Changes in the onset and intensity of wind-driven upwelling and downwelling along the North American Pacific coast

Authors

  • Brian Bylhouwer,

    Corresponding author
    • School of Resource and Environmental Management, Simon Fraser University, Burnaby, British Columbia, Canada
    Search for more papers by this author
  • Debby Ianson,

    1. School of Resource and Environmental Management, Simon Fraser University, Burnaby, British Columbia, Canada
    2. Department of Fisheries and Oceans, Institute of Ocean Sciences, Sidney, British Columbia, Canada
    Search for more papers by this author
  • Karen Kohfeld

    1. School of Resource and Environmental Management, Simon Fraser University, Burnaby, British Columbia, Canada
    Search for more papers by this author

Corresponding author: B. Bylhouwer, Stantec Consulting Ltd., Dartmouth, NS, B3A0A3, Canada. (bbylhouw@sfu.ca)

Abstract

[1] The timing, duration, and intensity of wind-driven upwelling and downwelling along the North American Pacific coast play an integral role in coastal circulation and basinwide ecosystem composition. It has been suggested that global warming will cause changes in these winds. Here we develop a new set of objective criteria to unambiguously determine the onset, duration, and intensity of upwelling and downwelling seasons due to local wind forcing. We use these criteria to examine and better characterize temporal trends in wind-driven coastal currents over the previous 60 years and relate them to global warming and large-scale climate oscillations in the coastal ocean between northern California and Vancouver Island (37°N and 51°N). We find an exceptionally variable onset of upwelling at all locations. Some significant temporal trends are found in summer onset and upwelling intensity time series near the Juan de Fuca Strait and off the coast of Oregon. Positive phases of the Pacific Decadal Oscillation are correlated to later and shorter upwelling seasons with weaker upwelling. Warm phases of the El Niño Southern Oscillation are associated with a later onset of summer upwelling south of Oregon and with more intense downwelling throughout the study area. Our analysis identifies strong interannual to interdecadal variability, and emphasizes the importance of time series length when isolating physical temporal trends influenced by large-scale oscillatory behavior of the climate.

1. Introduction

[2] The California Current System (CCS) from Vancouver Island to northern California has been well studied due to its high levels of primary productivity and its commercially viable fisheries such as anchovies, sardines, shrimp, mackerel, hake, sablefish, and salmon [Ware and Thomson, 1991; Thomson and Ware, 1996; Ware and Thomson, 2005; Mackas, 2006; Hickey and Banas, 2008; Chavez and Messie, 2009; Peterman and Dorner, 2012]. Productivity in the CCS is primarily driven by nutrients delivered by upwelling [Huyer, 1983]. In some regions, productivity may also be influenced by winter “preconditioning,” either through anomalously strong winter downwelling that reduces shelf nutrient inventories (and productivity during the following spring and summer) [Ianson and Allen, 2002; Tortell et al., 2012], or similarly from late winter upwelling events that can enhance nutrients and subsequent productivity [Schroeder et al., 2009; Black et al., 2011]. It is anticipated that the timing and intensity of upwelling and downwelling heavily influence the composition and vitality of the marine biota, and this influence has been demonstrated for the upwelling season [e.g., Ware and Thomson, 1991; Mackas et al., 2001; Botsford et al., 2006; Barth et al., 2007; Borstad et al., 2011].

[3] Both upwelling and downwelling processes in the CCS also influence the regional carbon cycle. Upwelling occurs primarily in the summer season when equatorward winds cause offshore advection of surface coastal water, resulting in the episodic shoaling of cold, carbon- and nutrient-rich, and oxygen-poor intermediate depth (∼100 m) water at the coast [Pond and Pickard, 1983; Smith, 1994]. CO2 outgassing can occur immediately following periods of upwelling, but the subsequent primary production turns surface waters into a carbon sink [Ianson and Allen, 2002; Hales et al., 2005]. Poleward winds dominate the northern CCS during winter, and cause downwelling, or a depression in the pycnocline at the coast, as surface waters are forced onshore [Smith, 1994]. This movement of the pycnocline pushes subsurface shelf waters, which are generally rich in carbon (and nutrients limiting to phytoplankton production) off the coastal shelf into the adjacent ocean basin [Hales et al., 2005]. The net annual carbon budget is influenced by the intensity of upwelling relative to downwelling, as well as the bathymetric characteristics of the coastal shelf [Ianson et al., 2009].

