Nonlinear interactions between the Amazon River basin and the Tropical North Atlantic at interannual timescales

Abstract

We study the physical processes involved in the potential influence of Amazon (AM) hydroclimatology over the Tropical North Atlantic (TNA) Sea Surface Temperatures (SST) at interannual timescales, by analyzing time series of the precipitation index (P-E) over AM, as well as the surface atmospheric pressure gradient between both regions, and TNA SSTs. We use a recurrence joint probability based analysis that accounts for the lagged nonlinear dependency between time series, which also allows quantifying the statistical significance, based on a twin surrogates technique of the recurrence analysis. By means of such nonlinear dependence analysis we find that at interannual timescales AM hydrology influences future states of the TNA SSTs from 0 to 2 months later with a 90–95% statistical confidence. It also unveils the existence of two-way feedback mechanisms between the variables involved in the processes: (1) precipitation over AM leads the atmospheric pressure gradient between TNA and AM from 0 to 2 month lags, (2) the pressure gradient leads the trade zonal winds over the TNA from 0 to 3 months and from 7 to 12 months, (3) the zonal winds lead the SSTs from 0 to 3 months, and (4) the SSTs lead precipitation over AM by 1 month lag. The analyses were made for time series spanning from 1979 to 2008, and for extreme precipitation events in the AM during the years 1999, 2005, 2009 and 2010. We also evaluated the monthly mean conditions of the relevant variables during the extreme AM droughts of 1963, 1980, 1983, 1997, 1998, 2005, and 2010, and also during the floods of 1989, 1999, and 2009. Our results confirm that the Amazon River basin acts as a land surface–atmosphere bridge that links the Tropical Pacific and TNA SSTs at interannual timescales. The identified mutual interactions between TNA and AM are of paramount importance for a deeper understanding of AM hydroclimatology but also of a suite of oceanic and atmospheric phenomena over the TNA, including recently observed trends in SSTs, as well as future occurrences and impacts on tropical storms and hurricanes throughout the TNA region, but also on fires, droughts, deforestation and dieback of the tropical rain forest of the Amazon River basin.

Appendix

1.1 Appendix 1: Recurrence technique

From the mutual information analysis we selected the lag value that corresponds to the one where mutual information steep changes to a slower decrease in its steep as the optimum value for τ from Fig. 13 we observe that for the P-E index τ = 1, also Fig. 13 shows that the value of m that corresponds to the lowest percent of false neighbors in the reconstructed phase-space is 4. The same procedure was used for all the time series (not shown).

The lagged dependence analysis for a 1% of the recurrence rate maintains the same results previously described for the feedback mechanism between the AM and TNA regions, and in some of the lags increasing the confidence of the results, this increase in confidence for some of the lags may be a result of the finer threshold that cleans spurious proximities between the trajectories and that may reduce the effect of noise in the time series (Fig. 14a–c).

1.2 Appendix 2: Complementary recurrence analysis

We compute a complementary recurrence analysis of the two-way mechanism, and analyze the dependence between the variables in both trajectories. In Fig. 15 we illustrate how the two-way feedback mechanism operates among the variables involved in process (such plot is furthered explained in Fig. 6 of the manuscript). The proposed two-way mechanism is supported by the significant correlations between the variables for lags from −2 to 2 months. Such results confirm that the mechanism acts as a two-way process, such that the TNA STTs affects the AM hydrology and vice versa. The recurrence results supports the well-known influence of the TNA SSTs on the surface winds S → W (Chung et al. 2002) is presented in Fig. 15.

Fig. 13

Mutual information with respect to time delay τ and the false nearest neighbors with respect to the dimension m for the AM P-E index

Fig. 14

Recurrence lagged dependence between the four variables according to the direction defined in the mechanism of feedback. The arrow between the names of the variables denotes which is the leading variable, the gray area represents the 90% confidence area, the blue dashed lines represent the 95% confidence intervals and the red line represents the calculated RMD between the variables. The recurrence threshold ε was based on a fixed 1% of the recurrence rate $RR=1/{N^2}\mathop \sum \nolimits_{i,j} {R_{ij}}$ for all the time series

Fig. 15

Recurrence lagged dependence between G → P, W → G and S → W. The arrow denotes the direction of influence between variables, the gray area represents the 90% confidence area, the blue dashed lines denote the 95% confidence intervals, and the red line represents the calculated RMD between variables. The recurrence threshold ε was based on a fixed 5% of the recurrence rate $RR=1/{N^2}\mathop \sum \nolimits_{i,j} {R_{ij}}$ for all the time series, as in Fig. 6

In Fig. 16 we present the two-way relations between the variables G and S, as well as P and W. Once the surface winds are affected by the TNA SSTs, then the pressure gradient (G) between the TNA and AM is affected. Cooling or heating of the TNA SSTs are also related with changes in G. The relation S ↔ G can be seen also as a two-way relationship where the SSTs drive the atmospheric pressure gradient for several months (0–1, and 4–8) while the pressure gradient drives the SSTs during a period of 2 months (0–2). According to the recurrence analysis P does not seem drive W, and therefore any connection between those two variables over the AM and the TNA is to be mediated by other variables in a nonlinear way as evidenced in the recurrence analysis. The influence of W over P is significant during the entire year confirming the well-established fact that there is a direct influence of zonal winds in the transport of moisture from the ocean to the continent to influence convective process in the AM (Yoon and Zeng 2010; Moraes-Arraut et al. 2011; Poveda et al. 2014).

Fig. 16

Recurrence lagged dependences supporting the two way influences between G ↔ S (left panel) and P ↔ W (right panel). The arrow denotes the direction of influence between the variables, the gray area represents the 90% confidence area, the blue dashed lines denote the 95% confidence intervals, and the red line represents the calculated RMD between variables. The recurrence threshold ε was based on a fixed 5% of the recurrence rate $RR=1/{N^2}\mathop \sum \nolimits_{i,j} {R_{ij}}$ for all the time series, as in Fig. 6