Domain Editor-in-Chief: Oliver Chadwick; Geography Department, University of California, Santa Barbara, US


Introduction

The intensification of the global hydrological cycle that follows the atmospheric warming and moistening (Held and Soden, 2000; Allen and Ingram, 2002; Trenberth et al., 2005; Donat et al., 2016) is a notable footprint of the Anthropocene with severe and pressing consequences on human society and ecosystems (e.g., Rockström et al, 2014). However, even though mean global precipitation is expected to increase in response to the anthropogenic forcing, the patterns of precipitation change are far from being uniform and extensive regions are getting dryer, notably within the subtropics (Milly et al., 2005; Held and Soden 2006; Cai et al., 2012; Collins et al., 2013; Schewe et al., 2014; He and Soden, 2017). Most of these regions have limited water resources today and are therefore threatened by future rainfall decline and drought-related hazards (IPCC, 2012; Hagemann et al., 2013; Schewe et al., 2014; Jiménez Cisneros et al., 2014; Mekonnen and Hoekstra, 2016). Subtropical Chile may be viewed as a paradigmatic case of such vulnerable regions. A particularly consistent rainfall decline pattern is simulated in the southeast Pacific sector by current climate models in response to anthropogenic forcing (Collins et al., 2013; Christensen et al., 2013; He and Soden, 2017), some cases projecting deficits of up to 40% towards the end of the century (Polade et al., 2017). This regional effect of climate change would directly affect central-southern Chile, the region that concentrates most of the country’s population and economic activities.

Changes in the hydrological regime in Chile is recognized as a major risk for the country’s future development (MMA, 2016). Chile’s economy over relies on natural resources, being the most material-intensive among the member states of the Organization for Economic Cooperation and Development (OECD/ECLAC, 2016). The major economic sectors – notably agriculture and mining – and an important fraction of electric power generation (about 30%) depend on the fresh water supply via precipitation, which therefore constitutes a critical factor for the country’s societal functioning. Hence, potential loss of water resources via rainfall deficit and enhanced evaporation makes Chile particularly vulnerable to climate change (Vicuña et al., 2012, Demaria et al., 2013, Magrin et al., 2014; CR2, 2015; MMA, 2016; Bozkurt et al., 2018).

If global greenhouse gases (GHG) emissions continue unabated, the future climate in Chile could look like the condition observed from 2010 to the present. In this period, central-southern Chile experienced the longest and more extended drought since ca. 1900, with just one or two similar events in the last millennium (Garreaud et al., 2017). These recent years have been characterized by annual rainfall deficits ranging from 15 to 45%, a matching reduction in surface and underground water resources, detrimental impacts on vegetation and very active fire seasons (Garreaud et al., 2017; González et al., 2018). Unsurprisingly, this drought was perceived by most of the population, although its impacts in rural areas were more direct and more detrimental than in urban settings (Aldunce et al., 2017).

Notwithstanding the robustness of the drying scenario shown by climate model simulations in the southeast Pacific sector and central-southern Chile (e.g., Polade et al., 2017), the historical and future hydrological changes in this region have been sparsely evaluated within the global climate change context. Previous studies have reported drying patterns along the southwest coast of South America (e.g., Aceituno et al., 1993; Minetti et al., 2003; Haylock et al., 2006; Quintana and Aceituno, 2012) but few of them have assessed the influence of large-scale climate change on these trends (Vicuña et al., 2010; Cai et al., 2012; Purich et al., 2013). Vera and Díaz (2015) showed that the austral summer drying observed during most of the 20th century in southern Andes is consistent with that modelled in response to external climate forcing and to changes in GHGs, particularly, while no signal of change was found in simulations including only natural forcing. Contrasting an observational dataset and model simulations, Boisier et al. (2016) analyzed a recent period of particularly large rainfall decline in central Chile (1979–2014). The causes of the observed trends were found to be of both natural and anthropogenic origin, the latter accounting for about one third of the total signal. However, these results were limited to a relatively short recent period, and no distinction was made between seasons or between major anthropogenic drivers.

Here we present an integrated view of the hydrological changes in Chile. We describe first the climatology, variability, and long-term (1960–2016) trends based on an updated network of rain-gauge and river streamflow data. We then assess the large-scale context and the role of different drivers of change trough the analysis of a large ensemble of climate model simulations. Before concluding, we discuss the sensitivity of regional precipitation to global anthropogenic forcing, as well as the influence of natural variability on the historical changes.

Datasets and methods

Local precipitation and streamflow data

The observational data used in this study include local rain-gauge and streamflow records from the Weather Service (DMC) and Water Bureau (DGA) of Chile. This dataset was collected and homogenized by the Center for Climate and Resilience Research (available at: http://explorador.cr2.cl). Following the method of Boisier et al. (2016), we applied a quality-control and gap-filling procedures on the monthly precipitation and streamflow time-series from stations having records since at least 1980. After this pre-processing, records from approximately 600 rain and 400 streamflow gauging stations were available for short-term statistics (we use the 1980–2010 period to compute climatologies). The number of stations is drastically reduced to about 200 (precipitation) and 100 (streamflow) for longer-term trend calculations (1960–2016). To make a direct comparison between both variables, the river discharge measured as water volume flow were converted to mm per time unit. This conversion was performed by normalizing the streamflow data by gauge-contributing catchment area, which is provided by the CAMELS-CL dataset (Alvarez-Garreton et al., 2018).

Climate model data

To evaluate the historical and future effect of anthropogenic forcing on large-scale circulation and precipitation in the southeast Pacific region, we adopted a large dataset of fully-coupled simulations performed with Global Climate Models (GCMs; see Tables 1 and Table S1), most of them from the Climate Model Intercomparison Project Phase 5 (CMIP5; Taylor et al., 2012). Following the diagnoses presented, we separate the model dataset on three groups.

A first group of simulations from six GCMs is used to assess the effect of changing atmospheric concentration of GHGs and of stratospheric ozone (O3) on precipitation and sea-level pressure (SLP) during the 1960–2005 period, including three types of simulations. Two out of these three types of simulations correspond to runs forced with all external drivers (labelled historical in CMIP5) and runs prescribing the evolution of GHGs solely (historicalGHG). The third simulation type analyzed within this group is taken from CMIP5 ‘miscellaneous’ family (historicalMisc), whose configuration depends on the model (Table S1). In some cases, historicalMisc corresponds to simulations forced by a time varying O3 only. In other cases, the historicalMisc simulation includes all anthropogenic forcing excepting O3, cases in which the O3-driven climate change is derived as the difference between historical and historicalMisc. An additional numerical experiment, equivalent to historicalMisc with fixed O3 forcing, was performed with MIROC-ESM-CHEM. More than one run per simulation type were considered for most GCMs (Table S1). Further details on the model configuration and on this group of simulations can be found in Eyring et al. (2013).

