Draft 1 Vulnerability of northern water supply lakes to changing climate and demand

Arctic regions face a unique vulnerability to shifts in seasonality, which influences the summer recharge potential of freshwater reservoirs caused by decreased precipitation and increased evaporative stress. This pressure puts small remote northern communities at risk due to limited existing freshwater supply. The lack of baseline knowledge of existing supply, demand, or reservoir recharge potential increases this risk. We therefore address this knowledge gap through a water resource assessment of municipal supply over a 20 year planning horizon in two communities in Arctic Canada using a novel heuristic model and existing data sources. We generated climate and demand scenarios to identify the mechanisms of drawdown as well as examine the influences on replenishment. We found a pronounced vulnerability to reduced winter precipitation and (or) increased ice thickness of reservoirs. Our heuristic supply forecasts indicate an immediate need for freshwater management strategies for northern communities in Ca...


Introduction
Expanding populations, resource development, and limited infrastructure and capacity has made sustainable water resources a primary issue for northern peoples.While warming in northern regions is generally predicted to coincide with an increase in precipitation (Rawlins et al. 2010), there is a large variability between northern regions in model predictions and observations (Prowse et al. 2006;Bring and Destouni 2014).Since northern watersheds D r a f t 3 historically experience relatively small amounts of runoff or recharge during the dry summer months due to low precipitation (Schindler and Smol 2006;Woo et al. 2006) the natural reservoir capacity of these regions to support human populations is environmentally limited.Recent environmental change has further strained northern water supplies due to observed reductions in precipitation, reduced long term snow cover, and an increase in the high sun season leading to evaporative drawdown (Woo 2010;Carrol et al. 2011).In particular, it has been reported that a decrease in winter precipitation has destabilized water balances from reduced snow and ice melt contributions to lakes (Bouchard et al. 2013;Hodson 2013;Lantz and Turner 2015).As a consequence, the influence of environmental change on the sustainability of northern water supplies is a concern (Alessa et al. 2008;Evengard et al. 2011).Recession of northern water supply lakes has already been noted (Pokiak et al. 2005), and northern residents have identified decreased summer water levels interfering with obtaining available drinking water and traditional subsistence harvest (Goldhar et al. 2014).Since water governance systems in northern regions are at an early-stage of evolution, especially in the Canadian Arctic, we seek to examine key vulnerabilities and facilitate the development of freshwater policy for accessible freshwater.
A significant challenge faced by northern policy makers is the lack of baseline data or quantitative studies on the influence of environmental change and growth on water supplies that would inform decision making.Lack of capacity and expertise has also left many communities without basic knowledge of their own freshwater supply.For example, the Government of Nunavut, Arctic Canada, has no specific agency or person(s) in charge of freshwater resource policy and management, no water management policy framework, and no policy development or planning for climate change adaptation with regards to freshwater resources.These gaps in knowledge have been identified in the draft Nunavut Land Use Plan (NLUP); "…there is not an D r a f t official water management strategy or policy currently in place for Nunavut…" (Nunavut Planning Commission 2014:18) and by the Kivalliq Inuit Association (KIA); (there is…) "no requirement for additional research to ensure the protection of freshwater quality, quantity, or flow.The NLUP should include freshwater quality, quantity, and flow in its list of goals for protecting and sustaining the environment…" (Kivalliq Inuit Association 2015:10).
Without knowledge of the current and future water supply needs, northern communities' ability for informed municipal planning or emergency preparedness is severely limited.Thus, we seek to address the knowledge gap in how water supply capacity will be impacted by examining the influence of population growth and climate change through water supply forecasting.We chose to examine two remote northern communities that are under continued growth pressures to 1) assess the trends in climate for the respective regions, 2) calculate the demand on local water supplies under growth and climate forecasts, and 3) project long-term forecasting to the capacity of their municipal supply to meet the demands of the communities in the future.We utilize a novel quantitative process that minimizes the field measurements required to collect watershed data, and employ data processing and modelling techniques through a heuristic method of climate forecasting and hydrological simulation which can provide the supply forecasts needed to advance water management policy in the region.

Methods
The vulnerability of water supply sources was assessed by utilizing hydrological simulation to forecast the accessible water volumes under a range of plausible climate scenarios.
A Long Term Forecasting (LTF) model was formulated to evaluate the ability of each supply lake in fulfilling municipal consumption requirements over a 20-year planning horizon.In constructing the LTF model, four parameters were chosen for use in characterizing plausible D r a f t future scenarios of municipal water supply based on the availability of data; air temperature, snowfall, rainfall, and water consumption.Daily values of air temperature and precipitation are required to calculate the amount of water in the watershed available for evapotranspiration or storage in the reservoir.Daily values of discharge were then used to calculate the amount of water extracted for municipal use.Thus, varying the values of the four LTF parameters, eightyone unique climate and consumption scenarios can be simulated using available meteorological and demand data.

