Drought impacts on water quality and potential implications for agricultural production in the Maipo River Basin, Central Chile

ABSTRACT Droughts can have serious negative impacts on the water quality needed for irrigated agriculture. The Metropolitan region of Chile is a relevant producer of high-value crops and is prone to droughts. Standardized Drought Indices were used to characterize meteorological and hydrological droughts for the period from 1985 to 2015. To understand the relationship between droughts and water quality, we evaluated the correlations between daily discharge and surface water quality observations. The threshold level method was used to compare physicochemical parameters during hydrological drought periods with the Chilean water quality thresholds for agricultural uses. A significant (p < 0.05) negative relationship between discharge and electrical conductivity and major ions was found in most of the basin. Hydrological stations located in irrigation districts exceeded the official thresholds for these parameters during hydrological drought periods seriously threatening irrigated agriculture of the region.


Introduction
Drought is a natural phenomenon that can occur in all climatic zones and is defined as a sustained and regionally extensive occurrence of below-normal natural water availability (Tallaksen and van Lanen 2004). In the context of climate change, an increase in global temperatures is projected that will impact evapotranspiration and specific humidity of the air, affecting the ability of the atmosphere to store water, with direct effects on the magnitude, frequency, intensity and spatio-temporal distribution of precipitation (Wang et al. 2016 and references therein). As a consequence, the frequency and severity of extreme events such as droughts and floods would increase, being droughts the slowest to develop, but the longest to last (Bond et al. 2008, Yeh et al. 2015 As a natural hazard, drought can cause economic and social damages (Bond et al. 2008). Agriculture is the primary land use across the globe; it is also a major economic, social and cultural activity, and it is highly sensitive to climate variations (Howden et al. 2007). It is predicted that, by 2050, global food demand will rise, indicating that crop production must double (Tilman et al. 2011). A large fraction of the supplementary food must come from irrigated agriculture (40% of the world's food supply is produced on irrigated lands) (Döll andSiebert 2002, Wang et al. 2016). Currently, irrigation represents nearly 70% of freshwater withdrawals from rivers, lakes, and aquifers and it is under constant pressure created by decreasing water availability and increasing water demand (Assouline et al. 2015). Climate change is expected to intensify the existing pressure on water availability and will affect agricultural systems particularly in semi-arid environments (IPCC 2014). According to Iglesias and Garrote (2015), the potential impacts on the agricultural production in these areas are: (a) an increase in water demand due to higher rates of crop evapotranspiration in response to rising temperatures; (b) intensification of water shortages, especially in the spring and summer due to increased irrigation requirements; and (c) deterioration of water quality due to high temperatures and low levels of runoff in some regions, inducing additional stress to irrigated areas, among others.
Meteorological drought that spreads through the hydrological cycle can reduce surface and ground water levels (triggering hydrological drought) and can lead not only to reduced water availability but also to a deterioration of water quality (Mishra andSingh 2010, Mosley 2015). This condition also intensifies water shortages by lowering the amount of usable water within a region. Although drought impacts on water quantity are well known (Van Vliet and Zwolsman 2008, Safavi and Malek Ahmadi 2015 and references therein), the consequences on water quality are not fully recognized yet (Zwolsman and van Bokhoven 2007). The low flows and water levels during hydrological droughts can affect water quality of freshwater systems by, for instance, increasing its residence time, reducing the flushing rate of water bodies with limited dilution of point source emissions and the interruption of sediment, organic matter and nutrient transport (Van Vliet and Zwolsman 2008, Mishra and Singh 2010, Mosley 2015, Palmer and Montagna 2015. Water quality characteristics such as the increase in total dissolved solids, and their constituent ions and the decrease of dissolved oxygen, could cause agronomic problems to irrigated agriculture (Sagasta andBurke 2005, Maestre-Valero andMartínez-Alvarez 2010). Therefore, it is an important variable to consider in the development of irrigation (Assouline et al. 2015).
In Latin America, the extra-tropical Andes have been identified as a key region affected by climate change. The accelerated retreat of glaciers may have contributed to observe negative trends in streamflow, and the drier areas are likely to suffer from salinization and desertification on agricultural lands (IPCC 2007(IPCC , 2014. Chile is a country that currently experiences multi-year droughts (Rangecroft et al. 2016). According to climate change projections, the country will likely see increases in drought frequency due to increases in temperature across the country as well as reductions in precipitation in the central and southern areas (Magrin et al. 2014). This trend may lead to an intensification of the existing aridity in northern Chile and their advance towards the south, enhancing the risk of water scarcity in the central area (FAO 2010).
Irrigated agriculture in drought-prone arid and semi-arid regions, due to their biophysical characteristics, is particularly vulnerable to this whole climatic, economic and human pressure. Moreover, there is a lack of case studies in Latin America related to drought impacts on water quality (Mosley 2015) and most importantly, its relation to irrigated agriculture. Therefore, the overall objective of this research was to evaluate the impact of drought on water quality and irrigation dependent agricultural production in the Maipo River basin, Central Chile. Specific objectives were to: (1) determine the spatiotemporal distribution of meteorological drought in the basin and explore its relationship with hydrological drought; (2) assess the impacts of hydrological drought on basic surface water quality parameters; and (3) identify the potential effects of drought-driven water quality deterioration on irrigated agriculture.

