Impact of climate change on the streamflow in northern Patagonia

Water resources availability is strongly affected by climate change. According to the last report of the Intergovernmental Panel on Climate Change (IPCC), the global water cycle exhibited an overall strengthening since at least 1980 (Gulev et al. 2021). This strengthening was attributed to human influences affecting the global energy budget, such as the increased concentrations of greenhouse gases. Human-induced climate change has significant implications for changes in atmospheric circulation, resulting in substantial changes in precipitation patterns, albeit with regional heterogeneity in trend patterns. The changes in the spatial and temporal patterns of precipitation and evaporation rates have caused variations in streamflow in major rivers of the world (Dai et al. 2009; Zhang et al. 2023). Moreover, the increase in global temperatures has led to significant changes in the hydrological cycle components associated to the cryosphere, including reductions in snow cover, snow depth, and glacier volume, together with an early snowmelt pulse (Huss et al. 2017; Shannon et al. 2019).

The main rivers in the northern region of Argentine Patagonia, known as the Comahue region, originate in the Andes ranges within the latitudinal band between 37° and 41°S. The streamflow in these rivers is controlled by variations in winter precipitation, snowmelt, and glaciermelt (Vich et al. 2014). The main socio-economic activities in the northern Patagonia region directly rely on the water availability of the Limay, Neuquén, and Negro rivers (Raggio & Saurral 2021). Up to 25% of the hydropower production of Argentina is generated by hydropower plants such as El Chocón and Piedra del Águila, located on the Limay river, and the Cerros Colorados Complex, located on the Neuquén river, amounting to approximately USD300 million per year (Seoane & López 2007; Romero et al. 2020; González et al. 2023). Irrigation for agricultural production, oil and mining exploitation, tourism, and sport fishing, also depend on the available water resources (Pessacg et al. 2020). For instance, intensive agriculture in the High Valley (Alto Valle) along the Negro river accounts for more than 75% of the total national production of apples and pears (Brendel et al. 2020).

Due to the socio-economic relevance of regional water availability, several studies have focused on developing statistical precipitation forecasts over northern Patagonia, particularly for the austral winter months (González et al. 2010; Romero et al. 2020). These seasonal forecasts are linked to streamflow generation and enable decision-makers to optimize the operation of dams and irrigation channels. Nevertheless, water authorities face not only fluctuations in precipitation occurring on seasonal or interannual timescales but also long-term changes that pose a threat to water security. For instance, a declining long-term trend in precipitation and streamflow over northern Patagonia has been identified (Seoane & López 2007; Lauro et al. 2019; Brendel et al. 2020), particularly for minimum annual streamflow, leading to an increase in the frequency of hydrological droughts (Rivera et al. 2018). This decreasing trend over the last century has been attributed to changes in large-scale circulation patterns, such as the poleward expansion of the Hadley Circulation and the consequent displacement of the extratropical storm tracks (Rivera et al. 2020). Part of this expansion mechanism, which modifies hemispheric circulation patterns, can be explained by the increase in greenhouse gas concentrations (Mindlin et al. 2020; Villamayor et al. 2021). Henceforth, the influence of anthropogenic climate change on the modulation of regional streamflow and, thus, water availability is of substantial significance.

Future changes in the magnitude, variability, and temporal distribution of streamflow are dependent mainly on the projected patterns of temperature, precipitation amount, phase, and timing. Considering the future climate projections for northern Patagonia, it is expected a significant reduction of precipitation over the headwaters of the main rivers in the region, particularly during the cold months (Almazroui et al. 2021; Raggio & Saurral 2021). Projected temperature increases are robust even under low-emissions scenarios (Almazroui et al. 2021), compromising water reservoirs associated with glaciers and permafrost (Forni et al. 2018; Hock et al. 2019). The expected changes in temperature and precipitation will increase the frequency and severity of droughts, especially under the more severe emissions scenarios (Spinoni et al. 2020), with a direct impact on streamflow generation. Irrigated agriculture, hydropower generation, and water supply for urban and rural populations will undoubtedly be affected by projected streamflow reductions. In this context, quantifying the expected changes in regional streamflow over the 21st century is of paramount importance for understanding and anticipating potential water stress episodes and developing policy measures to adapt to these changes.