<Figure 1>
Detailed water supply forecasts were completed on the two largest communities of the Arctic territory of Nunavut, Canada (Figure 1).The capital city of Iqaluit, Nunavut, has experienced a 40 % increase in population (~7543 in 2015) since the creation of the territory in 1999 (Nunavut Bureau of Statistics 2014).The local government has acknowledged the growing demand on the only water reservoir, Lake Geraldine, which underwent dam modifications in 2010 to increase its storage capacity.Municipal operators are currently exploring alternative water sources for resupply (exp Services Inc. 2014).Rankin Inlet is the second largest community with a 2015 population of ~2800, an increase of 27 % since the formation of the territory (Nunavut Bureau of Statistics 2014).Lake Nipissar serves as the sole municipal water reservoir in the city, which was deemed insufficient to support the current population (Stantec Consulting Ltd. 2014).Bathymetric surveys for Lake Geraldine and Lake Nipissar were completed in the summer of 2008 and the fall of 2009 respectively, however, the watershed of Lake Nipissar had not been accurately delineated, nor was there a water level gauge present in D r a f t 6 the lake (Budkewitsch et al. 2011a,b).A seasonal replenishment pipeline was installed in Rankin Inlet to assist with summer replenishment from the nearby Char River, which became operational in 2015.The design of the replenishment pipeline aimed to sustain the viability of Lake Nipissar over a 20yr design period.

Data Sources
Meteorological data was collected for each community (Environment and Climate Change Canada 2016a).Both hourly and daily records were acquired in order to capture daily precipitation, temperature, and hourly wind speed.Precipitation data was corrected for measurement bias in order to generate an adjusted and homogenized climate record between 1981and 2015(Goodison et al. 1998;;Smith 2006;Devine and Mekis 2008).From this adjusted dataset, 30yr climate normals  for average annual temperature, total snowfall, and total rainfall were calculated along with the average number of days of each snowfall and rainfall (Table 1).If a particular month contained more than three consecutive missing daily records, or more than five in total, the month was omitted.Subsequently, if a month was determined to be missing, the entire year was omitted from the 30yr climate normal calculation (World Meteorological Organization 1989).The mean annual air temperature, total snowfall, and total rainfall anomalies for 1981-2015 were computed for each year by subtracting the relevant average from the annual values.The total snowfall and total rainfall anomalies were then normalized by dividing by the baseline values and expressed as a percentage of the normal value.

<Table 1>
The Nunavut Water Board (NWB) provided monthly withdraw quantities for each town via their annual water license reports (Nunavut Water Board 2016).Reported periods were 1998 D r a f t and 2006-2014for Iqaluit, and 2009and 2011-2015 in Rankin Inlet.This value represented the total volume of water extracted from each supply lake for municipal use.Using the published population estimates from the Nunavut Bureau of Statistics (2014), these quantities were normalized to a 'litres per capita per day' (LCD).

Climate and demand forecasting
In order to calculate the evapotranspiration potential during hydrological simulation, a daily precipitation forecast was required which experienced both precipitation and nonprecipitation days.Thus, a Characteristic Climate Year (CCY) was generated and used as a baseline pattern for creating climate forecasts.Once established, this pattern was repeated with amplitudes that matched the forecasts for annual air temperature, total snowfall, and total rainfall.To determine whether any day in the CCY experience precipitation, the probability of precipitation over a trace amount of 0.2 mm was calculated for each matching day in the 1981-2010 climate normal period.For example, if there were 15 records of precipitation > 0.2 mm that occurred on a January 1st day from 1981-2010, the probability of precipitation on the CCY day of January 1st was 50 %.This was repeated for all 365 days of the CCY.A cutoff probability threshold was established individually for snowfall and rainfall events in order to select the total annual days of snowfall and rainfall in the CCY to match the 30yr climate normal, where only days which experienced a probability above the calculated threshold value were included as precipitation days.The thresholds used were 40 % for snowfall and 50 % for rainfall for Iqaluit, and 36 % for snowfall and 50 % for rainfall for Rankin Inlet (see supplemental).For days over the cutoff threshold, their value of precipitation was set to the average of their corresponding days with precipitation on record.For example, January 1st in the CCY would be set to the D r a f t average of the 15 January 1st days from 1981-2010 which experienced >0.2 mm of precipitation.
A coefficient was applied to scale the daily snowfall and rainfall amounts to match the 30yr climate normals (Iqaluit: Snowfall coeff: 0.946, Rainfall coeff: 1.034; Rankin: Snowfall coeff: 0.799, Rainfall coeff: 0.697).This process resulted in a yearlong dataset of daily air temperature and precipitation which matched all the defining characteristics of the 30yr climate normals.This CCY dataset would be used as the baseline pattern for generating future climate forecasts.