Study area
This work was conducted in the Metropolitan region of semiarid central Chile, where extended irrigated zones are highly dependent on water withdrawal from surface flows (Rosegrant et al. 2000). The irrigation districts have a wide variety of annual and perennial horticultural and vineyard crops that are supplied by the channel networks from the main rivers (Meza et al. 2012). Furthermore, agriculture represents a 74% of the water use, that is shared with drinking water needs from almost 40% of the Chilean population, and significant industrial and hydroelectric development (DGA 2003, MINAGRI & CNR 2009).
The Maipo River basin is the most populated region of Chile with almost seven million inhabitants including the Santiago metropolitan area. It supplies 70% of potable water and 90% of irrigation water demand of one of the most important producing areas of wine and cash crops in Chile, mostly concentrated in the Central Valley (DGA 2004(DGA , 2007. It is located in Central Chile (Fig. 1), between latitudes 32º55′-34º15′S and longitudes 69°55′-71°33′W, covering an area of 15 274 km 2 . The Maipo River flows 250 km to the Pacific Ocean from the Maipo Volcano (5623 m.a. s.l.). It is characterized by a Mediterranean climatic regime with a long dry season and a winter period (April to September) concentering more than 75% of annual precipitation, along with cold weather in the Andes. Mean annual total precipitation is around 380 mm, and mean annual temperature is 14°C (DGA 2014). Its hydrological regime is nivo-pluvial. According to the Chilean Water Code of 1981 (República de Chile 1981), the Maipo River is divided into three sections ( Figs. 1 and 2). The upper section of the basin is nival, with a significant increase of streamflow in spring months due to snow-melt in the Andes. The middle section is nivo-pluvial with two annual periods of higher discharge (spring and winter), and in the lower section, the regime is pluvial, with large volumes of streamflow associated with precipitation. Surface runoff it is estimated as 58 × 10 6 m 3 , of which 90% is generated by winter precipitation (DGA 2014). The geology in the region is characterized by rocks that range from Triassic to Tertiary in age. The statigraphic units in the Coastal Cordillera are mainly discordant Cretaceous sedimentary and volcanic rocks of continental origin, and of marine origin in intercalations towards the west. In the Andean Cordillera, the units are mainly concordant Jurassic and Cretaceous marine and continental sediments, associated with extensive evaporites (rocks formed by the evaporation of water), composed mainly of gypsum. There are also fluvio-alluvial-glacial deposits of Quaternary age in the Central Valle (Jorquera et al. 2015 and references therein). The geochemistry of bedrock geology in the Maipo River is dominated by outcropping evaporites that are associated with high SO 4 2levels in waters. The stream water in this area has high Ca, Cl, K, SO 4 2and Na concentrations, with a pH range of 7-8.5 (Jorquera 2013).

Hydroclimatic data
Monthly precipitation data of the Maipo River basin was extracted from the Climate Hazard Group InfraRed Precipitation with Station Data (CHIRPS version 2) satellite product. CHIRPSv2 data have a 0.05º spatial resolution with the satellite data calibrated with in situ situation data to create gridded rainfall time-series appropriate for trend analysis and seasonal drought monitoring. CHIRPSv2 spatial coverage spans from 50ºS to 50ºN (over all longitudes) with a spatial data record extending from 1981 to the present (Funk et al. 2015). The performance of this dataset has been previously evaluated in Chile with good results (Zambrano-Bigiarini et al. 2017). Daily discharge and physicochemical data (in general one measurement per season) from nine hydrological stations of the river basin were collected from the Chilean National Water Agency (DGA) ( Table 1). We used the period of record from 1985 to 2015. Maps corresponding to the hydro meteorological network and general characteristics of the basin were obtained from DGA official web page. 1 All  Errázuriz et al. (1998) and the 90-m digital elevation model (DEM) downloaded from Consortium for Spatial Information, CGIAR (http://srtm.csi.cgiar.org) (Jarvis et al., 2008).  maps were processed in ArcGIS (ArcGIS 10.4, ESRI, Redlands, CA, USA). Fig.1 shows the spatial distribution of the selected stations.

Standardized precipitation index (SPI)
The standardized precipitation index (SPI) (Mckee et al. 1993) is the most commonly used drought index due to its limited data requirements and flexibility to assess precipitation deficits over user-defined accumulation periods (Barker et al. 2016). It was recommended as the standard index worldwide by the World Meteorological Organization and the Lincoln Declaration on Drought (Hayes et al. 2011, Stagge et al. 2015, Van Loon 2015. The SPI was computed following the methodology of Mckee et al. (1993) using a gamma probability distribution. Negative values of SPI correspond to lower than the mean precipitation, and positive values indicate higher than the average rainfall. For this study, three SPI time scales were used: 3 (SPI-3), 6 (SPI-6) and 12 (SPI-12) months. The spatial and temporal distribution of meteorological drought across the river basin was represented by sub-basins: Maipo Alto, Maipo Alto-Maitenes, Maipo Medio, Mapocho Alto, Mapocho Bajo and Maipo Bajo. Within each sub-basin, polygons of lower-order sub-basins were used and precipitation was averaged and added for all lower-order sub-basins upstream each discharge station (see Section 3.2.3); therefore, the analysis was made with matching draining areas for precipitation and discharge data. For each lower-order sub-basin and time scale, a historical series of SPI was obtained. Finally, drought characterization was undertaken for each sub-basin and drought index: frequency (number of drought events), duration (number of months with negative values below the threshold), and intensity (lowest SPI and SSI values of the drought event).