Another dataset, including outputs from 13 GCMs, is used to contrast the historical changes driven by all external forcing (natural and anthropogenic) and by natural forcing only (incl. changes in solar irradiance as well as in aerosol concentration driven by major volcanic eruptions). The two types of simulations correspond, respectively, to the CMIP5 historical and historicalNat runs (Taylor et al., 2012).

Finally, a third group of GCM data is used to evaluate the historical and future precipitation changes, as well as to quantify the uncertainties of these changes. This group consists of a large ensemble from 34 GCMs, driven by all anthropogenic forcing. Within these simulations we have included CMIP5 historical runs from 1960 to 2005 and projections for 2006–2099, with several runs per simulation and GCM in many cases (Table S1). The model projections assessed correspond to simulations following the Representative Concentration Pathway (RCP) 8.5, scenario that consider a large radiative forcing of 8.5 W m–2 toward 2100 (rcp85; Taylor et al., 2012). We note that our study looks at the historical precipitation changes principally. Model rcp85 projections help us to quantify the modelled climate sensitivities to anthropogenic forcing, and the uncertainties of change under GHG emission rates similar to the current ones. To provide a range of plausible future changes in Chile is not the goal of this paper, so we do not include other scenario runs.

We recall that all GCMs involved in CMIP5 include time-varying O3 in their historical and future-scenario simulations, accounting for the O3 depletion in the last decades of the 20th century and the O3 recovery expected throughout the 21st century (Eyring et al., 2013). However, the models differ in the way O3 is considered. In some models, O3 is calculated within a fully-interactive chemistry and the concentration of this gas varies, within other factors, in response to Ozone Depleting Substances (ODS). In the other cases, the concentration of this gas is either computed by a semi-offline scheme or simply prescribed, in most cases following the scenario of Cionni et al. (2011).

Calculation of local and regional means trends

The historical hydro-meteorological changes in Chile are assessed using 1960 as the starting year for trend calculations. This choice is constrained by the need of (1) accounting for a period long enough to minimize the role of natural, decadal-scale climate variability (e.g., Boisier et al., 2016), and (2) having a critical number of local observations to make robust regional-scale statistics (few meteorological and streamflow stations have been operating since the first half of the 20th century).

The slope of a given variable regressed on time is used to quantify temporal trends. Records of observational sites having at least 90% of data available during the target period are included in the trend analysis. The standard error estimated for the regression parameters, and the 95% confidence interval derived for slope, are used to evaluate the statistical significance of a given trend. The null hypothesis of having no trend is rejected when the slope has low probability of being zero (p-value < 0.05).

Estimates of trend components caused by the Southern Annular Mode (SAM) and the Pacific Decadal Oscillation (PDO) are also included in the analysis (hereafter we will refer to these estimates as SAM- and PDO-congruent trends). We first compute a linear regression between detrended time-series of the assessed variable and that of the SAM and the PDO indices, to get the sensitivity to the changing phases of these phenomena. The sensitivity parameters obtained are then multiplied by the temporal trend of the corresponding mode of variability to derive the trend components. We adopted the index of Mantua et al. (1997) to characterize the PDO. Four SAM indices are included to assess the historical evolution of this phenomenon and its uncertainty. Three of them are constructed as the difference between the zonal mean SLP at 40°S and 65°S, using the monthly SLP data of two long-term reanalyses (NOAA-20CR and ERA-20C) and of the HadSLP2 product (Tables 1). The fourth index of the SAM included here is that of Marshall (2003), which is based on SLP recorded at key sites in the southern hemisphere. The same SLP-based definition of the SAM is used to get the mean state and evolution of this phenomenon in GCMs.

Table 1

Datasets used in this study. DOI: https://doi.org/10.1525/elementa.328.t1

Data type Period Source and reference Description

Local observations
Precipitation 1960–2016 DMC, DGA Monthly rain-gauge records
Streamflow 1960–2016 DGA, CAMESL-CL (Alvarez-Garreton et al., 2018) Monthly streamflow records, watershed boundaries
Reanalysis and products
Sea-level pressure 1960–2016 NOAA-20CR (Compo et al., 2011), ERA-20C (Poli et al., 2013), HadSLP2 (Allan and Ansell, 2006) The two reanalyses of the 20th century and the HadSLP2 dataset are used evaluate the historical trends in SLP trends and of the SAM index.
Global simulations
Historical all forcing 1960–2005 CMIP5 (historical; Taylor et al., 2012) This dataset includes full-forced historical runs from 34 GCMs. See supplementary Table S1 for details.
Historical natural only 1960–2005 CMIP5 (historicalNat; Taylor et al., 2012) Set of simulations from 13 GCMs accounting for natural climate forcings only. See supplementary Table S1 for details.
Historical GHG only 1960–2005 CMIP5 (historicalGHG; Taylor et al., 2012) Set of simulations from 6 GCMs accounting for changes in GHG forcing only. See supplementary Table S1 for details.
Historical O3 only 1960–2005 CMIP5 (historicalMisc; Taylor et al., 2012), others Set of simulations from 6 GCMs accounting for changes in stratospheric O3 only, or for all forcings excepting O3. See supplementary Table S1 for details.
Projections RCP8.5 2006–2099 CMIP5 (rcp8.5; Taylor et al., 2012) Future simulations under the RCP8.5 scenario from 34 GCMs. See supplementary Table S1 for details.

The trend calculation is applied in the same way in time-series of regionally-averaged variables. In this study, we focus on a comparison between the observed data and the large-scale precipitation simulated in GCMs over a domain spanning from 30°S to 48°S. This domain, including Mediterranean-climate and wet regions in central-southern Chile, is where model simulations show a predominant drying tendency in response to anthropogenic forcing (Collins et al., 2013). The model data are averaged over a box spanning these latitudes and the longitudes 80–70°W. Given the dissimilar distribution of precipitation and streamflow stations across Chile, the regional averages of local observations are first computed over smaller domains. The 30–48°S latitudinal band is divided north-south into four sub-regions of 4.5-degree of latitude. The averages over the sub-domains are then re-averaged over the larger domain, ensuring a better spatial representation.

