Environmental flow assessment in the context of climate change: a case study in Southern Quebec (Canada)

Water and sanitation access, health and well-being, responsible consumption and production, sustainable cities and communities, terrestrial life sustainability, partnerships and climate change mitigation measures are parts of the 17 sustainable development goals to reach before 2030 (United Nations, Sustainable Development Goals; UN-SDG 2016). ‘Sustainable management of water implies that, as part of water resources management activities, sufficient water is left for ecosystems so that they can continue to provide services to society into the future’ (Sood et al. 2017). In the context of low flow management, defining environmental flows can allow communities to take action to achieve some of these objectives. Environmental flows correspond to the ‘quantity, timing, and quality of freshwater flows and levels necessary to sustain riverine ecosystems which, in return, support human cultures, economies, sustainable livelihoods, and well-being’ (Arthington et al. 2018). Four categories of methods exist to calculate environmental flows: hydrological, hydraulic, habitat simulation and holistic (Tharme 2003; Linnansaari et al. 2013). Holistic methods, such as the Ecological Limits of Hydrological Alterations (ELOHA) framework (Poff et al. 2010), can provide guidance by combining scientific and social processes and build consensus around the rules for prescribing environmental flows. This study will focus on hydrological and hydraulic methods, which can be included in some of the methodological steps of holistic methods, such as ELOHA, to assess the impacts of climate change (St-Hilaire et al. 2021).

Environmental flow calculations based on hydrological methods are using historical discharge time series from impounded (Richter et al. 1996) or unregulated (Daigle et al. 2011; Berthot et al. 2020) rivers, as well as simulated flow data from hydrological models. In the context of climate change, historical flow analyses may prove to be inefficient to infer future environmental flow requirements, if the hydrological regime under study is evolving. Some of the model scenarios suggest that climate change may result in the presence of a trend (non-stationarity) in flow time series. Many climate change scenarios, including the most probable ones, indicate that low flows are becoming non-stationary, i.e. they tend to decrease for at least 25% of the global land surface (Döll & Zhang 2010). According to the IPCC (Intergovernmental Panel on Climate Change 2021), there is a strong relationship between the intensification of the global water cycle and the global temperature rising. In North America, an increase in air temperature of 1.5 °C may be associated with an increase in frequency of heavy precipitation (and floods), and severe agricultural and ecological stress; knowing that those hydrological events will be more variable within regions, seasons and from year to year (IPCC 2021). Lastly, there is high probability of earlier onset of spring snowmelt, with higher peak flows at the expense of summer flows in snow-dominated regions globally (e.g. regions such as Southern Quebec; IPCC 2021).

In this context, hydrological models allow us to consider land use changes, climate change scenarios (e.g. Gombault et al. 2015) or hydropower activities (Minville et al. 2010). Some models are more suitable for studying large areas with multiple drainage basins (Mortsch et al. 2000; Guay et al. 2015). Others have been developed to study the sustainability of river ecosystems (Mawdsley et al. 2009; Palmer et al. 2009), water surface quality (Whitehead et al. 2009) and allow for informed decisions on water management and mitigation policies (Vescovi et al. 2009; Carvalho et al. 2019). Uncertainties related to (1) the natural climate system itself; (2) the processing of data from climate projections feeding the hydrological simulations; and (3) the hydrological model and its calibration basis of available observations must be considered (Cyr 2012). For instance, numerous climate change scenarios can be compared which quantify some of the uncertainty associated with extreme flow scenarios (Gain et al. 2011). To simulate future flow trends to 2100, Gain et al. (2011) suggested that selecting a few global climate models (GCMs) and feeding future air temperature and precipitation scenarios as inputs to hydrological models may be sufficient to develop flow-weighted ensemble modelling.