Standardized streamflow index (SSI)
The standardized streamflow index (SSI) (Modarres 2007, Vicente-Serrano et al. 2012 was computed in a similar way to the SPI: using monthly streamflow, fitting a probabilistic distribution to the data and transforming it to a normal distribution (Van Loon 2015). There is no consensus regarding the probabilistic distribution of streamflow data (Vicente-Serrano et al. 2012, Svensson et al. 2017. In this study we used the two-parameter Gamma distribution for streamflow, suggested by Mckee et al. (1993) as the procedure can be applied to other variables related to drought (Modarres 2007, Shukla and Wood 2008, Nalbantis and Tsakiris 2009 and it fitted well to our data. The SSI was calculated for four different time scales: 1 (SSI-1), 3 (SSI-3), 6 (SSI-6) and 12 (SSI-12) months in each station from monthly historical streamflow followed by a drought characterization.
In this study, a drought event was defined for values below -0.84, for the SPI starting in the rainy season (April-September) and for the SSI at least three consecutive months according to the criteria from the Chilean government Water Agency, DGA, to determine exceptional droughts (Table 2) (DGA and DICTUC S.A. 2009, DGA 2012). The "SCI" package for R (Gudmundsson and Stagge 2016), was used to calculate SPI and SSI.

Relationship between meteorological and hydrological drought
The relationship between meteorological and hydrological drought in the Maipo River basin was assessed by applying the seasonal autoregressive integrated moving average (SARIMA)-based cross-correlation using the Pearson correlation coefficient from the SPI time series at different time scales of the lower-order sub-basins derived from the upstream precipitation of each discharge station and the SSI. A SARIMA model is notated as (p,q,d)(P,D,Q), where p is the autoregressive (AR) order, d the differencing order, q the moving average (MA) order and P, D, and Q are the seasonal autoregressive, differencing and moving average order, respectively. The goodness-of-fit was determined based on the Akaike information criterion (AIC), the corrected Akaike information criterion (AICc) and the Bayesian information criterion (BIC). The autocorrelation function (ACF) and partial autocorrelation function (PACF) were applied to determine the AR and MA order of the time series. Prior to cross-correlation, the two time series were pre-whitened by the SARIMA model to remove autocorrelation within each individual series and make nonstationary time series stationary (Box and Jenkins 1976). All analyses were conducted in R 3.4.3 (R Core Team 2014).

Threshold level method
The threshold level is a method that simultaneously characterizes streamflow droughts regarding duration and deficit volume, defining droughts as the periods during which the flow values fall below a predefined threshold level and continues until the threshold is exceeded again (Yevjevich 1967, Hisdal et al. 2004. For hydrological droughts, monthly or longer time resolutions could underestimate cases when severe droughts are shorter than 1 month, especially when it is analysed for a specific activity. In these cases, streamflow deficit characteristics operating on shorter time resolutions like daily discharge series could be more appropriate to obtain a more detailed information (Fleig et al. 2006). Generally, a truncation level can be fixed or varying over the year (Hisdal 2004) and derived from percentiles of the flow duration curve (70th to 95th) which reflects the deficit volume of water in discharge or water levels in natural and artificial reservoirs (Tallaksen et al. 1997, Van Loon 2015. For perennial rivers, those percentiles represent the streamflow that equaled or exceeded 70-95 percent of the time, respectively (Fleig et al. 2006). In this study, daily streamflow of the selected discharge stations and its related physicochemical data were used to determine the impact of hydrological drought on water quality by the threshold level method. Due to the strong seasonal pattern of the river basin, the variable monthly threshold was selected. The 80th percentile (Q80) was calculated for dry periods with 3 days of minimum duration and inter event time, to eliminate minor droughts and pooling mutually drought events (Zelenhasić andSalvai 1987, Van Loon et al. 2010), and the 20th percentile (Q20) was also considered to establish the behavior of the physicochemical parameters in wet periods (extreme flows) (Hrdinka et al. 2015). The wet and dry periods were compared with the normal stages which fall between both thresholds (Q20-Q80).

Physicochemical parameters and thresholds for agricultural water use
Surface water quality data from the period 1985-2015 of differing temporal resolution (in general one measurement per season) were evaluated for all discharge stations. The Chilean regulation NCh 1333/78 "Water quality requirements for different uses" (INN 1987) was used as a reference to select the physicochemical parameters and to compare their values to the established thresholds of water quality for irrigation. These parameters were also chosen according to previous drought and water quality studies and having at least 20 years of data. Measured parameters included: temperature (ºC), pH, dissolved oxygen (O 2 ; mg/L), electrical conductivity/salinity (μS/ cm) and related ions, nutrients and metals, such as Zn, Cu and As. The DGA follows the sampling and analytical procedures described in the "Standard Methods for the Examination of Water and Wasterwater" (APHA AWWA WEF 1995), details associated with these analysis are described in (Oyarzun et al. 2006). The database was provided with the correspondent daily discharge value. To establish a significant data set showing potential effects of drought on water quality, the data were grouped according to the respective threshold value (Q80 and Q20) and then compared. To examine, at a significant level (P < 0.05), the degree of association between measured water quality parameter and daily discharge, we used Spearman's correlation coefficient. Table 3 shows the characterization of meteorological and hydrological droughts. Temporal variations of meteorological drought are shown in Fig. 3. It can be noted that all sub-basins depicted similar patterns in the frequency and duration of meteorological drought events. Common drought episodes occurred during the years 1988/ 89, 1990/91, 1996/97, 1998/ 99, 2011/12 and 2014/15. All the sub-basins reached the extremely dry category, whereas the lowest SPI-3 and SPI-6 (<-2.05) values were obtained for the period 2009-2015 in five of the six subbasins. The lowest value in Maipo Alto Maitenes was observed in 1989. None of the sub-basins showed the extremely dry category in the SPI-12 accumulation period. Also, most of the sub-basins experienced the longest drought event in the period 1988-1999 and 2010-2014. During 16-18% of the study period the basin experienced meteorological drought. The frequency of drought events ranged between 8 and 22 times during the period of record.

