Streamflow in south-central Chile (SCC, ∼ 37–42∘ S) is vital for agriculture, forestry production, hydroelectricity, and human consumption . With more than 55 % of Chilean agriculture and forestry production delivered from this region , the drought episodes that occurred in the last few decades and the associated hydrological deficit have had damaging effects . These drought episodes have been linked to a significant decreasing trend in regional precipitation , with amounts 40 % below the 1901–2005 mean . SCC is expected to undergo important climate changes. Analysis of the multimodel ensemble for the scenario RCP4.5 presented in the Atlas of Global and Regional Climate Projections indicates 10 to 30 % reduction in spring and summer (October to March) precipitation by 2016–2035 and 2046–2065 relative to 1986–2005. The same projection forecasts 0.5 to 2 ∘C warming for summer (December–February). Drier and warmer summers may make SCC more vulnerable to water scarcity, given that this is the season of highest water demand in this region .

Current and projected increases in water demand in this region are likely a source of high uncertainty for future scenarios of water use . However, the lack of long-term observational records of hydrometeorological variables makes difficult the development of accurate estimations of natural variability, useful for determining the severity of the recent observed deficit in streamflow . Therefore, there is a need to extend the observational record to better understand long-term (e.g., centuries to millennia) variability and trends in streamflow, thus providing useful information for hydrological assessment of governmental and private planning initiatives within watersheds of SCC .

The Río Imperial, a major river in SCC, drains an area of 12 763 km2, extending from ∼ 37 ∘40′ S to ∼ 38 ∘50′ S (Fig. 1). The river begins at the confluence between the Cholchol and Quepe rivers, in the ninth Chilean administrative region, known as Araucanía. Summer streamflow is controlled by rainfall, as most snowmelt-dominated discharge (usually occurring in late austral spring) has vanished . This river is currently utilized for irrigation, fishing, tourism, and transportation (in certain sections). For instance, in 2001 4 % of the basin area was potentially usable for irrigation agriculture, with some projections suggesting that this proportion will grow to ∼ 10 % . Additionally, the Río Imperial has high hydroelectric potential, ranked 11th in the country, with 455.8 MW . An emerging sociohydrological problem in this basin is related to the allocation of water rights by the government: there is more streamflow allocated than the current river discharge. What seems to prevent a water availability crisis is that not all the allocated water is being exploited . Facing scenarios of water scarcity may obligate some users to fully claim their rights, likely triggering a regime shift with uncertain implications. However, the ability to ponder the possible consequences of these predicted changes is limited because accurate calculations of extreme scenarios are essential but largely nonexistent. For this reason, long-term evaluations of Río Imperial discharge are urgently needed.

Tree-ring analysis has proven to be a useful tool for supplementing available observations of streamflow. These proxy records can provide time series at yearly resolution, capable of delivering historical information on natural variability, estimations of return periods of extreme events such as droughts, and on the correlation with large-scale climatic forcings . In Chile, dendrohydrology has already been utilized for streamflow reconstruction in four major rivers: Maule (35∘30′ S; ), Biobío (37∘10′ S; ), Puelo (41∘39′ S; ), and Baker (47∘40′ S; ). To date, however, these studies have not yet been fully utilized for watershed planning and management. In order to provide more information on long-term streamflow variability and trends in SCC and thus contribute to improving water resource management in this key region for the country's economy, here we report findings of a tree-ring-based reconstruction of the summer streamflow of Río Imperial. This is the first reconstruction of streamflow in summer, the season with the highest water demand. Our research focused on expanding current understanding of long-term streamflow, thus providing more evidence of both regional coincidence among studies and seasonal/local particularities in streamflow. Therefore, our objectives were to (a) determine whether or not current summer streamflow changes in the Río imperial are unprecedented at the multicentury scale, (b) establish if the return period of extreme years of high and low streamflow has changed in the last few centuries, and (c) estimate the correlation between the multicentury variability of the Río Imperial streamflow and climate modes of natural variability such as the El Niño Southern Oscillation (ENSO), the Southern Annular Mode (SAM), and the Interdecadal Pacific Oscillation (IPO). Given that this is the fifth streamflow reconstruction for a Chilean river and that available results from previous studies have yet to be utilized for water resources planning, in our discussion section we also (a) summarize arguments related to physical processes that may support our summer reconstruction as a more robust time series compared to some of the previous studies, and (b) in light of our results, evaluate strengths and weaknesses of a recently enacted method to calculate minimum ecological discharge, hoping to engage authorities, managers, and stakeholders in considering proxy-data for improving water resources management and research.