These trends can be identified using different statistical tests such as the non-parametric Mann–Kendall test (Burn & Elnur 2002; Ehsanzadeh & Adamowski 2007), which is one of the most frequently used tools to achieve this. Using 7-day low flows, Ehsanzadeh & Adamowski (2007) observed a decreasing trend in flows over time in eastern Canada, with winter and summer low flows shifting to earlier dates. Chen et al. (2011) predicted an increase in winter flow and a decrease in summer flow in Southern Quebec rivers due to climate change. Similar scenarios were mapped in the Quebec hydro-climatic Atlas (Department of Environment and Fight Against Climate Change; DEFACC 2018). Chen et al. (2011) highlighted also the importance of accounting for uncertainty in downscaling methods. Several flow regionalization studies in Quebec have been proposed using multivariate analysis and taking into account flow variability (Assani & Tardif 2005; Daigle et al. 2011), eco-geographical parameters (Assani et al. 2006) and climate indices (Assani et al. 2011). More recently, six hydrological indices, representing the five hydrological characteristics listed as key components of the hydrograph by Poff et al. (2017), were selected to construct groups of spatially discontinuous, hydrologically homogenous hydrometric stations (Berthot et al. 2020). These groups are different from current spatially contiguous hydrographical regions used by the provincial water resources managers, and also different from ecozones (Hulley et al. 2014) or ecoregions (Belzile et al. 1997), including ecological data.

According to Poff et al. (1997), the flow regime impacts the river ecosystems integrity by influencing the water quality, energy sources, physical habitat and biotic interactions. Thus, the river ecosystems are adapted to their natural flow regime (Poff et al. 1997). Flow-habitat relationships can often be established through the hydraulic conditions associated with different flow values. In Berthot et al. (2021), the wetted perimeter associated with environmental flows was calculated, as a proxy to investigate the impact of the occurrence of low flows, equal to the metric values of environmental flows, in terms of riverine physical habitat availability. In this study, the wetted perimeter associated with simulated environmental flow values associated with a few climate change scenarios was investigated to provide information on how climate change may impact aquatic habitat availability. Berthot et al. (2020) ranked potential environmental flow metrics from most (highest flow) to least conservative considering three temporal scales (inter-annual, winter, summer) and two spatial distributions (spatially contiguous hydrographic regions and the aforementioned homogenous groups determined by low flow multivariate analysis). Candidate metrics were compared to percentages of mean annual flow (MAF) that are used as benchmarks for river ecosystem protection (Tennant 1976). These metrics are used in the present study, with these objectives: (1) to investigate the possible evolution of low flows associated with climate change scenarios for the spatially discontinuous low flow regions defined in Berthot et al. (2020); (2) to analyse how possible changes in low flow metrics may impact river hydraulics and aquatic habitat using the wetted perimeter as a proxy, as defined in Berthot et al. (2021); and (3) to discuss the use of thresholds to understand environmental flow values. The relevance of using frequency analysis for the assessment of environmental flows will also be discussed.

Two main hydrological methods are currently used during low flow periods in the Southern Quebec rivers (Canada): the 7Q2 flow metric (7-day average flow with a 2-year return period) as a threshold to manage water withdrawals (DEFACC 2015) and the 7Q10 flow metric (7-day average flow with a 10-year return period) to meet environmental discharge objectives (e.g. dilute treated or accidental wastewater discharged in water courses; DEFACC 2007). A total of eight environmental flow metrics were calculated, as in Berthot et al. (2020; Table 1). They were chosen because of their relevance for application in hydrological and climatic contexts similar to those in Quebec.

The 7Q10, 7Q2, Q95, Q90, LQ50 and the 70%LQ50 flow metrics were calculated and considered for three temporal scales: summer, winter and annual periods. The summer period is from July to September and the winter period from January to March. Only the AQ50 and 70%AQ50 flow metrics were calculated for the summer period. The 7Q2 and 7Q10 flow metrics were calculated using frequency analysis. The fitted distribution was the Generalized Extreme Value, as presented in Berthot et al. (2020). The MAF metric was used as a standardization parameter (Caissie & El-Jabi 2003). The 10% MAF, 25% MAF and 30% MAF were used as one worst and two fair river ecosystems thresholds (Tennant 1976; Caissie & El-Jabi 1995).