Spatio-temporal distribution of meteorological and hydrological drought
Although the hydrological droughts were less frequent (3-12 events) compared to meteorological droughts (8-22 events), these were longer, obtaining periods up to 64 months under hydrological drought. This pattern was also observed in the percentage of time in which most of the basin experienced hydrological drought (20-23%, with the exception of station 9D, 10%), being higher than meteorological drought (16-18%). Also, the longest drought periods in most of the basin occurred in recent years (2010)(2011)(2012)(2013)(2014)(2015), while in meteorological droughts, the longest and most intense droughts were in the period of 2010-2013 and also 1988-1999. All stations, except 1A, reached the most extreme drought class. As is also shown in Table 3 for both SPI and SSI, the frequency of drought events was higher at shorter accumulation periods. Hydrological drought characteristics based on SSI exhibited more spatial variability than the meteorological drought. The stations located in Maipo Alto Maitenes (1A) and Maipo Bajo (9D) showed the shortest hydrological drought duration from the basin. Stations 3B, 5C, 7C and 9D had most of the lowest values in the period 1994-1997. In contrast, stations 1A, 2A, 4B and 6C showed the lowest values in the period 2009-2015. Station 8C presented its lowest value (-2.62) in both periods. Figure 4 shows the correlation coefficients for SPI and SSI at all temporal scales. Positive relationships were obtained for most of the stations and the highest correlation of SSI in all temporal scales was identified with the highest accumulation period SPI-12 (Fig. 4). Also, correlation values tend to increase from the upper to the lower part of the basin. This pattern might show the significant high dependence of the streamflow on snowfall during the previous hydrological year. Fig. 5 provides an illustrative overview of the analysis. It shows the temporal variation of the SSI-1 and the SPI at 12-monthly scale in the station 4B, which are the time scales with one of the highest correlations (R = 0.77) and the station 1A with one of the lowest correlations (R = 0.43). Also, the figure shows the dominance of dry hydrological events during the last 10 years.

Hydrological drought and water quality
Spearman correlation coefficients between daily discharge volumes and the related physicochemical parameter from the period of record 1985-2015 are shown in Table 4. Mixed responses were observed for temperature and dissolved oxygen, and between 30% and 50% of the stations did not show significant correlations with discharge in both parameters (p < 0.05). On the one hand, significant negative relationships are found in the stations located in the lower part of the basin 8C and 9D for temperature (-0.24 and -0.72; p < 0.05), which have mixed and pluvial regime. On the other hand, stations Table 3. Characteristics of meteorological and hydrological drought events . * Drainage area considered as the representative SPI for each sub-basin. with a nival regime such as 1A, 2A, 3B, and 5C showed positive relationships (0.42, 0.46, 0.61 and 0.60; p < 0.05) with temperature. In the stations showing positive relations with temperature, significant negative relationships were found for dissolved oxygen in two of them (3B and 5C) (p < 0.05). Nonetheless, this parameter showed one of the lowest range of significant correlations within the whole set of physicochemical parameter data.

Sub-basin
Significant negative relationships were observed for pH, electrical conductivity, Na, Mg, Ca, SO 4 , Cl, K, NO 3 and PO 4 . For these parameters, most of the highest concentrations were found in drought periods. The percentage of response to this characteristic is variable in the basin ranging from 100% to 20% of the stations (Mg and NO 3 , respectively). Nitrate was the parameter showing the lowest correlation, supported by the fact that it did not show strong variations in its concentration due to periods of variable streamflow. The station 7C was the only site displaying a significant positive relationship of PO 4 (0.26; p < 0.05). For electrical conductivity, only the stations 5C and 9D from Mapocho Alto and Maipo Bajo sub-basins did not show a significant relationship with discharge, finding relatively constant values of this parameter in dry, regular and wet periods. The highest significant negative correlations (p < 0.05) for electrical conductivity (r s = -0.94) and related ions such as Na (r s = -0.92) and Cl (r s = -0.94) were found for the same station (3B); this behavior was a continuous pattern at most of the locations. In contrast, significant positive relationships (p < 0.05) were found for Al, As, Cu, Mn, Fe, and Zn within a range of 90% to 60% of the stations showing this pattern (Fe and Mn, respectively). Most of the highest concentrations of these parameters were observed in wet periods. The station 6C was the only site with a significant negative relationship of As and Zn. Table 5 shows the water quality threshold values (NCh 1333/ 78) compared to maximum, mean and minimum water quality values for three key station within the three subareas of the Maipo located in irrigation districts: 4B, 8C, and 9D (Fig. 1) and selected crops tolerance to salinity (FAO 1985) and chloride. At these stations, the parameters were compared with thresholds from official regulation for irrigation water uses.

