A decade-long (2010–2020) period with precipitation deficits in central–south Chile (30–41
Persistent climatic anomalies may alter catchment response to precipitation. Thus, catchment dynamics under unusual multiyear precipitation deficits might not be correctly predicted based on the interannual variability over the last decades. This applies even when past decades include severe but shorter dry conditions (Saft et al., 2016a). In other words, stationarity as commonly assumed for streamflow projections under climate change might be an invalid assumption (Blöschl and Montanari, 2010), which poses challenges for achieving realistic structures and parameters in hydrological models (Duethmann et al., 2020; Fowler et al., 2016).
Non-stationary catchment response modulates hydrological functioning. This applies particularly to drought propagation, i.e. the process leading to soil moisture droughts and hydrological droughts (streamflow and groundwater deficits) under dry meteorological conditions (Van Loon et al., 2014). While meteorological droughts are mainly controlled by regional precipitation, soil moisture and hydrological droughts are also controlled by catchment characteristics. Therefore, under similar meteorological conditions, the severity of hydrological droughts can vary significantly within a climatic region (Van Lanen et al., 2013). Most drought-related impacts on, for instance, agriculture, ecosystems, energy, industry, drinking water, and recreation depend primarily on groundwater and streamflow deficits (Van Loon, 2015). Therefore, understanding the geographical variation in drought propagation provides critical information for drought-hazard adaptation and mitigation (Van Loon and Laaha, 2015). In addition to such spatial variability, non-stationary catchment responses to precipitation would lead to a temporal variation in drought propagation.
This temporal aspect is becoming increasingly important since many regions around the globe are experiencing unprecedented long dry spells due to climate and circulation changes, causing unforeseen impacts on water supply (e.g. Schewe et al., 2014). Recent evidence has shown that protracted droughts may propagate differently within the same catchment (i.e. same landscape characteristics and governing runoff mechanisms) under similar precipitation deficits and temperature anomalies than shorter dry events. For example, studies in south-eastern Australia have reported changes in catchment functioning (Fowler et al., 2018; Saft et al., 2015, 2016b; Yang et al., 2017) during the Millennium drought that took place for more than a decade (1997–2010). More recently, Garreaud et al. (2017) reported an unprecedented decrease in annual runoff during a multiyear drought in central–south Chile, the so-called megadrought (MD). The amplified response of streamflow to a drought signal may be due to variations of drainage density related to depleted groundwater levels within the catchment (Eltahir and Yeh, 1999; Van De Griend et al., 2002), a factor also emphasized by Saft et al. (2016b).
The MD experienced in Chile since 2010 (and continuing to date) offers a great opportunity to understand the potential impacts of global changes on hydrology and water supply over wide ranges of hydro-climatic regions and landscape characteristics. The persistency and geographical extension of the MD have few analogues in the last millennia, and its causes have been partially attributed to anthropogenic climate change (Boisier et al., 2016, 2018; Garreaud et al., 2017, 2019). This uninterrupted sequence of years with precipitation deficits has impacted various sectors, including coastal ecosystems (Masotti et al., 2018), natural vegetation (Arroyo et al., 2020; Garreaud et al., 2017), fire regimes (Gonzalez et al., 2018), and water supply (Muñoz et al., 2020).
To deepen the understanding of the impacts of persistent droughts on water
supply, we explore the mechanisms causing the larger-than-expected hydrological deficits in central–south Chile during the MD. We complement
previous analyses of the MD in Chile (Garreaud et al., 2017; Muñoz et al., 2020) by incorporating 4 more years into the MD period and by focusing on drought propagation over 106 catchments located between 30 and 41
For a dry year within a long drought, we can distinguish three cases: (i) stationary drought propagation, when the streamflow deficits are similar to those observed in isolated years (single-year drought) with similar precipitation deficits; (ii) intensified drought propagation, when streamflow deficits are larger than those observed in years with similar precipitation deficits; and (iii) attenuated drought propagation, when streamflow deficits are lower than those observed in years with similar precipitation deficits. Based on previous studies relating groundwater dynamics to non-stationary catchment response to droughts (Carey et al., 2010; Eltahir and Yeh, 1999; Fowler et al., 2020; Saft et al., 2016b), we hypothesize that in catchments with longer hydrological memory (i.e. catchments where water is retained for a longer time in different storages such as aquifers and snowpack), the propagation of drought during multiyear precipitation deficits is intensified (i.e. larger streamflow deficits than those observed in years with similar precipitation deficits), when compared to single dry years.
