Reconstructing the annual mass balance of the Echaurren Norte glacier ( Central Andes , 33 . 5 ◦ S ) using local and regional hydroclimatic data

Despite the great number and variety of glaciers in southern South America, in situ glacier mass-balance records are extremely scarce and glacier–climate relationships are still poorly understood in this region. Here we use the longest (> 35 years) and most complete in situ mass-balance record, available for the Echaurren Norte glacier (ECH) in the Andes at ∼ 33.5 S, to develop a minimal glacier surface massbalance model that relies on nearby monthly precipitation and air temperature data as forcing. This basic model is able to explain 78 % of the variance in the annual glacier massbalance record over the 1978–2013 calibration period. An attribution assessment identified precipitation variability as the dominant forcing modulating annual mass balances at ECH, with temperature variations likely playing a secondary role. A regionally averaged series of mean annual streamflow records from both sides of the Andes between∼ 30 and 37 S is then used to estimate, through simple linear regression, this glacier’s annual mass-balance variations since 1909. The reconstruction model captures 68 % of the observed glacier mass-balance variability and shows three periods of sustained positive mass balances embedded in an overall negative trend over the past 105 years. The three periods of sustained positive mass balances (centered in the 1920s–1930s, in the 1980s and in the first decade of the 21st century) coincide with several documented glacier advances in this region. Similar trends observed in other shorter glacier mass-balance series suggest that the Echaurren Norte glacier reconstruction is representative of larger-scale conditions and could be useful for more detailed glaciological, hydrological and climatological assessments in this portion of the Andes. Published by Copernicus Publications on behalf of the European Geosciences Union. 928 M. H. Masiokas et al.: Reconstructing the annual mass balance of the Echaurren Norte glacier Figure 1. Panel (a) depicts a map of the Central Andes of Chile and Argentina showing the location of the Echaurren Norte glacier (ECH), Piloto Este glacier (PIL) and several smaller glaciers with mass-balance records in the Pascua Lama (PAS) and Cordillera de Colanguil (COL) areas. The locations of the snowpack and streamflow stations discussed in the text are also shown (Tables 1 and 2). In panel (b) a general view of the El Yeso area, showing the location of ECH, El Yeso Dam and the associated meteorological station is given. Laguna Negra is a natural lake that receives the meltwater from ECH. Base image acquired on 5 January 2014 and downloaded from Google Earth. Panel (c) provides a closer oblique view of Echaurren Norte as observed in 2014 and in the early 1970s (outlined in red and based on Peña and Narbona, 1978). Note that the glacier has remained in roughly the same position but has thinned markedly over the last decades. Panel (d) shows seasonal variations in temperature and precipitation at the lower reaches of ECH (3700 m a.s.l.) extrapolated from the El Yeso meteorological station (see Sect. 2.2 for details). Note that the bulk of precipitation occurs during the coldest months of the year (December–March precipitation only accounts for ∼ 5 % of the mean annual totals).

).In panel (b) a general view of the El Yeso area, showing the location of ECH, El Yeso Dam and the associated meteorological station is given.Laguna Negra is a natural lake that receives the meltwater from ECH. Base image acquired on 5 January 2014 and downloaded from Google Earth.Panel (c) provides a closer oblique view of Echaurren Norte as observed in 2014 and in the early 1970s (outlined in red and based on Peña and Narbona, 1978).Note that the glacier has remained in roughly the same position but has thinned markedly over the last decades.Panel (d) shows seasonal variations in temperature and precipitation at the lower reaches of ECH (3700 m a.s.l.) extrapolated from the El Yeso meteorological station (see Sect. 2.2 for details).Note that the bulk of precipitation occurs during the coldest months of the year (December-March precipitation only accounts for ∼ 5 % of the mean annual totals).