The tree-ring reconstruction for the Río Imperial has revealed new information on and insights into long-term streamflow variability, thus allowing for the assessment of changes in summer water availability in the SCC region, including the long-term significance of recent extreme hydrological events throughout the region. This streamflow reconstruction corresponds to the fifth record for Chilean rivers, but the very first focused on summer streamflow. Some of our findings agree with those of previous studies, which strongly suggests that our reconstruction represents well the summer hydroclimate of a large region in SCC. This is manifested by the good skill of our reconstruction in simulating instrumental records along the wider region 35–42∘ S (Fig. S2), which further supports findings by , who stated that this region configures a hydroclimatic cluster.

The noticeable decreasing trend post-1980 observed in the instrumental record as well as in our reconstruction, which was to a certain extent identified in all previous reconstructions , corroborates the occurrence of an ongoing and unprecedented long-lasting summer low flow in SCC, which also includes the recent drought detected for central Chile . We note that the post-1980 period represents ∼ 50 % of the composite instrumental time series, while it only represents ∼ 10 % of the reconstructed period; we posit this highlights the uniqueness of the post-1980 period. Thus, this period is noteworthy as the one with the lowest summer streamflow of the whole reconstruction. This low-flow period gradually emerged as the most acute as the time window was broadened (Table 5), despite the fact that no year during this timespan ranked high in the lowest percentile of the reconstruction. As shown above, the decreasing trend observed in our results has already been detected in other rivers of the region, specifically south of ∼ 37 ∘30′ S . The implications of this finding for water management and planning purposes are related to the ability to detect the effect of regional climate changes in the area. The increasing number of low-flow years may indicate a regime shift in the summer dynamics, which can be exacerbated with predicted increased warming and precipitation reduction in SCC . Our results indicated a decreasing trend in summer streamflow with a smaller natural variability when compared with the rest of the tree-ring record.

