Climate and water availability indicators in Canada: Challenges and a way forward. Part II – Historic trends

Abstract

Climate variability is recognized as an important influence on the availability of water throughout Canada, and projected climate change is anticipated to alter the amount, timing and distribution of water. This is Part II of a three-part (Parts I and III, this issue) analysis of water availability in Canada. Part II surveys current research, primarily Canadian in origin, on historical trends in climate and hydrologic indicators relevant to assessing water availability. Information on hydro-climate trends is not evenly distributed across Canada. Hydrologic trend research focuses on the North, British Columbia and the Prairies (Saskatchewan) with some research in Quebec, very little in Ontario and minimal analysis for Atlantic Canada. Overall, there is less research on trends in climatological indicators (drought, evapotranspiration, soil moisture); generally, the focus is on the Prairies. Hydrologic trends from basin-scale case studies are reported but inter-comparison is constrained by different periods of analysis. Trends vary by region. Generally, in the Prairies annual streamflow is decreasing, while results for the Yukon, BC, Ontario and Quebec are mixed. There is no clear signal on drought trends. Drought results are influenced by location, data (instrumental or paleo-climate), period of record and treatment of components in the drought index. For historic trend studies, observations with long duration, spatial coverage and minimal human-influences are crucial. These limitations in data affect streamflow assessment, but analysis of evapotranspiration and soil moisture trends is constrained by additional challenges of measurement, modelling and few data sets. Future research recommendations include combining climate change trend assessment with potential influences of large-scale atmospheric circulation patterns on inter-annual hydro-climatological variability. There is a need to explore the uncertainty associated with and methodological challenges in statistical analysis of hydro-climate time series. At present, there have been no rigorous detection and attribution studies of hydrology indicators in Canada.

Les effets importants de la variabilité du climat sur la disponibilité de l’eau au Canada sont reconnus, et on s’attend à ce que les changements climatiques projetés influent sur la quantité d’eau et sur la distribution de l’eau. Le présent article est le deuxième d’une analyse en trois parties (partie I et partie III, la série présente) de la disponibilité de l’eau au Canada. La partie II porte sur les études en cours, principalement au Canada, des tendances historiques des indicateurs climatiques et hydrologiques qui sont pertinentes à des fins d’évaluation de la disponibilité de l’eau. L’information sur les tendances hydroclimatiques n’est pas répartie de façon uniforme au Canada. La recherche sur les tendances hydrologiques s’intéresse principalement aux régions nordiques, à la Colombie-Britannique et aux Prairies (Saskatchewan), certaines analyses se centrent sur les tendances au Québec, très peu en Ontario et une infime partie dans le Canada atlantique. Il y a globalement moins d’études des tendances observées pour les indicateurs climatiques (sécheresse, évapotranspiration, humidité du sol); en règle générale, l’accent est mis sur les Prairies. Les tendances hydrologiques qui émergent des études de cas à l’échelle d’un bassin sont signalées, mais il est difficile d’établir des comparaisons entre ces tendances parce que les périodes d’analyse diffèrent. Les tendances varient d’une région à l’autre. Dans les Prairies, le débit annuel affiche une tendance généralement à la baisse tandis que le Yukon, la Colombie-Britannique, l’Ontario et le Québec présentent des résultats mixtes. Aucune tendance à la sécheresse ne se dessine clairement. Les résultats concernant les sécheresses dépendent de l’emplacement, du type de données (paléoclimatiques ou recueillies à l’aide d’instruments), de la période d’enregistrement et du traitement des éléments pris en compte dans l’indice de sécheresse. Certains aspects sont cruciaux pour les études des tendances historiques, notamment la longue durée de l’enregistrement des observations, la couverture spatiale et un nombre minimum d’effets anthropiques. Ces limites relatives aux données influent sur l’évaluation du débit, mais d’autres facteurs, comme la prise de mesures, la modélisation et la rareté des ensembles de données imposent des contraintes additionnelles à l’analyse des tendances de l’évapotranspiration et de l’humidité du sol. Les recommandations à l’égard des futures études comprennent notamment l’évaluation combinée des tendances des changements du climat et des effets possibles de patrons de circulation atmosphérique à grande échelle sur la variabilité hydroclimatologique interannuelle. Il est également nécessaire d’examiner les incertitudes qui entachent les analyses statistiques des séries chronologiques de données hydroclimatologiques et les méthodologies de ces analyses. Aucune étude rigoureuse n’a été menée jusqu’à maintenant sur la détection et l’attribution des indices hydrologiques au Canada.

Introduction

Assessment of the availability of water in Canada is an important component in effectively managing this important resource sustainably. Climate variability is recognized as an important influence on the availability of water throughout Canada, and projected climate change is anticipated to alter the amount, timing and distribution of water. Key questions emerge with respect to understanding the implications of a changing climate on water availability in Canada, including: how has the resource changed over time and area, what is the current status, and how is it likely to change in the future? This is the second component in a three-part series of papers (Cohen et al. 2015; Koshida et al. 2015) that explore these questions. It focuses on historical trends in hydrologic and climate indicators that are germane to understanding changes in surface water availability. The paper builds upon Part I (Koshida et al. 2015) which reviews literature on defining water availability – relevant indicators, their purposes, and the strengths and weaknesses of these metrics. The current paper (Part II) explores methods for detecting trends and associated caveats and methodological challenges of assessing trends in water availability. In addition, key regional trends for Canada in annual, winter and summer runoff/streamflow from basin-scale studies and drought assessments are synthesized. Trends in groundwater-based indicators are not included in this review. However, such a topic would be timely given the importance of groundwater as a resource in Canada and the growing literature on groundwater trends (Rivard et al. 2009; Allen et al. 2014). Part III (Cohen et al. 2015), focuses on assessing future implications of climate change, including a review of some common approaches for developing and assessing future scenarios of water availability and emerging potential trends and impacts. In this paper as well as the other two, a way forward is outlined in addressing important research questions related to trends and future scenarios of water availability in Canada in light of a changing climate.

Hydrology-based indicators

Studies to assess climate-driven trends in river flow/streamflow have been undertaken at the catchment, regional, national, continental and global scale (e.g. Yulianti and Burn 1998; Ouarda et al. 1999; Douglas et al. 2000; Whitfield and Cannon 2000; Zhang et al. 2001; Rood et al. 2005, 2008; Stewart et al. 2005; Burn et al. 2008; Déry et al. 2011, 2012). These assessments have addressed a number of themes, including:

•	Detecting whether there are statistically significant trends in indicators of river flow and other hydro-climatic indicators (Burn 1994; Lettenmaier et al. 1994);
•	Establishing whether there are statistically significant direct links between changes in hydro-climate indicators with changes in climate variables such as air temperature and precipitation (Westmacott and Burn 1997; Whitfield 2001; Burn and Hag Elnur 2002; Cunderlik and Burn 2004; Burn et al. 2008; Saint-Laurent et al. 2009; Déry et al. 2009b; Assani et al. 2010);
•	Detecting the changes in hydrology and then attributing changes to anthropogenic climate change informally (Yulianti and Burn 1998; Stott et al. 2010; Trenberth 2011) and with formal techniques (Maurer et al. 2007; Barnett et al. 2008; Hidalgo et al. 2009);
•	Linking variability in regional water supply to large-scale atmospheric and oceanic phenomena using indices such as the North Atlantic Oscillation (NAO), El Ninõ Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), and Arctic Oscillation (AO), for example (Déry and Wood 2005; Kingston et al. 2006; Fleming et al. 2007; Burn et al. 2008; Brabets and Walvoord 2009; Assani et al. 2010; St. Jacques et al. 2010; Nalley et al. 2012);
•	Also, some have combined these themes for broader assessments (Burn et al. 2004; Abdul Aziz and Burn 2006).

Trend analysis requires the careful selection and preparation of data for analysis. This includes screening of observations based on an assessment of location, representativeness, human influences, and data quality (e.g. rating curves), and subsequent checks to determine whether there are factors affecting data homogeneity. Long-duration time series are critical for trend detection, since this increases the power of statistical tests to detect trends (Chen and Grasby 2009; St. Jacques et al. 2010) and the signal-to-noise ratio of trends (Fleming 2010). Researchers have had to balance the need for temporal coverage with spatial coverage since as the length of record increases, the number of stations and their spatial distribution decreases (Harvey et al. 1999; Burn and Hag Elnur 2002; Khaliq et al. 2008; Burn et al. 2010). There are spatial gaps in hydrologic data suitable for trend detection, most notably in the north, particularly the Arctic Archipelago, central Prairies and urbanized areas, while coverage is dominated by stations in the south that also have longer periods of record (Déry et al. 2009a; Burn et al. 2010). Some researchers have discussed extending the period of record by filling in missing data using standard procedures, before estimating trends and significance (Wilby 2006; Khaliq et al. 2009; St. Jacques et al. 2010).

In climate change signal detection, determining whether trends in a hydrologic time series are statistically significant is a challenge since one is trying to distinguish between natural variability and any emerging upward or downward trends in the data. Kundzewicz and Robson (2004) provide a review of methods to detect change in hydrological records. The assessment of a hydrologic series can be undertaken for an individual station (point scale) or more broadly in the context of an observation network or representative drainage basins or climatic regions (field scale). The process includes: choosing the indicators for analysis, selecting the hydrometric stations, evaluating trends, and assessing trend significance, typically through a null-hypothesis significance test.

A number of parametric and non-parametric statistical techniques have been applied for detecting or otherwise characterizing trends in hydrologic time series, including Spearman rank correlation, the Mann–Kendall (M–K) test, linear regression, Bayesian and non-Bayesian change-detection algorithms, various forms of low-pass filtering, resampling methods and signal-to-noise ratios (Matalas and Langbein 1962; Ouarda et al. 1999; Burn and Hag Elnur 2002; Khaliq et al. 2009; Déry et al. 2012; Fleming and Weber 2012). Khaliq et al. (2009) inter-compare four statistical methods (two rank- and two slope-based) and demonstrate issues associated with the detection of trends in annual mean flows in Canada. The M–K nonparametric technique has emerged as the most frequently used test in Canadian studies, and in many studies it is the only statistical method used (Monk et al. 2011).

The detection and evaluation of the significance of trends can be affected by the nature of hydro-climatic time series including distribution of the data, seasonality, autocorrelation (serial and spatial correlation) and underlying assumptions for implementing statistical tests (von Storch 1995; Douglas et al. 2000; Kundzewicz and Robson 2004; Cohn and Lins 2005; Khaliq et al. 2009). For example, not accounting for positive serial correlation within a time series can result in a more ‘liberal’ hypothesis test and a resultant overestimation of significant trends detected (von Storch 1995, 17; Yue et al. 2002). A number of methods such as pre-whitening, bootstrap resampling, and variance correction have been used to address the issue (Hirsh and Slack 1984; Kulkarni and von Storch 1995; Hamed and Rao 1998; von Storch and Navarra 1999; Darken et al. 2000; Burn and Hag Elnur 2002; Yue et al. 2002; Cunderlik and Burn 2004). However, recent literature has questioned an implicit assumption associated with the trend tests – streamflow behaves as an independent and identically distributed (IID) random variable or autoregressive (AR) process (e.g. Yue et al. 2002; Fleming and Clarke 2002; Cohn and Lins 2005; Koutsoyiannis and Montanari 2007; Déry et al. 2009b). It points to the need to explore the uncertainty associated with, and methodological challenges in, statistical analysis in hydro-climate research.

Trend analysis has been conducted for varying time periods, but given the variability of climatic and hydrologic indicators, the choice of period of analysis can significantly influence results. Chen and Grasby (2009) demonstrate that two common statistical tests in hydroclimate trend detection, the M–K and Thiel–Sen, are sensitive to natural quasi-periodic oscillations of a decade or longer. Factors influencing the temporal trend detection are the length of the historical data set, the magnitude of the cycle and at what position in the oscillating time series the analysis begins. They recommend that trend assessment studies should use time series that are longer than 60 years, and analysis should not start at the highest or lowest position in a cycle. For example, early work by Déry and Wood (2005) in northern Canada reported a declining trend in river discharge for the period 1964–2003; however, by extending the period of record to 2007, Déry et al. (2009a) found a trend reversal – an increase in streamflow. The uncertainty in regional trend analysis resulting from different observation periods can be explored by calculating the trend significance indices for regional trends from different scenarios of partially overlapping common observation periods and their record lengths shifted on a time scale (Cunderlik and Burn 2004), or trend analyses based on shorter periods of record can be set in the historical context of longer time series (Wilby 2006; Burn et al. 2012).

The underlying premise in many hydrology trend detection studies is that there is a direct relationship between trends in climate and hydrology – similarities between trends and patterns detected in climatological variables such as temperature and precipitation and hydrologic indicators related to streamflow, for example, imply that these trends are related (Burn and Hag Elnur 2002). However, hydrologic temporal trends related to climate change impacts are not always uniform across large areas and, in some cases, differing regimes exist, confounding the expected/theorized climate change–hydrology relationships (Khaliq et al. 2009). For example, several studies in western Canada have demonstrated that adjacent glacial and non-glacial watersheds show systematically different responses to long-term climatic trends (Fleming and Clarke 2003; Stahl and Moore 2006; Fleming and Weber 2012).

Trends in hydrologic indicators have been linked with trends in climate indicators (e.g. temperature, degree-days, precipitation [rain, snow], snow cover, ice cover) to detect the effects of climatic change. However, simultaneous changes in other factors influencing hydrologic response, such as changes in land use, land cover, and land and water management practices, confound the detection process (Burn et al. 2008). Methods to differentiate among these factors influencing the behaviour of hydrologic trends would enhance the robustness of the attribution process (Burn et al. 2008). Jones (2011) introduced additional ecological and social processes (e.g. effects of vegetation responses to past human-caused disturbances and climate variability and change) that can complicate the analysis, and developed a checklist to guide trend assessment studies.

There may be merit in also exploring broader, more indirect processes driving inter-annual hydro-climatological variability, such as large-scale, atmosphere–ocean patterns to reveal relationships between circulation and variability in climate and streamflow, to supplement trend analysis for monotonic climate change influences (Woo and Thorne 2003). These patterns of low-frequency climate variability may introduce uncertainty in trend analysis of long hydro-climatic time series at the local and regional scale (Cunderlik and Burn 2004; Woo et al. 2006). Moreover, the low-frequency variability (e.g. long-duration severe meteorological or hydrological drought) may prove more challenging for adaptation than responding to short-duration, high-frequency variability in hydro-climate hazards (Khaliq et al. 2008; St. Jacques et al. 2010).

Researchers have explored the correlation of a wide range of indices with hydro-climate indicators for North American and Canada-wide influences (Rood et al. 2005; Gobena and Gan 2006; Bonsal and Shabbar 2008), and for various regions (Coulibaly et al. 2000; Déry and Wood 2005; Fleming et al. 2007; Woo and Thorne 2008) and specific watersheds in western, northern and eastern Canada (Burn 2008; Brabets and Walvoord 2009; Thorne and Woo 2011; Déry et al. 2012; Mazouz et al. 2012; Peters et al. 2013). A range of hydro-climate indicators, at different temporal scales, were used, including: snow cover (Brown and Goodison 1996), snow water equivalent (Hamlet et al. 2005), nival and glacier-influenced streamflow (Fleming et al. 2006), annual streamflow (Anctil and Coulibaly 2004; Déry et al. 2012), seasonal streamflow (Coulibaly and Burn 2005) and low-flow and high flow extremes (Ehsanzadeh and Adamowski 2008; Khaliq et al. 2008; Assani et al. 2010, 2012). Strong correlation between the teleconnection patterns and streamflows can identify notable changes in hydro-climatic behaviour, such as those found in the 1950s and 1970s related to periodic cycles (Anctil and Coulibaly 2004; Coulibaly and Burn 2004; Assani et al. 2010).

Stewart et al. (2005) illustrated the value of incorporating trend detection with an exploration of the potential influences of large-scale circulation anomalies (e.g. PDO for western North America) on streamflow changes. They were able to develop evidence that advances in the timing of spring streamflow were not solely influenced by PDO, but that a significant component of the change could be related to warming in the region reflecting global temperature trends. St. Jacques et al. (2010) explored the complexities of trend assessment and analyzed for the relative effects of climate change, PDO and human impacts on annual streamflow trends of southern Alberta rivers.

Trends in hydrology relevant to water availability

Annual, winter and summer trends from recent basin-scale studies in Canada are presented in Table 1. Direct inter-comparison amongst studies is constrained due to different periods of record analyzed; however, some general patterns emerge, with trends varying by region. Studies of watersheds in the Prairies are generally reporting decreasing trends in annual streamflow. Yukon, BC, Ontario and Quebec have mixed trends. Large declines in summer runoff or streamflow have been documented for the Peace, Athabasca, Oldman, and South Saskatchewan rivers. Streamflow in the Winnipeg River has increased primarily due to increases in November–April flow, with March streamflow increasing 101% (St. George 2007). This trend differs from changes reported by Déry and Wood (2005) and Rood et al. (2005) in other watersheds; however, flows of recent years have generally been lower than those of the 1960s and early 1970s.

Table 1. Observed trends in hydrologic indicators from basin-scale case studies (watersheds listed from western to eastern Canada).

River	Annual runoff or streamflow trend	Winter runoff or streamflow trend	Summer runoff or streamflow trend	Source
Yukon River (YT)	→1944–2005;	↑1944–2005; due to warm Pacific Decadal Oscillation, 1976–2005.	↓ During warm Pacific Decadal Oscillation, 1976–2005.	Brabets and Walvoord (2009)
	↑ during warm Pacific Decadal Oscillation (PDO),
	1976–2005
White River, Alsek River; glacier in basins (YT)	↑1975–1999			Fleming and Clarke (2003)
Dezadeash River, no glacier in basin (YT)	↓1953–1999			Fleming and Clarke (2003)
Big Creek, no glacier in basin (YT)	↓1975–1999			Fleming and Clarke (2003)
Campbell, Stave, Cheakamus and other rivers in south coastal BC	↑Increase,	↑ Dec.–Jan.,↓ Feb., 1984–2007		Fleming and Weber (2012)
	1984–2007
Cowichan River (BC)			↓ August, 1961–2006	Fleming (2010)
Fraser River (BC)	Variability increasing			Déry et al. (2012)
Castle River near Beaver Mines (BC)	↓1946–1948,			Rood et al. (2005)
	1951–2002
Bridge River (BC)	Possible small	Small ↑ 1984–2007	Possible small ↑ June–July, ↓August, 1984–2007	Fleming and Weber (2012)
	↑1984–2007
Columbia River (BC)	Possible small	↑1984–2007	Possible small ↑ June, ↓ August, 1984–2007	Fleming and Weber (2012)
	↑1984–2007
Columbia River at Fairmont Hot Springs (BC)	↓1946–1995			Rood et al. (2005)
Peace River (BC)	→ 1984–2007	→ 1984–2007	→ 1984–2007	Fleming and Weber (2012)
Peace River at town of Peace River (AB)			↓ –57%, 1912–2003	Schindler and Donahue (2006)
Mackenzie River (BC, AB, YT, SK, NT)	↑ECMWF; ↓EC; 1961–2002^a			Yip et al. (2012)
Liard River (YT, BC, AB, NT)	↓annual mean, 1960–1999	↑1975–1999, and 1960–1999	↓June and August, 1975–1999	Burn et al. (2004)
	↑annual minimum, 1960–1999
	↓annual maximum, 1975–1999
Athabasca River below Fort McMurray (AB)			↓ –19% from 1958–2003; ↓–33.3% 1970–2003↓ –30% to –20%, 1958–2009	Schindler and Donahue (2006)Peters et al. (2013)
Red Deer River (AB)	↓1913–2002			Rood et al. (2005)
Bow River (AB)	Small ↓ 1911–2002; and 1912–2007			Rood et al. (2005); St. Jacques et al. (2010)
Elbow River below Glenmore Dam (AB)	↓1912–2001			St. Jacques et al. (2010)
Oldman River near Lethbridge (AB)	↓1912–2001			St. Jacques et al. (2010)
			↓ –42% from 1915–2003	Schindler and Donahue (2006)
South Saskatchewan River at Medicine Hat (AB)	↓1913–1930, 1936–2002;			Rood et al. (2005); St. Jacques et al. (2010)
	↑1912–2001 (naturalized, climate only)
South Saskatchewan River at Saskatoon (AB, SK)			↓ –84% from 1912–2003	Schindler and Donahue (2006)
North Saskatchewan River at Edmonton (AB)	↓1911–2007			St. Jacques et al. (2010)
Prairie unregulated streams	↓1966–2005		↓1966–2005	Burn et al. (2008)
Winnipeg River (MB)	↑+58% since 1924	↑since 1924	→since 1924	St. George (2007)
Hudson Bay (MB)	↓1964–1980s;	↑1964–2008	↓1964–2008	Déry et al. (2011)
	↑1980s–2008
Churchill (MB)	↓1964–2008			Déry et al. (2011)
Nagagami River (ON)	↑1954–2008			Nalley et al. (2012)
Missinaibi River (ON)	↓1954–2008			Nalley et al. (2012)
Black River (ON)	↑1954–2008			Nalley et al. (2012)
Richelieu River (QC)	↑1954–2008			Nalley et al. (2012)
Eaton River (QC)	Small ↓1954–2007			Nalley et al. (2012)
St. Lawrence River, south shore tributaries (QC)		↓1950–2000		Assani et al. (2012)
St. Lawrence River, tributaries north and west of Montreal (QC)		↑1950–2000		Assani et al. (2012)
Thirteen rivers in New Brunswick (NB)		↑11/13 (5 sig. at 70%), ↓2/13, 1969–2006	↓9/13 (5 sig. at 70%) 1969–2006	Arisz et al. (2011)

Notes:

^a Runoff calculated using WATFLOOD with input data from Environment Canada (EC) gridded data set and European Centre for Medium Range Weather Forecasting (ECMWF) reanalysis climate data set (ERA-40).

Analyses demonstrate statistically significant correlations between seasonal streamflow variation and key climate indices (Table 2). For example, some trends appear to be correlated with phases of the PDO, including the Yukon River (Brabets and Walvoord 2009), Liard River (Burn et al. 2004) and some watersheds in the Prairies (St. Jacques et al. 2010). For the Liard River study, Burn et al. (2004) found an increase in winter flow (December. to April), an increase in annual minimum flow, an earlier onset of spring runoff, and an earlier peak spring flow; in addition, increases in April flow and April and spring air temperature as well as January to March flows and winter temperature were correlated. Different forcing factors influence trends – earlier onset of spring freshet was related to warmer spring temperatures, but the increase in winter flows was related to the PDO. Coulibaly and Burn (2005) assessed seasonal trends in streamflow at 79 stations organized into regions representative of different timing and magnitude of streamflow and correlated trends with persistent, large-scale Northern Hemisphere climate patterns. A change point in streamflow related to climatic indices was detected in the 1970s similar to other researchers (Anctil and Coulibaly 2004; Coulibaly and Burn 2004; Nalley et al. 2012), but another strong streamflow shift was observed in the 1950s.

Table 2. Selected studies using teleconnection indices and links to climate and hydrologic variability.

Region/indices	Effect(s) on climate	Effect(s) on hydrology	Watersheds	Source
Western Canada
Southern Oscillation Index (SOI), Multivariate El Ninõ Southern Oscillation (ENSO) Index (MEI), Pacific/North American (PNA) Index	Higher temperatures and lower than normal precipitation	Lower peak flows in the snowmelt period Spatial correspondence of high annual maximum flows and winter SOI and annual precipitation correlation (local factors contradict – coastal, glacial and lake influenced streamflow)	150 unregulated or moderately regulated stations in western Canada	Woo and Thorne (2003)
Pacific Decadal Oscillation (PDO)	“Warm” or “positive” phase: Correlation varies with latitude; dominant winter and spring. Apr. to Oct. air temp. warmer; high winter precipitation in areas with less precipitation; lower snow water equivalent; warmer spring and earlier snowmelt	Latitude-dependent; response different in nival, pluvial or hybrid rivers: southern BC/Washington decrease in flow; Alaska increase in flow	Review using Washington, BC, Alaska results	Whitfield et al. (2010)
PDO	“Warm” or “positive” phase:snow in coastal zone; inland areas low snow accumulation	Winter streamflow increases in coastal areas Reduced spring runoff	Rivers of Alaska, Yukon, Northwest Territories, British Columbia and Alberta	Woo and Thorne (2008)
PNA	“Warm” or “positive” phase: associated with a deficit of snow cover in western North America			Brown and Goodison (1996)
PDO, SOI, North Atlantic Oscillation (NAO), ENSO	PDO positive phase dry; PDO negative phase wet; winter more precipitation	A strong negative relationship between PDO and annual streamflow in south and central Alberta; weak positive relationship northwestern Alberta	16 stream gauges primarily in Alberta	St. Jacques et al. (2010)
Arctic Oscillation (AO)	AO positive phase – late spring to early summer temperature positive anomalies	Annual glacial streamflow signal positively correlated with AO; nival streamflow is not. Positive phase AO events: positive glacial streamflow anomalies in late spring and early summer; earlier freshet in nival rivers	Eight nival and glacial watersheds in southwest Yukon and northwest British Columbia	Fleming et al. (2006)
The North
PDO	“Warm” or “positive” phase: increase in winter temperatures; snowmelt earlier in spring “Cold” or “negative” PDO phase: lower winter temperatures and more snowfall	Annual max. flood and spring freshet date earlier; increase in winter flow and decrease in summer flowLower and earlier minimum flows	Laird River (12 streamflow gauging stations); Jean-Marie River	Burn et al. (2004)
PDO	PDO “cool” or “negative” phase	Higher annual discharge	Rivers discharging to Arctic and Hudson Bay	Déry and Wood (2005)
PDO, NAO, North Pacific (NP) index, Atlantic Multidecadal Oscillation (AMO), ENSO, AO	Positive PDO, NAO and negative NP increase in temp. Jan. to March	No significant relationship ENSO and AO; earlier spring freshet	Liard, Peace and Athabasca River Basins of MacKenzie River; 26 streamflow gauging stations	Burn (2008)
Eastern Canada
NAO, AO	High NAO/AO – reduced temperatures and precipitation	Reduced streamflow		Kingston et al. (2006)
AMO, NAO, NINO3.4, AO, PDO, SOI	Increase in summer and fall precipitation on South Shore	Annual maximum flow: north shore rivers negatively correlated AMO; south shore rivers (north 47°N); negatively correlated to SOI, positively to NINO3.4; south shore rivers (south 47°N) positively correlated to AO 7-day low flows; increase winter and summer on South Shore; decrease on North Shore	15 tributaries of the St. Lawrence River 18 tributaries of St. Lawrence River	Assani et al. (2010)Assani et al. (2012)
NAO, AMO	Positive phase NAO: storm tracks more coastal influence; temperature and precipitation above average; early snowmelt	Timing of spring floods positive correlation with NAO – early spring floods; negative correlation with AMO	17 “natural” rivers in Quebec	Mazouz et al. (2012)

Axelson et al. (2009) extended trend analysis of naturalized streamflow data for the Oldman and South Saskatchewan rivers using tree ring reconstructions. The gauge record showed a significant decreasing trend in the annual flow attributed to water storage, diversion and consumption. Spectral analysis of the proxy hydrometric data revealed a significant multi-decadal (~65 years) and inter-annual (2–6 years) variability related to ENSO. Trends show the influence of human activity and biophysical changes. A Quebec study suggested that regional trends were influenced by changes in land cover, particularly south of the St. Lawrence River where agricultural surface area has increased (Assani et al. 2012). In southern Alberta, human impacts such as water withdrawals had an equal if not greater influence as hydro-climatic drivers on trends in water availability (St. Jacques et al. 2010). Reservoir operations for hydroelectricity generation in the Churchill River Basin influenced trends in seasonality of flow (Déry et al. 2011). Reduction in forest cover, due to a mountain pine beetle infestation, expanded forest harvesting and fires, has lessened water demand from vegetation and influenced runoff in the Fraser Basin (Déry et al. 2012). Changes in volume of glacier melt may be affecting summer flow in the Columbia and Peace Basins in British Columbia (Fleming and Weber 2012).

In the Arctic, where monitoring networks are sparse, a different approach to trend analysis was carried out for the Mackenzie Basin utilizing a hydrologic model (WATFLOOD) to calculate a runoff time series based on two sets of observed gridded climate data as input (Yip et al. 2012). An overall decrease in annual runoff for 1961–2002 was calculated using the Environment Canada (EC) gridded data set, while an increase was calculated using the European Centre for Medium Range Weather Forecasting (ECMWF) reanalysis climate data set (ERA-40). The EC data set contained generally larger increasing trends in monthly air temperatures and smaller increasing trends (or larger decreasing trends) in monthly precipitation than the ECMWF ERA-40 data set did. The ECMWF data set showed larger increasing trends in monthly ET and soil moisture storage. This type of uncertainty could influence scenario-based assessments of future climate change as well, which will be discussed further in Part III (Cohen et al. 2015).

Climate-based indicators

A large number of climate indicators can be used to inform assessment of water availability, but for this review the focus is primarily on detection of trends in drought – a lack of water – with some attention to evapotranspiration, soil moisture and precipitation (number of days without rain), particularly where these indicators provide insights on this climatological event. Assessments of historical climate trends have used climate station point data, gridded observed data (national or global), gridded model output, and paleo-climate reconstructions for long-term pre-instrumental conditions. The previous section on hydrologic trend analysis highlights issues related to data and analysis, which are also relevant to the analysis of climate-related trends – length and period of record used in the analysis and confounding effects of other human-caused effects such as land use and land-cover change and statistical techniques, for example. A summary of selected scientific papers exploring historical trends in drought is found in Table 3.

Table 3. Selected historic drought trends using instrumental and paleo-climate records.

Location	Annual Palmer Drought Severity Index (PDSI)	Annual Standardized Precipitation Index (SPI)	Source
Southern AB and SK, not MB	Fewer extended droughts during 1901–2005 (instrumental) than 1365–1900 (paleo record); 2.4 vs. 2.8 seasons	Fewer extended droughts during 1901–2005 (instrumental) than 1365–1900 (paleo record); 2.0 vs. 3.2 seasons	Bonsal et al. (2013)
Southern AB and SK (eastern half of Prairie Pothole Region)	Variable but mostly positive PDSI (1906–2000)		Millett et al. (2009)
Southern AB and SK (western half of Prairie Pothole Region)	Positive and negative PDSI (1906–2000)		Millett et al. (2009)
Mid-latitude Canada	Widespread drying (1950–2008)		Dai (2011)
Eastern North America	Little change in drought area (1948–2008)		Sheffield et al. (2012)

Drought

Global-scale assessments of drought have highlighted trends for specific regions, including the Canadian Prairies. Dai (2011) found that the Palmer Drought Severity Index (PDSI) showed widespread drying during the period 1950–2008 over much of Africa, Asia, southern Europe, eastern Australia, mid-latitude Canada and southeastern Brazil, whereas much of the continental United States, Argentina and, western Australia became wetter during the same period. The PDSI drying patterns were also qualitatively consistent with observed streamflow decreases in these regions over the same period. However, PDSI calculations by Sheffield et al. (2012) found little change in global drought area over the past 60 years (1948–2008). They suggest that previous calculations of global drought increase may be overestimated. They argue that the Dai (2011, 2012) assessments used PDSI models with simplified potential evapotranspiration (PE) calculations which did not take into account changes such as decreasing wind speed, global dimming and decreasing vapour pressure (which suggests relative humidity has been increasing). This controversy illustrates the challenges for detecting trends in a complex climatological event such as drought (e.g. meteorological or hydrological) using derived indicators.

Streamflow is an additional indicator that can be used for hydrologic drought assessment. Sharma and Panu (2008, 2012) modified an annual hydrologic drought assessment methodology for monthly drought analyses. Monthly streamflow sequences from 18–22 rivers in northwestern Ontario and eastern Canada were standardized, and patterns in monthly streamflow distributions allowed for the prediction of drought severity (duration and intensity). Using a multi-method assessment of drought and comparison of the results would provide insights into uncertainty, and more robust assessments. Implications of using different drought indices for assessments of future impacts are presented in Part III (Cohen et al. 2015).

The importance of period of record and data source for drought trend results is illustrated by Bonsal et al. (2013) who used PDSI and Standardized Precipitation Index (SPI) to assess the variability of summer drought duration and intensity in the Prairies. They concluded that for the southern Prairies, the 105-year instrumental record (1901–2005) identified fewer extended droughts than what occurred in the paleo-climate period (1365–1900). This suggests that twentieth-century extended drought conditions were relatively benign when compared to conditions in previous centuries.

Drought trends vary spatially and temporally, and results for stations in the same region can be significantly different. For example, historical trends across 18 Canadian and US stations in the Prairie pothole region, from 1906–2000, indicated variable but mostly wetter conditions (positive PDSI) during the twentieth century in the eastern half of the region (Milllett et al. 2009). No stations showed a significant negative PDSI (drier conditions) trend. The western half of the region exhibited a mix of trends, with two stations having significantly wetter conditions (Medicine Hat AB and Muenster SK) and two stations having significantly drier conditions (Ranfurly AB and Saskatoon SK). Over the period, the east–west moisture gradient increased: the west became drier and the east wetter. Decadal patterns showed considerable variability among the 18 stations. For example, Medicine Hat, Alberta, had near-normal PDSI values during the dry 1930s but consistently lower PDSI values during the 1920s and 1980s. Similar results were found at other Canadian stations, indicating the northern portion of the Prairie pothole region was more affected by the 1980s drought than the 1930s drought.

A key question is whether evaporation and evapotranspiration are changing and whether they are responding to observed increases in air temperature. Fernandes et al. (2007) assessed Canada-wide trends in actual evapotranspiration (AET) using data from 101 hourly stations in a physically based land surface model (Ecological Assimilation of Land and Climate Observations [EALCO]). Annual AET increased up to 0.73% per year at 81 out of 101 locations, and decreased by 0.25% per year at the remaining 20 locations for the period 1960–2000. Statistically significant increasing AET trends were detected in 35% of the locations, mainly in the Pacific and Atlantic coasts and Great Lakes/St. Lawrence zones. However, in sharp contrast to findings using simpler AET models, here and other studies (Hydrological Atlas of Canada 1978; Gan 1998), annual AET trends in the prairies climate zone were mixed, with increases and decreases and no locations showing statistically significant trends.

Burn and Hesch (2006) compared trends for observed pan evaporation and estimated potential evaporation (PE) for 11 sites on the Prairie Provinces for the months of May to September for different periods from the 1960s to early 2000s. The pan and potential evaporation exhibited both significant decreasing trends and increasing trends, but overall more significant decreasing trends. Only seven out of 53 cases show matching significant trends between pan and potential evapotranspiration for the same stations. The differences in trends between pan and potential evapotranspiration may be influenced by trends in wind speed and vapour pressure deficit on both variables. Wind speed exerted an influence on PE but not on pan evaporation. PE also showed stronger correlations with wind speed than with pan evaporation. The significant decreasing trends were concentrated during the months of June and July for both pan and potential evapotranspiration.

Trends in precipitation and maximum temperature can provide insight on conditions related to meteorological drought. Mekis and Vincent (2008) found that the trends in a range of indicators related to daily temperature and precipitation were more pronounced for the period 1900–2007 than for the 1950–2007 period. For example, the maximum number of consecutive dry days decreased in all locations by 10.7% for the 1900–2007 period, and decreased by only 5.3% for the 1950–2007 period (p = 0.05). Mekis and Vincent (2011) have updated annual and seasonal trends for various precipitation indices. Trends for southern Canada (south of 60°N) were analyzed for the period 1900–2009, and trends for all of Canada were analyzed from 1950–2009. Annual rainfall in Canada increased by 8.7% from 1900–2009, and 12.5% from 1950–2009, respectively. Annual snowfall increased by 6.8% from 1900–2009, and 4% from 1950–2009, respectively. However, the changes in snowfall are neither temporally nor spatially consistent. Annual snowfall showed a steady increase until the 1970s for both time periods, followed by a considerable decrease until the 1980s, with no change up to 2009.

Soil moisture (SM) is an important variable for calculating water availability, especially in the Prairies, but few SM monitoring stations exist. The lack of an extensive monitoring network that generates long-duration data sets makes it difficult to develop a large-scale assessment of soil moisture for Canada. Wen et al. (2011) reconstructed 60 years (1950–2009) of daily gridded soil moisture values for three soil layer depths (0–20 cm, 20–100 cm and 0–100 cm) for the Prairie Provinces using the Variable Infiltration Capacity (VIC) land surface hydrology model. Daily maximum and minimum air temperature and precipitation data for 1,167 climate stations were obtained from Environment Canada’s National Climate Archives. VIC was calibrated and validated with observed daily hydrograph data obtained for 12 watersheds from Environment Canada’s Water Survey of Canada. Wen et al. (2011) also used observed soil moisture anomalies from six sites in Alberta to validate the VIC model. The calibrated VIC model was used to calculate the Soil Moisture Anomaly Percentage Index (SMAPI) on daily and monthly time scales, and was used as a measure of agricultural drought severity. Wen et al. (2011) concluded that the VIC-reconstructed 60-year average of soil moisture accurately portrayed the climatology of the Prairie Provinces (e.g. Palliser Triangle region), and it was used to quantify and document the 2002 severe drought. No historical trend information or findings were included in this study, but the model could be used to reconstruct and update spatial and temporal distribution of soil moisture.

A way forward?

This review, focusing primarily on Canadian literature, demonstrates that there has been considerable research on trends in water availability, particularly using streamflow as an indicator. Yet, gaps and issues remain. These will be explored to provide insights on topics that can move the research agenda forward.

The analysis of trends in water availability does not provide a pan-Canadian perspective. Analysis using streamflow as an indicator has been concentrated in BC, the Prairies, the North and Quebec. Drought analysis using climate-based indicators is even more limited and has focused almost exclusively on the Prairies. Yet other regions of Canada have vulnerability to drought and may become more exposed under a changing climate. Assessment of the evolving nature of drought would be useful for these regions as well. For example, historical (and future) water budget information and drought indices such as PDSI, SPI (and SPEI) can be calculated for climate stations across Canada as long as there is a complete monthly (or weekly) data set of temperature and precipitation data. Limitations in data affect streamflow trend assessment and are, in part, the reason for the restricted spatial coverage of analysis. Analysis of evapotranspiration and soil moisture trends is constrained by challenges of measurement, modelling and few available data sets. Similarly, drought assessment also demonstrates the influence of spatial complexity, data source (instrumental or paleo-climate), period of record, and the drought index chosen and the treatment of climate components in the index.

Land use and land cover change, and water management activities that modify water distribution and availability within natural or regulated systems and/or extreme events that alter river courses, and possibly affect stage/discharge curves, lead to questions about the integrity of streamflow observations, and affect confidence in trend analysis. Some researchers (e.g. Burn et al. 2008; Khaliq et al. 2009; Jones 2011) have recommended assessment and pre-treatment of hydrologic data time series using accepted standards before trend analysis (see St. Jacques et al. 2010). Similar concerns related to non-climatic influences on climate data (e.g. station automation, site relocation, etc.) have led to homogeneity testing and adjustment of temperature and precipitation time series in order to detect and correct artificial shifts and create homogenized climate data (Beaulieu et al. 2008, 2010; Vincent et al. 2012). Testing for homogeneity and homogenization of hydro-climate time series data should be key methodological steps prior to testing for trends to remove potential bias and increase robustness of the analysis. The most common statistical technique employed in trend detection studies in Canada is the M–K (Monk et al. 2011). Some researchers have suggested that the utilization of multi-statistical methods for testing significance of trends may result in more extensive assessment and robust conclusions. There is a need to explore the uncertainty associated with and methodological challenges in statistical analysis in hydro-climate research, for example, related to autocorrelation, and assumptions about the data.

Trends in streamflow variability have been linked to large-scale atmospheric oscillations (e.g. PDO, AO, ENSO, NAO). New insights may be gained by combining the influences of these large-scale atmospheric circulation patterns on hydro-climatological variability with detection of a climate change signal (monotonic change). Little research has been undertaken in this area, but studies by Stewart et al. (2005) and St. Jacques et al. (2010) offer examples of approaches.

There are no “formal” detection and attribution studies for hydrology indicators in Canada; in fact, there are only a few in North America, and they focus on basins in the mountainous northwest US. Formal detection and attribution is a rigorous approach that links trends detected in observational time series with output from global climate modelling. These assessments are limited by a number of challenges, including: the characteristics of the indicators (e.g. runoff, ET and derived parameters such as PDSI), constraints in modelling selected indicators in climate models, and limited spatial and temporal coverage of observational data sets (Bindoff et al. 2013).

Different observing periods and spatial domains used in water availability trend studies challenge inter-comparison of results, contribute to the sometimes contradictory trends in the same region, and make it difficult to integrate findings in a robust pan-Canadian assessment. While consensus on methods – data assessment, length of record, indices and statistical tests – would allow for systematic basin-scale analysis of trends for Canada, it would constrain the capacity to select methods suitable to the particular context and research problem. However, assessments exploring these methodological issues across a number of basins would be warranted.

Most assessments of trends in hydro-climate are directed at exploring whether trends can be detected in the hydrologic regime using indices related to annual, seasonal, and monthly flow, as well as extremes such as flood or low flow. The intent is to determine whether there is a detectable signal of climate change in the time series, and to describe that change. An emerging need is to inform decision making (particularly in an adaptive management context for water and ecosystem management) on the evolving environment. This could be achieved by developing new or assessing existing indices relevant to a particular decision-making context, such as in Monk et al. (2011).

Acknowledgements

We would like to thank Barrie Bonsal, Donald Burn, Stephen Déry, Yonas Dibike, Sean Fleming, D.K. Kang, Wendy Monk, Terry Prowse, René Roy, Blaise Gauvin St-Denis, Jeanine St. Jacques, and Elaine Wheaton for comments on earlier drafts of an unpublished report presented at the 2013 Canadian Meteorological and Oceanographic Society–Canadian Water Resources Association–Canadian Geophysical Union (CMOS–CWRA–CGU) Congress in Saskatoon, which led to this series of papers. We would also like to thank Jenna Disch for assisting with tables and the reference list, and Kit Szeto, editor Diana Allen and the anonymous reviewers for comments on this series of papers. The opinions expressed here are those of the authors, and not necessarily those of Environment Canada.