A regional mean trend analysis based on local precipitation and streamflow data is done for two domains in central (30–39°S) and southern (39–48°S) Chile, respectively (indicated in Figure 4). The analysis of these smaller regions is done to distinguish changes in the Mediterranean and wet climate regimes in Chile, which roughly correspond to the central and southern boxes defined. Moreover, the streamflow records available for trend analysis are only available in central Chile. The regional means over these two smaller boxes in central and southern Chile, respectively, are computed from a pair of sub-regions of 4.5-degree of latitude.

Given that precipitation and streamflow records may include missing data, the regional averages of these variables are constructed from anomalies rather than from net values. Hence, a missing monthly value in a given station is estimated to have the mean anomaly off all other stations within the corresponding sub-region. All anomalies are computed with respect to the climatological mean of the 1980–2010 period (Figure 1).

Figure 1 

Precipitation and streamflow climatology in Chile. Mean annual precipitation (a) and normalized streamflow (b) in local observational sites. The contribution of austral winter semester (April–September) to annual totals for each variable are shown in panels (c) and (d), respectively. All values are computed in stations having at least 20 years of records within the period 1980–2010. DOI: https://doi.org/10.1525/elementa.328.f1

To further alleviate the scale mismatch between the local and GCM data, when compared, observed and simulated rainfall trends are expressed as changes relative to the corresponding climatological means.

Calculation of uncertainties driven by internal variability and climate sensitivity

Following a similar approach than Hawkins and Sutton (2011), we assess the modeled uncertainties for the precipitation trends presented here. We make use of the above-described large model dataset available for the CMIP5 historical and rcp85 simulations, including several runs per GCM (see Table S1). The trend differences between runs performed in a single model are used to quantify uncertainty due to internal variability. The between-runs variance obtained on average from all GCMs having more than one run per simulation type is our estimate of the model-mean spread due to interval variability (SIA2). A total variance (STOT2) is also computed from the complete ensemble of simulations. We then consider that STOT2 results from the internal variability (SIA2) and a climate sensitivity (SCS2) components. That is:

STOT2=SIA2+SCS2
(1)

Hence, the residual variance (STOT2 minus SIA2) is our estimate of the spread fraction driven by the different climate sensitivities in GCMs. Finally, we use the standard deviation as the spread metric of these two components, which are considered to be SIA2/STOT and SCS2/STOT, respectively.

Precipitation and streamflow mean state and variability in Chile

Because of its elongated shape, and the relative influence of the South Pacific and the Andes Mountains, continental Chile features several climatic zones. This heterogeneity manifests primarily as a strong north-south precipitation gradient, forming contrasting environments such as the extremely arid Atacama Desert and regions in southern Chile receiving more than 3000 mm/year on average (Figure 1). The orographic effect on precipitation induced by the Andes defines another noticeable character of the hydrological regime in Chile, in particular across the central and southern regions of the country, where precipitation on the mountain slopes can be two-to-three times larger than at lower elevations (Viale and Garreaud, 2015).

Precipitation in Chile takes place mostly within frontal systems embedded in the southern hemisphere westerly wind belt, which reach the continent from the southeast Pacific (e.g., Falvey and Garreaud 2007). As such, the seasonal variation in the latitudinal position and intensity of the storm track produces a marked seasonality in most regions of the country (Figure 1c). This seasonality is at its peak at subtropical latitudes (25–35°S), where more than 90% of the annual precipitation occurs during the austral winter semester (May to September), and weakens southwards towards a regime with a more even winter and summer precipitation in austral Chile. The precipitation regime of the Andean Plateau, at the northeastern limit of the country (Chilean Altiplano), is different from the rest of the country. The summer season peak that characterizes this regime responds to moisture advection from the interior of the continent and enhanced convection with high temperatures, which are partially controlled by the South American Monsoon (Aceituno 1998; Garreaud et al., 2003).

The amplitude and geographical patterns of annual, basin-wide mean, water flow derived from streamflow data (see Datasets and methods) match closely those of precipitation (Figure 1b), except in the arid regions with very low runoff coefficients. However, both variables differ in their seasonality, notably along the Andes foothills. The river discharges in these regions, with a flow maximum shifted towards spring and summer, characterize the predominantly snow-melt source of the high mountain watersheds. The spring and summer time release of water stored in the seasonal snowpack is of critical value for arid and semi-arid regions in Chile, providing most of the water supply for human consumption, irrigation and other economic activities (e.g., Vicuña et al., 2012).

Figure 2 synthesizes relevant aspects of the inter-annual precipitation variability in Chile. To dampen the biases imposed by the contrasted rainfall regimes described above, the typical amplitude of the year-to-year fluctuation is quantified through a coefficient of variation, derived as ratio of the standard deviation of the annual precipitation to the climatological mean (Figure 2a). This metric reveals a strong variability in central and northern Chile, where the amplitude of typical annual anomaly exceeds 50%. From a statistical point of view, the large variability affecting these arid and semi-arid regions responds to the fact that the annual accumulation integrates precipitation from only few events and during a short period of the year (Figure 1c), being more prone to large departures from the average. By contrast, regions with rainfall during most of the year, as in southern Chile, have a larger probability of experiencing different conditions from season to season, smoothing year-to-year fluctuations. Given these different degrees of variability, the detection (and attribution) of long-term hydrological signals represents a harder task in the northern regions than in the southern counterpart.

Figure 2 

Interannual variability of precipitation. Standard deviation of annual precipitation normalized against the mean annual accumulation (a). Covariance of annual precipitation with indices of ENSO SST 3.4 (b) and of the southern annual mode (c). R stands for Pearson’s correlation between the corresponding variables. DOI: https://doi.org/10.1525/elementa.328.f2

Beyond the seasonal aspect, there are important external controls on the precipitation regime in Chile, exerted by large-scale modes of variability. Within these modes, El Niño-Southern Oscillation (ENSO) and SAM are known as major sources of climate variability in southern South America at seasonal and inter-annual time scales, notably for precipitation in Chile (Aceituno 1988; Rutllant and Fuenzalida, 1991; Montecinos and Aceituno, 2003; Silvestri and Vera, 2003; Gillet et al., 2006; Garreaud et al., 2009). The regional rainfall sensitivity to the phase and amplitude of these large-scale phenomena, particularly the one related to SAM, is a key factor to account for when evaluating long-term changes in Chile and is, therefore, re-assessed with the observational dataset used here.