Drought driven water quality impacts on irrigated agriculture
Physicochemical data were grouped according to the corresponding thresholds levels (Q80, Q20-Q80 and Q20) and the threshold for each parameter according to the Chilean regulation for irrigation water uses (dashed lines). Related curves represent its relationship with daily discharge (solid black line) of the station 4B, as shown in Fig. 6. Stations 4B and 8C showed significant positive relationships with dissolved oxygen, which tends to decrease during low flows (Fig. 6, station 8C not shown). Although pH had a negative relationship with discharge for station 4B, its values remained in the allowable range for agricultural water use (5.5-9). As heavy metals like As and Cu could increase with high streamflow (Table 4), during drought periods their concentrations were not surpass-  1986  1987  1988  1989  1990  1991  1992  1993  1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014   The SPI of each sub-basin is represented by the largest drainage area of the related discharge stations (Table 3). Pink shading indicates the -0.84 threshold value defined for drought events. Blue shading shows the major drought events (longer than 3 months).
ing the established limit. However, variables such as electrical conductivity, SO 4, and Cl showed an exponential behavior while the streamflow decreased (station 4B and 8C), and reached their maximum mean concentration during hydrological drought periods, markedly exceeding its relative threshold ( Fig. 6; Table 5). Most of these maximum and minimum values related to water quality for agricultural water use were reached in recent periods of extreme low flows (2010-2015).

Spatio-temporal drought characterization
The drought events identified in the scope of this study coincide with those described in previous publications (Garreaud et al. 2017, Oertel et al. 2018. Although the basin has a precipitation gradient from the east (Andes mountains), decreasing in the central valley and rising again to the west (Pacific Ocean coast), similar meteorological drought frequency, duration and intensity values were found for all the sub-basins. This result could be mostly related to the shared Mediterranean climate of the central Chilean region, in which the differences in altitude related to temperature and precipitation are not strongly influencing the meteorological drought pattern. Furthermore, the basin can be considered as a homogeneus region in terms of the underlain precipitation probability distribution, regardless of the mean annual precipitation at any place across the basin (Núñez et al. 2011). The same applies to other arid regions in Latin America like Mexico (Núñez et al. 2016). Similar results were reported by Van Loon and Laaha (2015), where comparable meteorological drought conditions were found in catchments from Austria mainly governed by a continental climate type. Also, Barker et al. (2016) found comparatively little differences for shorter accumulation periods in SPI for different catchment types in the UK. The monthly streamflow deficit (SSI-1) in the basin ranged from 6 to 30 months; from which the longest corresponds to the station 4B located in the central valley, reflecting the anthropic intervention. Longer drought events were observed for recent years (2010-2015), indicating that the rainfall following droughts was unable to generate the surpluses needed to restore hydrological conditions that were present prior to the drought and to compensate the increasing demand from the activities within the basin (agriculture, industrial and urban supply) (Lorenzo-Lacruz et al. 2013a). The apparent long droughts (from 3 to 12 months accumulation period), are the result of strong long-term decreasing temporal trends in streamflow (Barker et al. 2016), being the SSI-12 the measure of the annual drought severity. The decrease in the annual streamflows since 2010 in the region was also reported by the Center for Climate and Resilience Research (CR) 2 (2015).
In contrast to meteorological drought characteristics, hydrological droughts detected by SSI show differences among the sub-basins. The station located at the higher sub-basin (1A) shows a lower drought duration compared to the middle zones of the basin. The hydrological drought periods in the basin were not restricted to the rainy season (winter-autumn) but also to the snowmelt months (springsummer). This pattern can be explained by the relationship between precipitation and snowpack accumulation, and the impacts on spring and summer discharge in rivers with a nival regime in the Central Andes (Masiokas et al. 2006). Carrasco et al. (2005) reported for central Chile that the number of days with precipitation have decreased after the 1970s, suggesting that glaciers are exposed to more days with atmospheric conditions appropriate for melting. Also, the same study showed an increase in elevation of the 0°C isotherm in winter and summer for the period 1974-2001, and that warming has concentrated during periods with no precipitation, enhancing melting of glaciers that lead to its recession.
According to Van Loon et al. (2010) the combined effect of higher temperatures in winter with lower precipitations determines the development of hydrological droughts in snow affected regions: when the recharge from snowmelt is often too low to end the drought, and it continues into summer having an adverse impact on water resources. Also, when winter temperatures are above 0°C, the snow accumulation is limited, and a continuous snow cover does not develop.

Relationship between meteorological and hydrological drought
Positive correlation coefficients were found in most of the basin between SPI and SSI in which as SPI accumulated period increases, the correlations with SSI increase. The highest correlation of SSI at all temporal scales was found with the SPI-12 accumulation period. This pattern increases from the upper to the lower part of the basin, denoting the dependence of the streamflow on snowmelt and the previous year of precipitation especially downstream. Some authors have evaluated the propagation of meteorological drought through the hydrological cycle, linking different time scales of SPI to SSI-1, which represents the streamflow deficits. Lorenzo-Lacruz et al. (2013b) found similar results, attaching the response to the highpermeable geologies of the catchments characterized by limestone lithology in the South of the Iberian Peninsula which results in groundwater recharge delaying the streamflow response to meteorological variability. In the Maipo River basin, the regions close to the Andes mountains present an extremely low permeability due to the presence of granitic rocks. However, the central depression corresponding to a tectonic pit and the region closest to the cost, with a large influence of the presence of Cretaceous volcanic and sedimentary rocks, present a permeability from medium-high and medium-low, respectively (DGA 2004). The low correlation in the station 1A (1500 m a.s. l.), the highest altitude analysed in this study for discharge, could be linked to its relationship with snow, which is a key factor explaining the variability of the hydrological response. Further analysis (data not shown) of this station revealed that it continues to increase its correlation up to SPI-24 and then starts to decrease. López-Moreno et al.   1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

Drought impacts on water quality
In the following sections, surface water quality and its relation to discharge and especially low flows are discussed. However, it is important to highlight that the seasonality of the low flows (which was not restricted to summer droughts), the mixed regime of the basin, and the anthropogenic activities, strongly influence the behavior of the parameters.