Introduction
The extratropical Andes, between ∼ 23 and 55 • S, contain a large number and variety of glaciers ranging from small glacierets at elevations of over 6000 m in the high, arid Andes of northern Chile and Argentina, to large outlet glaciers that reach the sea in the humid southwestern portion of Patagonia and Tierra del Fuego.Altogether, these ice masses concentrate the largest glacierized area in the Southern Hemisphere outside Antarctica and are highly valued as sources of freshwater, as indicators of climatic change, as tourist attractions and as environmental and cultural icons in different sectors of the Andes.As reported for other mountainous areas of the globe, glaciers in southern South America display a widespread retreating pattern that has usually been attributed to warmer, and sometimes drier, climatic conditions in this region (Villalba et al., 2003;Rignot et al., 2003;Rivera et al., 2000Rivera et al., , 2005;;Masiokas et al., 2008Masiokas et al., , 2009;;Le Quesne et al., 2009;Pellicciotti et al., 2014).Quantitative assessments of regional glacier mass-balance changes and glacier-climate relationships are, however, seriously hampered by the scarcity and short length of in situ glacier massbalance data and proximal climate records within the Andes.The latest publication of the World Glacier Monitoring Service (WGMS, 2013) reports annual mass-balance measurements for seven extratropical Andean glaciers (five in Argentina, two in Chile).Four of these records start in 2010 and are for small glaciers and glacierets located ca. 29.30 • S, two records are located between 32 and 34 • S and start in the midto-late 1970s and the remaining record from Tierra del Fuego (54.8 • S) starts in 2001.Discontinued, short-term glacier mass-balance measurements (see e.g., Popovnin et al., 1999) and recent programs initiated at new sites (e.g., Rivera et al., 2005;Rabatel et al., 2011;Ruiz et al., 2013) complete the network of direct glacier mass-balance data currently available in southern South America.Although not optimal in terms of spatial coverage, arguably the single most important limitation of this network is the short period of time covered by consistent, reliable records.Of the two longest massbalance series mentioned above (Echaurren Norte and Piloto Este glaciers in the Central Andes, see Table 1.1 in WGMS, 2013), only the series from Echaurren Norte (ECH) in Chile (Fig. 1a-c) provides a complete record spanning more than 35 years.In fact, this series constitutes the longest direct glacier mass-balance record in the Southern Hemisphere (see Escobar et al., 1995a, b;DGA, 2010 andWGMS, 2013) and is thus a "reference" glacier in the WGMS global assessments.The mass-balance record from the Piloto Este glacier (PIL; located ca. 100 km to the north in Argentina; Fig. 1a) covers the 1979-2002 period and contains several data gaps that have been interpolated using various techniques (Leiva et al., 2007).
Many studies dealing with recent climate and glacier changes in southern South America have pointed out the shortness, poor quality or absence of climatic records at highelevation sites or in the proximity of glaciers in the Andes (Villalba et al., 2003;Rivera et al., 2005;Masiokas et al., 2008;Rasmussen et al., 2007;Falvey and Garreaud, 2009;Pellicciotti et al., 2014;Vuillle et al., 2015).Given the lack of suitable data, many climatic assessments have used records from distant, low elevation weather stations and/or gridded data sets to estimate conditions and recent climate variability within the Andean range.It is interesting to note, however, that the amount of hydroclimatic information (in particular from solid and liquid precipitation, and hydrologic variables) is comparatively better for those portions of the southern Andes that support large populated centers and where the water provided by the mountains is vital for human consumption, agriculture, industries and/or hydropower generation.In these areas, mainly between ca. 29 and 42 • S, local and national water resource agencies have monitored a well-maintained network of hydrologic and meteorological stations for several decades (see e.g., Masiokas et al., 2006Masiokas et al., , 2010)).The data from the stations in this region are slowly becoming publicly available and are substantially better in terms of quantity and quality than those for the less populated, more inaccessible areas in southern Patagonia or in the Desert Andes of northern Chile and Argentina.
The Central Andes of Chile and Argentina between ∼ 31 and 35 • S (see Lliboutry, 1998) have a mean elevation of about 3500 m, with several peaks reaching over 6000 m (Fig. 1a).The climate of this region is characterized by a Mediterranean regime with a marked precipitation peak during the cold months (April to October) and little precipitation during the warm summer season (November to March; Fig. 1d).Almost all of the moisture comes from westerly Pacific frontal systems, precipitating as rainfall in the Chilean lowlands and as snow in the Andes to the east (Miller, 1976;Aceituno, 1988;Garreaud, 2009).The snow accumulated in Comparison between the annual mass-balance series of ECH and regional records of maximum winter snow accumulation and mean annual river discharges in the Andes between 30 and 37 • S (see Fig. 1).The regional records are expressed as percentages with respect to the 1981-2010 mean values.Variations in annual total precipitation at Santiago are also included to highlight the strong common hydroclimatic signal in this region.
the mountains during winter remains frozen until the onset of the melt season (usually October-November), producing a unimodal snowmelt-dominated regime for all rivers originating on either side of the Andes at these latitudes (Masiokas et al., 2006;Cara et al., 2016).This relatively simple configuration entails some potential benefits for the study and understanding of the hydroclimatic and glaciological processes in this region: first, the strong covariability between total rainfall amounts measured in central Chile and winter snow accumulation and river discharges recorded in the Andes (see Fig. 2) allows for the use of a relatively limited number of station records to capture the main regional hydroclimatic patterns.The strong common signal among these variables also offers the possibility of inferring or reconstructing selected instrumental data (e.g., winter snow accumulation, which begins in 1951) using data from other well-correlated variables with a longer temporal coverage (e.g., Andean streamflow records which are available since 1909).Masiokas et al. (2012) used these relationships to extend Andean snowpack variations using central Chilean rainfall records and precipitation-sensitive tree-ring width series.
In contrast to the well-known similarities between precipitation (solid and liquid) and surface runoff, the spatial and temporal patterns of high-elevation temperature records in the Central Andes of Chile and Argentina are still poorly understood.Falvey and Garreaud (2009) presented a detailed assessment of temperature trends over the 1979-2006 period along the western margin of subtropical South America, reporting a notable contrast between surface cooling (−0.2 • C decade −1 ) in coastal stations and a warming trend of ca.+0.25 • C decade −1 in the Andes only 100-200 km inland.However, only two land stations were available with long enough records above 2000 m (i.e., El Yeso and Lagunitas stations in Chile at 2475 and 2765 m, respectively), but radiosonde data from the coastal station Quintero (ca.33 • S) showed comparable positive trends for the free troposphere www.the-cryosphere.net/10/927/2016/The Cryosphere, 10, 927-940, 2016 (Falvey and Garreaud, 2009).This lack of high-elevation surface-temperature data also restricted the recent assessments of Vuille et al. (2015), who focused their elevationdependent temperature trend analyses on the region north of 18 • S because data were too sparse farther south.The station El Yeso (33 • 40 36 S, 70 • 05 19 W) is located only 10 km south of ECH (Fig. 1b).Mean daily and monthly temperature and total precipitation measurements from this station have been available since 1962 but contain several months with missing data prior to 1977 (temperature) and 1975 (precipitation).Since 1977, both series are practically complete and updated on a regular basis.To our knowledge, in the entire extratropical Andes there is no other operational meteorological station with such a long and complete record of temperature and precipitation variations less than a few kilometers from a glacier, which moreover contains the longest ongoing mass-balance monitoring program in the Southern Hemisphere.This rare combination of relatively long, complete climate records near a well-studied glacier site clearly highlights the importance of this unique location for varied glaciological and climatological investigations in the southern Andes.
In this contribution we use seasonal mass-balance records from ECH plus locally and regionally averaged monthly hydroclimatic data to model and reconstruct annual glacier mass-balance changes over the past 105 years.Since only the glacier-wide seasonal and annual mass-balance components are available for ECH, one of the main objectives of the study was to explore the suitability of simple mass-balance models that require a minimum amount of input data (Marzeion et al. (2012); see also Kaser et al., 2010).Although this simplistic approach provides limited insight into the intricate physical processes involved in this glacier's intra-annual massbalance variations, it may, nonetheless, offer a useful starting point to address some basic (yet still poorly known) questions regarding the glacier's sensitivity to climate variations.We did not consider a data-intensive approach to measure and model the complex daily energy and mass-balance variations of this glacier (e.g., Pellicciotti et al., 2014), because of the lack of the high-resolution, in situ meteorological and glaciological measurements usually required in these types of analyses.Another primary objective was to use the available, well-correlated hydrological records from this region (Fig. 2) to extend the ECH annual mass-balance record and evaluate the fluctuations of mass balance over a much longer period than that covered by regular glaciological measurements.Comparisons with other shorter mass-balance series and with a record of glacier advances in this region suggest the resulting time series contains a discernible regional footprint.Overall, we believe the findings discussed below constitute a substantial improvement in the understanding of the main patterns and forcings of the glacier mass-balance changes in this region and provide a useful background for more detailed glacio-climatic assessments and modeling exercises in this portion of the Andes.