[4] Climate change could influence the intensity of wind-driven upwelling winds, and thereby have a substantial impact on future productivity and carbon cycling [Snyder et al., 2003; Auad et al., 2006; Merryfield et al., 2009; Bakun et al., 2010]. Global warming has been linked to enhanced, upwelling-favorable, alongshore winds for major global eastern boundary currents [Bakun, 1990]. Previous analyses have confirmed this relationship south of 40°N in the CCS, between 1945 and 2008 [Hsieh et al., 1995; Schwing and Mendelssohn, 1997; Garcia-Reyes and Largier, 2010]. A meta-analysis of General Circulation Model (GCM) projections as well as Regional Circulation Model (RCM) analyses have also shown increasing upwelling winds along the entire CCS from 1990 to 2100 [Merryfield et al., 2009].

[5] Winter downwelling intensity has also been shown to be generally increasing over the last 50 years in the northern limit of the CCS [Foreman et al., 2011]. Some models have predicted an increase in winter downwelling-favorable winds through the year 2100; however, an ensemble of model predictions does not show a statistically significant trend [Merryfield et al., 2009]. Although the influence of large-scale climate oscillations, such as the El-Niño Southern Oscillation (ENSO), the Pacific Decadal Oscillation (PDO), or the North Pacific Gyre Oscillation (NPGO), could account for temporal trends in historical analyses, particularly due to the brevity of available time series, these relationships have not been well studied [e.g., Garcia-Reyes and Largier, 2010; Chenillat et al., 2012].

[6] Climate-induced changes to the timing of the onset of upwelling and downwelling seasons have been studied less systematically than overall intensity, despite their known effect on marine ecosystems [e.g., Sydeman et al., 2006; Schwing et al., 2006; Kosro et al., 2006; Holt and Mantua, 2009; Borstad et al., 2011; Black et al. 2011]. For example, shifts in the onset of either winter or summer season could cause detrimental mismatches between nutrient delivery and subsequent productivity of pelagic fish, benthic invertebrates, and/or marine mammals or birds, independent of any change in upwelling or downwelling magnitude [e.g., Beaugrand et al., 2003; Edwards and Richardson., 2004; Sydeman et al., 2006; Bakun et al., 2010].

[7] A variety of indicators have been used to estimate the onset of summer upwelling, including surface chlorophyll [e.g., Henson and Thomas, 2007]. Physical indicators such as alongshore winds or sea surface pressures are generally available over longer time periods and have been used more extensively [e.g., Schwing et al., 2006; Bograd et al. 2009; Foreman et al. 2011, Garcia-Reyes and Largier, 2012; Iles et al., 2012]. An ocean current index (current shear) off Vancouver Island [Thomson and Ware, 1996] and changes in distributions of sea surface temperatures [e.g., Lynn et al., 2003; Holt and Mantua, 2009] have also been used. No metric has been favored [Kosro et al., 2006], perhaps because some methods are best suited to specific portions of the CCS but not others [Holt and Mantua, 2009]. Timing of the downwelling season has also been evaluated [Thomson and Ware, 1996, Bograd et al., 2009; Foreman et al., 2011] but with less attention. It has been suggested that the onset of summer upwelling has been occurring later over the last several decades in the central-northern CCS [Bograd et al., 2009] and the northern limit of the CCS [Foreman et al., 2011].

[8] The goal of this research is to evaluate temporal changes in the seasonal onset, duration, and intensity of upwelling and downwelling seasons in the CCS for a relatively long period: 1948–2010. Using time series data sets of wind speeds from reanalysis as well as lighthouse and buoy data, we first develop and test an alternative, simple, and objective criteria based directly on reanalysis wind data to demarcate the winter, summer, and two transition seasons. Using our established criteria, we examine temporal changes in the onset and duration of each season, over a large portion of the CCS, from the northern tip of Vancouver Island to northern California. Finally, we compare these data with well-established patterns of climate oscillatory behavior and contrast our method and results with previous studies.