Few studies have assessed the impacts of future climate change on streamflow in northern Patagonia region. These studies commonly used a complex modeling chain, starting with bias-corrected statistically downscaled general circulation model (GCM) projections of future climate to drive hydrological models to produce future streamflow projections (Krysanova et al. 2018; Miller et al. 2021). An early study from Seoane & López (2007) estimated a projected streamflow reduction for the Limay river of nearly 10% by the end of the 21st century. However, more recent evaluations projected a reduction ranging from 20% (Pessacg et al. 2020) to 50% (Raggio & Saurral 2021), depending on the selected GCMs, hydrological models, and rivers. These studies underscore large uncertainties in future streamflow projections for northern Patagonia, and call for new evaluations to account for climate change impacts on regional water resources.

A valuable dataset to perform this kind of research is based on streamflow simulations from global hydrological models (GHMs) belonging to the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP). These simulations were commonly used to quantify the impact of climate change on water resources, addressing aspects such as the seasonal dynamics of river streamflow, the mean annual discharge and the frequency and intensity of streamflow extremes from both regional and global perspectives (Gosling et al. 2017; Krysanova et al. 2017; Zaherpour et al. 2018). For example, Ju et al. (2023) used a multi-model ensemble based on six GHMs to evaluate the uncertainty in terrestrial water storage projections at global scale. Using a similar methodological approach, Asadieh & Krakauer (2017) identified global changes in high and low streamflow extremes over the 21st century.

The objective of this research is to use the GHM simulations from ISIMIP to assess future changes in the streamflow of the main rivers in northern Patagonia (Limay, Neuquén, and Negro rivers). The GHMs are driven by multiple GCMs forced under two Representative Concentration Pathways (RCPs) that provide quantitative descriptions of greenhouse gases evolution throughout the 21st century. Based on GHM validation over the historical period and using a multi-model ensemble approach with the best GHMs, we aim to quantify the projected changes in surface water resources and assess their uncertainty. The structure of the paper is organized as follows: Section 2 describes the study area and data; Section 3 summarizes the methods used in the evaluation; Section 4 provides the main results obtained in the model validation, the future streamflow projections and its main drivers; the discussion of the main outcomes is included in Section 5; while the conclusions are summarized in Section 6.

The Negro river is formed by the Limay and the Neuquén rivers, which are born in the higher elevation of the Andes (Figure 1(a)). The Limay drainage network is complex, characterized by a series of large glacial lakes and a marked vegetation gradient, with rainforests in the west and shrublands in the east in response to the spatial changes in precipitation (Lallement et al. 2020). On the other hand, the Neuquén basin lacks natural lakes that regulate its streamflow, with sparse vegetation cover consisting in limited natural forest areas in the west and scattered shrublands in the east (Forni et al. 2018). These basins differences result in distinct annual streamflow and annual cycle features among the Limay and Neuquén basins (Berri et al. 2019).

Monthly streamflow records from the main rivers in the Comahue region, obtained from the National System of Hydrological Information (https://snih.hidricosargentina.gob.ar/) were used for the validation of simulated streamflow. The selected gauging stations are located in the Limay, Neuquén, and Negro rivers (Figure 1(a)), and the monthly streamflow records span from 1903 to 2017. Following previous studies (Rivera et al. 2018; Lauro et al. 2019), these records were subjected to quality control procedures to identify data outliers and periods with spurious sequences with data repetition. Data gaps were identified in the time series of the Limay and Negro rivers. To complete the streamflow records, we applied gap-filling routines based on data from donor stations in the same river or its main tributaries, requesting a coefficient of determination higher than 0.9. Most data gaps were completed by calculating linear regressions between the records from the donor stations and the target stations for infilling. After implementing these procedures, the time series selected for the study retained less than 1% of missing data. Due to data limitations, we used the 1961–1990 reference period for the Limay river and the 1976–2005 period for the Neuquén and Negro rivers.

Limay and Neuquén records are unaffected by river regulation or diversion upstream of the gauging stations (Pichi Picún Leufú Dam became operational after 1999). In contrast, Negro river is expected to exhibit a signal representative of both natural and anthropogenic influences in the streamflow records. It is beyond the scope of this study to separate both components or generate a naturalized streamflow record for Negro river.