Surface water temperature
Surface water temperature at stations located in irrigation districts and with a mixed or pluvial regime, showed a significant negative correlation with discharge. This behavior can be attributed to the inverse relationship between water temperature and river discharge with higher warming rates under low-flow conditions (Caruso 2002, Van Vliet andZwolsman 2008). Low flows could cause longer water residence times and reduced water volumes allowing higher water temperatures as a result of reduced thermal capacity and higher sensitivity to atmospheric and other heat inputs (Mosley et al. 2012). Positive relationships were observed mostly at stations located in higher zones with a nival regime which can be explained by the fact that higher flows at these elevations correspond to colder melting water from the Andes Mountains.

Electrical conductivity/salinity
Salinity expressed by electrical conductivity and major ions (Cl, Na, SO 4, Mg, Ca, K) had significant negative correlations with discharge and increased in almost all the stations analyzed during low-flow conditions. Similar results have been reported by Nosrati (2011) (Jorquera et al. 2015). Droughts cause substantial rises in salinity, and it has been reported an inverse relationship between low-inflow and high salinity periods, where a change in discharge flows can increase the conductivity concentrating the present dissolved contaminants (Prathumratana et al. 2008). The decreased dilution of saline groundwater inputs and high evaporation rates along to longer water residence times benefits salt accumulation (Caruso 2002, Mosley et al. 2012). In the case of the station 5C, that does not show a significant relationship with electrical conductivity, suggest that the natural conditions of the station (river of glacial origin, rich in ionic compounds) prevailed against the variable streamflow (DGA 2004). Also, the station 9D located in Maipo Bajo which showed a similar behavior could be mostly related to the dominance of groundwater outcrops and irrigation returns from the areas served by the channels in the irrigation districts (DGA 2003;DGA 2004).

Dissolved oxygen (DO)
The responses related to DO are variable but with lower significant relationships than the other parameters. We would expect decreased DO values under low flows due to a lower oxygen saturation concentration under higher water temperatures (summer droughts), higher rates of organic matter decomposition by microorganisms, and lower re-aeration rates (Prathumratana et al. 2008, Van Vliet and Zwolsman 2008, Palmer and Montagna 2015. However, this positive correlation was only observed at two stations located in irrigation districts. Increases in DO during drought periods, have been reported by Zwolsman and van Bokhoven (2007) due to an increase in primary production in the river water and by Attrill and Power (2000) as a result of the reduction in the number of storm events that lead to a less untreated sewage inputs or/and improved effluent quality from the wastewater treatment.

pH
Six stations showed significant negative correlations with pH.
In those cases where pH increased, it is likely that lower water volumes favored stagnant conditions for algae growth which along with plants use all the carbon dioxide dissolved (Van Vliet and Zwolsman 2008, Gołdyn et al. 2015 and references therein). In other cases, and depending on the water body and its location, increases in pH during low-flow conditions could be due to marine intrusion (Senhorst andZwolsman 2005, Prathumratana et al. 2008).

Nutrients
We found significant negative relationships for nitrate at two stations (1A and 5C), significant negative correlations of phosphate at two stations located in irrigation districts (4B and 8C), and one significant positive correlation at station 7C. This erratic behaviour can be related to the constant nutrient input from the different human activities with anomalously high concentrations mainly downstream in the Central Valley and the Coastal Cordillera in agricultural areas (Jorquera et al. 2015). It is expected that a higher nutrient content during droughts could have a connection with a reducing dilution capacity of the water body (Zwolsman and van Bokhoven 2007, Prathumratana et al. 2008, Van Vliet and Zwolsman 2008. On the contrary, deficient levels of dissolved nutrient concentrations in rivers could lead to rapid assimilation by algae where differences between nitrogen and phosphorus concentration could be linked to nutrient limitations and increasing denitrification process with higher water residence time (Cook cited Mosley et al. 2012). Also, low nutrient levels can be due to a reduced non-point source runoff inputs (Caruso 2002).

Metals
Significant positive relationships with discharge were found at most of the stations for essential (Zn, Cu, Fe and Mn) and nonessential (Al, As) metals (DGA 2004), as most of the concentrations of those variables increased during wet periods.
Only the station 6C showed a significant negative relationship with discharge. Similar positive correlations between Fe and discharge have been observed by Flores et al. (2017) in the Elqui river basin in North Central Chile, but with a more erratic behaviour for Cu and As. It is expected that several elements from regular anthropogenic activities can concentrate due to limited dilution (Zwolsman and van Bokhoven 2007, Van Vliet and Zwolsman 2008). However, for inorganic pollutants, an indirect relation could be assumed. The low flows along with elevated temperatures, could modify the rates of respiration, reaeration and retain water quality constituents which may be released during wet conditions (Mosley 2015). This condition means that components sediment can be delivered with high flows (thaw streamflow) after a drought period, enhancing their concentration in the river. Flores et al. (2017) suggested that turbulent flows associated with higher discharges could favour the transport of particulate forms of Fe in the Elqui basin. Furthermore, and according to Mosley et al. (2014), in the acidification process resulted from sediment exposure to air during lower water levels, it can release high and potentially toxic concentrations of dissolved metals.