<Table 2>
Climate forecasts were generated through analysis of available meteorological data and compared to IPCC Global Climate Model scenarios for Canada 75th percentile model for Nunavut 2016-2035 (Table 2).From this, patterns of change for annual average air temperature, total snowfall, and total rainfall from 1981-2015 were compared (see supplemental).A linear regression curve was fit to normalized 1981-2015 values of each climate parameter and trends were analyzed using a Mann-Kendall non-parametric trend analysis in the R statistical software package.For mean annual air temperature, a significant rate of change (p < 0.01) was observed for both Iqaluit (0.7 °C/dec) and Rankin Inlet (0. 9°C/dec).This was selected as the 'normal' rate of change and used to project behavior to 2035.This rate of change was also used as a baseline to calculate 'high' and 'low' climate scenarios for air temperature, which were multiples of the 'norm' rate of change (Table 2).Both snowfall and rainfall trends for Iqaluit and Rankin Inlet were not found to significantly differ from the average from 1981-2015.Therefore the snowfall and rainfall 'normal' rates of change were set to 0 % / dec (ie.No change from 30yr climate normal conditions).The rate of change for 'low' for snowfall and rainfall were set to the observed trends from 1981-2015 in order to depict a plausible scenario of climate conditions to compare to the normal trends (see supplemental).Although the observed historic trends The CCY series was used as the baseline values for each forecast scenario in order to maintain a consistent pattern of precipitation throughout the forecasted period, which was representative of the trajectory of the 1981-2015 climate normal for each town.The three modifier settings (low/normal/high) were applied to the baseline values to generate 20yr forecasts between 2016 and 2035.This method of generating climate forecasts provided plausible climate ranges generated specifically from observed local climate patterns (Figure 2).

<Figure 2>
Water consumption in the LTF represented the total volume of water extracted from each supply lake scaled to population.This value represented the total volume of water extracted from each supply lake for municipal use.In practice, total water consumption is comprised of multiple uses: residential, industrial, institutional, etc.We applied a water consumption model that aggregated these uses into one withdraw value, which was scaled to each city's population.This allowed for scaling based on both the demographic growth of the city and per capita changes in consumption.Three forecasts (high, medium, and low) were generated for both per capita consumption and population growth based on historic population and consumption data.In Iqaluit, the average per capita value over the reported period was 352 LCD, which was used as the 'medium' consumption setting.'Low' and 'high' settings were established by applying a -15 D r a f t 10 % and 14 % change to the average.The 'high' setting of 401 LCD mimicked the City of Iqaluit's designed consumption rate (Golder Associates Ltd. 2013) while the 'low' setting provided an evaluation of supply if efficiency measures were introduced.For Rankin Inlet, the reported average was 627 LCD.This rate was significantly higher than both that of Iqaluit and the Canadian residential average of 251 LCD (Environment Canada 2014).The notably high consumption values for Rankin Inlet are due to a high 'bleed rate' and leakage within their piped water utilidor system (Stantec Consulting Ltd. 2014).As such, we utilized 627 LCD for the 'high' consumption setting, and purposely assumed reductions of 15 % and 30 % as 'medium' and 'low' settings respectively.Thus, we assumed that water consumption in Rankin Inlet is already at its 'high' consumption value, and evaluated the effects of significant reductions on consumption in the future with infrastructure improvements.

<Table 3>
Per capita consumption rates were applied to the 2015 population estimates published by the Nunavut Bureau of Statistics and extended to generate forecasts for 2016-2035 (Nunavut Bureau of Statistics 2014).By combining matching per capita consumption levels and population forecast settings, total water consumption forecasts were generated (Table 3).