In line with previous results, a clear influence of ENSO and SAM on annual precipitation is observed in the central and southern regions of Chile (Figure 2b, c). The warm ENSO phase is associated to wet conditions in most of the country, but is a particularly important driver for the center-northern regions (25–35°S). In turn, the positive phase of the SAM leads to negative rainfall anomalies of particularly large amplitude in central-southern regions (35–45°S). At annual time scales, these modes explain about 30% of the rainfall variance in some regions, but more geographically defined and stronger correlation patterns are found when considering seasonal averages (not shown). We note that the indices used here for ENSO and SAM (Datasets and methods) are not independent (R2 ~ 15%), so the fractions of precipitation variance explained by these modes of variability are not completely additive. Moreover, the influence of both ENSO and SAM depend on the definition or index considered (notably in the case of SAM; see e.g. Ho et al., 2012).

Given the relatively small size of watersheds in Chile, the streamflow anomalies in a given hydrological year (considered typically from April to March) follow tightly the precipitation anomalies of the corresponding calendar year. Given this strong relationship, the analysis done to characterize precipitation variability (Figure 2) leads to very similar results when applied to streamflow data (not shown).

Observed trends in precipitation and streamflow (1960–2016)

Annual mean trends

To introduce the trend analysis, in Figure 3 we compare time series of annual precipitation from three rain-gauge stations in central-southern Chile, usually taken as references because of their reliable, complete and century-long records. Three time-series of watershed-normalized annual streamflow are also shown in Figure 3. In this case, the streamflow measuring sites were selected with the criteria of having long records (50+ years), being located at similar latitudes than the three rain-gauge stations assessed, and including Andean sectors within their contributing catchment areas. The differences in the mean precipitation and in the year-to-year variability, from Quinta Normal in Santiago (33.4°S, ~300 ± 150 mm yr–1) to Carriel Sur in Concepción (36.8°S, ~1000 ± 200 mm/yr) to El Tepual near Puerto Montt (41.4°S, ~1700 ± 300 mm/yr), illustrates the latitudinal gradient described previously. Further, the contrast between the rain-gauge mean precipitation and the normalized streamflow at similar latitudes evidences the increasing rainfall with height (the mean elevation of the corresponding watershed is indicated in Figure 3). Given rainfall-runoff mechanisms, a noticeable positive correlation is observed between precipitation and streamflow time series. This co-variability also reveals the control of large-scale phenomena on both variables in the central-southern Chile (note, e.g., the positive and negative anomalies reached during the El Niño – La Niña transition in 1997–1998).

Figure 3 

Precipitation and streamflow in selected locations of center-southern Chile. Time-series of annual precipitation measured at stations Quinta Normal(a), Carriel Sur(c) and El Tepual(e). Normalized annual streamflow at Rio Aconcagua En Chacabuquito(b), Rio Bio Bio en Rucalhue(d) and Rio Tolten en Villarica(f). Dashed lines shown the corresponding linear regression model for the 1960–2016 period. For each case, the temporal trend is indicated along with a two-sided p-value, indicating the probability of having a null slope. DOI: https://doi.org/10.1525/elementa.328.f3

A predominantly negative tendency within the available periods can be noticed in the precipitation and streamflow records shown in Figure 3. Qualitatively, the long-term rainfall decline at El Tepual is large in amplitude compared to the relatively weak inter-annual rainfall variability in this location. In contrast, at Quinta Normal there is no clear long-term signal emerging from a background of large rainfall variability. For a quantitative comparison, the linear trends computed between 1960 and 2016 are indicated in Figure 3. Consistently, no long-term changes were found for the rainfall data at Quinta Normal nor for the streamflow at Rio Aconcagua en Chacabuquito. All other cases show negative trends, of particularly large amplitudes in the rainfall records at El Tepual and in streamflow at Rio Bio Bio en Rucalhue (>5% per decade).

To have a robust diagnosis of hydrological changes at a regional scale, the linear trends were computed in all rain-gauge and streamflow records satisfying the minimum data requirement within the period assessed (Datasets and methods). The mean annual precipitation and streamflow trends obtained for 1960–2016, although with some important between-sites discrepancies in the amplitude and –in few cases– in the sign of change, confirm a dominating drying pattern in most central-southern Chile (Figure 4). Consistent with the weak signal-to-noise value seen in Santiago (Figure 3a), the rain-gauge records in central Chile lead mostly to negatives but non-statistically significant rainfall trends (–1.7% per decade on average; Figure 4a). An interesting result to highlight is that the streamflow data within the same region indicate a decline larger in amplitude (–3.8% per decade) than the one derived for precipitation (Figure 4b). Since runoff is a fraction of precipitation, an amplification in streamflow trends –seen in relative term– would be expected if there were no changes in other components of the water budget. The dependency of streamflow on other processes, including evapotranspiration, snow accumulation, vegetation changes and land use may further explain distinct hydrological responses to precipitation changes. Following this stronger signal, the streamflow data also show a larger fraction of cases with trends statistically different from zero. Yet, there is also a large spread within the streamflow trends derived in this region (a standard deviation of 4.3 mm per decade), including a number of sites with positive values.

Figure 4 

Changes in annual precipitation and streamflow. Linear trends of annual precipitation (a) and streamflow (b) based on local observations from 1960 to 2016. Large markers indicate locations where the corresponding trends are significantly different from zero, with a confidence interval of 95%. Boxes indicate the two domains used for regional averages, in central (30–39°S) and southern (38–48°S) Chile, respectively. *Trends are computed in stations having valid data in at least 90% of the period assessed. DOI: https://doi.org/10.1525/elementa.328.f4

In southern Chile, the few rain-gauge stations with long-term records indicate a larger and more robust regional drying signal (Figure 4a). Almost all stations show negative trends of amplitudes exceeding 5% per decade. Most of these trends are also significantly different from zero. Unfortunately, no streamflow gauging stations with enough data are available south of 40°S for an assessment of long-term river discharge changes in southern Chile.

In addition to the drying pattern observed in central-southern Chile, our results indicate positive rainfall trends north of 30°S and in Punta Arenas (52°S) in austral Chile (Figure 4a). In the first case, although the hypothesis tests do not indicate any robust signal of change, the positive rainfall trends agree in sign with the streamflow trends observed at similar latitudes (Figure 4b). As noted above, the large variability in the semi-arid northern Chile does not allow a proper assessment of change with the methods adopted here. Given the low rainfall frequency in this region, the episodes of precipitation may be considered as extreme events. Therefore, the changes in this regime must be approached with a suitable technique and dataset (daily means at least). In Austral Chile, it is the lack of long-term meteorological data that hinders a clear diagnostic of change at the scale of the region.