Glacier mass-balance data
The Echaurren Norte glacier (33 • 33 S, 70 • 08 W) is located within a southwestern-oriented cirque ∼ 50 km southeast of Santiago de Chile, in the headwaters of the Maipo River basin (Fig. 1a-c).ECH provides water to Laguna Negra, a natural lake that together with the nearby El Yeso artificial lake constitute crucial water reservoirs for extensive irrigated lands and for the metropolitan Santiago area in central Chile.
Mass-balance measurements started at this easily accessible glacier in the austral spring of 1975 under the auspices of Dirección General de Aguas (DGA), the institution in charge of monitoring and managing water resources in Chile.Summer and winter mass-balance data at ECH have been regularly measured until the present by DGA officials and have been reported in sporadic internal documents and scientific publications (Peña and Narbona, 1978;Peña et al., 1984;Escobar et al., 1995aEscobar et al., , b, 1997;;DGA, 2010).These records have also been reported to the WGMS, from which we obtained the 1975-2012 data used in this manuscript (annual massbalance data extend to 2013; see WGMS ( 2013) and http: //www.wgms.ch).The glacier has thinned in the last decades and presently consists of small remnants of both clean and debris-covered ice (Fig. 1c).Despite this evident ice-mass loss, the elevation range of the glacier has not changed much since measurements started in the mid-1970s.According to Peña and Narbona (1978) and Escobar et al. (1995a, b), in the first years of the mass-balance program the glacier covered an area of 0.4 km 2 distributed over a short elevation range between ca.3650 and 3880 m a.s.l.(Fig. 1c).Over the time period covered by the mass-balance records, no adjustment has been made to incorporate the changes in surface area of the glacier, thus the reported values are considered here as reference-surface mass-balance estimates (i.e., the mass balance that would have been observed if the glacier topography had not changed over the study period; see Cogley et al., 2011).
Mass-balance data from PIL from 1979 to 2002 and shorter time series from small glaciers and glacierets further north in this region are also available from the WGMS database (Leiva et al. (2007), Rabatel et al. (2011), WGMS (2013); see Fig. 1a and Table 1).Here we compare the cumulative annual mass-balance records of these glaciers as independent validation measures of the main patterns and temporal trends observed in the measured and modeled massbalance series from ECH.