2. Methods

2.1. Study Area and Data Sources

[9] Our study area encompasses the coastal region between northern Vancouver Island and San Francisco, approximately 37° to 51°N, out to the continental shelf break (Figure 1). We use daily wind data from the National Center for Environmental Prediction (NCEP) [Kalney et al., 1996] reanalysis data sets, provided by the National Oceanographic and Atmospheric Administration (NOAA) Office of Oceanic and Atmospheric Research Physical Sciences Branch. Three-hourly wind data were retrieved from North American Regional Reanalysis (NARR) [Mesinger et al., 2005], and hourly wind data were collected from Environment Canada (EC) and NOAA buoy and lighthouse stations (Figure 1 and Table 1). All data were standardized to 10 m height [Zoumakis, 1993] and vector averaged into daily values of wind speed and direction. Data were omitted if less than seven hourly wind measurements were recorded. Temporal coverage of observational data was poor even without this threshold (gaps as large as 1712 days; Table S1, supporting information), particularly with buoy stations. The number of individual gaps range from 17 to 175 per record, and the total number of days missing for an individual location range from 47 to 3860 (Table 1, supporting information). Observational data were therefore used only to ground-truth NARR and NCEP data for this study.

Figure 1.

Schematic of locations extracted from (a) North American Regional Reanalysis (NARR) (blue capital letters), (b) NCEP reanalysis (red Greek letters), and observational data (green roman numerals) used in this study. Exact locations are listed in Table 1. Asterisks are used to indicate the actual locations of observation locations IX–XIII and NCEP location ζ, to avoid overcrowding. The 200 m isobath is shown to roughly indicate the shelf break.

Table 1. Names and Coordinates of Data Sourcesa
StationLatitude (° N)Longitude (° W)WMO Buoy
  1. a

    NARR, North American Regional Reanalysis; NCEP, National Centres for Climate Prediction; WMO; World Meteorological Organization.

  2. b

    Lighthouse stations.

Observational Stations
XII49.73−127.92c46132
XI48.83−125.99c46206
X47.68−124.49desw1b
IX47.35−124.73u46041
VIII46.14−124.51u46029
VII44.64−124.5u46050
VI43.34−124.38caro3b
V41.85−124.38u46027
IV40.78−124.59u46022
III38.96−123.74ptacb
II38.24−123.3u46013
I37.76−122.83u46026
NARR Data
Z50.91−129.45 
Y50.5−128.83 
X50.09−128.21 
W49.68−127.59 
V49.27−126.96 
U48.86−126.34 
T48.45−125.72 
S48.08−125 
R47.5−125 
Q46.91−125 
P46.33−125 
O45.75−125 
N45.16−125 
M44.58−125 
L44−125 
K43.42−125 
J42.84−125 
I42.26−125 
H41.68−125 
G41.1−125 
F40.52−125 
E39.94−125 
D39.44−124.6 
C38.94−124.2 
B38.44−123.8 
A37.94−123.4 
NCEP Data
ζ48.57−125.63 
ε46.67−125.63 
δ44.76−125.63 
γ42.86−125.63 
β40.95−125.63 
α37.14−123.75 

[10] The NARR data set contains meteorological variables constructed by assimilating observations of temperature, wind speed, and pressure onto a system of grids using a three-dimensional forecasting model. These reanalysis data are produced by the National Center for Atmospheric Research (NCAR) and derived from Global Reanalysis data generated by the NCEP Department of Energy Atmospheric Intercomparison Project reanalysis (NCEP-DOE AMIP-II) [Kanamitsu et al., 2002]. Global reanalysis data are developed using continuously refined land-surface models on a 180 km wide grid [Mesinger et al., 2005]. The NARR project downscales NCEP-DOE AMIP-II outputs to regional-scale models with a finer 32 km grid. While wind data from NARR have failed to capture the declining trend in observed surface winds over the contiguous United States [Pryor et al., 2009], they compare more favorably with observations over coastal areas, agreeing to within 2 m s−1 and 31° direction [Moore et al., 2008].

[11] NARR data were chosen from 26 locations spaced roughly 0.5° apart and oriented with the shelf break along the entire study area, and were extracted at a three hourly time step from the beginning of the record (1979) to 2010. NCEP grid cells were chosen based on their proximity to the continental shelf. The coarser NCEP (2.5° ×2.5°) grid may obscure coastal wind phenomena [Bakun et al., 2010], but the data were included due to their superior temporal coverage, from 1948 to 2010.

[12] Previous comparisons between model data derived from NCEP and observations off the coast of Vancouver Island and Washington state found wind speeds that differed by 1.6–4.5 m s−1 and some bias in the principal axis of wind (∼10°) during the fall and summer [Tinis et al., 2006]. A substantial mismatch in the orientation or direction of winds could result in a biased estimate of upwelling and downwelling magnitude or occurrences.