Seasonal mean trends

We now look at the linear precipitation and streamflow trends separated by season and region. The focus is on two regions located in central and southern Chile, respectively (Figure 4; see also Datasets and methods). The seasonal mean precipitation and trends are shown both in absolute and relative terms for the two regions assessed in Figure 5. Given the strong rainfall seasonality in central Chile, the drying trend seen at annual scales in this region responds principally to changes during the rainy season (JJA), which averages –9.5 mm (–2%) per decade. Yet, the changes seen in relative terms indicate a stronger change in spring (SON) and summer (DJF). We note here that given the very dry summer condition in central Chile, any change compared to the climatology should be interpreted carefully.

Figure 5 

Regional and seasonal mean precipitation trends. Climatological mean precipitation (a, b) and linear trends for the period 1960–2016 (c, d), derived from rain-gauge records located in central (30–39°S, a, c) and southern (39–48°S, b, d) Chile. Grey bars and red dots indicate absolute and relative trend values. Error bars indicate the trend interquartile range obtained within the stations of the corresponding region. DOI: https://doi.org/10.1525/elementa.328.f5

In southern Chile, the negative rainfall tendency represents a robust signature in all seasons (Figure 5d). Seen in relative terms, the trends expose a clear seasonality of change in this region, with a summer decline (~8% per decade) two times stronger than in winter (~4% per decade). As we discuss in the section that follows, these seasonal differences very likely reveal the additive influence of GHG and stratospheric O3 forcings in this region.

Regional and seasonal mean trends in river discharge are also computed using the long-term records available in central Chile (Figure 6). To inquire into potential differences between rain and snowmelt-dominated regimes, the streamflow stations used for regional averages were split between those located below and above 500 m a.s.l. Even though the streamflow observed at low elevation accounts for the mountain hydrology, precipitation contribution becomes predominant and the mean streamflow seasonality follows closely the precipitation regime, with a clear maximum in JJA (Figure 6a). The 1960–2016 trends, in this case, are fairly consistent with those obtained for precipitation in central and southern Chile (Figure 5c), showing a change towards a dryer condition, of particularly large amplitude in DJF (~–8% decade) and SON (~–5.5% decade).

Figure 6 

Regional and seasonal mean streamflow trends. Climatological mean streamflow (a, b) and trends for the period 1960–2016 (c, d), derived from gauge records located in central Chile (30–39°S). Regional statistics are based on stations located below (a, c) and above (b, d) 500 m a.s.l. Grey bars and red dots indicate absolute and relative trend values. Error bars indicate the trend interquartile range obtained within the stations of the corresponding region. DOI: https://doi.org/10.1525/elementa.328.f6

The hydrological seasonal pattern of change is even more pronounced when looking at the streamflow data from stations located above 500 m a.s.l (Figure 6d). In this case, a strong drying trend in summer and spring contrasts with the weak to even positive changes observed in winter. This particular behavior may reveal the relative influence of the Andes warming on snow accumulation and melt, leading to a shift in the runoff maximum toward earlier dates in the year (Bozkurt et al., 2018).

Large-scale drivers of change, anthropogenic forcing and future scenario

PDO and SAM-induced precipitation trends in Chile

As shown in Figure 2, both ENSO and SAM have a large influence on precipitation in Chile. One can maintain two differing perspectives on this influence. On one hand, given that ENSO and SAM provide an important source of variability, they add noise to local climate series, making trend detection much harder. On the other hand, if well-known, a large-scale mode of variability such as ENSO and SAM can be quantified, reducing the uncertainties in the trend attribution. For instance, Boisier et al. (2016) have shown that low-frequency ENSO-like variability in the Pacific Basin, characterized through the PDO, explains about 50% of the rainfall decline observed in central Chile in the period 1979–2014. The remaining fraction of this trend was unlikely driven by other natural phenomena, but was consistent with the simulated response to the historical anthropogenic forcing. Following a similar methodology, we explore the causes of the long-term (1960–2016) precipitation trends in Chile.

We first look at the changes in precipitation that are estimated to be forced by both the PDO and the SAM (see Datasets and methods). It is important to recall here that the evolution of these two phenomena during the period assessed is of different nature. The anthropogenic forcing likely modifies the frequency and amplitude of cold and warm events of ENSO or ENSO-like cycles such as the PDO (Cai et al., 2014), but there is no clear evidence showing a pattern of change favoring a particular phase of these phenomena. As such, we consider the trends in the PDO index as natural, transitory changes of its phase. By contrast, beyond the inherent intra-seasonal variability of the SAM, all indices used to quantify this phenomenon evidence a drift towards it positive phase during the last decades. This trend has been shown to be driven by the major global anthropogenic forcings, in particular the stratospheric ozone depletion (Cai et al., 2003; Shindell and Schmidt, 2004; Arblaster and Meehl, 2006; Karpechko et al., 2008; Fogt et al., 2009; Polvani et al., 2011; Thompson et al., 2011; Gillett et al., 2013). Therefore, we interpret the SAM-congruent rainfall trends as mostly an anthropogenic signature.

The PDO and SAM-congruent precipitation trends from 1960 to 2016 are shown for central-southern Chile in Figure 7. In annual time scales, the PDO does not force any significant change in this part of the country because the PDO itself does not exhibit a clear trend between 1960 and 2016 (period that includes both positive and negative PDO phases). This result contrasts with the strong drying in Chile led by the same phenomenon in the period 1979–2014 (Boisier et al., 2016). On the contrary, the trends in annual precipitation that are estimated to be driven by the SAM between 1960 and 2016 show a clear drying signature in most regions of central-southern Chile (Figure 7b), of similar or even larger amplitude than the actual observed trends (Figure 4).

Figure 7 

Precipitation trends driven by the PDO and the SAM. Annual precipitation trends (1960–2016) in central-southern Chile congruent with the PDO (a) and the SAM (b). Total (grey), SAM-congruent (red) and PDO-congruent (orange) seasonal mean trends in central (30–39°S, c) and southern (39–48°S, d) Chile. DOI: https://doi.org/10.1525/elementa.328.f7

The seasonal disaggregation of trends is also shown in Figure 7, averaged over central and southern Chile. In both regions, the SAM evolution seems to play a major role conducting negative rainfall trends, notably in DJF and MAM. This seasonal character agrees well with the rainfall trends that are actually observed in southern Chile, which are of particularly large amplitude in these seasons (Figure 7d). A less clear relationship is seen in central Chile, particularly in MAM, where no robust precipitation trends are derived from rain-gauge records (Figure 7c). As the linear analysis suggests, the weak trend in this region and season may be partially explained by a counteracting effect of the PDO and the SAM. During summer (DJF), both the SAM and the PDO appear to act in the same direction leading to a large negative rainfall trend, a feature that matches the actual trends observed in both regions. We recall the very dry summer mean condition in central Chile, so the relative trend in this season and region must be interpreted carefully.