Minimal glacier mass-balance model
A minimal model only requiring monthly temperature and precipitation data (Marzeion et al., 2012) was used to estimate the interannual surface mass-balance variations of ECH and to explore the relative importance of temperature The Cryosphere, 10, 927-940, 2016 www.the-cryosphere.net/10/927/2016/and precipitation variability for the ECH records.In their publication, Marzeion et al. (2012) used gridded precipitation and temperature data to calibrate individual models for 15 glaciers with existing mass-balance measurements in the greater Alpine region.The climate data used here come from El Yeso, a permanent automatic weather station maintained by DGA and located ca. 10 km to the south and 1200 m lower than ECH's snout (Fig. 1b).The data are freely available at the DGA website (http://www.dga.cl) and contain practically complete monthly temperature and precipitation records since 1977 (only four missing months were filled using their long-term means).The mass-balance model can be defined as follows: where MB represents the modeled annual specific mass balance of the glacier, P i are monthly total precipitation values at the El Yeso station and α is a scaling parameter introduced to compensate for the precipitation gradient between the elevation of this station (rounded here to 2500 m) and the front of ECH (fixed at 3700 m in this analysis).We do not differentiate solid vs. liquid precipitation, because at this glacier (and in other high-elevation areas in this portion of the Andes) the bulk of precipitation occurs during the winter months and the fraction of liquid precipitation is usually minimal compared to the large proportion that falls as snow (see Fig. 1d).The use of total precipitation values also avoids the additional complexity and uncertainties involved in differentiating solid from liquid precipitation at this glacier, which is distributed over a very small altitudinal range (see also Fig. 1c).T i represents mean monthly temperatures at El Yeso extrapolated to the elevation of the glacier front using a constant lapse rate of −0.065 • C/100 m and T melt is the monthly mean temperature above which melt occurs.As indicated in Marzeion et al. (2012), the maximum operator ensures that melting occurs only during months with mean temperatures above T melt .The parameter µ is expressed in mm K −1 and was introduced to translate the monthly temperature records into monthly ablation values at the glacier.In order to estimate the parameters α and µ and validate the final model, we performed a leave-one-out cross-validation procedure (Michaelsen, 1987).In this approach, ECH data for each year between 1977 and 2012 (common period between the El Yeso data and the ECH mass-balance series) were successively excluded and the minimal mass-balance model (Eq. 1) was calibrated with the remaining values.At each step the parameters α and µ were first optimized to minimize the root mean squared error (RMSE; Weisberg, 1985) of the modeled values and then used to estimate the mass-balance data omitted that year.This resulted in 36 predicted values which were compared to the actual annual mass-balance observations to compute validation statistics of model accuracy and error.The exercise showed that the model parameters are relatively time stable: α ranged between 3.9 and 4.1 (mean value used here = 3.9), whereas µ varied between 89.0 and 91.0 mm K −1 (mean value used = 90.1 mm K −1 ).
The mean estimated value of α indicates that accumulation at the glacier is normally about four times larger than the annual precipitation recorded at El Yeso.The mean estimated value for µ is also reasonable and within the range of values reported by Marzeion et al. (2012) for the 15 glaciers with direct measurements in the European Alps (76-156 mm K −1 , see their Table 1).Finally, for the sake of simplicity, we prescribed T melt = 0 • C as suggested in Marzeion et al. (2012).

Glacier mass-balance reconstruction
In addition to modeling the interannual mass-balance variations of ECH using the temperature and precipitation data from El Yeso, we also used regionally representative hydroclimatic indicators to extend the observed glacier massbalance record prior to 1975.The use of these indicators (regionally averaged series of winter snow accumulation and mean annual river discharges; see Masiokas et al., 2006) was supported by visual comparisons and correlation analyses which showed strong, statistically significant positive associations not only with the winter record at ECH, but also with the annual mass-balance series of this glacier (Table 2 and Fig. 2).The correlation was also positive but weaker between the summer component at ECH and the regional snowpack and streamflow series.
The regionally averaged record of winter snow accumulation is based on eight selected stations located in the Chilean and Argentinean Andes between 30 and 37 • S (Fig. 1a and Table 3).The data set has been updated from the one used by Masiokas et al. (2012) and contains the longest and most complete snowpack records in this region.Prior to computing the regional average, the individual series were expressed as percentages from their 1981 to 2010 climatology mean values.A similar approach was used to develop a regional record of mean annual (July-June) streamflow variations.This series was calculated using monthly data from 11 gauging stations with the longest and most complete records in this portion of the Andes (Fig. 1a and Table 3).The resulting snowpack and streamflow composite records cover the 1951-2014 and 1909-2013 periods, respectively (Fig. 2).
The glacier mass-balance reconstructions are based on simple linear regression models where the predictand is the 1975-2013 ECH annual mass-balance series and the predictors are, alternatively, the regional 1951-2014 snowpack and 1909-2013 streamflow records depicted in Fig. 2. Given the relative shortness of the common period between the predictor and predictand series (39 years), the reconstruction models were also developed using a leave-one-out crossvalidation procedure (Michaelsen, 1987).Here, linear regression models for each year were successively calibrated on the remaining 38 observations and then used to estimate the predictand's value for the year omitted at each step.A simple linear regression model based on the full calibration data set  was finally used to reconstruct the mass-balance values over the complete period covered by the regional time series.The goodness of fit between observed and predicted mass-balance values was tested based on the proportion of variance explained by the regression models and the normality, linear trend, and first-and higherorder autocorrelation of the regression residuals.The uncertainties in each reconstructed mass-balance value in year t (ε reco(t) ) were calculated integrating the standard error of the regression estimate (SE regr ) and the standard error of the mean annual streamflow values used as predictors in the model (SE mean(t) ).This latter error is derived from the standard deviation of the regional record (σ ) and increases as the number of contributing streamflow series (n) decreases back in time (see Table 3).
An independent verification of the reconstructed massbalance records was undertaken by comparing the cumulative patterns of these series with the cumulative mass balances reported for the Piloto Este glacier and for other glaciers with shorter mass-balance series available in this portion of the Andes (Fig. 1a and Table 1).We also compared the ECH cumulative series (observed and predicted) with a regional record of glacier advances identified during the 20th century in the Andes between 29 and 35 • S. The latter record was compiled in a recent review of glacier fluctuations in extratropical South America and is based on direct observations, reports from documentary evidence, and analyses of aerial photographs and satellite images from this region (see Masiokas et al., 2009).The uncertainty of the cumulative series modeled for ECH (ε (T ) ) were calculated by propagating (adding) the individual errors estimated for each reconstructed value.