The ISIMIP project adopted climate and socio-economic input data for multiple impact models used to synthesize expected changes in various sectors considering different levels of global warming (Hempel et al. 2013; Warszawski et al. 2014). This approach supports adaptation and mitigation decisions requiring regional or global perspectives in the context of facilitating transformations for sustainable development (Rosenzweig et al. 2017). Given that the more recent phase of ISIMIP (ISIMIP3b) was under development at the time of conducting this research, we analyzed the impact models from the global water sector belonging to phase 2b of ISIMIP (ISIMIP2b). From the 17 available GHMs, we selected the models with monthly streamflow simulations for at least two emissions scenarios. This condition was satisfied by 10 GHMs, a number that we consider adequate for the evaluation of streamflow variations over the region and is also in line with previous research (Giuntoli et al. 2015; Krysanova et al. 2017; Veldkamp et al. 2017). The 10 GHMs are CLM4.5, H08, JULES-W1, LPJmL, MATSIRO, MPI-HM, ORCHIDEE, ORCHIDEE-DGVM, PCR-GLOBWB, and WaterGAP2-2c (see https://www.isimip.org/impactmodels/ for a summary of the main characteristics). All the GHMs simulate hydrological processes and river routing with a spatial resolution of 0.5° (Hattermann et al. 2017). It must be noted that only the WaterGAP2-2c model is calibrated with observed data from 1,319 stations, covering approximately 54% of the global land area (Krysanova et al. 2020). These GHMs were driven by climate simulations from four GCMs of the Coupled Model Intercomparison Project Phase 5 (CMIP5) experiment: GFDL-ESM2M, HadGEM2-ES, IPSL-CMSA-LR, and MIROC5 (see Frieler et al. 2017 for details). The outputs of these models were subjected to bias correction using the EartH2Observe, WFDEI, and ERA-Interim data Merged and Bias-corrected for ISIMIP (EWEMBI) reanalysis climate product as a basis, resulting in a final spatial resolution of 0.5° latitude by 0.5° longitude. From these models, we analyzed baseline (1976–2005) and future (2021–2050 for the near-term and 2070–2099 for the long-term) monthly temperature, precipitation, and sea level pressure simulations, to account for the main drivers of streamflow variability. The future projections were forced by two of the four available RCPs: the RCP2.6, a low-emission scenario with strong mitigation and in line with a 1.5–2 °C global warming level, and RCP6.0, a business-as-usual greenhouse gas scenario without explicit mitigation measures (Frieler et al. 2017).

A point-to-pixel analysis was performed to compare the time series of streamflow observations to the corresponding hydrological model pixel. We selected the streamflow simulations from the 10 GHMs for the grid points corresponding to the locations of the analyzed stream gauges (Figure 1(a)). To evaluate the performance of ISIMIP2b hydrological models in simulating the amount and variability of the regional streamflow, we used the following comparison statistics: the Pearson correlation coefficient (PCC), the Nash–Sutcliffe efficiency (NSE), the percent bias (PB), the modified Kling–Gupta efficiency (KGE), and the volumetric efficiency (VE). The PCC measures the strength of the linear relationship between the streamflow simulations and the stream gauges observations. The NSE (Nash & Sutcliffe 1970) determines the relative magnitude of the variance of the residuals compared to the variance of the observed streamflow values. The PB measures the average tendency of the simulated streamflow to be larger or smaller than the observed streamflow, where positive values indicate overestimation bias and negative values indicate underestimation bias. The KGE (Gupta et al. 2009; Kling et al. 2012) is designed as a more general index that compares the variability of the observed and simulated streamflow values by including information about the correlation between them and their standard deviations, as well as any bias present, expressed by the relation between the mean values. Finally, the VE (Criss & Winston 2008) represents the fractional mismatch of water volume at the proper time. A similar set of metrics was employed by Krysanova et al. (2017) for the comparison of simulated and observed streamflow in 12 large river basins worldwide. The optimal value for PPC, NSE, KGE', and VE is 1, while for the PB is 0. Following Duc & Sawada (2023), a NSE of zero is chosen as the boundary between good and bad streamflow simulation, while a KGE higher than −0.41 indicates a good model performance (Knoben et al. 2019). For the VE, we used the threshold of 0.33 as a reference for a good streamflow simulation (Ecrepont et al. 2019).