Greenhouse gases and stratospheric ozone forcings

The SAM-driven drying in central-southern Chile (Figure 7) suggests that the precipitation regime in this region is being affected by global anthropogenic forcing. To further support (or call into question) this inference based solely on local observations and reanalysis data, we looked into a large ensemble of climate model simulations.

In a first stage, we analyzed a set of simulations done with six GCMs, designed to quantify the impact of the GHG and O3 forcing separately, and a third group of simulations from the same models accounting for all the historical climate forcings (Datasets and methods). The multi-model mean trends in annual precipitation and in SLP obtained from these experiments are shown in Figure 8. In agreement to previous findings (e.g., Arblaster and Meehl., 2006), the ensemble of simulations gathered here shows a clear pattern of change in the southern hemisphere SLP between 1960 and 2005 (the historical period common to these runs). This pattern, like the positive SAM phase, is simulated in response to both the GHG and O3 forcings, with trends amplitudes of the same order (about 0.3 hPa per decade, Figure 8b, c). The additive effect of both drivers is fairly consistent with what the full-forced simulations show on average (Figure 8a), notwithstanding these simulations also account for other anthropogenic (changes in the aerosols concentration and in land use) and natural climate drivers.

Figure 8 

Simulated sea-level pressure and precipitation trends. Multi-model mean trends (1960–2005) in annual sea-level pressure (contour lines; solid and dashed contours indicate positive and negatives trends, drawn every 0.1 hPa decade–1) and precipitation (color shading). Results are from the ensemble of simulations from six GCMs including all historical forcings (a) the greenhouse gases (GHG) forcing only (b) and the stratospheric ozone (O3) forcing only. DOI: https://doi.org/10.1525/elementa.328.f8

The SLP trends driven by GHG and O3 represent a surface manifestation of an extratropical circulation response present in most of the troposphere and lower stratosphere (e.g., Karpechko et al., 2008; Son et al., 2010; Polvani et al., 2011). This response entails a poleward shift of the band of maximum meridional pressure gradient, westerly flow and baroclinicity. Accordingly, precipitation tends to decrease in the subtropics and mid-latitudes, and to increase south of 50°S, as shown by the simulations assessed here (Figure 8). The southeast Pacific and the southwest bound of South America, notably, are particularly affected by the drying tendency. Additionally, the decreased westerly flow at ~40°S impacts directly central-southern Chile by reducing the orographic rainfall component (Garreaud et al., 2013).

To synthetize the GHG and O3-driven effects on the large-scale circulation and on the rainfall regime in Chile, the simulated changes in the SLP-based SAM index and in precipitation averaged over the central-southern Chile domain (hereafter PCL, box in Figure 8), are shown in Figure 9. In the simulated period, the SAM index increases in all seasons when all climate forcings are accounted for, but shows a seasonal pattern of change characterized by a stronger signal in the austral summer (Figure 9a). This seasonal pattern is clearly led by the O3 forcing, as has been shown in previous studies (e.g., Gillett et al., 2013). The GHG forcing increases the SAM polarity in the same order of magnitude than O3, but with a weaker seasonality. It is important to highlight that both the GHG and O3-induced changes in the SAM vary substantially from model to model (see error bars in Figure 9).

Figure 9 

Simulated trends in the Southern Annular Mode (SAM) and in central-southern Chile (30–48°S) precipitation (PCL). Seasonal mean trends in the SAM (a–c) and in PCL(d–f) simulated by six GCMs between 1960 and 2005. Error bars indicate the multi-model mean trend and spread (standard deviation). Results are from simulations including all historical forcings (a, d), from simulations forced by GHG only (b, e) and from simulations forced by O3 only (c, f). Red bars in bottom panels indicate the SAM-congruent changes in PCL. DOI: https://doi.org/10.1525/elementa.328.f9

On average across the model simulations, the PCL response to all anthropogenic forcing indicates a year-long drying pathway, but of stronger amplitude in summer (~4% per decade, Figure 9d). The negative trend in PCL matches fairly well the additive effects of GHG (Figure 9e) and O3 (Figure 9f) drivers. Such as in the case of the SAM, the seasonality of change in PCL is dominated by the O3 component, which leads to a clear drying in summer and relatively weak changes through the rest of the year. Following the same procedure than for the observational data, the influence of the SAM on the PCL changes is estimated in models by using the inter-annual covariability of both variables (Datasets and methods). The resulted SAM-congruent trends suggest that a major component of the simulated drying in central-southern Chile is effectively directed by the positive SAM phase-like SH circulation changes (red bars in Figure 9df).

The major control of the SAM on the modelled PCL trends, and the consistent PCL changes shown by full-forced simulations and the GHG-and O3-forced simulations, point out to the anthropogenic forcing as the leading factor behind the modelled southeast Pacific drying. Previous studies have also highlighted the influence of anthropogenic forcing on long-term negative rainfall trends in this region, while no significant effect of natural forcings have been detected (e.g., Vera et al., 2015). Yet, in order to overcome doubts regarding the relative influence of external forcing on precipitation in the region and period assessed, we look also at the group of simulations prescribing natural climate drivers only (Datasets and Methods). In this case, we account for data from 13 GCMs from which we can compare the historical trends driven by all external forcings and by natural ones only (Table S1). This larger group of simulations roughly replicates the precipitation trend patterns shown in Figures 8 and 9 when all external forcing is accounted for (see Figure S1 in supplementary material). By contrast, natural-only runs do not exhibit any robust response in the region of interest. Neither is simulated a seasonal effect of external natural forcing in the rainfall regime in Chile, therefore reaffirming that the modelled changes are mainly driven by the anthropogenic climate drivers described above.

Observed versus simulated precipitation trends, and future scenario in Chile

As the group of simulations analyzed here show, the positive SAM-like circulation trend is a robust response to both the GHG and O3 forcings in models, and this response would have contributed to a drying tendency in the southeast Pacific region, particularly in central-southern Chile. This simulated drying occurs in all seasons, but has a maximum in summer due to the seasonal effect of O3 depletion. Qualitatively, this modelled signature of climate change matches fairly well the precipitation and streamflow trends observed in central-southern Chile since 1960.