Minimal glacier mass-balance model
The 1975-2012 winter and summer values observed at ECH are depicted in Fig. 3a.The winter series shows a long-term mean of 2.54 m w.eq.and a larger range of variability (SD 1.24 m w.eq.) than the summer series, which fluctuates around a long-term mean of −2.93 m w.eq.(SD 0.72 m w.eq.).The observed and modeled annual mass-balance series are remarkably similar (Fig. 3b) and show a strong positive correlation (r = 0.883, rmse = 0.77 m w.eq.), indicating that 78 % of the variance in the ECH record can be accounted for by the minimal model presented in Eq. ( 1).Both series show sim-ilar, slightly negative linear trends and negative means (−0.35 and −0.34 m w.eq.for the observed and modeled series, respectively) over the 1977-2012 interval.

Attribution assessments
In order to test which climate variable (temperature or precipitation) has a stronger influence on the annual mass-balance variations at ECH, the glacier mass-balance model was also run alternatively replacing the temperature and the precipitation monthly data by their long-term average values.The results from this analysis (Fig. 3c) suggest that precipitation variations constitute the dominant forcing modulating annual glacier mass balance at this site.Regardless of their different absolute values, the precipitation-driven estimates (blue dashed line in Fig. 3c) show a strong positive correlation (r = 0.882) and remarkable similarities with the ECH annual mass-balance series (red line).In contrast, the temperaturedriven estimates (dark-red dashed line) show a poorer correlation with the ECH record (r = 0.240) and a substantially lower inter-annual variability, which only barely follows the variations in the annual mass-balance series.To evaluate if the influence of temperature had been underestimated in the full model (where the parameters α and µ can compensate for each other), both parameters were also optimized individually using a leave-one-out approach and considering each term of Eq. ( 1) as separate models.In this case the parameters showed almost exactly the same mean values (3.8 for α and 90.3 mm K −1 for µ) as those obtained using the full model (3.9 and 90.1 mm K −1 for α and µ, see Sect.2.2), suggesting that the poor performance of temperature is not due to the interaction of the parameters in the mass-balance model.

Annual mass-balance reconstruction 1909-2013
Figure 4a shows the reconstruction of the ECH annual massbalance series based on the regional record of mean annual streamflows.The snowpack-based mass-balance reconstruction is not shown as it is significantly shorter than the streamflow-based series and shows virtually the same variations over their overlapping interval.The streamflowbased regression model (Table 4) is able to explain 68 %  of the variance in the annual mass-balance series over the 1975-2013 period and shows no apparent sign of model misspecification, offering the possibility of reliably extending the information on glacier mass-balance changes back to 1909.This reconstructed mass-balance record is almost three times longer than the mass-balance record currently available at ECH and shows a strong year-to-year variability embedded within several periods of overall positive or negative conditions (Fig. 4a).In particular, positive mass-balance conditions were reconstructed between 1914 and 1941, in the 1980s and in the late 1990s to the early 21st century.In contrast, the clearest sustained period of negative mass balances occurred between the 1940s and the 1970s.
The cumulative values of the streamflow-based massbalance reconstruction show very good correspondence with the observed cumulative series and an overall negative trend between 1909 and 2013 (Fig. 4b).Within this centurylong negative trend, a prominent period of extended positive mass balances can be observed between the mid-1910s and the early 1940s.After 1941 and during the following four decades, the cumulative mass-balance series shows an impressive decline that is interrupted in 1980 by a ∼ 10-year long period of sustained positive conditions (Fig. 4b).Since the early 1990s and until 2013 the cumulative mass-balance series resumes the negative tendency, only interrupted by a short-lived period of positive conditions in the first years of the 21st century.It is important to note, however, that ascribing absolute values to this reconstructed cumulative series is complicated and should be used with caution due to the large uncertainties involved and the fact that the model is calibrated using reference-surface mass-balance estimates (Cogley at al., 2011).Between 1975 and 2013 the lower elevation of the glacier did not change much (see Fig. 1c) and therefore the reference-surface and the conventional massbalance estimates are probably roughly equivalent.However, for earlier decades and without historical information on the glacier area and frontal position, it is difficult to estimate the impacts of changing glacier geometry on the actual mass balance of this glacier.1).