The performance of the GHMs in the representation of streamflow characteristics was evaluated using the metrics described in Section 3.1. For this purpose, a rating index introduced by Jiang et al. (2015) was employed, which provides a comprehensive hierarchical appraisal of model performance. Each model was ranked based on its performance relative to each metric, allocating a rank of 1 for the best-performing model to 10 for the worst. Then, the ranks were aggregated across all the metrics and the three selected rivers. Similar approaches were performed by Balu et al. (2023) and Rivera & Arnould (2020) for the selection of the most suitable GCMs. From the rating index, a multi-model ensemble with the best GHMs was selected to quantify the projected changes in streamflow using simulations forced by the RCP2.6 and RCP6.0 scenarios in the near-term (2021–2050) and long-term (2070–2099) future horizons relative to the baseline period of 1976–2005. The mean, median, and range of the change in future streamflow projections were quantified in the results as percentages relative to the baseline period. The uncertainty in streamflow projections was identified from the dispersion of all possible combinations of GHM/GCM, considering also the range of changes from the multi-model ensemble average of the GCMs for each GHM, and for the multi-model ensemble of the GHMs for each GCM. This approach will help identify the main sources of uncertainty in streamflow projections over the study area.

To assess the accuracy of the streamflow simulations, we employed a quantitative approach involving the comparison of the observed and simulated hydrological cycles during the baseline period selected for each river. It must be noted that the comparison is made based on the hydrological simulations resulting from the ensemble of the available GCMs. This ensemble was created using a simple arithmetic average.

From the evaluation of the GHMs, a common feature for the selected rivers is that most of the models adequately represented the winter peak in streamflow. Nevertheless, the models fail to reproduce the second peak during late spring, associated with snowmelt contribution, which is particularly important for the Neuquén river. Previous studies showed that this limitation could be explained by a poor estimation of air temperature over the higher elevations of the Andes by the GCMs, a factor that creates a misrepresentation of the snow accumulation during winter and the following contribution as streamflow during spring (Raggio & Saurral 2021). Another striking feature is the capability of the models to reproduce the annual cycle of the Negro river, considering that is highly regulated by large dams located over the Limay and Neuquén rivers (Figure 1). Some of the GHMs include the implementation of dams and reservoirs in the streamflow simulations, for example, the H08, LPJmL, MATSIRO, PCR-GLOBWB, and WaterGAP2-2c, while others include only a routine for irrigation, as the CLM4.5 and the MPI-HM models (see https://www.isimip.org/impactmodels/). Changes in land use and land cover and reservoir operations are responsible for a significant increase in water availability over the lower portion of the Negro basin during both summer and winter (Veldkamp et al. 2017). Therefore, this result highlights the need for an adequate representation of streamflow regulations by the GHMs and also supports our decision of using observed streamflow over Negro river without removing the water management component.

Considering the rating index, from the initial pool of GHMs, we finally selected the CLM4.5, H08, MATSIRO, MPI-HM, and WaterGAP2-2c for the evaluation of future streamflow conditions over the study area. It must be noted that, among these selected models, H08, MATSIRO, and WaterGAP2-2c incorporate human activities for streamflow regulation, while the models CLM4.5 and MPI-HM include an irrigation scheme in their formulations.

After selecting the final set of GHMs to represent streamflow in the study area, we assessed the projected changes based on simulations forced by the RCP2.6 and RCP6.0 scenarios in the near-term (2021–2050) and long-term (2070–2099) future horizons relative to the baseline 1976–2005 period. Given the models' challenges in adequately representing the snowmelt peak, we focused on the projection of mean annual streamflow. However, considering that the projected changes can be linked to shifts in the precipitation patterns, particularly the winter rainfall peak, the changes in annual streamflow can be attributed mostly to changes in the winter months.

By evaluating the range of the multi-model projections in Supplementary material (Tables S1 and S2), we can account for the contribution of GHMs and GCMs to the uncertainty in future streamflow projections for the Limay river. We found that the largest source of uncertainty is associated with the GCMs, a factor that can be linked to the representation of winter precipitation amount and variability. The uncertainty signal is lower for the RCP6.0 scenario, particularly considering the end of the century (Figure 5; Supplementary material Tables S1 and S2).