To test this hypothesis, we characterized the historical precipitation and streamflow deficits at the catchment scale and followed Saft et al. (2015) to evaluate annual precipitation–runoff (P–R) relationships and identified those catchments where drought propagation during the MD was maintained, intensified, or attenuated with respect to their historical behaviour. We analysed catchment memory from observed hydrometeorological data and from the hydrological processes simulated by a bucket-type model calibrated for each basin. We related catchment hydrological memory with shifts in P–R relationships during the MD and with drought propagation for different types of drought, from extreme single-year droughts to moderate but persistent droughts (including the MD). Finally, we addressed a general question with practical implications: what is worse in terms of water supply – a single year with extreme precipitation deficits or several consecutive years with moderate deficits?
Characteristics of the study domain and location of catchments.
The study area corresponds to central–south Chile, spanning 9 out of 16 administrative regions between 30 and 41
Monthly streamflow values were computed when 15 or more days had valid data.
For those months, a mean monthly value was computed from the available daily
values and then aggregated into the total number of days within the month to
get total monthly runoff. Subsequently, gaps in monthly streamflow time
series were filled based on a procedure previously used for monthly precipitation data (Boisier et al., 2016). The method uses multivariate regression models, taking advantage of the streamflow co-variability among multiple gauging stations in the study region (within or across basins). Using this approach, data for a station can be filled if the missing data do not exceed 25 % of the period. Each missing month is independently assessed based on an ensemble of multivariate models based on covariant records from other stations as predictors. A given linear model is used if it shows a predictive power (coefficient of determination,
We computed catchment-scale solid to total precipitation fractions and daily estimates for snowmelt over the period April 1981 to March 2020 based on the ECMWF surface reanalysis ERA5-Land dataset, available at a spatial resolution of 9 km (Muñoz-Sabater, 2019a).
As a basis for the hydrological modelling (Sect. 3.1.2), we computed hypsometric curves for each catchment based on ASTER GDEM (Tachikawa et al., 2011).
The CAMELS-CL dataset includes catchment characteristics such as topography and geology and hydrological signatures such as the baseflow index, representing the slow catchment response to precipitation. The anthropic-related data provided include land cover for the year 2016, the location of reservoirs, and water use rights granted within the basins.
From the 327 catchments located between 30 and 41
Given the latitudinal extent and terrain complexity, the study region features very different hydroclimate regimes (Fig. 1). Annual precipitation ranges from less than 100 mm to the north to near 3000 mm in the southern part. Precipitation also increases substantially towards the west due to the orographic effect exerted by the Andes on the predominantly westerly atmospheric flow (Viale and Garreaud, 2014).
To characterize the different hydrological regimes of the study catchments,
we classified them based on the hydro-climatic metrics summarized in Table 1, which represent the main seasonal hydro-climatic characteristics of a catchment. The classification was based on a
Hydro-meteorological basin features used for classification.
In addition to the analysis of the observations-based dataset from CAMELS-CL, we run the HBV model (Bergström, 1972; Lindström et al., 1997) to simulate streamflow and other fluxes for each of the 106 study catchments. With these simulations, we seek to improve our understanding of runoff mechanisms from a process-based perspective, particularly regarding the role of snow and groundwater in runoff generation. The HBV is a bucket-type model that simulates the main hydrological processes in a catchment through a number of routines. In the snow routine, snow accumulation and melt are simulated based on a simple degree-day approach. A variable fraction of all melted and rainfall water is retained in the soil depending on the current soil water level. The remaining part is transferred to the groundwater routine. In this routine, groundwater storage is represented by two boxes, an upper soil box representing faster groundwater release to total streamflow and a lower soil box representing a slower groundwater release to total streamflow, both with linear outflows. Finally, the simulated outflows from the groundwater stores are routed using a simple routing scheme, leading to the total streamflow. Besides streamflow, time series of a number of other fluxes and storages can be obtained from the model, such as actual evapotranspiration (ET), soil water storage, or the different streamflow components.