Hydrological Simulation
Evaluating the effect of each LTF parameter on water supply was processed by performing a water balance analysis.The amount of water stored in each supply lake represented the net result of total water input to the watershed subtracted by water removed via evapotranspiration and water withdrawn for municipal use.The simplified description of each watershed was evaluated through the use of the Hydrological Engineering Center's Hydrologic Modelling Software package (HEC-HMS).This allowed for the input of detailed geo-spatial D r a f t characteristics of the watershed to be combined with meteorological and consumption schedules in order to determine the dynamic storage volume of each supply lake.Detailed field data such as watershed contours, soil type, hydraulic conductivity, surface roughness, and dynamic active layer depth distribution were nonexistent, and are known to be limited in availability in potential study areas within northern regions.In their place, typical values were used to represent the watersheds and provide input data for the evapotranspiration routine within HEC-HMS (see supplemental).The overall model was simplified based on typical conditions found in similar northern regions described by Dingman (2008) andQuinton et al. (2005).For example, Iqaluit and Rankin Inlet are underlain by continuous permafrost and baseflow in continuous permafrost watersheds is negligible when compared to precipitation runoff (Woo et al. 2006).Simplification of the geospatial models within the HEC-HMS software allowed for higher sensitivity to be placed on the influence of precipitation, lake withdraw, and bathymetry within each watershed; datasets with higher implications for forecasting and community planning.The formulation of the hydrological model did not account for changing geological conditions, such as the deepening of the active layer and associated vegetation changes with warming temperatures (Wrona et al. 2016) over the 20yr forecasting period.Although this would have an impact on seasonal runoff values, incorporating this level of complexity into the model would counter the heuristic objectives of the method.Thus, we make the assumption that our model parameters calibrated on observed data are representative of future conditions, however, we acknowledge the inherent uncertainties in the modelling process that are likely to occur due to changes in catchment-mediated processes and unknown changes of parameters in a changing climate (Peel and Blöschl 2011).

D r a f t
The computation of snowmelt was conducted using a Temperature Index (TI) model.The HEC-HMS software is currently limited to only the Temperature Index method in snowmelt calculations.The TI approach applied a fixed amount of snowmelt for each degree-day above the specified Base Temperature (BT).This method allows for a reduction in data required for computation compared to a surface energy balance.Reported daily precipitation values are discriminated as snowfall or rainfall depending on the reported air temperature.The precipitation threshold temperature (PX) value conducts this discrimination, where temperatures below PX (2 °C) report snowfall.Above the base threshold temperature (BT) of 1 °C, melt occurs which is distinguished between wet melt rate and dry melt rate.If there is a precipitation event which exceeds a rain rate limit of 0 mm/day while a snowpack is present, the wet melt-rate of is applied (3.96 mm/°C-day).When no rainfall occurs, an ATI melt-rate function is applied (1.96 mm/°Cday), which can be modified by a time based pattern or antecedent coefficient.Due to a lack of in-situ experimental data, and to facilitate a simplified assessment for future heuristic application, the ATI-Metlrate was held stable as a function of time and experienced temperature differential.This same methodology applied to the ATI-Coldrate Function, which described the cooling of the snowpack and remained at a constant value throughout the simulation period and with respect to experienced temperature differentials.Due to the presence of continuous permafrost throughout the watershed, the ground melt component of the TI was set to 0 mm/day.
Water removed from the watershed due to evapotranspiration was estimated using a modified Thornthwaite Temperature Index model (Dingman 2008), allowing for daily values of estimated evapotranspiration to be forecasted using only air temperature and approximations of daily sunshine hours and without the need for complex forecasts of longwave and shortwave radiation.A constant value of 0.3 mm/day of sublimation was applied during days when snow D r a f t 13 was present and a crop coefficient of 0.65 was used.A constant sublimation rate helped to simplify the data requirements for sublimation evapotranspiration calculations when snow was present on the ground.The actual daily volume of water removed through evapotranspiration can be less than estimated evapotranspiration as actual volume is limited by the availability of water in Canopy storage.Canopy storage volume is replenished during precipitation events, and diminished when no precipitation occurs.The following formula was used to compute the PET forecasts (Dingman 2008); When SWE is present on surface: When no SWE is present on surface: Where PET is in mm/day, D is the length of sunshine hours per day, and e is the saturation vapour pressure at the mean daily temperature (Ta).
To account for the lake volume seasonally sequestered in ice, a physical model of ice growth was used; where heat conducted into the ice is assumed to equal heat released through phase change by the growth of ice.Allocation was provided for the transport of observed snowfall off of the lake ice.A coefficient (α = 15) was used during melt conditions to increase the melt rate from air temperature in order to represent melt rate contribution from radiation.The LTF output forecasted the accessible volume of water; the liquid water below an ice layer and above the elevation of the pump intake.The Water Survey of Canada (WSC) provides data records of water depth from their hydrometric station installed in Lake Geraldine (Iqaluit) for 2007-2016 (Water Survey of Canada 2016).This observed data was used to calibrate the settings employed by the HEC-HMS simulations.As Lake Geraldine experienced a modification to its dam in 2010 affecting maximum capacity, simulations were split into two periods, 2007-2010 and 2011-2014 (NWB consumption data was only available up to the end of 2014 during calibration).Over these periods, the root mean squared error (RMSE) between simulated lake-D r a f t level and HMS simulation was 0.2202 m.As lake depth is currently 13.73 m, this error represented approximately 4.4 % of the total volume of Lake Geraldine (Figure 3).