[13] A number of climate indices representing oscillatory climate behavior are associated with the study area, including the ENSO, PDO, NPGO, Northern Oscillation Index (NOI), Pacific-North America Oscillation (PNA), and the Arctic Oscillation (AO). ENSO, PDO, NPGO, and NOI have all been shown to influence alongshore winds and therefore nutrient availability throughout our study area [e.g., Ware and Thomson, 1991; Mantua et al.; 1997; Schwing et al., 2006; Di Lorenzo et al., 2008; Black et al., 2011]. Correlations between alongshore wind behavior and the NOI were weaker than they were for the other three indices; therefore, only ENSO, PDO, and NPGO were used to consider the effect of climate variability on changes in onset, duration, and intensity of upwelling and downwelling seasons.

[14] Monthly NPGO values from 1950 to 2010 were collected from http://www.o3d.org/npgo/. The NPGO index is defined as the second empirical orthogonal function (EOF) of sea surface height anomalies (SSHa) over the eastern North Pacific Ocean (180°W–110°W; 62°N–25°N) and exhibits decadal scale oscillatory behavior. A positive phase of the NPGO is associated with enhanced equatorward wind stress in the CCS at mid-low latitudes south of 38°N [Di Lorenzo et al., 2008].

[15] Monthly PDO values between 1950 and 2010 were acquired from the Joint Institute for the Study of the Atmosphere and the Ocean Climate Data Archive (JISAO CDA) (http://jisao.washington.edu/data_sets/). The PDO index also exhibits decadal scale oscillations and is defined as the first EOF of sea surface temperature anomalies (SSTa) in the Pacific Ocean north of 25°N; a positive PDO value is associated with a deepening of the Aleutian low-pressure system off the southern coast of Alaska and enhanced poleward winds along the eastern Pacific Ocean coast [Mantua et al., 1997].

[16] Monthly ENSO values from 1950 to 2010 were also collected from the JISAO CDA. The JISAO Global-SST ENSO index is defined as the SST difference between the tropical (equatorward from 20°N and 20°S) and extratropical (poleward from 20°N and 20°S) Pacific Ocean [Penland et al., 2010]. An El Niño (positive) phase of ENSO is generally characterized by abnormally low pressure south of Alaska beginning in Northern Hemisphere winter. This low pressure leads to enhanced poleward winds throughout our study area [Rasmusson and Wallace, 1983; Hsieh et al., 1995].

2.2. Calculating Alongshore Stress

[17] The magnitude of alongshore wind stress (AWS) is related to coastal winds by [Pond and Pickard, 1983]

display math(1)

where inline image is the wind stress exerted by the component of the daily average wind speed, inline image, parallel to the coastline; inline image represents air density; and inline image is a dimensionless drag coefficient formulated by Trenbreth et al. [1990] as

display math(2)

[18] Upwelling is driven by the alongshore component of the wind speed. The principal axis of rotation (PAR) was calculated at each location to define the local coastline orientation [e.g., Tinis et al., 2006; Garcia-Reyes and Largier, 2010]. The variance ratio (VR) was also calculated at each location. VR values range from 1 < VR < ∞, with increasing values corresponding to higher degrees of wind anisotropy [Tinis et al., 2006].

2.3. Demarcating Seasons

[19] Estimating wind stress using equation (1) for all stations on a daily time step shows a clear seasonal signal, where downwelling favorable winds prevail during winter months and upwelling favorable winds dominate during the summer (e.g., Figure 2a). To determine the onset and conclusion of the upwelling season objectively, we considered the total cumulative AWS over an annual period (e.g., Figure 2b). To reduce the contribution of high-frequency variability, we first applied a 30 day running average (RA) to the daily-wind stress data. We then estimated the season onset and conclusion by associating them with a percentage of the total cumulative AWS for the year. For example, summer onset was associated with 10% of the total cumulative annual negative AWS from 1 January, with the conclusion associated with 90%. We refer to these values as the cutoff percentages (COP). The COP ensures that isolated or transient events do not trigger a false early onset, or late termination, of season (such as a strong upwelling event during the late winter downwelling season, e.g., Black et al. [2011]).

Figure 2.