The HBV model has been implemented in several software packages. Here we used the version HBV light (Seibert and Vis, 2012). The model was calibrated using a genetic algorithm (Seibert, 2000) with parameter ranges similar to those suggested earlier (e.g. Seibert and Vis, 2012). The 14 free parameters values were derived after 3500 model runs. For each catchment, 100 independent calibration trials were performed based on the non-parametric variation of Kling–Gupta efficiency, NPE (Pool et al., 2018), which resulted in ensembles with 100 parameter sets.
To characterize the slow groundwater contribution to runoff for each basin, we computed a groundwater index (GWI) as the mean annual outflow from the lower soil box (GW) normalized by the mean annual simulated streamflow. Note that GWI provides a measure of the groundwater flux simulated by HBV (i.e. not the groundwater storage). In contrast to the baseflow index provided in CAMELS-CL, which is computed from a low-pass filter applied to streamflow observations and thus represents the response timings to precipitation, the GWI represents timings and also the source of the water.
The hydrological memory of a catchment is the composite of response times associated with the physical mechanisms transferring and storing water through the basin (Fowler et al., 2020). Such response times have been qualitatively related to the presence of aquifers, lakes, and snow (Van Loon and Van Lanen, 2012). Thus, there is no unique way to quantify hydrological memory. For example, catchment memory has been assessed based on soil moisture and groundwater dynamics (Agboma and Lye, 2015; Peters et al., 2006), on streamflow recession curves (Rodríguez-Iturbe and Valdes, 1979), on lag correlations between soil moisture and other fluxes within the catchment (Orth and Seneviratne, 2013), and on recovery times from droughts (Yang et al., 2017).
In this study, we assessed hydrological memory based on the following indices:
Seasonal streamflow memory is represented by the Seasonal GW memory is represented by the Annual streamflow memory is represented by the
Meteorological and hydrological droughts were characterized by the observed
annual precipitation and streamflow anomalies at the catchment scale, respectively. For each basin, the relative anomaly of streamflow (
Stationarity in drought propagation during the MD was assessed by following the procedure suggested by Saft et al. (2015) to identify significant shifts in annual rainfall–runoff relationships over Australian catchments during the Millennium drought. Saft et al. (2015) showed that prolonged rainfall (liquid precipitation) deficits resulted in shifts in rainfall–runoff relationships at the catchment scale, and Saft et al. (2016b) related the shifts to catchment characteristics (aridity index and rainfall seasonality) and soil and groundwater dynamics. The physical mechanisms likely associated with these factors were discussed by Saft et al. (2016b) but not explicitly modelled.
In this study, we computed annual P–R relationships between annual
precipitation (solid and liquid) and runoff time series for each catchment
and performed a global test to validate linear model assumptions with the
R package gvlma (Peña and Slate, 2006). From the 170 catchments fulfilling the criteria (i) to (iii) described in Sect. 2, we selected 106 for which the linear assumptions in P–R relationships were fulfilled and where the annual rainfall explained more than 50 % of the variance in annual runoff (
For each catchment, we tested if the P–R relationship during the MD (April 2010 to March 2020) was different to the P–R relationship computed with the previous period (April 1979 to March 2010), by performing the analysis of variance model from the R package aov (R Core Team, 2018) to the intercept parameter from the linear regressions (see Eq. (1) from Saft et al., 2015). From this analysis, we defined two types of cases: (i) catchments with a significant shift in P–R relationship at a 0.1 significance level and (ii) catchments that did not experience a significant shift (test
In addition to computing the shifts in P–R relationships (Sect. 3.2.2), which represent an overall catchment response during the MD period, we analysed annual drought propagation over the entire period of record (April 1979 to March 2020) based on annual precipitation and streamflow anomalies (Sect. 3.2.1). Drought propagation during a hydrological year was represented by the contrast between streamflow and precipitation anomalies.