By evaluating the range of the multi-model projections in Supplementary material Tables S1 and S2, we found that the largest source of uncertainty is associated with the GCMs and, thus, to the representation of winter precipitation amount and variability. The uncertainty signal is lower for the RCP6.0 scenario, particularly considering the end of the century (Figure 7, Supplementary material Tables S1 and S2).

The most relevant streamflow reductions are projected for the end of the century, particularly considering the RCP6.0 scenario, with a decrease in mean annual streamflow projected for the multi-model ensemble ranging from −24% for the Limay river to −32% for the Neuquén river. This result aligns with the projected increases in the frequency of days with low flows over northern Patagonia (Giuntoli et al. 2015). Considering the results for the Limay river, the projected streamflow reduction is higher than the estimation performed by Seoane & López (2007), which ranged roughly between −11 and −18.5% for the end of the century. Using regional climate models from the CORDEX Project and the InVEST water yield model, Pessacg et al. (2020) found a projected decrease in the water yield of Limay river of around −20% for the period 2071–2100. Our results show that some hydrological simulations exceed −30% of reduction in mean annual streamflow by the end of the century for the RCP6.0 scenario, and an even higher decrease for the streamflow of the Limay river under the RCP8.5 high-emissions scenario was projected with the VIC hydrological model (Raggio & Saurral 2021). The most relevant streamflow reductions are projected by the multi-model ensemble for the Neuquén river, particularly considering the CLM4.5 and MPI-HM hydrological models. Some of the analyzed simulations project a streamflow decline of almost −50% relative to the reference period. In line with our results, previous studies identified a reduction in the available water for irrigation of agricultural areas of the Neuquén river basin (Forni et al. 2018). Similar reductions in the mean annual streamflow were projected by the VIC hydrological model under the RCP8.5 scenario (Raggio & Saurral 2021). The streamflow of the Negro river responds to the projected changes in both the Limay and Neuquén rivers, leading to a reduction of approximately −27% in the mean annual streamflow projected by the multi-model ensemble under the RCP6.0 scenario by the end of the century.

By evaluating the range of the multi-model projections, we found that GCMs are mostly responsible for the uncertainty in the streamflow projections, a result that aligns with previous research at global and regional scales (Giuntoli et al. 2015; Krysanova et al. 2017). The only exception, i.e. GHMs as the main source of uncertainty, was found for the Neuquén river considering the RCP6.0 scenario and the long-term future. This result can be linked to the role of temperature in the snowmelt processes over the basin, as suggested in previous studies (Giuntoli et al. 2015; Raggio & Saurral 2021).

The projected decreases in mean annual streamflow can be attributed to several factors, some of which are associated to future precipitation changes in northern Patagonia. Figure 8 shows the projected annual precipitation changes estimated by the multi-model ensemble of GCMs for the 2021–2050 and 2070–2099 periods, forced with the RCP2.6 and RCP6.0 scenarios. For the near-term future, both scenarios exhibit a similar spatial pattern of precipitation change, with projected decreases ranging from −5 to −10% in the headwaters of the main rivers of the region. The projected precipitation decrease is reduced from 0 to −5% for the 2070–2099 period under the RCP2.6 scenario and can be interpreted as the atmospheric response to the mitigation policies included in this scenario (van Vuuren et al. 2011). This aligns with the reduced decrease in regional streamflow projected for the same scenario at the end of the century (Figures 5–7). The largest precipitation reductions are projected by the RCP6.0 scenario (Figure 8), particularly over the headwaters of the Nequén river, which exhibits the most relevant projected decreases in streamflow for the 2070–2099 period (Figure 6). This projected trend in precipitation is expected to result in longer and more severe drought events over the region, with serious implications for the development of hydrological droughts (Asadieh & Krakauer 2017; Spinoni et al. 2020; Ferreira et al. 2023). The projected drying signal for the region can be linked to the projected changes in the mean sea level pressure field across the Southern Hemisphere, particularly during the winter rainy season. For the selected future periods and scenarios, there is a projected increase in the sea level pressure during winter over South America south of 40°S and the South Pacific Ocean, indicating a strengthening of the semi-permanent anticyclone, particularly during 2070–2099 and considering the RCP6.0 scenario (Supplementary material, Figure S1). Additionally, there is a strengthening of the subpolar low-pressure band, a factor that contributes to an intensification of the southern westerly wind belt, leading to a precipitation decrease over northern Patagonia (Quintana & Aceituno 2012; Rivera & Arnould 2020). The result is also consistent with a poleward expansion of the Hadley Circulation and a positive trend in the Southern Annular Mode, factors that were linked to anthropogenic forcings and thus are dependent on the greenhouse gas emission scenarios (Grise & Davis 2020; Rivera et al. 2020; Garreaud et al. 2021).