A direct comparison of the observations-based regional mean trends with those simulated is presented in Figure 10. For this comparison, the observation-based rainfall trends are averaged over the whole central-southern Chile domain (Datasets and Methods), and we use the full-forced simulations carried out within CMIP5, including runs from 34 GCMs (Table 1 and S1). The rainfall trends averaged over the central-southern Chile domain summarize the results described previously for each region separately. Such as the results based on the attribution experiments (synthesized in Figure 9), the focus is on the SAM and PCL trends separated by season. The time frame used as reference for observations (1960–2016) is contrasted to the same period in GCM simulations, which includes historical runs until 2005 and the RCP-8.5 scenario runs for 2006–2016. To look at the historical changes in a long-term perspective, Figure 10 also shows the trends projected to end of the 21st century by the same group of models and under the same emission scenario.

Figure 10 

Observed and simulated trends in the Southern Annular Mode (SAM) and in central-southern Chile (30–48°S) precipitation (PCL). Observations-based (a, d) and simulated (b, e) seasonal mean trends in the SAM (top) and in PCL (bottom) between 1960 and 2016. Panels (c) and (f) show the SAM and PCL trend simulated for the period 2016–2099. Model results are based on full-forced historical and RCP8.5 simulations from 33 CMIP5 GCMs (error bars indicate multi-model mean and standard deviation). Red bars in bottom panels indicate the SAM-congruent changes in PCL. Orange bars in panel (d) indicate the PDO-congruent trend based on observations. DOI: https://doi.org/10.1525/elementa.328.f10

Like the results described previously, both the observed and simulated PCL indicate a major influence of the positive SAM-like circulation changes on the drying trends in Chile. The increasing polarity of the SAM shown by CMIP5 models (Figure 10b) matches closely the change in the SAM index obtained from the subset of models used to disentangle the contribution of GHG and O3 (Figure 9a). The SAM index derived from four products based on observations (see Datasets and methods) indicates a broadly consistent behavior with that modelled, characterized by a strengthening of this phenomenon in all seasons, but of clear larger amplitude in austral summer (Figure 10a). In other seasons, there is less clear agreement between SAM trends derived from products and models, notably in SON. This partial mismatch may be due to a significant contribution of natural cycles to the actual trends, to different responses to external climate forcing, or to systematic biases in SLP reconstructions (Fogt et al., 2009).

The sign and seasonality of the observed PCL change show a remarkable agreement with those obtained from CMIP5 ensemble (cf. Figure 10d and 10e). Such as the model results, the observations-based PCL indicates a negative trend in all seasons, but with an intensity markedly larger in summer than in winter. However, despite this correspondence, there is a large difference in the trends’ magnitude. The PCL trends based on observations are dramatically stronger than those obtained on average from the CMIP5 GCMs. This disagreement, which may have large implications on defining suitable adaptation strategies for the future changes of precipitation in Chile, is discussed further in the section that follows.

Due to the limitations for ODS emissions accorded at the Montreal Protocol, the Antarctic total ozone column has stabilized in the last decade (Solomon et al., 2016), and is expected to recover towards the end of the century (Cionni et al., 2011). This pathway, accounted for within the CMIP5 RCP scenarios (Eyring et al., 2013), manifests clearly in the modelled SAM projections, lessening or reversing the summer signal of the recent decades (Barnes et al., 2014). Accordingly, the climate projections assessed here show a consistent pathway of change in the SAM and in precipitation in central-southern Chile. On average across the GCMs, the SAM polarity continues to strengthen through the 21st century in response to the strong GHG change scenario included in RCP8.5 (Figure 10c). Yet, compared to the historical period, the future increase in the SAM is dampened in austral summer due to the O3 recovery, leading to a less pronounced and, in some cases, even reversed seasonal pattern of change. The projected increase in the SAM polarity in models modulates the seasonal pattern of change in central-southern Chile precipitation, which, in contrast to the historical period with marked seasonal differences, is simulated to decrease in the future at similar rates in all seasons (Figure 10f).

Discussion

How can the large bias between the observed and modelled drying trend in central-southern Chile be interpreted? A first issue to consider is the scale mismatch and representativeness of spatial averages based on local observations. It should be noted that the strong drying signal shown in Figure 10 depends substantially on the rainfall records in southern Chile (south of 40°S), a region with limited observations and important uncertainties (Figure 4). Another, particularly worrying reason to explain the modelled vs. observed difference would be a systematic underestimation of the southeast Pacific rainfall response to anthropogenic forcing in current GCMs. A third plausible option is that the actual negative rainfall trend has an important source of natural variability in addition to anthropogenic factor (e.g., Muñoz et al., 2016).

To further explore this last point, we look closer at the uncertainty in annual PCL trends using the large ensemble of historical and future CMIP5 simulations. Given that we account for more than one realization per GCM and simulation type assessed, we can quantify the contribution of both internal variability and climate sensitivity to the dispersion obtained for the ensemble of runs (Datasets and methods). Figure 11a shows the modelled mean trends in annual PCL, using 1960 as starting time and different years from 1980 to 2099 as the end for trends calculation. The standard deviation between the trends of the ensemble as well as the estimated contribution of natural variability to the this spread is shown in Figure 11a (dark grey). The remaining fraction of the ensemble spread is considered to account for the different PCL sensitivities to the anthropogenic climate forcing (light grey in Figure 11a).

Figure 11 

Uncertainties in observed and simulated changes in central-southern Chile precipitation. Multi-model mean (black) and observed (red) trends in central-southern Chile annual precipitation (δPCL), computed between 1960 and the year indicated in the x-axis (a). Shaded area indicates the simulated trend uncertainty (mean ± 1.0 standard deviation) derived from the ensemble of model simulations (Datasets and methods). Dark and light grey shading show de model spread fraction driven by natural variability and climate sensibility, respectively. Panel (b) shows the modeled (black) and observed (red) δPCL normalized by 1.96 times the corresponding standard error of the trend (seδ). A normalized δPCL of magnitude larger than 1.0 corresponds to a trend statistically different from zero, with a confidence level of 95%. This threshold, indicated by the dashed line in panel (b), allows deducing the time of emergence (TOE) of changes in PCL. DOI: https://doi.org/10.1525/elementa.328.f11

As expected, the uncertainty in the modelled PCL change is reduced as the period considered for trend calculations increases and the role of internal variability weakens (Figure 11a). Thereby, the uncertainties in the PCL projected to the end of the 21st century respond, principally, to differences in the GCMs’ climate system characterization (the ‘model uncertainty’ after Hawkins and Sutton, 2011). We note that this spread, related to the distinct model PCL sensitivities to anthropogenic forcing, is quite small compared to the precipitation change uncertainty in other regions of the globe (e.g., Boisier et al., 2015). Indeed, under a common socioeconomic scenario –RCP 8.5 in this case–, the direction of PCL change (drying) is systematic across the GCMs assessed. Notwithstanding the consistency on the modelled century-long PCL projections, natural variability remains an important source of uncertainty for trends computed in shorter time periods, in particular between 1960 and the present, warning us about the potentially mixed nature of the actual PCL trend. A similar caveat has been pointed out in other regions having large natural climate variability (e.g., Deser et al, 2012).