In particular, we focused on catchment responses during two types of events: (i) single-year extreme and severe droughts and (ii) multiyear moderate and mild droughts. For each catchment, precipitation anomalies were classified by following the drought classification thresholds provided by McKee et al. (1995). In this way, annual anomalies between 0.5 and 0.067 quantiles were classified as mild to moderate droughts. Annual anomalies below the quantile 0.067 were classified as severe to extreme droughts. These thresholds are currently being used by the Public Works Ministry to declare water scarcity decrees (DGA resolution no. 1.674 from 2012; DGA, 2012).
To analyse the hydrological memory effect on the propagation of extreme and severe droughts, we separated these events based on the precipitation anomaly of the preceding year (below or above the median).
Based on the classification scheme described in Sect. 3.1.1, we identified
72 pluvial and 34 snow-dominated basins. Some of their main characteristics
are presented in Fig. 2. Most of the snow-dominated basins are located in
central Chile (30–35
Characteristics of pluvial and snow-dominated basins obtained from
CAMELS-CL dataset. Panel
To visualize the importance of groundwater processes within the study catchments, and their relationship with snow processes, in Fig. 2d we relate
the GWI computed from HBV simulations with the SF derived from ERA5-L for
each basin (Table 1). These variables come from independent datasets and show a significant correlation (
In addition to the climatic characteristics (e.g. precipitation, snow fraction, and aridity), GW contribution to runoff depends on physical factors (e.g. geology, topography, soil properties, and soil drainage density), which may explain the large scatter in Fig. 2d. In fact, the geologic characteristics vary across basins, as can be seen in Fig. S1 in the Supplement. The most common geologic classes in snow-dominated basins are acid volcanic rocks (main class in 59 % of basins), followed by acid plutonic rocks (main class in 18 % of basins) and pyroclastics (main class in 18 % of basins). In pluvial basins, there is greater heterogeneity in geologic classes, with 22 % of basins dominated by pyroclastics and 19 % of basins by acid plutonic.
Additional characteristics of snow-dominated and pluvial catchments,
including
Regarding the land cover, 68 % of the snow-dominated catchments are mainly covered by barren soil and snow, while the rest is mainly covered by shrubland. None of these land cover classes is directly associated with anthropic activities. 94 % of the snow-dominated basins have less than 5 % of their areas covered by croplands.
The region where pluvial basins are located features a higher heterogeneity of land cover classes, compared to the Andean region of central Chile (where snow-dominated basins are located) (Alvarez-Garreton et al., 2018). The dominant land cover classes in pluvial basins are native forest (main class in 61 % of pluvial basins) and shrubland (main class in 11 % of pluvial basins). Anthropic-related land cover classes dominate the rest of the pluvial catchments: forest plantation in 11 % of basins, grassland in 9 % of basins, and cropland in 7 % of basins.
The correlation between autumn–winter precipitation (
Hydrological memory represented by different indices. Panel
Figure 3a indicates that
Interestingly, the correlation between
The control of
To remove the effect of the precipitation of the current year, Fig. 3c
presents the dependency of the P–R regression residuals from Sect. 3.2.2
(i.e. annual streamflow not explained by the current precipitation) to the
precipitation from the previous year. The
The relationship between hydrological memory beyond 1 year and catchment storage dynamics driven by GW and snow processes is further explored in Fig. 4. The GWI derived from HBV simulations (Sect. 3.1.2) and SF (Table 1) are used to summarize GW and snow processes within a catchment in Fig. 4. These plots show that both snow and GW mechanisms contribute to the hydrological memory, which is consistent with the time lags these processes introduce to the pathways of precipitation within a basin. The adopted approach to compute the memory of a catchment, based on seasonal streamflow and GW flows at the catchment outlet, represents the composite response times of all catchment mechanisms. Therefore, there might be other factors contributing to the overall response time of a catchment, including topography, soil properties, geology, drainage area, and water table levels (Robinson and Ward, 2017), which may explain the large scatters in Fig. 4.