Human-induced climate change is expected to generate a reduction in the streamflow of the main rivers (Limay, Neuquén, and Negro) of the Argentinean portion of northern Patagonia, which account for a significant part of the total hydropower production of the country and allow the irrigation for the main agricultural activities in the region. This was analyzed considering a set of 10 GHMs from the ISIMIP2b dataset, driven by bias-corrected GCMs used to produce future precipitation and temperature projections under the RCP2.6 and RCP6.0 emission scenarios. After a validation procedure, the initial set of GHMs was reduced to five models, adequately representing the winter streamflow peak but not capturing the late-spring snowmelt pulse. The multi-model ensemble projects a streamflow decrease during the near-term future (2021–2050) of around −10 to −15% compared to the 1976–2005 baseline period, with no dependence upon the selected scenario and river. The uncertainty is given mainly by the different precipitation projections by the GCMs used to drive the GHMs for the streamflow projections, with dispersion in the projected changes of around 20%. Streamflow projections for the long-term future (2070–2099) show divergent responses to the selected scenarios, with a mean reduction of −5 to −6% considering the RCP2.6 scenario but large dispersion in results from individual simulations. This leads to projected streamflow decreases ranging from −20% to slight streamflow increases relative to the reference period. The projected mean reductions using the RCP6.0 are between −20 and −40%, with the Neuquén river showing the most important streamflow decreases. The projected deficit in precipitation in the headwaters of the main rivers in the Comahue region, in response to an increase (decrease) in surface pressure over Patagonia and the South Pacific Ocean (the subpolar low-pressure belt) during winter months, was identified as one of the main drivers of the streamflow decrease over the 21st century.

The results of this study serve as a warning for regional water authorities already facing the challenges of managing reduced water levels for hydropower generation, human consumption, and irrigation for agricultural production. It is important to highlight that this is the first study analyzing such a large and diverse set of streamflow simulations over northern Patagonia. The best GHMs identified in this study can be applied to specific basin-scale research, such as quantifying changes in hydrological extremes. Given that the primary source of uncertainty is associated with the GCMs, it is advisable to consider alternative models to drive the GHMs, particularly those from the latest CMIP6. The adoption of improved bias correction techniques applied to the GCMs has the potential to decrease uncertainty in future projections. Furthermore, it may result in a more accurate representation of snowmelt processes during late spring, a crucial factor limiting the ability to provide precise information regarding the timing and magnitude of the snowmelt pulse, especially over Neuquén river basin.

In light of the projected reduction in surface water availability, the creation of hydrological drought management plans must be developed to ensure water security for local communities and environmental needs, aspects often not considered in regional water management strategies. To achieve this, improved modeling of streamflow processes and a thorough understanding of their future evolution is of paramount importance for providing robust information for decision-makers and water managers.

This work was supported by the National Agency for Scientific and Technological Promotion (ANPCyT) and the Juan Agustín Maza University (UMaza) under grant PICTO-UUMM-2019-00004, and the National Scientific and Technical Research Council (CONICET) under grant PIBAA 2022–2023 (28720210100485CO).

Streamflow records from the analyzed rivers can be accessed online through the National System of Hydrological Information from the Secretariat of Infrastructure and Water Policy of Argentina: https://snih.hidricosargentina.gob.ar/. Streamflow simulations from the Inter-Sectoral Impact Model Intercomparison Project phase 2b can be accessed at https://data.isimip.org/search/tree/ISIMIP2b/OutputData/water_global/.

The authors declare there is no conflict.