Our results additionally posited the SAM as the most influential climate mode driving the variability of the Río Imperial's summer streamflow, especially during the second half of the 20th century. The correlation we found between SAM and the reconstruction establishes the tree-ring time series as independent new proof of the importance of the SAM in modulating precipitation and streamflow in a large region of central and southern Chile . The SAM has been extensively described as a strong modulator of precipitation variability in SCC . The tree-ring reconstruction showed a statistically significant correlation with the SAM reconstruction presented in , especially for the long-term trend; we consider that this result confirms that the reconstruction of summer streamflow for the Río Imperial captures characteristics of the regional hydroclimate. The reconstruction featured in contains two of the tree-ring chronologies presented in our study (PIN and LAN), in a pool of more than 1000 series developed from different species of the Southern Hemisphere. Our summer streamflow reconstruction is key in further supporting the SAM–streamflow relationship, because in this season and overall, snowmelt is a minor contributor to discharge (see Fig. 2). This makes rainfall the principal source of water for the river. During the last 75 years or so, the SAM has been shifting to a positive phase, unprecedented in the last 1000 years . A positive phase means that the pressure gradient between the midlatitudes and the polar latitudes is negative, that is, more positive or less negative in the midlatitudes relative to polar latitudes. This gradient shifts the core of the southern westerly winds towards the south, limiting the effect of storm tracks over midlatitudes. In the past, the SAM has been shown to be negatively correlated with the frontal activity and hence precipitation amounts, especially for the spring season . Given the relatively short delay between precipitation and discharge in the region (e.g., ), this relationship may well explain the decreasing summer streamflow trend observed for recent decades, as detected in the September–December negative correlation between the AAO in relation to both instrumental records and the tree-ring reconstruction (Table 5). This way, the post-1980 trend in summer streamflow of the Río Imperial may be linked to the aforementioned positive phase of the SAM . This evidence suggests an unprecedented decreasing trend of the Río Imperial's summer streamflow for at least the last 3 centuries, which may well be extrapolated to the last few millennia considering the correlation with the SAM. However, it is important to consider results according to on the relationship between precipitation, frontal activity, the ENSO, and the SAM in the Southern Hemisphere. In that study, ENSO (defined as the second empirical orthogonal function (EOF) from monthly anomalies at the 500 hPa geopotential level) had higher correlation with frontal activity in spring at the interannual timescale, although the correlation was also statistically significant with the SAM (defined as the first EOF from monthly anomalies of the 500 hPa geopotential level). also demonstrated that precipitation in SCC is significantly correlated with frontal activity. These studies suggest that the impact of the SAM on precipitation in the study area can also be associated with ENSO dynamics. For example, the low flow for the summer 1999 was concomitant with a high AAO and a strong La Niña event in 1998–1999. Recent studies have found that in-phase (e.g., positive ENSO and positive SAM) and out-of-phase occurrences of these climate modes can lead to varying strength of the associated atmospheric circulation , possibly controlling the intensity of their impacts on precipitation and hence streamflow. In fact, in a study analyzing the changes in rainfall in central and south-central Chile found that ENSO and AAO are not independent modes of variability but rather they both affect precipitation regimes. It is also important to consider possible different seasonal effects of these modes of variability, for example those found by , who, utilizing a 11-year moving average of a 300-year tree-ring streamflow reconstruction for the upper Biobío river, determined a negative correlation between the Pacific Decadal Oscillation (PDO) and streamflow during the snowmelt season (October–March) while a positive correlation between the SAM and that seasonal streamflow. Thus, we believe that the SAM–streamflow statistical relationship we found should be tested further in order to develop more accurate predictive models of extreme interannual dynamics. Using output from the CMIP5, predicted that the positive phase of the SAM will continue in some climate change projections, likely (but not certainly, given the difficulties in simulating the ENSO–SAM interactions) impacting precipitation in the midlatitudes. In this scenario, the SAM could provide more accurate statistical models of return periods.

Another relevant aspect of our reconstruction is that, like previous studies, it seemed to better represent low-flow conditions rather than high flows. discussed this finding in their reconstruction of the Río Maule, arguing, following , that this was likely a consequence of a smaller sensitivity of tree growth to precipitation above a certain threshold. However, the R2adj values of previous studies do not completely support that argument as the highest R2adj (54 %) was found in the Río Baker streamflow reconstruction , the southernmost and the wettest landscape of all sampled locations in Chile. In terms of R2adj, our reconstruction ranks second (47 %), followed by Biobío (45 %, ), while Puelo and Maule rank together with the lowest explained variance (42 %). Notice that the two highest ranked reconstructions represent summer to fall streamflow, when the soil moisture is at its lowest in the annual cycle. summarized empirical and theoretical arguments to better understand, on the one hand, the physical process linking tree growth and streamflow and, on the other hand, what determines whether tree-ring reconstructions reproduce dry conditions. In that study, the authors presented data from the Oldman River basin (∼ 49∘ 40′ N), a cold semiarid watershed in western Canada. They suggested the likelihood of spurious self-correlation in the statistical models relating streamflow and tree-rings because precipitation may be considered as a confounding factor. The mechanism these authors suggested is that soil moisture modulates the relationship between tree growth and other hydrometeorological variables (e.g., evaporation or runoff), more so in areas (and perhaps seasons) where (when) soil moisture is relatively constant. Following these arguments, one should expect streamflow reconstructions to better represent years or periods of water scarcity rather than high flows. indeed found higher correlations between low flow and tree-ring width in the Oldman River basin. Our analysis, and the fact that the two of the previously reported reconstructions for Chile with the highest R2adj aimed to include summer months, suggest that reconstructions focusing on summer streamflow (at least partially as in the case in ) have physical meaning and not just statistical significance. This renders the Río Imperial reconstruction as a representative time series of summer streamflow for the last 301 years, with a high capacity for capturing the occurrence of dry years. However, further research is needed because the Río Puelo reconstruction , also aiming for summer to fall streamflow, did not rank as high as our reconstruction and the one for Río Baker.