Relationship between hydrological memory and indices of groundwater and snow storages (GWI and SF, respectively). Panel
Given the marked north to south precipitation gradient (Fig. 1), droughts
characteristics in this section are analysed at the catchment scale but
following a latitudinal order. In the following sections, we analyse drought
propagation per hydrologic regime. Heat maps in Fig. 2 illustrate the
precipitation and streamflow annual anomalies (Fig. 5a and b) and
Relative annual anomalies of catchment-scale precipitation
The relative precipitation anomalies are consistently higher in catchments
north of 35
The annual streamflow relative anomalies (Fig. 5b) present larger values and
larger variations in space and time than precipitation, which is due to the
lower absolute values compared to precipitation and the dependency of
streamflow on local terrestrial characteristics. If we compare the spatial
patterns of anomalies (Fig. 5b) and
To further characterize the hydro-climatic anomalies during the MD, Fig. 6
presents the frequency distribution of 10-year mean precipitation, observed
streamflow, and simulated ET for each basin, with the mean values during the
MD plotted by red dots. These plots indicate that the MD has been extremely
unusual in terms of precipitation and streamflow. The average precipitation
during the MD is within the first decile for 91 % of the study catchments
(96 out of 106). The average runoff during the MD has been more extreme than
precipitation deficits, with 96 % of catchments (102 out of 106) presenting 10-year mean runoff values within the first decile. These values represent the minimum value over the last 4 decades for some basins located north to 32
Box plots of 10-year mean precipitation
Given the relatively small scale of catchments in Chile, runoff in most of
them has a strong dependency on the interannual precipitation variability,
explaining typically
Annual runoff (
Number of catchments with a significant shift and no shift in P–R relationships during the MD. The shift magnitude in catchments with change is given in italics.
For the 65 catchments showing a change, the historical P–R regressions consistently underestimate the runoff deficits during the MD. 95 % (62 out of 65) of the catchments with a significant change in P–R relationship during the MD, had a negative shift, that is, an intensification in drought propagation. For similar precipitation deficits to other dry years, observed streamflow during the MD in snow-dominated catchments was up to 57 % lower than predicted by the historical P–R relationship. In pluvial catchments, these shifts reached up to 37 % (Table 1).
For those catchments with a significant change, higher GWI values are associated with larger shifts in P–R relationships (
While these results provide insights about changes in P–R relationships during a multiyear period, they do not indicate if the changes are progressive, i.e. if the basins progressively generate less water for a given precipitation amount, compared with their historical behaviour. We address this in the following section, where annual drought propagation is analysed in detail.
Figure 8 presents the time series of annual precipitation and streamflow
anomalies averaged across snow-dominated catchments (Fig. 8a) and pluvial
catchments (Fig. 8b) over the last 4 decades. Drought propagation is
represented by the difference between average runoff and precipitation anomalies (secondary
Precipitation and runoff anomalies in snow-dominated
In snow-dominated basins, the difference between runoff and precipitation anomalies consistently increases (becomes more negative) in the second year of precipitation deficits, showing that in catchments with longer hydrological memory, consecutive years with precipitation deficits are associated with intensified drought propagation. This plot also provides insights about hydrological recovery, understood as the hydrological condition after a meteorological drought has ceased (Yang et al., 2017). While 2016 had near-average precipitation in snow-dominated basins, there was probably not enough water entering the system over enough time to recharge groundwater systems up to levels such as those before the MD (similarly to the conceptual drought propagation illustrated in Fig. 3 from Van Loon, 2015). This is reflected by the larger streamflow deficits in 2016 compared to 2008, even when the above-mean precipitation values in 2008 following the deficits in 2007 are comparable to those in 2016 and 2015, respectively. This can be related to the catchments' memory (Sect. 4.2) and the 7-year (2009 to 2015) precipitation deficits prior to 2016, which probably prevented a full hydrological recovery after a single year of above-average precipitation. These results are consistent with large recovery times reported for semi-arid Australian catchments following extreme droughts (Fowler et al., 2020; Yang et al., 2017). In this way, hydrological memory would be an explanatory factor for both the intensification in drought propagation and a delayed hydrological recovery.
For pluvial catchments (Fig. 8b), prior to 2010, the propagation of
meteorological to hydrological drought has ranged from around 0 % to 15 % (i.e. streamflow anomalies have been, on average, up to 15 % lower than
precipitation anomalies), independently of the precipitation deficits of
previous years. Such propagation was observed even in the driest 2 years
of the historical record, 1996 and 1998. After 2010, there are 2 years (2012 and 2016) where drought propagation has been intensified up to 25 %.