Overall perspective of hydrological drought impacts on water quality
Impacts on water quality and their magnitude depend on the specific characteristics of the water body: water body type (for instance river, lake, estuarine) size, depth, shape, water regime, water residence time, the human activities in the catchment among others (Van Vliet and Zwolsman 2008, Delpla et al. 2009, Li et al. 2016). In addition, according to Zwolsman and van Bokhoven (2007) the impact of droughts on water quality would be greater when the water quality is already poor. Although the reductions in nonpoint source pollution during drought periods can improve the quality of the receiving water body (Caruso 2002), the decreased dilution of point source pollution during low-flow conditions can increase its concentration (Zwolsman and van Bokhoven 2007). Sources of diffuse pollution in the Maipo River basin are pesticides (mainly applied during summer), solid household waste or minor wastes from agriculture, clandestine landfills along rivers and canals and infiltration from black wells, especially in rural areas (DGA 2008). Fertilizers are the principal source of nitrate in tributaries of the Maipo River in the central valley and high values of K, Mg, NO 3 and P are also related to the presence of septic waste and detergent in sewage (Jorquera et al. 2015). According to DGA (2004), liquid industrial waste has been a significant source of deterioration in the basin, turning it in the most contaminated basin in the country. An additional effect is that during a drought, higher temperatures will increase mineralization and releases of nitrogen, phosphorus, and carbon from soil organic matter which will be discharged during runoff and erosion from intense precipitations after a drought period transporting the contaminants to the water bodies (Delpla et al. 2009). Therefore, water quality problems can not be limited only to the drought period itself. Drought-rewetting cycles may impact water quality for a longterm, as it enhances decomposition and flushing of organic matter into streams (Evans cited Delpla et al. 2009, Huang et al. 2015. This wet and dry cycles also can increase the bioavailability of heavy metals due to less stable forms of metal deposits (Claessens and van der Wal cited Verweij et al. 2010).
In general terms, across the stations, electrical conductivity/ salinity has been one of the most direct drought-related quality parameters and can be suggested as a characteristic variable for decision-making. Palmer and Montagna (2015) recommend salinity-based predictions to dry regions, which can have more drought-like inflows more frequently than natural droughts due to its water demand. In addition to electrical conductivity/ salinity, total suspended solids and alkalinity have been also proposed for monitoring impacts (Prathumratana et al. 2008). Under continuous climatic variations driven by global warming and the developing research in its impacts on water quality, it is important to involve comprehensive and efficient governmental measures respecting this aspect. Also, be aware of its implications to develop drought mitigation and adaptation measures for water quality impacts.