The HYDROTEL model was used to provide the simulated flow data set. HYDROTEL is a semi-distributed hydrological model (Fortin et al. 1995, 2001a, 2001b; Fortin & Royer 2004) used along with a physiographic and meteorological database provided by a geographic information system (PHYSITEL; Rousseau et al. 2011). HYDROTEL simulates two types of computational units for various hydrological processes: (1) at the scale of relatively homogeneous units (RHHUs), snow cover build up and melting; potential evapotranspiration; vertical water balance; and overland flow routing using a geomorphological unit hydrograph; and (2) at the river segment level, stream discharge routing (Fortin et al. 2001a). Daily flows were simulated by HYDROTEL using air temperature and precipitation outputs from regional climate models under two Representative Concentration Pathways (RCP4.5 and RCP8.5; IPCC 2014). A total of three GCMs from RCP4.5 and three GCMs from RCP8.5 were selected (Table 2). Among the RCP8.5 climate scenarios, two GCMs were considered as boundary conditions for the Regional Canadian Model (CRCM5). In Table 2, the IPSL-CM5A-LR is the Institute Pierre Simon Laplace Model (Glisaclimate 2021a). The IPSL-CM5A includes the Nucleus for European Modelling of the Ocean model (NEMO 2021) with low resolution (-LR). The INMCM4 is the Russian Institute for Numerical Mathematics Climate Model Version 4 (Glisaclimate 2021b). It is composed of an atmospheric model and an oceanic general circulation model (Volodin et al. 2010). The CanESM2 is the Canadian Earth System Model version 2 which consists of the physical coupled atmosphere-ocean model CanCM4 coupled to a terrestrial carbon model (CTEM) and an ocean carbon model (CMOC; Canada 2021). The CMCC-CESM is the Centro Euro-Mediterraneo per I Cambiamento Climatici Model (Glisaclimate 2021c). The CESM acronym is for the Carbon Earth System Model. This last model includes atmosphere, ocean, land surface and vegetation modules (CMCC 2021). Those models were selected based on the variability of the median of the 7Q2 flow metric values estimated from hydrological scenarios generated by HYDROTEL, fed by the climate model outputs. Environmental flow metrics listed in Table 1 were calculated and compared with the same flow thresholds, for both historical time series and future synthetic time series from GCMs. It should be noted that the simulated flows were not post-processed. In spite of the fact that those simulations are likely biased, relative changes can be investigated under the hypothesis that model biases remain constant.

The Quebec Department of Environment and Fight Against Climate Change (DEFACC) provided the simulations for 284 identified hydrometric stations. All sites were selected, mainly to compare the AQ50, 70%AQ50, LQ50, 70%LQ50, Q90 and Q95 flow metric results. Because long (i.e. >20 years) time series are required to perform frequency analysis and estimate the 7Q2 and 7Q10 flow metric values, the number of hydrometric stations for which this analysis could be completed varied, as shown in Table 2. It should be noted that the variability in 7Q2 and 7Q10 flow values has been previously presented in the hydro-climatic Atlas of Quebec (DEFACC 2018).

Table 2 shows the number of hydrometric stations selected to calculate the 7Q10 and 7Q2 flow metrics when the hypothesis of independence, homogeneity and stationarity were all validated. In Table 2, the pre-selected GCMs are presented as well as their associate qualifier and their reference letter for the next figures. The aim was to have one optimistic, one intermediate and one pessimistic climate scenario per regional climate model. The qualifiers are linked to the emissions forecasts. The pessimistic scenario is the one where greenhouse gas emissions would continue to increase at the same rate as in recent years. Simulated flow values were compared with the HYDROTEL calibration period of 1981–2010. The simulated flow values were calculated for two temporal slices: 2020 (2011–2040) and 2050 (2041–2070). Hereafter, these time slices will be referred to as the horizons 1990, 2020 and 2050. Except for the 7Q10 and 7Q2 flow metrics, 284 hydrometric stations were selected to calculate the Q95, Q90, AQ50, 70%AQ50, LQ50 and the 70%LQ50 flow metrics for the three time slices. Those hydrometric stations are distributed over the eight hydrographical regions of Southern Quebec (Figure 1).

The evolution of the river sites was investigated over three time slices. Also, it was proposed to reconfigure the hydrological regions and discuss the results within newly defined ‘low flow regions’, using two multivariate analyses. A Principal Component Analysis was used to select the six hydrological indices listed in Table 3 (from Berthot et al. (2020)), to explain low flow differences within the selected river sites. Then, an Ascendant Hierarchical Clustering (Sokal & Sneath 1963) was used in this study to define homogeneous groups of stations (the term ‘HC’ will be used for the Hierarchical Clustering groups; refer to Berthot et al. (2020)). Then, the exercise consisted of tracking the evolution of each HC from 1990 to 2050 with the help of spatial markers: the Gaspésie region (HC1), the Saguenay River (HC2), the southernmost coast area of the St. Lawrence River (HC3), the same region plus along the St. Lawrence River and/or the western part area (HC4), with the western rivers (HC5), the north-eastern most river (HC6) and along the 48th parallel (HC7). Results below will also be for an HC8 group for horizons 2020 and 2050, including two rivers in the southern part.