<Figure 3>
There was no WSC hydrometric station installed in Lake Nipissar (Rankin Inlet) and as such, no method of calibration for the Rankin Inlet simulations.Therefore, the HEC-HMS parameters calibrated for Lake Geraldine were applied to the watershed of Lake Nipissar.The water balance of Lake Nipissar in Rankin Inlet had an added input component from a seasonal replenishment pipeline, which transferred water from the Char River.This pipeline completed its first replenishment season in 2015 transferring 243,637 m³ over 85 days at an average rate of 0.034 m³/s.This pumping rate was 85 % that of the designed maximum capacity of 0.040 m³/s.Thus, we utilized a pump rate of 0.034 m³/s and 99 day replenishment schedule to match the maximum specifications of transferred water (Stantec Consulting Ltd. 2014).The seasonal replenishment schedule was held constant during all presented forecast scenarios.

Worst-case Scenarios
In addition to the 20yr water supply forecasts generated through the standard application of the LTF protocol, non-standard 'worst-case' simulations were also completed to investigate the impact of climate anomalies and infrastructure limitations.We examined the influence of a 33yr low of both total snowfall and total rainfall occurring, and the associated influence on water supply into the future.Meteorological records identified a 33yr low of snowfall occurred in 1998 for Iqaluit, and a 33yr low for rainfall in 2013.For Rankin Inlet, these years were 1989 and 1988, respectively.A composite yearly record of the 33yr low of total snowfall and total rainfall was generated and then applied at varying individual years on top of the standard LTF forecasts.

D r a f t 16
Settings of 'high' consumption, 'low' air temperature, 'low' snowfall, and 'low' rainfall were used as the baseline forecast settings, and individual years were replaced with the 33yr low value.For example, when the LTF supply forecast of 'high' consumption, 'low' air temperature, 'low' snowfall, and 'low' rainfall for Lake Geraldine was run, the 2017 year was replaced with the newly generated 33yr low dataset, and then the scenario re-simulated.By applying this 33yr low climate anomaly to each individual year preceding the date of water shortage for each town, the impact of a climate anomaly had on advancing the date of shortage was investigated.
We also investigated the impact of changes to the seasonal replenishment schedule of Lake Nipissar.While the standard LTF results for Rankin Inlet incorporated a consistent rate of seasonal replenishment from Char River, were ran simulations where individual years experienced no replenishment.For example, on top of the standard scenario of 'high' consumption, 'low' air temperature, 'low' snowfall, and 'low' rainfall for Lake Nipissar, replenishment pumping from the Char River (alternate water replenishment source) was turned off.This was repeated for each prior year to predict future water shortages introduced from a failure of summer water replenishment from the Char River.Rankin Inlet also experienced notably high water consumption in 2011.Thus, we also investigated the impact of another year of 962 LCD consumption, where an individual year of 'high' consumption between 2017 and 2022 was replaced with 962 LCD.

Climate and Demand Forecasts
The process of precipitation bias correction revealed climate trends different to those of uncorrected data.This was pronounced for Iqaluit, where adjusted snowfall was distinctly higher D r a f t after applying the bias corrections based on the gauges utilized (Table 1).Warming trends were observed since 1981 in both cities.While a 1.8 %/dec increase of snowfall was observed in Iqaluit, summer precipitation decreased by 6.9 %/dec.In contrast, an increase in both snowfall and rainfall at 5.7 and 3.4 %/dec respectively was observed for Rankin Inlet (Table 2; supplemental).

<Figure 4>
The primary output of the LTF was forecasts of accessible volume, which illustrated the range of accessible volume vs time curves for all climate/demand scenarios.This demonstrated the range of accessible reservoir volumes for the climate scenarios associated with each demand setting.Reported water consumption in Iqaluit ranged from 304-392 LCD.The 2015 consumption for Iqaluit (392 LCD) was similar to our modelled 'high' forecast.This consumption level is projected to cause end-of-winter shortages as early as 2025 (Figure 4a).
Likewise, even under lower consumption (e.g., 352 LCD) end-of-winter shortages are projected as early as 2031.We calculate that consumption would have to be reduced to 299 LCD for a viable supply past the 20yr planning horizon.The water supply of Iqaluit is forecast to enter a state of drawdown, where the water supply will no longer fully recharge by 2020, 2025, and 2034 for the high, normal, and low consumption settings respectively.There is a distinct tipping point in the nature of the accessible water volumes, which can be observed when the reservoir enters drawdown (Figure 4).For example, the upper boundary of the 'high' consumption setting in Iqaluit enters drawdown in 2023 when the summer recharge ceases to refill Lake Geraldine to D r a f t the maximum dam capacity prior to freeze up.In previous years, summer recharge exceeded dam capacity, which results in the reservoir filling to maximum capacity on the date of freeze up.
From 2023 onward, summer recharge no longer reaches dam capacity, and the trend of accessible volumes forecasted rapidly depletes.This tipping point is also observed for the lower boundary of the 'high' consumption setting and the upper boundary of the 'medium/current' consumption setting.This response can be used as an indicator of water supply integrity by municipal water managers; if the summer recharge does not result in maximum reservoir capacity up to the date of freeze up, rapid depletion can be expected in upcoming years.Climate anomalies also have a large influence on the amount of accessible water due to a dependence on precipitation runoff for the recharge.If a climate anomaly, such as a 33yr low in snowfall and rainfall occurred between 2017-2024 (e.g.similar to 1998 or 2009), the date of end-of-winter shortage could advance by 8 years depending on the consumption during this period (Figure 5a).
If a 33yr low of snowfall and rainfall were to occur in Iqaluit in 2017 under a 'high' consumption scenario, modelled results forecast that water shortages could occur as early as 2018.