These larger streamflow deficits during the MD may be due to different factors, including the large ET in 2012 and 2016 (positive anomalies and
Regarding the overall response during the MD, since P–R relationships do not explicitly account for variations in ET, the positive ET anomalies during specific years of the MD (2012, 2016 and 2018, Fig. S2) may partly explain the P–R shifts identified in some pluvial catchments (Fig. 7). Towards the south of the study region, basins move from being water-limited to energy-limited (Alvarez-Garreton et al., 2018). Therefore, ET in pluvial catchments is modulated by both the available water and the available energy, in contrast to snow-dominated catchments, where ET is primarily driven by precipitation (these are water-limited basins). This suggests that ET may be a factor influencing the intensification of drought propagation during the MD in pluvial catchments. However, despite higher ET in 3 years during the MD, the average ET during the 10-year MD period has been lower than other 10-year windows (Fig. 6c). Therefore, the hydrological memory beyond 1 hydrological year in pluvial basins (Fig. 3) is likely another factor contributing to the P–R shifts in pluvial catchments.
Figure 9 presents the observed and simulated drought propagation during single-year severe and extreme droughts and during persistent mild and moderate droughts, for snow-dominated (Fig. 9a and b) and pluvial catchments (Fig. 9c and d). Consistent with the hydrological memory in snow-dominated catchments, we observe that drought propagation in these basins is highly dependent on the meteorological conditions from the previous year (Fig. 9a), which define the initial condition of soil water storages. If a severe or extreme drought happens after a wet year, such as 1981, 1985, 1988, 1998, 2007, and 2009 (Fig. 9a), drought propagates without amplification; i.e. streamflow deficits are lower than precipitation deficits. By contrast, if the extreme drought happens after a dry year, such as 1995, 1996, 2013, and 2019, meteorological droughts are amplified by nearly 20 % (difference between median streamflow and median precipitation deficits in Fig. 9a). This is also observed in persistent but moderate droughts, when under similar precipitation deficits, the surface water supply decreases after 1 year with below-average conditions.
Observed and simulated annual drought propagation for consecutive
dry years. Panel
If we look at streamflow predictions, these plots indicate that the HBV model represents catchment response for extreme and persistent droughts well, consistently outperforming the prediction from P–R regressions. It should be noted though that HBV allows for memory effects only to a certain extent given the groundwater storage capacity defined by calibration.
Pluvial basins in the study region have shorter hydrological memory compared to snow-dominated catchments, which leads to a more similar behaviour under extreme meteorological droughts occurring after a wet and a dry year (difference in streamflow to precipitation anomalies between 10 %–20 %; Fig. 9c). However, even when these basins are largely controlled by precipitation during the same year, there is some memory that is over 1 year (Fig. 3c), which may be influencing the observed decrease in streamflow generation after 2 years of consecutive precipitation deficits (Fig. 9d). This effect is well captured by the HBV model, while the annual P–R relationship tends to overestimate observed runoff. This indicates that a good representation of fluxes such as ET, soil moisture, and groundwater dynamics, allows catchment response to persistent droughts and to extreme droughts to be foreseen.
Hydrological memory is related to slow groundwater and subsurface flows transferring precipitation from previous seasons, especially from winter (Figs. 3 and 4). If the winter snowpack is large, hydrological memory is further extended, not only by the season lag resulting from the snowmelt contribution to streamflow in spring and summer, but also by the slow GW contribution during the next autumn, when the snow has already melted.
We showed that the importance of GW contribution to runoff is positively
correlated with snow accumulation (Fig. 2d). This relationship between snow
and groundwater contribution to downstream streamflow supports the GW
recharge conceptualization recently proposed by Taucare et al. (2020) for the Western Andean Front in central Chile. In contrast to former assumptions of no GW recharge in high elevated Andean areas, Taucare et al. (2020) demonstrated the existence of GW circulation in fractured rocks originating from rain and snowmelt above
This conceptualization is also supported by the groundwater recharge mechanisms driven by snowmelt shown by Carroll et al. (2019) in a Colorado River headwater basin. Snowmelt infiltrates in situ, where ET is low compared to precipitation, the soil storage is shallow, and there is a low permeability bedrock underneath. Infiltrated snowmelt is routed through the steep topography as shallow ephemeral interflow, which supports large recharge rates in topographic convergence zones, where ET is still moderate and ephemeral stream channels in the upper basin appear (Anderson and Burt, 1978). At high elevations where snowmelt starts, topography and soil permeability would have a larger control in groundwater recharge than precipitation. That is, the rates at which snowmelt infiltrates are more sensitive to catchment characteristics than to the snowpack volume (Carroll et al., 2019). This suggests that during dry years, the portion of snowmelt directly contributing to runoff would decrease, while the portion of GW from snowmelt infiltration would increase.