Figure 2 shows box plots depicting environmental flow metrics variability for inter-annual (a), summer (b) and winter (c) periods, for three horizons (1990, 2020 and 2050), and for the six models (A to F, where A, B, C and E are GCMs and D and F are two Regional Climate Models). Flow values were standardized (divided by MAF) for comparison with MAF thresholds. The flow metrics were ordered from the lowest to the highest for the horizon 1990. First, the minimum and maximum values presented more variability in extreme flows than for median flow values. Minimum flow metric values decreased from 1990 to 2050 for summer (-S) and annual flow metrics and maximum flow values increased from 1990 to 2050 for winter (-W) and annual periods. The winter and the LQ50 low flow metrics increased from 1990 to 2050. The variance of most flow metrics is expected to increase by 2050 with a greater occurrence of extreme low metric values. The increase in variance appeared higher between the horizons 2020 and 2050 than between 1990 and 2020. Also, the variance of flow metrics seemed higher for intermediate (B and E) and pessimistic (C and F) climate change scenarios than optimistic (A and D). In general, inter-annual flow metrics median values (Figure 2(a)) were more or less the same for the three horizons and remain <25% MAF. Summer flow metrics median values (Figure 2(b)) are expected to decrease, while winter flow metrics median values (Figure 2(c)) are expected to increase, considering all the selected scenarios, except for the winter 7Q10 and 7Q2 flow metrics E and F scenarios. The AQ50 flow metric still provided the highest values during summer periods (≈25% MAF) and the LQ50-W flow metric median values increased from <25% MAF in 1990 to >30% MAF in 2050.

Figure 3 displays the spatial distribution of the flow value differences (%) for the 7Q2, Q90, LQ50 and AQ50 flow metrics, from horizons 2020 to 2050. Note that the spatial patterns are similar for 7Q2 and 7Q10, Q90 and Q95, LQ50 and 70%LQ50, and of course, between AQ50 and 70%AQ50. It was, therefore, decided to average the simulated flows of the six scenarios for ease of conveying a general result. Two main groups emerged from the 7Q2 flow metric map (Figure 3(a)). Differences of +5 to +15% between 2020 and 2050 (green) are expected along the St. Lawrence River from the northernmost coast in the east to the southernmost coast in the west. The second group includes stations for which −5 to +5% differences (yellow) are expected, mainly in the Gaspésie region (east) and in the Saguenay region on the north shore of the St. Lawrence River, as well as numerous stations located in the southwestern portion of the study area. However, in the sub-region identified in the zoomed square, a very wide range of differences was observed, ranging from <−35 to >35%. In Figure 3(b) and 3(c), the Q90 and LQ50 flow metric maps show that three groups of rivers can be delineated in the southern part of the area: −15 to −5% (orange), −25 to −15% (red) and −35 to −25% (purple). The latter two ranges of differences are predominant in Figure 3(d) for the AQ50 flow metric map. In addition to the spatial distributions observed in the 7Q2 flow metric map (Figure 3(a)), a clear divide was observed on both sides of the 48th parallel in the Q90 and LQ50 flow metric maps (Figure 3(b) and 3(c)) and on both sides of the St. Lawrence River (AQ50 flow metric map; Figure 3(d)). Results displayed in Figure 3 suggest that the flow metric calculated for the shortest period (AQ50) showed less variability among rivers than the inter-annual flow metrics (7Q2, Q90 and LQ50), and provided a rather clear spatial clustering of rivers.