Rankin Inlet
In 2011, the NWB reported total withdraws from Lake Nipissar at 962 LCD.This value is 2.7 times that of the average in Iqaluit and 3.8 times that of the Canadian residential average of 251 LCD (Environment Canada 2014).Under a high consumption setting of 627 LCD, end-ofwinter shortages are projected as early as 2022 (Figure 5).If we assume maintenance and/or conservation efforts are able to reduce consumption by 15 % (533 LCD), end-of-winter shortages may still occur as early as 2028.Our results forecast that consumption would have to be reduced by 30 % to 439 LCD for the Rankin Inlet water supply to remain viable past 2035.These D r a f t projections include the Char River seasonal replenishment pump operating at 85 % capacity, and still project near term shortages.Lake Nipissar has a watershed area ~30 % that of Lake Geraldine and experiences relatively less sensitivity to runoff.If a climate anomaly, such as a 33yr low in snowfall and rainfall occurred (e.g.similar to 1989), the date of end-of-winter shortage could advance by 1 year depending on the consumption during this period (Figure 5b).

<Figure 5>
Alternative Supply Evaluation Our LTF allowed for custom climate and demand scenarios to be simulated, such as the alternative supply and replenishment of Lake Nipissar from the Char River in Rankin Inlet.We calculated that any interruption to the seasonal replenishment of Lake Nipissar or extreme climate anomalies could advance the date of water shortage by 1-2 years.For example, even if only the 2016 season of replenishment was skipped, Rankin Inlet could face a water shortage in 2020 (Figure 5b).The replenishment pipeline was designed to sustain consumption requirements when operating at 85 % capacity, and a water consumption rate of 469 LCD (Stantec Consulting Ltd. 2014).We forecasted this rate and presented it as the LTF 'low' consumption scenario, reflecting viability past 2035.However, we calculated that Lake Nipissar is currently in a state of drawdown (Figure 4b).If the annual average consumption remained higher than 627 LCD (our 'high' forecast scenario), short-term shortages will likely occur.For example, if the consumption rate experienced in 2011 of 962 LCD were to occur in any year between 2016 and 2019, Rankin Inlet could face end-of-winter shortages in 2020 even with seasonal replenishment operating at 85 % capacity (Figure 5b).This vulnerability would be compounded further by longer than normal winters resulting in later thaw of lake ice.

Climate and Demand Forecasts
Our heuristic hydrologic model was designed to produce forecasts using near term data for site specific conditions.The difference in trends in both towns highlighted the importance for consideration of local level conditions as opposed to exclusively relying on regional or global models of climate change.Likewise, while RCP regional climate models depict an overall increase in winter precipitation for Nunavut, the observed local trends currently differ from this trend.The variability in observed patterns for snowfall is likely a result of the high degree of spatial variability in Arctic regions relative to other snow covered regions (Liston 2004).Derksen and Brown (2012) found that observed reductions in June snow cover extent since 2005 were far below model projections.Ultimately, the influence of future environmental change is likely to depend on local geography and socio-economic factors, such as infrastructure and development (Instanes et al. 2016).The near term vulnerability of Lake Geraldine and Lake Nipissar to climate anomalies and high consumption events illustrate the immediate need for a real time monitoring program to be established.