Comparison with other glacier records
Examination of the main patterns in the reconstructed cumulative mass-balance series shows a good correspondence with a regional record of glacier advances identified in the Central Andes over the past 100 years (Masiokas et al., 2009;Fig. 4c).In most cases, the glacier advances are concentrated during or soon after the periods of sustained positive mass balances reconstructed or observed at ECH.This situation is particularly clear in the 1980s and 1990s, where a large number of glacier advances were identified during and/or immediately after the peak in mass balances that culminated in 1989 (Fig. 4b and c).Glacier advances were also identified in the 1930s, 1940s and 1950s, likely associated with the extended period of positive mass balances that culminated in the early 1940s.A few well-documented advances identified in this region between 2003 and 2007 may be associated with the minor peak in cumulated mass balances observed at the turn of the 21st century (Fig. 4b and c).
The cumulative variations in the modeled and observed mass-balance series from ECH are also very similar to those observed in the 1979-2002 cumulative record of PIL, providing additional support for the overall reliability of the rewww.the-cryosphere.net/10/927/2016/The Cryosphere, 10, 927-940, 2016 constructed time series (Fig. 5).The cumulative tendency of PIL appears to be smoother than the ECH series, but still shows slightly positive or near-equilibrium conditions between the late 1970s and the mid-1980s followed by a sharp decline until the turn of the 21st century.The cumulative series from other glaciers located further north in the Pascua Lama and Cordillera de Colanguil areas (Fig. 1a and Table 1) only cover the last decade or so of the ECH record.However, in all cases their overall tendency is similar and markedly negative, reflecting the sustained unfavorable conditions that these ice masses have endured in recent years.It is interesting to note that the smaller glaciers (Table 1 and Fig. 5) are the ones consistently showing the steepest negative cumulative trends whereas the largest glacier (Guanaco glacier, with ca.1.8 km 2 in 2007) shows the least negative trend.

Discussion and conclusions
Compared to other mountainous glacierized areas, the extratropical Andes in southern South America contain one of the least complete networks of in situ glacier mass-balance and high-elevation climate records in the world.This scarcity of basic information in this extensive and glaciologically diverse region has been highlighted on many occasions and several recent studies have attempted to overcome this limitation by estimating mass-balance changes through remote sensing and/or modeling approaches of varied complexity and spatial coverage (e.g., Casassa et al., 2006;Radić et al., 2013;Lenaerts et al., 2014;Pellicciotti et al., 2014;Schaefer et al., 2013Schaefer et al., , 2015)).With such limited data availability, the few existing glacier mass-balance records become particularly relevant as they provide crucial information and validation measures for many glaciological, climatological and hydrological analyses.
In this paper we analyzed an up-to-date compilation of the longest and most complete in situ glacier mass-balance and hydroclimatic records from the Andes between 29 and 37 • S to address some basic (yet poorly known) glaciological issues in this region.First, we show that it is possible to estimate annual glacier mass-balance changes using very simple modeling approaches.Results from a minimal model requiring only monthly temperature and precipitation data (Eq. 1) revealed that up to 78 % of the variance in the ECH annual mass-balance series between 1977 and 2012 could be captured simply using available records from the El Yeso station, ca. 10 km from the glacier (Figs.1a and 3b).Winter precipitation variability appears to be the dominant forcing modulating annual mass balances at ECH, with temperature variations likely playing a secondary role (Fig. 3c).This is particularly interesting because it contrasts with the findings in other regions where the recent glacier behavior is generally more strongly related to changes in temperature instead of precipitation (e.g., Marzeion et al., 2012).However, although Peña and Narbona (1978) also noted a dominant influence of the winter accumulation term on the resulting annual mass balance of this glacier, the results should be assessed with caution given the simplistic nature of our model and the various factors that ultimately affect the annual mass balance at this site.For example, more detailed assessments should also consider the impact of sublimation on the mass balance of glaciers in this high-arid portion of the Andes (McDonell et al., 2013;Pellicciotti et al., 2014).
To test the reliability of the temperature records used to model the glacier mass-balance series, we correlated the El Yeso monthly temperature record with ERA-Interim gridded reanalysis temperatures for the 700 mb geopotential height (roughly 3000 m a.s.l.) and also with a 0 • C isotherm elevation series available from central Chile (Fig. 6).The El Yeso temperature record shows strong positive correlations with ERA-Interim gridded data over an extensive region that includes central Argentina, central Chile and an adjacent area in the Pacific Ocean (Fig. 6a).The El Yeso temperature record also shows clear similarities and a positive significant correlation with the 0 • C isotherm elevation series over the 1977-2004 interval (Fig. 6b and c).The independence of these three data sets indicates that the El Yeso mean monthly temperature data are reliable and that the poor performance of this variable in the mass-balance modeling exercise is not related to the overall quality of the temperature series.Although this issue is beyond the main purposes of this study, more complex modeling approaches are also needed to evaluate whether climate data at higher temporal resolution (instead of monthly values as used here) are capable of capturing a larger percentage of the mass-balance variations observed at ECH.
Annual mass-balance variations observed at ECH can also be reproduced or estimated accurately through simple linear regression using regionally averaged winter snowpack or annual streamflow records as predictors (Fig. 4a).This is due to the existence of a strong common hydroclimatic signal in this region, which results in very similar interannual variations in winter snow accumulation, mean annual river discharges and glacier mass-balance changes such as those measured at ECH (Fig. 2).This simple approach allows extending the information on glacier mass-balance changes several decades prior to the beginning of in situ measurements (back to 1909) and offers the opportunity of putting the existing glacier record in a longer-term perspective.Many of the extreme values reconstructed in this study have been documented in historical reports and recent analyses of instrumental hydroclimatic data.For example, the extreme positive values of 1914 and 1919 coincide with extremely wet winters in central Chile (see e.g., Fig. 2; Taulis, 1934;Masiokas et al., 2012), whereas the period with above-average balances centered in the 1980s or the negative conditions between the 1940s and 1970s have been identified, respectively, as the snowiest and driest intervals during the instrumental era in this region (Masiokas et al., 2010).Examination of the main intra-to multi-decadal patterns in this extended series also indicates that the sustained negative mass-balance conditions reported for ECH in recent years are not unusual and were probably surpassed by more negative and longer periods between the 1940s and 1970s (Fig. 4a).However, the impact of a few consecutive years of negative mass balances are more serious today than several decades ago because of the low volume of ice remaining and the poorer overall "health" of the glacier.
The cumulative series of the reconstructed mass-balance values (Fig. 4b) shows a steep negative trend that is consistent with the recent loss of ice reported for other glaciers in this region (Fig. 5; Escobar et al., 1995a) Rivera et al., 2000;Masiokas et al., 2009).This negative trend has been temporarily interrupted by periods of sustained positive mass balances that, in most cases, precede or coincide with recent glacier readvances identified at these latitudes in the Andes (Masiokas et al., 2009;Fig. 4c).The clearest example is the relationship between the peak in cumulative mass balances in the mid-to-late 1980s and the 11 documented glacier advances in the following decade.It is also interesting to note that several of the glacier events that occurred after periods of positive mass balances have been identified as surges (Helbling, 1935;Espizua, 1986;Masiokas et al., 2009;Pitte et al., 2016).The well-known surges of Grande del Nevado glacier (in the Plomo massif area) in 1933-1934, 1984-1985 and 2004-2007 are particularly noteworthy as they consistently occurred near the culmination of the three periods with overall positive mass balances in the 1920s, 1930s, 1980s and in the first decade of the 21st century (Fig. 4b).In agreement with the progressively smaller magnitude of these peaks in the cumulative mass-balance series, the three Grande del Nevado surges also showed a decreasing power and transferred progressively smaller quantities of mass from the upper to the lower parts of the glacier.Two recent surges of Horcones Inferior glacier in the nearby Mt.Aconcagua area also occurred in the mid-1980s and again between 2002 and 2006, suggesting a possible connection between the development of surging events and the periods with overall positive massbalance conditions in this region (Pitte et al., 2016).
The fact that only limited information is available for ECH, together with the use of reference-surface massbalance estimates, (see Sect. 2.1) poses interesting yet complicated questions regarding the applicability of this series in related glaciological and/or climatological assessments.Since reference-surface mass-balance variations are more closely related to changes in climate than the conventional mass balance of a glacier (Cogley et al., 2011), the reconstructed series discussed here is arguably more relevant to climate-change-related studies rather than hydrological studies.If the purpose is to evaluate the hydrological contribution of this ice mass over the last century, then conventional massbalance estimates are necessarily required to take the changing glacier geometry into account.In any case, and considering the relevance of the observed ECH series for regional, hemispheric and global mass-balance studies, a reanalysis (Zemp et al., 2013) of the entire mass-balance record would probably produce important worthwhile information to assess the hydrological impact of the recent ice-mass losses in this semi-arid region (e.g., Ragettli et al., 2014).This issue is particularly relevant due to the extended droughts experienced in recent years and the increasing socioeconomic con-flicts over the limited water resources (almost entirely originating in the mountains) arising on both sides of the Andes.
Keeping these caveats in mind, the common pattern of strongly negative mass balances, the similarities with the few available glacier chronologies and the regional nature of the predictors used in the ECH reconstruction suggest that this series may nonetheless be considered representative (in relative terms) of the mass-balance changes during recent decades in other less-studied areas in this region.Reliable data from a larger number of glaciers, together with additional studies of the glacier-climate relationships are, however, still needed to support this hypothesis and to identify, for example, the main climatic forcings behind the recent glacier shrinkage observed in the Central Andes of Chile and Argentina (Masiokas et al., 2009).This is a challenging issue due to several factors, including the serious lack of glacier mass-balance series and high-elevation climate records, the complex dynamic response of individual glaciers to similar changes in climate and the great variety of glaciers existing in this region (Pellicciotti et al., 2014).The results discussed in this study offer a useful starting point to address the various pending issues mentioned above and will hopefully stimulate further glaciological, climatological and hydrological research in this poorly known mountainous region.