Potential effects of drought driven water quality deterioration on irrigated agriculture
The production of fruit trees has great importance at the national level, especially grapes and avocado plantations mainly destined to exportation (MOP 2007). According to INE (2007), the local agricultural land is cultivated according to the order of importance of crops and fruit plantations, annual and permanent forage, vegetables, and vineyards which concentrate mostly in the central Maipo Valley. The most important and extensive agricultural areas according to crop type are located in Melipilla, Maipo and Cordillera provinces, where two of the evaluated stations are located. The exceedance of the Chilean thresholds at these stations during hydrological drought periods could be linked to potential effects on irrigated agriculture as described in the following section: Commonly, relevant water quality parameters for agriculture include salt content, sodium concentration, presence or quantity of macro and micro nutrients and trace elements, alkalinity, acidity and hardness of water (Bauder et al. 2008(Bauder et al. , 2016. Nonetheless, according to its priority, salinity is the most important factor for assessing irrigation water quality (Ghassemi et al. 1995). The electrical conductivity and major ions analyzed at two stations of the irrigation districts had a significant negative correlation with discharge (Table 4). According to DGA (2004), the values of electrical conductivity in the Maipo River surpass the regulation under normal conditions due to the presence of salts in natural conditions from leachates, groundwater outcrops in the lower parts of the basin and by the effect of the wastewater discharge. However, the periods of low flows can alter to a large extent the concentration of salts harming the water quality for irrigation purposes. The values of electrical conductivity vs. discharge had an exponential relationship with streamflow at stations 4B and 8C, reaching values of 2213 μS/cm (station 8C) water that can only be used by salinity-tolerant plants. Avocado is highly sensitive to the salts in irrigation waters (<750 μS/cm) and although in the metropolitan region it is common the use of rootstocks more resistant to the salinity for obtaining better yields, this concentration could affect its productivity greatly (INDAP 2007). Also, the use of saline water in the land can cause, from the evaporation process, the salinization of the soil (Sagasta andBurke 2005, Multsch et al. 2017). Soil salinization can reduce soil productivity in early stages; it also has a lethal effect on vegetation in advanced stages, and high salt concentrations inhibit the uptake of water by plants (Sagasta and Burke 2005). The worst panorama in this case could be a salt crust on the soil surface, making agriculture impossible (Multsch et al. 2017).
Regarding the major ions impacts, the high proportion of Na, sodicity, in relation to Ca and magnesium Mg ions, can degrade the soil structure resulting in a more erodible and impermeable soil that could reduce plant growth (Sagasta and Burke 2005). In the Maipo River and under normal conditions, chloride concentration decreases, from the mountain range to the ocean; whereas in the Mapocho sub-basin chloride concentration registers an increase due to the great anthropic influence (DGA 2004). Higher chloride concentration can burn plant leaves during irrigation (Dezfuli et al. 2013). In the case of strawberry crops, they are highly sensitive to salts especially sodium and chloride, and saline soils up to max 2500 μS/cm, although it is suggested a maximum of total salt of 400 ppm (INDAP 2007). However, in drought periods only chloride have reached at least the half of the concentration limit for this crop, obtaining the same values for sodium (from 145 to 350 mg/L as maximum values) in the three stations. The same behavior is observed for sulphate, where is surpassed the threshold in the three stations. The presence of sulphate in the sub-basins of the Maipo River basin is attributable to the exploitation of plaster in the area, with an increasing tendency downstream, which could be explained by the evapotranspiration associated with the intensive use of water in irrigation (DGA 2004). Sulphate and Cu concentrations exceeding water irrigation thresholds in the basin have been also reported by Pizarro et al. (2010) with an increasing trend over time. This increase is probably a response to dynamic processes resulting from the interaction between mining pollution, erosion, and natural processes such as weathering or flooding.
Although dissolved oxygen is not included in the regulation, according to Bhattarai and Chérif cited Maestre-Valero and Martínez-Alvarez (2010), dissolved oxygen concentration is an important irrigation water quality parameter, due to low values can cause oxygen deficiency in the root zone of plants inducing poor root and plant performance, making them prone to disease, with slow growth and low yields. Therefore, the significant positive relationship found in two of the three stations, although low, could suggest an increased risk during low-flow conditions for the irrigation areas. Regarding pH, plants irrigated with great changes in water pH can affect rhizosphere pH, impacting plant growth, photosynthesis and nutrient absorption (Zhao et al. 2013). The station 4B was the only one having a significant negative relationship with discharge, although it did not surpass the range of the threshold. The crops that are sensitive to more basic pH could be at risk during low-flow conditions. Nonetheless, essential and nonessential metals parameters had a significant positive relationship with discharge, thus, the high flows after droughts should be studied more carefully to determine if the increase in the concentration of pollutants are due to settlement in the river bottom during low-flow conditions. Higher Mn concentration can cause phytotoxicity which is mostly related to the reduction of biomass and photosynthesis inhibition (Al-Isawi et al. 2016). Also, an arsenic concentration above the threshold may result in land degradation regarding crop production and food safety due to it is absorbed by crops adding it to the dietary intake and in a long-term reduce yields because of toxic accumulation (Heikens 2006).
In summary, the results suggest that in dry years the whole water cycle is affected by the reduced reservoir recharge and snow storage that will extend to the zones of the basin that depend on them. The important precipitation input from the months of the previous hydrological year(s), could also represent the retreating in the snow cover for the thaw months affecting the recovery of the low flows from the rainy season. Summer and spring months in particularly dry years could be warmer (Carrasco et al. 2005) and this results in a higher demand from the different activities (e.g. higher rates of evapotranspiration which means more irrigation demands). Low flow conditions combined with inputs from economic activities, deteriorates the water quality of the main rivers from which the water will be distributed through channels to the irrigation districts, causing a major risk for agriculture.

Conclusion
We used standardized indices to analyse drought in different sections of the Maipo River basin and assessed their relationships with water quality variables. The longest hydrological drought periods in most of the basin occurred in recent years (2010)(2011)(2012)(2013)(2014)(2015). The highest correlation of SSI with SPI-12 denotes the dependence of the streamflow on the previous year of precipitation. In general terms, across the stations, electrical conductivity/salinity has been one of the most direct droughtrelated water quality parameters and can be suggested as a characteristic variable for decision-making. The electrical conductivity and major ions analyzed in the irrigation districts increase during drought periods, reaching values that had surpassed the Chilean regulation for agricultural water uses. This increase can affect sensible crops produced in the basin such as avocado plantations. For inorganic pollutants, a more deep analysis should be carried out to determine if there is a relationship between drought re-wetting cycles and the increase of its concentration in wet periods. It is important to highlight that this basin is highly influenced by human activities that could impact the behavior of the physicochemical parameters being difficult to attribute it to a unique source. Therefore, this study stands out the importance of tracking the anthropogenic contribution to water quality under drought conditions within the framework of climate variability and change. Deeper and more frequent monitoring is needed to differentiate and detail the changes during the different seasonal droughts and the role of natural conditions. Although the present study focuses on central Chile, its results can be of benefit to all Mediterranean areas of the world. Because of their similar climates, these regions have much in common: cash crops, major wine industries, urban development pressure, facing water-shortage through months of summer drought in the present, and the forecasted climate change impact to make them hotter and drier in the future. To conclude, this study gives a valuable overview of drought-driven water quality deterioration and its potential impacts on irrigated agriculture, indicating that further water management strategies should consider monitoring water quality during drought periods.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
Funding for field research was provided by the BMZ/DAAD (German Federal Ministry for International Cooperation and Development) in the scope of the project "Education and Research related to the Water Energy and Food Nexus at River Basin Scale (EDUNEXUS)". A Master scholarship from the Mexican Council of Science and Technology (CONACYT) was granted to M.D Peña-Guerrero.