Figure 4 shows the results of the multivariate (ACC and HCA) clustering of stations for horizons 1990 (a), 2020 (b) and 2050 (c). Again, the average of the simulated flows of the six scenarios is presented for ease of reading. There were seven low flow regions (HC1 to HC7) for horizon 1990 and eight for horizons 2020 and 2050. Non-parametric ANOVA (Kruskal–Wallis) and post hoc (Wilcoxon–Mann–Whitney) tests were used, and confirmed that each HC group was significantly different from the others (p-value <0.05). According to the selected scenarios, HCs will likely change (i.e. stations will be grouped differently) because the flow characteristics used to compute the principal components are expected to change (Table 4). Part of these changes is associated with changes of the month during which the lowest flows occur (Figure 5).

Figure 5 shows the spatial distribution of the month with the lowest flows. Results were calculated for the inter-annual (Figure 5(a)), summer (Figure 5(b)) and winter (Figure 5(c)) periods of horizon 2050. In Figure 5(a), 90% of river sites had their lowest flow month during winter for the horizon 1990, 75% by 2020 and 54% by 2050. For the inter-annual period, these percentages are higher than for the one-month forward or backward shift, varying from 3 to 9% of the river sites considered in this study, depending on the time horizon. There were three main sub-regions of change for the inter-annual period (Figure 5(a)). The two northern groups included rivers with the lowest monthly flow during the winter periods. According to the scenarios selected for the present study, the month with the lowest flows would change from March to February for eight rivers on the eastern part of the study area and four river sites from March to February in the western portion. A third group of stations on the western side would be subjected to a shift from winter to summer, i.e. from February to August in most cases.

Figure 5(b) and 5(c) focus on intra-seasonal changes. For the summer period, 23% of the river sites are characterized by a shift forward by one month in 2020 and 35% in 2050, while 16% have shifted backward by one month in 2020 and 10% in 2050. For the 2050 horizon, the lowest monthly flow values advanced from July to August in the southern circled area and near the Saguenay River. The lowest monthly flow values shifted earlier by one month from September to August in the north-eastern area and from August to July in the Gaspésie region. For the winter period, 1% of the river sites have moved forward by one month in 2020 and 1% in 2050, while 9% have moved back by one month in 2020 and 15% in 2050. For horizon 2050, the lowest monthly flow values shifted forward from January to February in the southern circled area. The lowest monthly flow values dropped by one month from March to February in the north-eastern and in the western areas.

Figure 6 shows environmental flow metrics and their associated wetted perimeter values for two selected sites that are representative of two different homogeneous hydrological groups, and their evolution for the six climate change scenarios. The metrics were ranked from the lowest to the highest flow metric values for the 2020 horizon. The dashed horizontal lines indicate the so-called Tennant thresholds (10% MAF, poor; 25% MAF and 30% MAF, fair) and the QMC wetted perimeter threshold as a value that may be indicative of fair physical habitat availability. The metrics are distinguished between the summer (‘-S’) and winter (‘-W’) periods and the inter-annual period (no letter). Figure 6 allows us to compare how the wetted perimeter (solid lines) evolve as a function of ranked environmental flow metrics (dashed lines). Differences occur mostly for the more conservative environmental flow metrics (right end of graphs) and are greater for summer environmental flow metrics than for inter-annual metrics. The 7Q10 and 7Q2 flow metrics missing values are due to the non-validated hypothesis of independence, homogeneity and stationarity (Table 2).

Figure 6(a) shows how summer environmental flow metric values may decrease significantly in the future (−15% for the AQ50 flow metric). For Figure 6(b), the LQ50-W flow metric will remain the most conservative in winter by 2050, while the 30% MAF should be applied instead of the AQ50 and LQ50S environmental flow metrics in summer. According to the selected scenarios, the wetted perimeter values will decrease during the summer period or increase during the winter period as seen in Figure 6(b) but not in Figure 6(a) (Aux Écorces River). The Ouelle River (Figure 6(b)), a particularly shallow and warm river (Daigle et al. 2015), is characterized by relative wetted perimeter values much lower than the other two examples. In this river, a decrease of 10% in the wetted perimeter value could be expected for the AQ50 metric by 2050. In such cases, looking at both flow and wetted perimeter thresholds may impact the choice of the more conservative environmental flow metric. As expected, RCP8.5 climate change scenarios (D, E and F) provided lower flow values than RCP4.5 (A, B and C). The values for D and C climate change scenarios seemed close.

