Observed precipitation in the Paraná-Plata hydrological basin: long-term trends, extreme conditions and ENSO teleconnections

Abstract

The Paraná-Plata basin is the second largest hydrological basin in South America and is of great importance for the countries of the region (Argentina, Bolivia, Brazil, Paraguay and Uruguay). The present study focuses on the long-term trends in basin-scale precipitation with special emphasis on the role of distribution changes in extreme large-scale precipitation events and on the characteristics and evolution of ENSO teleconnections over the last 50 years. First, we defined a Paraná-Plata basin total precipitation index (PTPI) as the precipitations spatially averaged over the hydrological basin. On interannual time scales, such an index is mainly representative of anomalous monsoon precipitations in the northern part of the basin and large convective precipitation anomalies in the center of the basin (Paraguay-southern Brazil-Uruguay-northern Argentina) typical of the canonical ENSO teleconnection pattern. Our major findings clearly highlight a positive trend in yearly averaged PTPI mainly from the late 1960s to the early 1980s with a strong dependence from month-to-month. The largest precipitation increase is observed from November to May in southern Brazil and Argentina. A close examination of PTPI distributions during the two halves of the period 1950–2001 shows that the changes in the mean state from 1950–1975 to 1976–2001 result from significant changes in each calendar month mean state and in the tails of the PTPI anomaly distributions in May with lesser and weaker large-scale dry events and stronger large-scale wet events. Further studies will be needed to assess whether the observed trend in large-scale extreme precipitation conditions can be related to natural climate variability or anthropogenic activities and whether it is associated to changes in local/regional extreme events. The stronger wet conditions in different months seem to be associated to changes in ENSO characteristics (amplitude, propagation, spatial structure, ...) since the 1982–1983 El Niño. Indeed, spatial ENSO teleconnections (stronger in November and April–May) have greatly evolved from 1950–1975 to 1976–2001. Moreover, we demonstrate that there is a strong modulation and displacement of the teleconnection patterns from one event to another, impeding the definition of robust statistical relationship between ENSO and precipitation in the Paraná-Plata basin (except maybe over a very limited area near the common border between Paraguay, Argentina and Brazil). Finally, the non-antisymmetrical patterns of precipitation between El Niño and La Niña conditions and the non-linear relationship between precipitation and either Niño3.4 or Niño1+2 sea surface temperature indices show that linear statistical forecast systems are actually of very limited use for impact predictions on society on a local or regional scale.

Appendix

In the present study, we made use of the non-parametric Mann-Whitney Pettitt Test to identify the change point (from year to year) in the monthly time series of the Paraná-PLata Precipitation Index. A brief description of the test follows.

Basically, a time series of length T (x₁, x0₂, ..., x _T ) is subdivided into two samples of respective lengths (x₁, ..., x _t ) and (x_t+1, ..., x _T ). Two indices U(t) and V(t) are then defined so that:

$$ V(t) = \sum\limits_{j = 1}^{j = T} {{\text{sign}}\left( {x_t - x_j } \right)} $$

(where sign(x)=1 for positive values of x, −1 for negative values of x and 0 for x equal to zero) and

$$ U(1) = V(1)\;{\text{and}}\;U(t) = U(t - 1) + V(t)\;{\text{for}}\;t = 2, \ldots ,T $$

Finally, the most significant change-point is found where the absolute value of U(t) is maximum. Moreover, at each point, it is possible to test the significance probability of the value where:

$$ p(t) = 1 - \exp \left( {\frac{{ - 6 \times U(t)^2 }} {{T^3 + T^2 }}} \right) $$

This test has been found to be relatively effective in identifying change-points in a time series. Moreover, in the case of a long-term trend, it identifies the center of the period of the trend as a change point. Finally, if the time series is dominated by a cycle, this test (alike many similar tests) will be more or less effective depending on the length of the cycle relative to the length of the time series under study. In particular, if the length of the tested time series is longer than the cycle period, the test may be inadequate and the time series should be subsampled prior to using the test. In our case, we found the Mann-Whitney Pettitt Test to identify well the different periods when changing conditions were observed and therefore to be well adapted to our needs.