Rainfall variability and trends of the past six decades (1950–2014) in the subtropical NW Argentine Andes

Abstract

The eastern flanks of the Central Andes are characterized by deep convection, exposing them to hydrometeorological extreme events, often resulting in floods and a variety of mass movements. We assessed the spatiotemporal pattern of rainfall trends and the changes in the magnitude and frequency of extreme events (≥95th percentile) along an E-W traverse across the southern Central Andes using rain-gauge and high-resolution gridded datasets (CPC-uni and TRMM 3B42 V7). We generated different climate indices and made three key observations: (1) an increase of the annual rainfall has occurred at the transition between low (<0.5 km) and intermediate (0.5–3 km) elevations between 1950 and 2014. Also, rainfall increases during the wet season and, to a lesser degree, decreases during the dry season. Increasing trends in annual total amounts characterize the period 1979–2014 in the arid, high-elevation southern Andean Plateau, whereas trend reversals with decreasing annual total amounts were found at low elevations. (2) For all analyzed periods, we observed small or no changes in the median values of the rainfall-frequency distribution, but significant trends with intensification or attenuation in the 95th percentile. (3) In the southern Andean Plateau, extreme rainfall events exhibit trends towards increasing magnitude and, to a lesser degree, frequency during the wet season, at least since 1979. Our analysis revealed that low (<0.5 km), intermediate (0.5–3 km), and high-elevation (>3 km) areas respond differently to changing climate conditions, and the transition zone between low and intermediate elevations is characterized by the most significant changes.

Appendix: Quantile regression

Quantile regression, which is often used in climate-related studies (e.g., Friederichs and Hense 2007; Sankarasubramanian and Lall 2003; Elsner et al. 2008; Bremnes 2004) is an extension of median regression based on estimating the value of the parameter vector β from the set of allowable vectors that minimizes the mean-loss function:

$${\text{L}}_{\uptau} \left( {\upbeta,{\text{y}}} \right) \, = \frac{1}{n}\mathop \sum \limits_{i = 1}^{n} p_{\uptau} \left\{ {{\text{ y}}_{\text{i}} {-} \,\upmu\left( {{\text{x}}_{\text{i}} ,\upbeta} \right) \, } \right\}$$

(1)

where y_i (i = 1, …, n) are the response values, µ is the estimate of the τ quantile, and x_i and β are the covariate vector (in our case the time) and parameter vector, respectively. The loss function is p_τ (·), where:

$${\text{p}}_{\uptau} \left( {\text{z}} \right) = \left| {\text{z}} \right|\left\{ {\uptau \, \cdot{\text{ I}}\left( {{\text{z}} > 0} \right) + \left( {1 -\uptau} \right)\cdot{\text{ I}}\left( {{\text{z}} < 0} \right)} \right\}$$

(2)

and I(·) is the indicator function, which is 1 when the argument is true and 0, if not. The loss function is non-negative taking a minimum value of zero only when z = 0. Given a series of samples with µ constant (intercept-only model), the resulting value of β (a scalar in this case) that minimizes the total loss function occurs only when µ is equal to the τ quantile of the response. If the model fits well, a plot of fitted versus actual values will show that τ percentage of observed values should be less than the fitted values, with 1 − τ percentage of the observed values greater than that of the fitted values (Yu et al. 2003). The total loss function is an unbiased sample estimate of the expected value of p_τ[Y − µ(x · β)], and the minimization over β is a consistent estimate of the minimization of this expected value. For the fit, we employ a linear model for the regression function of the form:

$$\upmu =\upbeta_{0} + \mathop \sum \limits_{i = 1}^{p}\upbeta_{\text{i}} \cdot{\text{ x}}_{\text{i}}$$

(3)

where x_i is climate covariate i and there are p of them. Uncertainties associated with quantile regression coefficients at the 95 % confidence level were estimated applying boot-strapping techniques with 200 minimum numbers of iterations (Hahn 1995).