Given the crucial role that summer streamflow plays in a number of economic activities, such as agriculture and human consumption in the basin, the tree-ring reconstruction we have presented here can provide water managers and stakeholders with improved support for decision-making and regional planning. For example, the post-1980 generalized regional decrease in streamflow, both instrumental and reconstructed (e.g., ), paralleled paradigmatic legal and institutional changes in national water management, whereby the policy to assign water rights (and hence rights to use a proportion of the streamflow) transited into a model “with strong private property rights, broad private economic freedoms, and weak government regulation” (verbatim ). There is growing evidence of increasing water conflicts in SCC despite recent reform in this legal framework of water rights allocation . Some have argued that at least part of the problem resides in the insufficient hydrological information for users, managers, and decision makers . The relevance for water management in Chile is related to the mechanism for granting water rights, since it relies on the average river flow. By using this system, water rights have often exceeded the actual flow of rivers, specifically for the summer, forcing authorities to decree zones of water scarcity

An example of a decree for watersheds in the north of Chile can be found here https://goo.gl/S3n8d3 (last access: 17 April 2017).

The significance of our findings for planning is related to the need to consider a long-term view of the natural variability that defines the discharge regimes across the region. Planning governmental responses for periods of low flow based upon the 1, 2, or 5-year events detected in the available instrumental record may lead to ineffective mitigation actions, because those events may not be the most extreme on record. Considering the implications of our results, we assert that the extended record we have provided can be utilized as a complementary source of information for water resources managers to more accurately determine the range of “worst-scenario” or “more-recurrent” droughts, thus helping the planning and implementation of more efficient mitigation policies. There are several examples where dendrohydrology has provided key information for the decision-making process. presented an example in which a series of hydroclimatic scenarios derived from tree-ring chronologies were formally included in a government-supported hypothetical decision-making exercise where relevant stakeholders were tasked with developing adaptation measures for drought episodes. In that paper, and in the context of Canadian water policy, these authors stated that dendrohydrological research is a legitimate source of information for water resource planning and management. Another study, which aimed to characterize long-term Athabasca River streamflow in Alberta, Canada, found that whereas a trend analysis of instrumental data depicted declining regional flows, a tree-ring reconstruction showed periods of severe and prolonged low flows not captured by the instrumental record ; suggesting that worst-case scenarios estimated from historical gauge data likely underestimate the potential magnitudes of natural droughts in that region . described another application, where the Denver Water Board, the oldest and largest water supplier in the State of Colorado, USA, wanted to predict drought scenarios using reconstructions capable of capturing an episode that occurred during the 1950s as well as other more recent droughts. asserted that dendrohydrology can help in determining drivers of periods of extreme hydrological episodes in the Upper Colorado River basin.

The evidence presented in our study regarding the relatively dry post-1980 period, the higher frequency of droughts detected in the present for the Río Imperial relative to previous centuries, and the SAM projected to follow a positive phase leading to less precipitation altogether suggest that the regional hydrology is moving toward a new regime of more frequent (although not necessarily more intense) droughts. In a similar manner, as has been documented in several recent studies targeting other regions of the world (e.g., , and , among others), our reconstruction of natural hydroclimatic variability of the Río Imperial can provide a valuable framework to ponder the impact of anthropogenic climate change on future hydroclimatic changes in SCC, thus leading to a prudent water resource management strategy. Furthermore, water managers could use the reconstructed flows of the Río Imperial to plan mitigation strategies for drought events with return periods of 5, 10 and 20 years. As this hydrological application of dendrochronology becomes widely accepted as valid and water managers realize that this kind of historical information can guide operational strategies in the forthcoming decades , the possible regime shift needs to be incorporated into adaptation and mitigation plans associated with the regional impacts of future global climate changes. This will avoid, or at least mitigate, the negative consequences of these changes on the management of water resources for municipal and agricultural water supplies, electrical power, ecological habitats (e.g., ), and impacts on the economic activities in the region as water sources become stressed.