The large snowmelt infiltration rates at high elevations during snowmelt season, combined with the absence of precipitation events (precipitation is concentrated in the winter season), would explain the higher GWI in snow-dominated basins.
High snow fraction ratios also indicate an important spatial variation of precipitation within the basin, with most precipitation falling in the upper part of the basin and, therefore, travelling longer paths to reach the outlet where streamflow is recorded. Longer paths lead to longer travel times (i.e. longer hydrological memory), which may support groundwater recharge from interflow.
Regarding drought propagation, our results show that precipitation has an influence for longer times in basins where snow and GW processes dominate the hydrological response. Snow-dominated catchments feature longer memory than pluvial basins and have been more affected by the persistency of precipitation deficits during the MD, causing a significant change in overall catchment response to the precipitation over the last decade. These results complement previous findings in Australia (Saft et al., 2015, 2016a, b), providing new evidence of the vulnerability of catchments to drying climatic trends. In particular, our results reveal that snow processes are tightly associated with the intensification in drought propagation, a conclusion that was not drawn by the analysis of the Australian Millennium drought given the lower elevation of the Australian basins analysed (mean elevations below 1500 m a.s.l.).
The role of snow in hydrological memory and its impacts on water provision during droughts highlight the need to further understand the seasonal characteristics of drought propagation. The effects of spring–summer precipitation deficits may be different from those of autumn–winter deficits (Berghuijs et al., 2014; Jasechko et al., 2014). This is particularly important in central–south Chile, where shifts in precipitation and streamflow seasonality due to anthropogenic climate change have already been detected in observations (Boisier et al., 2018; Bozkurt et al., 2018; Cortés et al., 2011).
Our results also provide insights regarding model representation of hydrologic mechanisms. The HBV model overcomes some of the limitations of diagnosing the progressive water deficits only from shifts in annual P–R relationships. A model that can capture catchment memory is more suited to represent deficits in streamflow under persistent drought, compared to the simulations from P–R regressions that only consider the precipitation of the same year (Fowler et al., 2020). Nevertheless, the HBV model is a simplified representation of actual hydrological processes, which limits its capability to simulate long-term memory effects. Memory effects in the HBV model are caused by soil water storage as this store can accumulate precipitation deficits. The groundwater stores in HBV cannot drain below the level where streamflow ceases and, thus, represent only the dynamic storage in a catchment (Staudinger et al., 2017), which means that memory effects caused by groundwater stores are limited in the model. This is consistent with the limitations of conceptual bucket-type models to simulate long, slow hydrological processes demonstrated by Fowler et al. (2020). The effects of multiyear drought cannot accumulate in models where the time constants (usually months) are shorter than the observed memory (seasonal and longer). In this way, catchments that are more prone to a non-stationary hydrologic response under persistent droughts pose greater challenges for projecting the impacts of a drying climate. This highlights the need to advance towards robust modelling frameworks in order to achieve reliable streamflow predictions under drier climate projections.
The proposed approach focuses on understanding the causes of intensified drought propagation by analysing the basins' runoff mechanisms and the hydrological memory within the basins. However, anthropic factors such as irrigated agriculture, reservoirs, and water abstractions for human consumption may also contribute to different catchment responses to precipitation during persistent droughts. While the water used by natural ecosystems is related to the incoming precipitation (e.g. in water-limited basins, ET is proportional to the precipitation), the water used to supply human demands may remain similar to pre-drought conditions if the associated infrastructure allows for it (e.g. deep pumping wells or large reservoirs). This may cause a decrease in runoff, which is not directly related to precipitation deficits. In this study, we aimed at removing some of these anthropic effects by filtering out catchments with reservoirs, but there are still anthropic activities within the analysed sample. In particular, pluvial basins feature human-induced land cover classes such as cropland, forest plantations, and grassland (Sect. 4.1). Previous findings have related these classes to differences in water provision, particularly during dry years (Alvarez-Garreton et al., 2019). Hence, drought propagation within these basins is likely a combined result of climatic conditions, hydrological mechanisms, and anthropic effects. In contrast, the sample of snow-dominated catchments is less prone to anthropic effects given their location in high elevated areas of the Andes cordillera (no human-related dominant land cover classes within the basins; Sect. 4.1). This suggests that the intensification of drought propagation in these basins is mainly due to climatic conditions and hydrological mechanisms.
