 and 1992 following the 1982 El Chichón and 1991 Mount Pinatubo eruptions (figure 2(e)), which is consistent with findings of reduced Amazon River discharge following large tropical eruptions (Iles and Hegerl 2015). The W-CLIM post-eruption decrease magnitudes are much smaller than in W-FULL, which incorporates Pacific trade wind anomalies related to the 1982–83 and 1991–92 El Niño events.

The Amazon region wet-season precipitation increase simulated in W-FULL is driven by increased moisture convergence via atmospheric dynamical changes, through a strengthened Walker circulation. However, within this large-scale framework, we note that the relatively low-resolution IPSL-CM5A-LR model does not directly simulate some of the convective-to-synoptic-scale features important for Amazon region wet-season precipitation, such as coastal-generated squall lines (Greco et al 1990, Garstang et al 1998). More work is needed to investigate the role of these higher-resolution processes over the Amazon Basin, for instance using dynamical downscaling (Ramos da Silva and Haas 2016).

Biases in IPSL-CM5A-LR may account for some of the spatial discrepancies between the observed and modeled changes over the Amazon Basin. The double ITCZ and excessive eastern Pacific cold tongue biases, common to many models (Li and Xie 2014, Zhang et al 2019), translate into too-weak southerly cross-equatorial winds (Hu and Fedorov 2018) and may explain the spurious moisture advection trend from the equatorial Pacific in W-FULL. Some circulation differences may also stem from the model's coarse orographic representation of the Andes Mountains (Insel et al 2010). Additional examination of pacemaker-type simulations such as those through the Decadal Climate Prediction Project (Boer et al 2016) can help assess the robustness of our results in a multi-model framework.

Insufficient tropical North Atlantic SST warming in our simulations may cause discrepancies in the equatorial Atlantic wind field and explain the lack of moisture advected by the model mean flow. The weak tropical North Atlantic warming may also account for W-FULL's lack of skill in simulating Amazon region dry season (July–October) precipitation and low-season river discharge (table S4 and figure S6), which are negatively associated with the tropical Atlantic interhemispheric SST contrast through the position of the ITCZ (Yoon and Zeng 2010, Fernandes et al 2015). The model could be missing the observed warming from declining North American and European sulfate aerosols (Cox et al 2008, Hua et al 2019), which do not vary in our simulations, or from increased ocean heat transport convergence to the tropical North Atlantic (Servain et al 2014).

Despite these caveats, our study provides strong evidence for the influence of the tropical Pacific on the intensification of Amazon Basin wet-season precipitation and high-season river discharge from the early 1990s to the early 2010s. As the Amazon region faces increasing anthropogenic stressors over the coming decades, understanding the role of decadal ocean-atmosphere variability will be important to reduce uncertainties in climate-related hydrological impacts. For instance, how might a continuation of the recent positive IPO phase since around 2013 (Meehl et al 2016, Cha et al 2018) affect Amazon Basin wet-season precipitation and high-season discharge given further tropical Atlantic warming? Additional comparative studies prescribing tropical Pacific wind stress could help in understanding the ongoing hydroclimate changes in the vital Amazon region.

A R F and M A B were supported by the UK NERC-funded SMURPHs project (NE/N006143/1). A R F, G G, and M K were also supported by the French ANR under the program Facing Societal, Climate and Environmental Changes (MORDICUS project, Grant ANR-13-SENV-0002). This work was granted access to the HPC resources of TGCC under the allocation 2015-017403 and 2016-017403 made by GENCI. GPCP was produced as part of the GEWEX effort under the WCRP. The ERSST and GPCP data were provided by the NOAA/OAR/ESRL PSL, and ERA-I was provided by ECMWF and C3S. The Amazon discharge data at Óbidos were provided by the SO HYBAM Amazon Basin Water Resources Observation Service monitoring network (www.so-hybam.org). We appreciate the constructive feedback from three anonymous reviewers.

The data that support the findings of this study are available upon reasonable request from the authors.