Supply Forecasts
Under all consumption scenarios considered the climate forecast settings of 'lowest air temperature, snowfall, and rainfall projected the lower boundary of the forecasted volume.This represented climate scenarios with the least mean-annual air temperatures, and lowest snowfall and rainfall estimates.Freshwater systems in northern regions have already been shown to be vulnerable to desiccation under reduced snowfall (Bouchard et al. 2013), and our results demonstrate this is also a primary vulnerability for the sustainability of municipal water supplies D r a f t under low snowfall conditions.Under future warming projections, reduced thickness (Brown and Duguay 2011), and further reductions in winter precipitation (Derksen and Brown 2012) may increase the risk to municipal water supplies.Likewise, precipitation runoff was the primary source of water for Iqaluit's water supply in summer, and as such, accessible volumes were noticeably affected by climate simulations under reduced precipitation.This sensitivity was more apparent during the lower consumption scenarios.
While warmer forecasts demonstrated higher evaporation losses, cooler forecast scenarios resulted in thicker lake ice cover, thereby removing a greater amount from accessible volumes during winter.The seasonal sequestering of accessible lake volume in ice significantly impacted each community's end-of-winter supply.Each town experienced a maximum ~2 m of lake ice depending on climate conditions.With a total useable depth of 9.73 m above intake in Iqaluit, 35 % of the accessible volume was sequestered in ice when the water supply lake was at its spillway elevation.It is important to note, even under warmer climates, variability in weather can increase the risk to supply under reduced snowfall and potential years of increased ice-thickness (Prowse et al. 2011).Thus, during years of warm summers and cold winters, there may be pronounced vulnerability to diminished end-of-winter supply if reservoirs do not fully recharge prior to ice formation.The shallow nature of Arctic supply lakes leave them more susceptible to the effects of lake ice and an overall lower rate of recharge (Instanes et al. 2016).For example, Lake Nipissar has a maximum reported useable depth of only 4.6 m measured above the intake, with ~56 % of the accessible volume stored in ice at the end of the winter (Stantec Consulting Ltd. 2014).We found that this percentage only increases as water levels drop.As such, monitoring of ice-conditions on municipal supply lakes in northern regions may be essential to elucidate longterm trends.

D r a f t 22
We found a pronounced short term vulnerability of both towns' water supplies.This underscored the immediate importance of conducting water supply forecasts in northern communities.Infrastructure projects require significant time for mobilization and execution in remote and especially northern communities.For example, the initial Lake Nipissar water supply assessment was completed by FSC in 2010 and the resultant pipeline did not begin replenishment until 2015 (Stantec Consulting Ltd. 2014).If a similar 5 year lead time is required to respond to the our results, Iqaluit or Rankin Inlet would only have 1-4 years before severe water shortages would occur.In extreme climate conditions, Iqaluit may have less than 5 years before reaching critical water shortages (Figure 5).A long term water security strategy is essential to incorporate these findings and better plan for the future.Our ability to attain baseline water supply forecasts should reduce exposure to future water shortages, and this data could be used for water management planning before emergency conditions are likely to occur.

<Figure 6>
The viability of the water supply for Rankin Inlet is directly tied to the availability of water for seasonal replenishment.Reductions in replenishment volume, or a missed season of replenishment, will result in an earlier than expected date of water shortages.The replenishment pipeline intake was located at the outflow of a lake on the Char River.There was a lack of hydrometric records of Char River creating uncertainty regarding the expected flow rates throughout the replenishment season.Likewise, community consultation on this municipal project was minimal.As such, there is much debate behind the feasibility of the Char River remaining a viable summer replenishment source due to extremely low flow conditions observed D r a f t (Figure 6).The designed replenishment was based on a limit of 10 % of the river flow rate to preserve the integrity of the downstream ecosystems (Stantec Consulting Ltd. 2014).With reported minimum summer river flows of 0.131 m³/s, this 10 % limit represented only 33 % of the designed pumping rate (Stantec Consulting Ltd. 2014).This highlights the vulnerability of the chosen intake location to low flow conditions.Likewise, climate is projected to become increasingly variable for northern regions, and years of significantly reduced summer precipitation may impede recharge potential (Prowse et al. 2011).Our methodology can be expanded to provide flow forecasts for alternative sources like Char River.Through this application, more informed decisions can be made regarding infrastructure investments to avoid expenditures into temporarily solutions.

Conclusion
Our examination of the influence of climate and demand shows significant short term vulnerability for the water supplies of Iqaluit and Rankin Inlet.Under current consumption, Iqaluit is at risk of end-of-winter shortages as early as 2018.Likewise in Rankin Inlet, unless substantial conservation efforts are employed immediately, end-of-winter shortages could occur as early as 2020.The climate and growth scenarios employed through the LTF protocol were developed for applications where historical in situ data was limited and in situations of limited local funding and capacity for conducting water resource assessments.While these limitations restrict in-depth fieldwork and detailed assessments, they do not preclude the immediate need for supply forecasting.The simplistic and heuristic nature of this method allows for broad application to other regions where small or remote communities face similar limitations in local capacity and funding.This application reaches beyond northern and Arctic regions and can  ).RCP models provide climate projections of plausible future conditions in response to a range of anthropogenic forcing, which range from low emissions and mitigation (RCP 2.6), intermediate emissions (RCP4.5), to a high emission scenario (RCP8.5).
Table 3: Settings for the forecast of demand for Iqaluit and Rankin Inlet, Nunavut.Modelled scenarios for Iqaluit are based on average conditions (Medium), with 'Low' and 'High' settings representing a -15 % reduction and a 14 % increase to the current average water consumption respectively.For Rankin Inlet we assume reductions to the current consumption value of 15 % (Medium) and 30 % (Low) from the current 'High' average consumption.