Our analysis of 106 basins along central Chile indicates that larger solid precipitation fraction within a catchment leads to increased slow GW contribution to runoff, thus connecting precipitation anomalies in a given winter with streamflow until autumn of the next year. In this way, snow-dominated catchments have a memory that strongly connects streamflow generation over consecutive hydrological years. In pluvial basins on the other hand, hydrological memory is shorter, and the annual streamflow is mostly explained by the precipitation of the current year.
These different hydrological memories have led to contrasted drought propagation during the MD. The MD in central–south Chile has been extraordinary because of its persistence (10 years to date) and extended
spatial domain (
Catchments with longer hydrological memory showed larger shifts in P–R relationships during the MD, compared to their historical behaviour, revealing the intensification of drought propagation during multiyear droughts. For snow-dominated catchments, after 1 year of precipitation deficits, the surface water supply – under equivalent precipitation deficits – significantly decreases. That is, the basins progressively generate less streamflow for a given precipitation amount compared with the historical behaviour because snowmelt infiltrates into depleted levels of GW and does not reach the catchment outlets during the same hydrological year. In pluvial basins, on the other hand, there is a weaker decrease in the water supply after consecutive years of precipitation deficits.
What is worse – an extreme single-year drought or a persistent moderate drought? We have shown that for any type of drought, hydrological memory and initial storage conditions are key factors modulating catchment responses. In snow-dominated basins located in the Andean semi-arid region of Chile, catchments strongly depend on both the current and previous precipitation seasons. In absolute terms, single-year extreme droughts induce larger absolute streamflow deficits (i.e. less water supply). However, moderate but persistent deficits induce a more intensified propagation of the meteorological drought (larger streamflow deficits relative to precipitation). The worst scenario would be an extreme meteorological drought following consecutive years of below-average precipitation, as occurred in 2019. In pluvial regimes, initial conditions and hydrologic memory are still important factors to represent catchment response fully. However, water supply is more strongly dependent on the meteorological conditions of the current year. Therefore, an extreme drought would have a higher impact on water supply than a persistent but moderate drought.
Snow-dominated catchments store water that is then released during dry seasons, a characteristic particularly valuable in regions with a limited water supply and a severe risk to droughts, such as in central Chile. We have demonstrated that these basins are prone to intensify the propagation of persistent droughts, which pose additional challenges to water management adaptation in central Chile given the drying projected trends for the region.
The CAMELS-CL catchment dataset was obtained from the Center for Climate and Resilience Research website at
The supplement related to this article is available online at:
This research was conceived and design by CAG, JPB, and RG. JS and MV implemented the HBV model. CAG wrote the paper with input from all co-authors. All the authors have been involved in interpreting the results, discussing the findings, and editing the paper.
The authors declare that they have no conflict of interest.
This research has been developed within the framework of the Center for Climate and Resilience Research (CR2, ANID/FONDAP/15110009) and the joint research project ANID/NSFC190018. Camila Alvarez-Garreton also acknowledges support by ANID/FONDECYT/1201714. We thank the editor Markus Hrachowitz and the referees Anne Van Loon and Gemma Coxon for their constructive comments that greatly improved the original manuscript.
This research has been developed within the framework of the Center for Climate and Resilience Research (grant no. ANID/FONDAP/15110009) and the joint research project ANID/NSFC190018. Camila Alvarez-Garreton also acknowledges support by ANID/FONDECYT/1201714.
This paper was edited by Markus Hrachowitz and reviewed by Anne Van Loon and Gemma Coxon.