Figure 1 .
Figure 1.Panel (a) depicts a map of the Central Andes of Chile and Argentina showing the location of the Echaurren Norte glacier (ECH), Piloto Este glacier (PIL) and several smaller glaciers with mass-balance records in the Pascua Lama (PAS) and Cordillera de Colanguil (COL) areas.The locations of the snowpack and streamflow stations discussed in the text are also shown (Tables1 and 2).In panel (b) a general view of the El Yeso area, showing the location of ECH, El Yeso Dam and the associated meteorological station is given.Laguna Negra is a natural lake that receives the meltwater from ECH. Base image acquired on 5 January 2014 and downloaded from Google Earth.Panel (c) provides a closer oblique view of Echaurren Norte as observed in 2014 and in the early 1970s (outlined in red and based onPeña and Narbona, 1978).Note that the glacier has remained in roughly the same position but has thinned markedly over the last decades.Panel (d) shows seasonal variations in temperature and precipitation at the lower reaches of ECH (3700 m a.s.l.) extrapolated from the El Yeso meteorological station (see Sect. 2.2 for details).Note that the bulk of precipitation occurs during the coldest months of the year (December-March precipitation only accounts for ∼ 5 % of the mean annual totals).

Figure 2 .
Figure 2.Comparison between the annual mass-balance series of ECH and regional records of maximum winter snow accumulation and mean annual river discharges in the Andes between 30 and 37 • S (see Fig.1).The regional records are expressed as percentages with respect to the 1981-2010 mean values.Variations in annual total precipitation at Santiago are also included to highlight the strong common hydroclimatic signal in this region.

Figure 3 .
Figure 3. Panel (a) shows winter and summer values observed at ECH between 1975 and 2012.Panel (b) shows annual mass-balance series observed at ECH and modeled using El Yeso climate data (red and black lines).The estimated uncertainties of the modeled values (±2 RMSE) are shown with gray shading.In panel (c) Annual mass balances observed at ECH (red line) are compared to mass balances modeled using full variability in temperature but climatological monthly precipitation (dark-red dashed line), and full variability in precipitation but climatological monthly temperatures (dark-blue dashed line).Note the greater similarities between the observed series and the precipitation-based mass-balance estimates.

Figure 4 .
Figure 4. Panel (a) shows a comparison between the annual massbalance record observed at ECH (red line) and the reconstructed series derived from regionally averaged streamflow data (blue line).The estimated uncertainty of the reconstructed series (±2 ε reco ) is indicated by gray shading.Panel (b) shows the cumulative record of the observed and reconstructed ECH mass-balance series (darkred and dark-blue lines).The initial value of the observed ECH cumulative record was modified to match the corresponding value in the reconstructed series.The aggregated errors in this series (see Sect. 2.3) are also shown by gray shading.Panel (c) depicts glacier advances identified in the central Andes of Chile and Argentina during the past 100 years (see text for details).Events are grouped into 10-year intervals.

Notes: adj r 2 =
adjusted coefficient of determination used to estimate the proportion of variance explained by regression; F = F ratio for ANOVA test of the null hypothesis that all model coefficients are 0; SE = standard error of the estimate; RMSE = root mean squared error of regression.b 0 = constant of regression model; b 1 = regression coefficient; DWd = Durbin-Watson d statistic used to test for first-order autocorrelation of the regression residuals; Port.Q = Portmanteau Q statistic to test if high-order autocorrelation in the regression residuals is different from 0; ns = results are not statistically significant at the 95 % confidence level; * = statistically significant at the 99 % confidence level.

Figure 5 .
Figure 5.Comparison between the cumulative patterns in the observed and reconstructed records from ECH and other glaciers with available direct mass-balance data in the Dry Andes of Chile and Argentina (Fig. 1 and Table1).

Figure 6 .
Figure 6.Panel (a) depicts a map showing the correlations (p < 0.1) between mean warm season (October-March) temperatures at the El Yeso station and gridded warm season ERA-Interim mean temperatures for the 700 mb geopotential height level over the 1979-2012 period.The black star marks the location of the El Yeso station.The diagram in (b) shows variations of mean monthly temperatures at El Yeso and the mean monthly elevation of the 0 • C isotherm (ZIA) derived from radiosonde data from the Quintero coastal station.To facilitate the comparison, both series are expressed as anomalies from their mean seasonal cycles.Panel (c) shows a scatterplot of the El Yeso temperature and ZIA anomalies depicted in (b).Note the positive, highly significant correlation between these two variables.ZIA data were provided by J. Carrasco from Dirección Meteorológica de Chile.

Table 1 .
Basic information of the glacier mass-balance series used in this study.

Table 2 .
Correlation analyses between the ECH mass-balance series and regional hydroclimatic records.The number of observations used in each correlation test is indicated in parenthesis.

Table 3 .
Stations used to develop regionally averaged series of mean annual river discharges and winter-maximum snow accumulation for the Andes between 30 and 37 • S. Mean annual streamflow values refer to a July-June water year.
Masiokas et al. (2013)climatology values for each station are expressed as mm w. eq. for snowpack and as m 3 s −1 for streamflow.In the case of the San Juan and Cachapoal rivers, the mean values used correspond to the 1981-2007 and 1981-2001 periods, respectively.Data sources: (DGA) Dirección General de Aguas, Chile; (DGI) Departamento General de Irrigación, Mendoza, Argentina; (SSRH) Subsecretaría de Recursos Hídricos, Argentina.SeeMasiokas et al. (2013)for further details.

Table 4 .
Summary statistics for the simple linear regression models used to estimate ECH annual mass balances using regional snowpack and streamflow records.