D r a f t
significance, the trends provided a lower boundary of plausible climate conditions to assess the response of the supply lake.The rate of change of 'high' for snowfall and rainfall was set to the modelled forecast values obtained by the RCP 8.5 model of the IPCC Global Climate Model for Nunavut (Environment and Climate Change Canada 2016b).With this combination of historic 30yr climate normals, locally assessed climate trends, and global climate model trends, the response from a full scope of plausible climate trends can be analyzed.
These two factors were adjusted iteratively to match observed ice measurements.Model calibration utilized ice-thickness, time of freeze-up and thaw as reported in 1995(Trow  2004) and 2009 (Stantec Consulting Ltd. 2014).The following formulae are used to calculate ice thickness on a daily basis.During ice growth:During ice melt:Where dXi/dt = daily ice growth (m/day), Tw = water temperature (°C) assumed to be 0 °C, Ta = mean daily air temperature (°C), Raf = air film resistance (m²C/W), ST = constant % of SWE remaining after wind forced relocation, Xs = depth of snow (m), ρs = snow density (kg/m³), ks = thermal conductivity of snow (W/mC), Xi = thickness of ice cover (m), ki = thermal conductivity of ice (W/mC), L = latent heat of fusion (J/kg), ρi = ice density (kg/m³), and α = air temperature melt factor.
management strategies in any global location.Thus, short term risk can be identified to direct planning, adaptation, and infrastructure responses to changing climate and demand.

Figure 1 :
Figure 1: Map of Nunavut, Canada, showing the study areas of Iqaluit and Rankin Inlet.

Figure 2 :
Figure 2: Data analysis process for modelled parameters.

Figure 3 :
Figure 3: Comparison of Lake Geraldine observed and simulated water depth (m.a.s.l.).The

Figure 4 :
Figure 4: Long Term Forecast of accessible water volume (1000 m3) for a) Iqaluit and b)

Figure 5 :
Figure 5: Long term forecast indicating the effect a 33yr low of snowfall and rainfall occurring

Figure 6 :
Figure 6: Outlet of the Char River, Rankin Inlet, Nunavut as observed a) August of 2007, b) Map of Nunavut, Canada showing the study areas of Iqaluit and Rankin Inlet.
Comparison of Lake Geraldine observed and simulated water depth (m.a.s.l.).The break in the simulated data from 2009-2011 was chosen to avoid false simulation error while dam modifications were in progress.
Long Term Forecast of accessible water volume (1000 m3) for a) Iqaluit and b) Rankin Inlet.Three consumption scenarios are indicated with the influence of air temperature, snowfall, and rainfall forecasts displayed by the shaded area.
Long term forecast indicating the effect a 33yr low of snowfall and rainfall occurring in any single year from 2017-2021 on the accessible volumes of a) Lake Geraldine (Iqaluit) and b) Lake Nipissar (Rankin Inlet) as indicated by the shaded area under 'high' consumption.The effect of Lake Nipissar without seasonal alternative water supply replenishment, and a single year of 962 LCD (similar to 2011) occurring after 2015, is also shown.Figure 5 401x105mm (300 x 300 DPI) Outlet of the Char River, Rankin Inlet, Nunavut as observed a) August of 2007, b) August of 2006.

Table Captions Table 1 :
Climate Normals (1981-2010)for Iqaluit and Rankin Inlet, Nunavut.Adjusted values are based off of daily averages computed from hourly data.

Table 2 :
Settings for the Long Term Forecast of climate for Iqaluit (IQ) and Rankin Inlet (RI).IPCC Global Climate Model scenarios for Canada 75 th percentile Representative Concentration Pathways (RCP) model for Nunavut 2016-2035 are included (Environment and Climate Change

Table 1 .
Climate Normals (1981-2010)for Iqaluit and Rankin Inlet, Nunavut.Adjusted values are based off of daily averages computed from hourly data.

Table 2 .
Settings for the Long Term Forecast of climate for Iqaluit (IQ) and Rankin Inlet (RI).IPCC Global Climate Model scenarios for Canada 75 th percentile Representative

Table 3 .
Settings for the forecast of demand for Iqaluit and Rankin Inlet, Nunavut.