The purpose of this study was to evaluate the need to adapt methods for estimating environmental flows in Southern Quebec rivers, under scenarios of climate change. The results provided an opportunity to discuss the importance of including information other than discharge (i.e. wetted perimeter) to assess environmental flows in the context of climate change, even if the set of scenarios is limited and no hydraulic modelling was done. Hydrological methods, spatial and temporal scales and conservative thresholds that may provide insight in habitat quantity were investigated. According to the selected climate change scenarios, results showed that:

1. Summer low flows, as characterized by a number of environmental flow metrics, will become lower than winter low flows by 2050, for the six climate change scenarios;

2. The AQ50 flow metric, which has been deemed a fairly conservative environmental flow for decades, will decrease considerably by 2050 during summer (from −15 to −35% mainly along St. Lawrence and Saguenay rivers);

3. The LQ50 flow metric, either applied for a summer, winter or annual period, will provide the highest values for most of the six climate change scenarios;

4. The inter-annual flow metrics are influenced by the season with the lowest flows, which is expected to change for many stations;

5. The frequency of occurrence of extreme flow values will increase by 2050;

6. The southern part of the study area will be more affected than the northern part;

7. The percentage differences for most flow metrics between horizons will be increasing mostly between 2020 and 2050.

Flow metrics can be separated into two groups: those estimated from ranked discharge time series or the so-called flow duration curve (Q90, Q95, AQ50, LQ50, 70%AQ50 and 70%LQ50 flow metrics) and those estimated using frequency analysis (7Q2 and 7Q10 flow metrics). Frequency analysis requires that the hypothesis of independence, homogeneity and stationarity be validated. Stationarity is defined as the absence of trends in the observed time series. When using climate change scenarios to generate future hydrological scenarios, the likelihood of non-stationarity in synthetic time series is high. One way to circumvent this challenge is to compute the metrics for sub-periods within which stationarity is verified. This strategy was implemented in the present study with three periods being investigated at the 1990, 2020 and 2050 horizons. Despite this caution, these three hypotheses, mainly the stationarity, were not validated for the 284 sites considered in this study, which explains the varying sample size described in Table 2. St-Hilaire et al. (2021) reiterated the fact that environmental flow assessment usually required stationary condition to fit statistical distribution with constant parameter values to the empirical distribution of extremes. When this is done, the return period does not account for any trend in the timeline. Although there is a strong corpus of literature that describes how to perform non-stationary frequency analysis, its implementation is not without challenge. It is possible that the greatest challenge, as highlighted by St-Hilaire et al. (2021), is the interpretation by managers of a return period in a non-stationary context, although some reflections are being carried out (Poff 2017). For Poff (2017), the non-stationarity of the hydrological regime needs to be anticipated in the management of environmental flows, and he proposes in particular to look at non-flow based parameters to assess the functions, processes and structures of lotic ecosystems. In the present study, the evolution of the 7Q2 and 7Q10 flow metrics vary in the same direction as the other flow metrics, across horizons and for inter-annual and seasonal periods, except for the climate change scenarios E and F (Figure 2) for the 2050 winter period (Figure 2). Spatially, Figure 3(b) shows a wide range of variability in values for sites in the southern part of the study area. Thus, environmental flow metrics based on frequency analysis are not recommended for environmental flow assessment in the context of climate change.

The results made it possible to understand the evolution of environmental flows in terms of spatial and temporal scales. Observing the evolution of the spatial distribution of the flow values by HC (Figure 3 and Table 4) furthers our understanding of how climate change will impact rivers both locally and within regions. This is more explicit when we consider the evolution of the month with the lowest values (Figure 4), and observe that the period of lowest flow can change seasonally. The temporal scale was analysed in Figure 2, comparing box plots of inter-annual flow metrics versus seasonal flow metrics (−S and −W), for the six climate change scenarios. The spatial information showed that the values of the inter-annual flow metrics are more varied than the values of the summer or winter flow metrics. The combination of spatial and temporal results showed that it is preferable to assess environmental flows by low flow regions and seasons, in order to have a better understanding of the impact of environmental flow management on river ecosystems. Low flow regions can be used to classify groups of rivers according to the potential risk of negative impacts on the river ecosystems during low flow periods.

Throughout this study, Tennant's (1976) flow thresholds were used to discuss the results in relation to poor (10% MAF) flow values to protect river ecosystems and two fair values (25% MAF, Caissie & El-Jabi 1995; 30% MAF). The Tennant's flow thresholds were chosen because they are known internationally. For instance, results showed that the LQ50 flow metric values will get closer to the 30% MAF flow threshold during winter by 2050 and will be under the 25% MAF or close to the 10% MAF flow thresholds during summer by 2050. This is the opposite of today's trend and it raises the question of the short-term adaptation of river ecosystems to new flow regimes. These conservative thresholds clarify managers' interest in carrying out comprehensive hydrological, geographical and biological studies to estimate them. In addition, minimum and maximum flow thresholds can be used to provide a range of possible environmental flows as the sustainability boundaries proposed by Richter (2009) or the presumptive standards method (Richter et al. 2012). However, the assessment of flow thresholds is a huge task rendered more difficult because of the paucity of ecological information to match with the definition of environmental flow. Flow thresholds need to take into account a more complete biological context, in addition to fisheries interests, and have to be adapted to climate change, to manage the impacts on aquatic ecosystems (Meyer et al. 1999). To partially fill this information gap, the wetted perimeter threshold was proposed in a previous study (Berthot et al. 2021). This tool permits us to estimate the potential availability of physical habitat for river ecosystems. In this study, the six climate change scenarios (four GCMs: IPSL-CM5A-LR, INMCM4, CanESM2, CMCC-CESM and two RCMs from CanESM2-CRCM5) provide a first insight into the potential changes in habitat availability associated with selecting environmental flow metrics in the context of climate change in the Southern Quebec rivers.

Finally, recent studies on the impact of climate change on low flow periods used global warming temperature predictions (Marx et al. 2018) or eco-hydrological assessment models (Shrestha et al. 2019). Because it could be very expensive to acquire relevant ecological data for the whole territory (≈730,000 km² in the present study area), we suggest that water temperature could be a tool of interest (Daigle et al. 2019). Temperature is one of the so-called master variables in aquatic ecosystems that can be used in habitat simulation methods to calculate environmental flows. It is linked to air temperature and is, therefore, impacted by the effects of climate change. Hence, joint modelling of future environmental flow and temperature scenarios would provide a better understanding of the choice of an appropriate environmental flow metric. However, a thorough hydro-thermal analysis of low flows will require a better distribution of monitoring stations to adequately cover large territories such as Southern Quebec.

Climate change models uncertainties aside, the results of this study are relevant for forward thinking about adapting the management of environmental flows, according to the evolution of the hydrological context with climate change effects. Although many jurisdictions, such as Quebec, still rely solely on hydrological metrics, hydro-climate change scenarios can provide an additional hydrological tool and allow for the implementation of holistic methods such as ELOHA (St-Hilaire et al. 2021). This study showed that: (1) relatively homogeneous low flow regions are not matching with current hydrological regions and will shift in the future; (2) there is a need to calculate the flow metrics for the season(s) concerned by the management of water abstraction; and (3) using tools such as the wetted perimeter provides insight into the repercussions of flow changes in the physical habitat, mainly during the summer period. Assessing environmental flows for different climate change scenarios, for Southern Quebec rivers, mainly presented the limit of hydrological methods based on frequency analysis, which becomes more complex to use in the context of non-stationarity. Also, important changes are to be expected in the flow-ecological relationships due to the upward trend in winter and the downward trend in summer by 2050.

In this study, the choice was made to use the available database and tools and to draw on recent studies to address local needs. Some of the conclusions will likely be useful for many northern jurisdictions that face the same challenges. However, methods will likely have to be adapted: (1) the spatial and temporal scales used to compare the results might be adapted to the local or regional context; (2) it should be ensured that the wetted perimeter method can be applied, depending on the geomorphology of the river sections; and (3) the uncertainties link to the hydrological model and the climate change scenarios used will differ according to regions/model(s) used.

The authors wish to acknowledge the financial contribution of the Quebec Department of Environment and Fight Against Climate Change (2013–2020 Climate Change Action Plan and the Electrification and Climate Change Funds). The contribution of the Direction de l'Expertise Hydrique of the Quebec Department of Environment and Fight Against Climate Change in terms of expertise and data sharing is acknowledged. Our final thanks go to Rebecca Tharme, Alain Rousseau and Ali Assani for their review and inputs.

Data cannot be made publicly available; readers should contact the corresponding author for details.