The observations-based PCL drying trends between 1960 and a year later than ~1990 are of clear larger amplitude than that simulated in most GCMs (red line in Figure 11a). In line with the uncertainties shown by the CMIP5 ensemble for trends computed in relatively short time periods (less than ~50 years), the year-to-year variation of the observed PCL trends exposes the sensitivity of this metric to natural climate anomalies. In order to have a more robust metric of change, we normalized the value of a given trend by the trend error estimated for the corresponding regression slope (δPCL/1.96seδ; see Datasets and methods). These normalized trends represent signal-to-noise ratios, and show a more consistent pathway of change between models and the observations (Figure 11b). However, the observed normalized trend, of magnitude ~1.5 for 1960–2016, is still stronger than that derived for most CMIP5 models, even considering the range estimated for natural variability. In particular, this metric indicates that the long-term signal of PCL change emerges from natural variability near the present time (a trend is considered to be significantly different from zero when the magnitude of the normalized trend exceeds the unity; this threshold is indicated by a dashed line in Figure 11b). The time of emergence (TOE) in models is centered in ~2020, while the observed PCL indicates that the TOE has very likely been already reached.

This uncertainty assessment reinforces the conclusion that an anthropogenic change towards a dryer climate in central-southern Chile is underway. If the actual trends diagnosed with local rain-gauge records are correctly accounting for the strength of a regional-scale drying, then these results also suggest that the actual rainfall sensitivity in central-southern Chile to anthropogenic forcing is larger than those simulated in most CMIP5 models.

Concluding remarks

In this study, we presented an analysis based on a recently assembled rain-gauge and streamflow dataset within Chilean territory, showing a robust drying trend over the period 1960–2016 in a large region spanning 30°S to 48°S (the central and southern part of the country). Our results indicate that precipitation trends in this region have emerged above natural variability during the last decade. This signal is mainly driven by the rainfall decline observed in wet regions south of 37°S. The more episodic nature of precipitation in central Chile (30–37°S) makes any detection of change difficult. However, the negative rainfall trend in this region clearly agrees with the more robust change observed in river discharge data.

We also described a distinct seasonal signature on streamflow and precipitation trends, which unveil leading mechanisms of change and the role played by different large-scale climate forcings. In particular, the strong summer drying observed in southern Chile, of about 8% per decade, is coherent with well-known positive SAM-like changes in the southern hemisphere circulation. As reported earlier and as reassessed here with an ad-hoc set of climate simulations, the positive SAM-like circulation change in summer is, to a large degree, caused by stratospheric ozone depletion, amplifying the effect of the same sign that is driven by the GHG forcing in all seasons. Hence, in addition to precipitation change detection, we conclude that the main character of the observed long-term drying signal in Chile is attributable to anthropogenic forcing.

The SAM-congruent precipitation change is relevant for climate projections in Chile. Since the stratospheric ozone content is expected to recover throughout the 21st century as a result of the Montreal protocol, one can expect a lessening of the summer drying in southern Chile due to this forcing mechanism. The other important mode of variability that modulates precipitation variability at decadal scales in Chile is the PDO. The positive-to-negative phase transition of the PDO from the end of the 1970s to 2000s has largely contributed to the rainfall decline in central Chile in this period (Boisier et al., 2016), but not to the longer-term drying trend analyzed in this work. Yet, future changes in this mode of decadal variability will certainly modulate precipitation on top of the secular anthropogenic drying trend.

Besides the agreement on the sign and on the seasonal pattern of precipitation change shown by the observational and model data, the magnitude of the actual drying quantified in this study is substantially larger than that modelled. Our analysis indicates that this bias is unlikely to be fully explained by natural variability, implying that the current climate simulations may be underestimating the drying amplitude in central-southern Chile. Further research, looking deeper into the physical processes governing the Southeastern Pacific precipitation response to O3, GHG and other climate forcings, is needed to constraint the uncertainties of the anthropogenic drying in Chile.

It is worth to recall that the changes in the precipitation regime are superimposed to a warming trend, so other aspects significant to regional hydro-climate and water availability should be considered, notably the increasing evapotranspiration and changes in annual cycle of snowmelt and runoff. Changes in temperature, in freezing level height and in rainfall extremes should also be accounted for when assessing risks of floods or droughts.

Regarding water availability in the coming decades, the take-home message for planning is that we can expect the anthropogenic drying trend to continue, although modulated on annual-to-decadal time scales by ENSO, the PDO and other sources of natural variability. The long-lasting and extended drought experienced in Chile in recent years serves as an analog for future climate. This phenomenon resulted in multiple often linked impacts affecting vegetation and watersheds (Garreaud et al, 2017), the biogeochemistry of coastal water (León-Muñoz et al., 2018; Aguirre et al., 2018), and the intensity of forest fires (Urrutia-Jalabert et al., 2018; González et al., 2018), stressing mutually-dependent effects of the drought on Chile’s socio-ecosystems. Responses from the public and private sectors generally considered this drought to be a transient event. Moreover, current governance arrangements, i.e., more than 40 different national agencies with jurisdiction over water resources make coordinated action difficult and inefficient. As indicated by several authors (e.g., Van Loon et al, 2016; Aldunce et al., 2017; Crausbay et al., 2017), facing drought in the Anthropocene will require new, holistic approaches to water governance and climate change adaptation.

Data Accessibility Statement

Precipitation and streamflow data for Chile are available at: http://explorador.cr2.cl/. The CMIP5 data were acquired from the Earth System Grid Federation. https://esgf.llnl.gov.

Supplemental files

The supplemental files for this article can be found as follows: