A century of hydrological variability and trends in the Fraser River Basin

This study examines the 1911–2010 variability and trends in annual streamflow at 139 sites across the Fraser River Basin (FRB) of British Columbia (BC), Canada. The Fraser River is the largest Canadian waterway flowing to the Pacific Ocean and is one of the world’s greatest salmon rivers. Our analyses reveal high runoff rates and low interannual variability in alpine and coastal rivers, and low runoff rates and high interannual variability in most streams in BC’s interior. The interannual variability in streamflow is also low in rivers such as the Adams, Chilko, Quesnel and Stuart where the principal salmon runs of the Fraser River occur. A trend analysis shows a spatially coherent signal with increasing interannual variability in streamflow across the FRB in recent decades, most notably in spring and summer. The upward trend in the coefficient of variation in annual runoff coincides with a period of near-normal annual runoff for the Fraser River at Hope. The interannual variability in streamflow is greater in regulated rather than natural systems; however, it is unclear whether it is predominantly flow regulation that leads to these observed differences. Environmental changes such as rising air temperatures, more frequent polarity changes in large-scale climate teleconnections such as El Niño-Southern Oscillation and Pacific Decadal Oscillation, and retreating glaciers may be contributing to the greater range in annual runoff fluctuations across the FRB. This has implications for ecological processes throughout the basin, for example affecting migrating and spawning salmon, a keystone species vital to First Nations communities as well as to commercial and recreational fisheries. To exemplify this linkage between variable flows and biological responses, the unusual FRB runoff anomalies observed in 2010 are discussed in the context of that year’s sockeye salmon run. As the climate continues to warm, greater variability in annual streamflow, and hence in hydrological extremes, may influence ecological processes and human usage throughout the FRB in the 21st century.


Introduction
The Fraser River Basin (FRB) drains 240 000 km 2 or one quarter of British Columbia (BC), Canada, forming one Despite the watershed's vastness, immense diversity and its population of ≈2.7 million people (Fraser Basin Council 2009), few studies have examined climate change impacts on streamflow variability and trends across the FRB. Some studies on the FRB report increasing trends in observed annual flows and the occurrence of earlier peak flows in the latter half of the 20th century (Morrison et al 2002, Ferrari et al 2007. Observed mean summer water temperatures of the lower Fraser River have warmed by ∼1.5 • C since the 1950s (Patterson et al 2007). With continued climate change, global climate model (GCM) simulations project that the Fraser River may exhibit more of a pluvial regime rather than a nival one owing to warmer winter air temperatures and phase changes in precipitation (Morrison et al 2002, Kerkhoven andGan 2011) and incur a rise of 1.4 • C in August water temperatures by 2100 (Ferrari et al 2007). Changes in the amount and timing of streamflows will also influence sediment and chemical (e.g., nutrients, carbon, and contaminants) fluxes in the FRB, which will have implications for water resources (e.g., water quality, channel and estuarine maintenance dredging) and aquatic habitats.
The success of migrating and spawning salmon populations relies heavily on water temperatures and flow levels. Fraser River sockeye experience high mortality rates during years of above normal river temperatures and discharge (Rand and Hinch 1998). Energy demands increase when water levels are high, especially in the lower Fraser River, leading to exhaustion and eventually death (Macdonald 2000). In addition, with some of the fall runs now occurring 3-6 weeks earlier (Cooke et al 2004), the migrating salmon are exposed to water temperatures that are ∼3 • C higher than normal (Patterson et al 2007), influencing their swimming performance, metabolic rates and overall survival (Lee et al 2003). Thus as climate continues to change and as anthropogenic developments such as impoundments and diversions progress, there is an urgent need to better understand the variability in river conditions as this is crucial in determining the implications for ecological processes, such as the success of salmon runs, and human values associated with them (Knowler et al 2003, Evendon 2004. To our knowledge, this is the first comprehensive study that examines the interannual variability and trends in observed streamflow at 139 locations across the FRB using hydrometric data spanning a century . Specifically, we investigate the hypothesis that environmental change (whether natural or anthropogenic) may be leading to altered hydrological variability or extremes as has been observed in other regions of Canada (e.g., Déry et al 2009). Furthermore, we investigate whether there are links between either topography or anthropogenic developments with observed trends and fluctuations in annual streamflow across the basin. Although establishing the full range of impacts of changing flow regimes on ecological functions in the FRB is beyond the scope of this letter, we suggest linkages between variation in flow conditions and salmon returns. We use this example to highlight the relevance of streamflow variability on this keystone species within the FRB as the annual stock success in turn plays a vital role in both ecosystem health and human activities in the basin.

Study area
The Fraser River is the largest Canadian river by drainage area flowing into the Pacific Ocean. It extends from 49 • N to 56 • N and 118 • W to 125 • W and encompasses 13 major sub-watersheds (figure 1). The basin has a mean elevation of ∼1320 m with the main stem of the Fraser River and its vast network of tributaries situated between BC's Coast Mountain and Rocky Mountain ranges. The Fraser River headwaters are in the Rocky Mountains near Jasper at the Alberta/BC border, from where it flows northwest through the Rocky Mountain Trench toward Prince George. There it turns sharply south where the Nechako River joins it and flows through the Fraser Plateau, collecting discharge from the Chilcotin River (figure 1). It passes through the Fraser Canyon west of the semi-arid interior plains before flowing westward at Hope and into the Strait of Georgia and Salish Sea near Vancouver, BC. The main stem of the Fraser River spans a distance of 1375 km and the watershed experiences a 3950 m drop in elevation from its highest point in the Rocky Mountains to its outlet at sea level (Benke and Cushing 2005). Although the proportion of land-use in the Fraser basin is only 2.2% urban and 0.6% agricultural (Benke and Cushing 2005), the area supports 53% of farmland in the province of BC (Fraser Basin Council 2009). These land-use values, along with the river's designation as a Canadian Heritage River in 1998, and the relationship between First Nations communities and local salmon runs across eight First Nations language groups, all highlight the range of human demands and values associated with the Fraser River (Evendon 2004, Fraser Basin Council 2006. Mean annual air temperatures across the basin range between 0.5 • C in the northwest by the Skeena Mountains and 7.5 • C in the south near the Okanagan region. Summer and winter mean air temperatures in the FRB range from 11 • C to 16.5 • C and −11 • C to −1 • C, respectively. Relatively dry conditions exist in the interior plateau, in the rain shadow of the Coast Mountains, where precipitation averages 400-800 mm yr −1 (Benke and Cushing 2005). In contrast, coastal and mountainous sections may experience precipitation in excess of 3000 mm yr −1 . Snowfall forms an important component of the annual precipitation, especially in the basin's northern and mountainous sections. For instance, the areally averaged 1 April snow water equivalent near Prince George (which is located on the northern-most section of the main stem) is 409 mm yr −1 for 1966-89 (Danard and Murty 1994). As a result of its highly varying terrain and climate, the FRB has an extremely diverse vegetative cover consisting of alpine, subalpine, montane and coastal forests as well as mixed drier forests and grasslands in the interior plateau (Benke and Cushing 2005). Glaciers cover 1.4% of the FRB and augment the summer flows of alpine rivers and headwater streams through glacial melt (Stahl and Moore 2006).
Most of the FRB is minimally impounded, with the Nechako River being the only major system that is regulated (figure 1). Flow regulation began in the early 1950s when the Kenney Dam was constructed (Hartman 1996), resulting in some of the impounded water being diverted to the coastal Kemano River watershed and the remaining water stored in a reservoir spanning 922 km 2 and containing a volume of 32.7 km 3 (Schiefer and Klinkenberg 2004). The area upstream of the Kenney Dam covers 15 600 km 2 or 7.2% of the FRB area as gauged at Hope, BC (see figure 1).

Data and methods
Daily streamflow data at 148 FRB gauges (nine gauges on the main stem and 139 on its tributaries) for the period between 1911 and 2010 were extracted from Environment Canada's Hydrometric Database (HYDAT) (Water Survey of Canada 2011; see figure 1 and supplementary data available at stacks.iop.org/ERL/7/024019/mmedia). The criteria used in selecting the hydrometric stations include: (a) a location within the FRB, (b) an area upstream of the gauge ≥ 100 km 2 , and (c) data availability for ≥10 yr. These criteria were chosen to exclude the many small (< 100 km 2 ) basins and/or short (n < 10 yr) observational records available in the FRB. Data availability varies markedly across sites and improves in the latter half of the study period (see supplementary data available at stacks.iop.org/ERL/7/024019/mmedia). The database includes hydrometric data for the Ootsa, Tahtsa, Tetachuck and Whitesail rivers in the Nechako River Basin for years prior to 1953 when the Kenney Dam was built. Although these four river systems were largely flooded as a result of the filling of the Nechako reservoir in the early 1950s, they provide valuable information on the natural hydrology in the upper reaches of the Nechako watershed. Since 1954, water diverted from the Nechako reservoir enhances the flow of the Kemano River that discharges directly to the Pacific Ocean, outside the FRB domain. Data from a hydrometric gauge at the Kemano Powerhouse (ID: 08FE002) were thus used to estimate the inter-basin transfer of water from 1954 to 2010. Apart from the site identification number, coordinates and upstream gauged area, information on flow regulation, mean basin elevation, and glacierized area were also extracted from HYDAT when available.
Time series of annual streamflow were then constructed based on the daily discharge time series. In addition, time series of the mean and coefficient of variation (CV) for 11 yr moving windows of annual streamflow were compiled. Gaps in the hydrometric time series were in-filled with the daily mean flow over the available period of record at each site. An average of 6.5% of the daily streamflow data at each site was in-filled in this manner (see supplementary data available at stacks.iop.org/ERL/7/024019/mmedia). Déry et al (2011) found this gap filling strategy attenuated streamflow trend magnitudes, especially when missing data were consecutive and occurred at the beginning or end of the time series. This study minimized such issues by using an inhomogeneous study period in some of the analyses thus maximizing data availability and spatial coverage. For nine sub-basins in the dataset, the hydrometric gauges were moved to a nearby location from their original site, leading to slight (<10%) changes in upstream gauged area. In these circumstances, the streamflow time series were combined for the 'paired' records by adjusting the hydrometric data for the missing contributing area at the gauge with the smallest basin extent (e.g., Déry et al 2005, Wei andZhang 2010). We thus assumed that the missing contributing area exhibited similar runoff rates to the gauged portion of the basin. Thus splicing of the hydrometric time series at the nine sites where gauges were moved resulted in a total of 139 time series of annual streamflow across the FRB. Homogeneity tests and analyses (such as in Tomé and Miranda 2004) were conducted on the basins affected by gauge relocations but few, if any, were found to be significantly impacted by these.
Discharge data were first divided by their respective basin areas to obtain time series and statistics of areal river runoff R (in mm yr −1 ) at each site of interest. This included the mean and CV in annual streamflow that were calculated for all sites of interest for their respective periods of data availability over calendar years 1911-2010 (see supplementary data available at stacks.iop.org/ERL/7/024019/mmedia). The CV was used in this study to quantify the interannual variability in streamflow as it facilitated comparisons between sites across the FRB where the variance in annual runoff varies by several orders of magnitude. Values of R and CV were then correlated with the available geographic information (latitude, longitude, gauged area, mean basin elevation and glacier coverage) and with record length. Correlation values were considered statistically significant when p < 0.05.
The Mann-Kendall test (Mann 1945, Kendall 1975) was used to assess monotonic trends in the time series of CV for 11 yr moving windows of annual streamflow where data were available over four time periods (median years 1920-2005, 1940-2005, 1960-2005 and 1980-2005). The 11 yr moving windows were generated from the time series of annual streamflow and the number of sites included in the analyses increased for the latter half of the study period when data availability improved (see supplementary data available at stacks.iop.org/ERL/7/024019/mmedia). For instance, the trend analysis for median years 1920-2005 includes only sites with available streamflow data spanning 1915 (first year of the first moving window) to 2010 (last year of the last moving window). An additional analysis was performed to explore trends in seasonal values of the CV in 11 yr moving windows of streamflow for 1980-2005. Here we considered winter to include the months of January, February and March, spring to span April, May and June, summer to include July, August and September, and autumn to span October, November and December.
We qualified trends as 'detectable' when their signalto-noise ratios |snr| > 1, where snr equaled the absolute trend slope divided by the standard deviation over a given period of interest (Déry et al 2011). Prior to the trend analyses, time series of CV for 11 yr moving windows of annual streamflow were pre-whitened to remove the effects of autocorrelation on the results (Yue et al 2002). This provided a measure of changing hydrological variability, or extremes, on a decadal time scale across the FRB (Déry et al 2009). Note that detectable trends in time series of CV for 11 yr moving windows of annual streamflow were not uniquely attributable as they depended on the evolution of its statistical components, the corresponding means and standard deviations. This study focused on trends in the CV of streamflow as we were more interested in detecting changes in hydrological variability over time. Field significance (at a 10% level) of trend results was established using the bootstrapping methodology of Burn and Hag Elnur (2002).
Here we randomly resampled each of the time series of the CV in 11 yr moving windows 1000 times and recorded the distribution of significant trends. Finally, we explored the relationship between the mean and CV in 11 yr moving windows of annual runoff over median years 1917-2005 for the Fraser River at Hope (ID: 08MF005). This gauge encompasses 93% of the total FRB area and has one of the longest, continuous records of the basin's streamflow.

Streamflow variability and trends across the FRB
The mean areal annual runoff (R) varies substantially across the FRB, with low values (R < 200 mm yr −1 ) in small streams in the central portion of the basin and in rivers to the lee of the Coast Mountains on the Chilcotin and Nechako plateaus ( figure 2(a)). High values (R > 1200 mm yr −1 ) occur in alpine streams that are glacier-fed and in the coastal watersheds near the mouth of the Fraser River where abundant rainfall augments runoff generated by snow and glacier melt at high elevations (see supplementary data available at stacks. iop.org/ERL/7/024019/mmedia). Interannual variability in streamflow, expressed by the CV, peaks in small streams in the central portion of the basin and in rivers to the lee of the Coast Mountains and is attenuated in alpine and larger watersheds. Although high interannual variability is observed on the regulated Nechako River, some of its naturally flowing tributaries (the Stellako and Nautley rivers) show similar patterns in CV values. A comparison of figures 2(a) and (b) reveals a clear negative relationship between R and CV values. In fact, the correlation coefficient between ln(R) and CV in annual streamflow for the 139 sites is −0.77 (p < 0.001). Values of R correlate negatively with latitude and positively with glacier coverage, whereas values of the CV in annual streamflow correlate negatively with gauged area, mean basin elevation, and glacier coverage (table 1). An additional analysis reveals the absence of a statistically significant relationship between the CV in annual streamflow and length of the observational records. Thus we find no evidence that record length suppresses the statistics of interannual variability in streamflow in the FRB.
Detectable trends for the CV in annual streamflow show remarkable spatial consistency over different time periods (figure 3). The paucity of hydrometric data for the first half The trend analysis at the seasonal level reveals more complex spatial patterns in the recent variability in streamflow across the FRB (figure 4). Over 1980-2005, there is a general tendency toward less variability in autumn and winter whereas there is a majority of positive trends in the CV of spring and summer streamflow. In autumn, streamflow variability increased across the North and South Thompson River watersheds while it decreased in the Chilcotin and Lower Fraser and Estuary watersheds. Annual trends mainly reflect those observed in spring and summer when streamflow rates peak in the FRB. All seasonal trend analyses exhibit field significance at the 10% level.
There is no significant relationship between the mean and CV in 11 yr moving windows of annual runoff for the Fraser River at Hope (figure 5). The past two decades show near-normal runoff values, but relatively high interannual variability.

Impact of anthropogenic developments on streamflow variability and trends across the FRB
Over 1954-2010, an average of 3.2 km 3 yr −1 was diverted from the Nechako watershed to the Kemano River to generate and supply hydroelectricity for an aluminum smelter in Kitimat, BC. This diversion leads to a 4.2% reduction (equivalent to a decline of 14.7 mm yr −1 in the annual runoff) on the Fraser River at Hope over that period. After the construction of Kenney Dam, the CV in annual streamflow for the Fraser River at Hope over two periods of equal duration rises slightly from 0.124 (1912-53) to 0.129 (1954-95). In the absence of naturalized discharge data for the Nechako River, however, it is unknown how flow regulation and water retention in this watershed affects interannual streamflow variability at downstream locations, such as on the main stem of the Fraser River. Apart from the Nechako River, several smaller tributaries of the Fraser River, especially in the North Thompson, South Thompson, Thompson-Nicola and Bridge-Seton subwatersheds, are regulated. The CV in annual streamflow for these systems averages 0.36, considerably higher than in naturally flowing rivers of the FRB (table 2). Limiting the comparison to watersheds < 10 000 km 2 and to the closest nearby gauge with natural flows yields mean CV values in annual streamflow of 0.40 and 0.26 for regulated and natural systems, respectively. Nonetheless, it is important to note that  (see table 2 and supplementary data available at stacks. iop.org/ERL/7/024019/mmedia). Thus flow regulation within the FRB does not necessarily lead to suppressed interannual variability in streamflow.

Concluding discussion
This study provides new information on streamflow variability and trends across the FRB that expands on the work of Foreman et al (2001) and others. Consistent with our results, Foreman et al (2001) reported a tendency toward Table 2. The mean and standard deviation (SD) of annual runoff (R) and the coefficient of variation in annual streamflow (CV) for natural and regulated rivers. Note that n provides the number of sites used in the analyses.

Rivers
Mean greater fluctuations in annual streamflow on the Fraser River main stem toward the end of the 20th century. Our findings expand on this study to show that small streams in BC's interior plateau contribute much less runoff than those in the higher elevations and along the BC coast. Interannual variability in streamflow varies markedly across the FRB, with low values in large and alpine watersheds and with high variability in small, low elevation streams and in the rivers to the lee of the Coast Mountains. Streamflow variability remains low (CV < 0.2) in mountainous watersheds where glacier melt moderates year-to-year fluctuations. Large basins (gauged area ≥ 50 000 km 2 ) and rivers downstream from major lakes (e.g., Francois, Stuart and Quesnel lakes), which attenuate hydrologic signals, also exhibit low interannual variability in streamflow. There is also a noticeable propensity for high interannual variability in streamflow to occur in streams and rivers with low runoff productivity and vice versa. Furthermore, the trend analyses suggest a tendency toward greater interannual variability in streamflow (i.e. greater hydrological extremes) across most of the FRB over recent decades, most notably during spring and summer. Air temperatures have warmed by ≈1 • C from 1895 to 1995 in BC, with greater increases during winter and spring in interior and northern regions of the province (BC Ministry of Water Land and Air Protection 2002). Furthermore, annual precipitation has increased by 2-4% per decade from 1929 to 1995 over the southern half of BC. These trends toward a warmer, wetter climatic regime but with less snowfall have imposed modifications on the surface hydrology of the FRB including greater interannual variability in river runoff as reported in the present study. Furthermore, more temperate wintertime conditions have led to the proliferation of the mountain pine beetle (Dendroctonus ponderosae) with mature pine tree mortality rates reaching 80% (e.g., Picketts et al 2012). Deforestation from this pest outbreak along with industrial wood harvesting reduces the interception of snow by the tree canopy, favors greater snowpack accumulation at the surface, and enhances runoff productivity through less water demand from vegetation (Boon 2012). Thus deforestation yields higher runoff during wet years than would be expected for pristine environments. Glacier melt during warm, dry years moderates runoff fluctuations from headwater catchments, offsetting low precipitation amounts (Fountain andTangborn 1985, Fleming andClarke 2005). Glaciers in BC are currently retreating and downwasting at an accelerating rate (Schiefer et al 2007), suggesting that many of the alpine watersheds may experience declining summer flows and hence greater interannual variability in streamflow (Stahl and Moore 2006). Large-scale teleconnection patterns such as El Niño-Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) strongly modulate the climate of the FRB (e.g., Thorne and Woo 2011). Polarity changes in ENSO and the PDO are projected to occur more frequently and with greater intensity in the 21st century (e.g., Müller andRoeckner 2008, Lapp et al 2012), yielding an increasing threat of hydrological extremes across the FRB, especially in glacierized watersheds. Rapid shifts between El Niño and La Niña conditions in the tropical Pacific Ocean over recent years are consistent with these model projections and with the amplifying hydrological variability across the FRB reported in this study. This will likely impose greater challenges for water resource managers as well as for humans, fauna and flora of the FRB. In addition to implications for water resources, one of the greatest on-going concerns of FRB streamflow is its role on migrating salmon and the likely effects of changing conditions on salmon stock sustainability.
The interannual variability in streamflow remains low in rivers such as the Adams (CV = 0.14), Chilko (CV = 0.13), Quesnel (CV = 0.14-0.17) and Stuart (CV = 0.22), where the principal salmon runs of the Fraser River occur (Eliason et al 2011). This low streamflow variability increases the likelihood of salmon surviving the migration back to their spawning grounds. Rand et al (2006) link the success of the early Stuart sockeye salmon population to river water levels, with years of higher than normal discharge (such as 1997) forcing salmon to exert greater amounts of energy during their up-river migration. In the lower Nechako River (downstream of the confluence with the Stuart River), flows are relatively shallow, especially since the construction of the Kenney Dam, affecting water temperatures and hence fish metabolism (Hartman 1996, Rand et al 2006. Thus in the FRB, low flows result in warmer water temperatures whereas high flows accompany strong currents, with both situations increasing energy demands for migrating salmon. Our trend analyses reveal a tendency toward greater interannual variability in streamflow and thus recent increases in the frequency and/or intensity of low/high flow conditions. Increased streamflow variability from environmental change may therefore hinder the survival rate of the Fraser River sockeye during their spawning migration. For instance, spring freshets will occur earlier and be of shorter duration as snowpacks decline (Cunderlink and Burn 2004). In late summer and autumn, discharge may be lower but remain higher during winter as more precipitation falls as rain, rather than snow (Healey 2011). Spawning conditions during late summer and autumn could therefore be compromised through warmer water temperatures and lower flows in the FRB coastal streams while in winter stronger freshets and storms would scour the spawning gravels before the fry are ready to emerge (Melack et al 1997).
The Fraser River experienced an unexpected record sockeye salmon run of >30 million up-river migrating fish in 2010, surpassing all other runs in the past century (Cone 2012). However, the excess abundance of 2010 was not homogeneous throughout the Fraser sub-basins. The majority of sockeye returned to the Adams and Chilko Rivers while other upstream returns were near or below expected values (Cone 2012). These events coincided with relatively low flows across northeastern sections of the FRB, most notably in the Upper Fraser and Quesnel watersheds, that were concomitant with an El Niño event that year (figure 6). In fact, the hydrometric gauges on the Fraser River at Marguerite (ID: 08MC018) and Quesnel River near Quesnel (ID: 08KH006), both upstream of the Chilko and Adams rivers, registered historical low values of annual runoff with departures reaching −1.7 and −2.0 standard deviations from the 1981-2010 mean values, respectively. Concurrently, in the upper reaches of the Nechako and Chilcotin watersheds where fish returns were abundant, positive anomalies in annual runoff were observed, illustrating the regional variability in streamflow generating processes and responses to large-scale climate oscillations (e.g., Fleming et al 2006, Thorne and Woo 2011). The relationship between these regional runoff anomalies and basin variations in the salmon returns should be explored further to determine if an historical pattern exists between these factors. Investigation of future effects is also of interest, for example to see whether the abundance of the 2010 sockeye salmon return will be expressed in the 2014 run when their offspring return and if flow variation influences their lifecycle. While the 2010 FRB flow regimes may have played a role in influencing the spatial patterns of returning spawners that year, fecundity, reproductive success and their offspring's success will be affected by hydrometric variability from spring 2011 through to spring 2012 when they are in the freshwater rearing stages.
Future work should explore relationships between streamflow conditions (e.g., water temperatures and levels, sediment and nutrient fluxes) and the factors that control these (e.g., precipitation and snowmelt) and the subsequent implications for water resources, aquatic ecosystems, salmon returns and human activities in the FRB.