(a) Five-year time series of 30 day RA (black line: negative indicates upwelling favorable wind, positive indicates downwelling-favorable wind) AWS at a typical NARR location, U, off Vancouver Island, to illustrate the performance of the objective demarcation criteria. Gray shading shows the summer (upwelling) season while hatched shading indicates winter (downwelling) season. Seasons were determined using the criteria outlined in section 2. (b) The percentage of cumulative total upwelling during 1992 at station U as a function of time (days). Dashed line indicates the 10% and 90% COP used to identify upwelling onset and conclusion, respectively. (c) Outcomes for summer onset (Julian Day) and (d) summer duration (in days) using demarcation criteria outlined in section 2 for station U.

[20] The same estimates were made for the onset and conclusion of downwelling seasons using instead the total cumulative positive AWS occurring between two summers. The winter onset is associated with 10% of the total cumulative positive AWS occurring between the last day of the preceding summer and the first day of the following summer. A 5 year time series of 30 day running mean AWS at location U illustrates the performance of the criteria (Figure 2a). Estimation of summer onset and duration at U is also shown for the entire data set (Figures 2c and 2d). The spring transition is simply defined as the period between the end of the winter downwelling season and the beginning of the summer upwelling. Similarly, the fall transition begins when summer upwelling ends and concludes at the onset of winter downwelling.

[21] Sensitivity of seasonal onset to the choice of RA (ranging from 5 to 60 days) and COP (2%, 10%, and 18%) were analyzed at NARR locations A, K, and W (Figures S1–S6, supporting information). Results were generally sensitive to COP (as expected) and less sensitive to RA. As a result of these analyses, an RA of 30 days and COP of 10% were chosen to minimize the possibility of transient wind events erroneously signaling a season onset or ending, without losing the ability to resolve seasonality completely. Sensitivity analyses also confirmed that varying the time at which to start adding up annual negative AWS by even 3 months had little effect on the ultimate choice of summer onset.

2.4. Time Series Analysis

[22] The time series of season onset, duration, and intensity (both upwelling and downwelling) for each of the 26 NARR stations were grouped together using a hierarchical cluster analysis to succinctly summarize results (the hcluster function in the R programming language) [R Core Team, 2012]. We define intensity as the total upwelling- or downwelling-favorable wind stress integrated over the season in question. The hcluster algorithm clusters the data from different stations based on a Euclidean distance within the parameter space [Lucas and Jasson, 2006]. The physical meaning of the groupings calculated by hcluster must therefore be determined by the observer.

[23] The result is a dendrogram of progressively more similar data sets. The final groupings presented here are the four main branches of the dendrogram. Timing, duration, and intensity results were each grouped separately. The time series within each group determined by hcluster were then averaged together, producing one time series per cluster.

[24] Time series of the ENSO, PDO, and NPGO indices were compared to the clustered NARR time series and NCEP-derived wind data of season onset, duration, and intensity via correlation analysis. To consider the possibility that preconditioning by climate oscillations influences upwelling and downwelling seasons, the onset, duration, and timing variables were compared with the average values for each index for the 4 months prior to the month in which the transitions occur. For example, if summer onset was predicted to occur in June for a given year, then the climate index values for March, April, May, and June of that year were averaged together. If summer onset occurred in May in a subsequent year, climate indices for the subsequent year would be averaged from February, March, April, and May.

[25] The correlations were examined for significance using Pearson's R test. Spearman's rank coefficient was considered as an alternative to Pearson's R as it can represent nonlinear monotonic relationships. Spearman's rank correlations were estimated between climate indices and NCEP data. When these coefficients were compared with the Pearson's R values for the same analysis, no large differences in correlation values (all but two correlations experienced changes of 0.1 or less) were found. An additional 7% of the correlations became significant using the Spearman's rank correlations, and in no instance was the sign of the result altered.

[26] The NCEP data cover 62 years while the NARR data only span 31 years. Thus, differences in temporal extent could impact estimations of trends and correlations with climate indices [e.g., Easterling and Wehner, 2009]. To test the effect of time series length, additional trend and correlation analyses were conducted using a truncated version (1979–2010 to match the NARR data) of the NCEP onset, duration, and intensity data.

3. Results

3.1. Assessing NARR and NCEP Data Relative to Observations

[27] The PAR values derived from observational, NARR, and NCEP data sources are in close agreement with each other for a given latitude (Figure 3a) and follow the orientation of the coast as expected (e.g., NW-SE orientations off northern California and Vancouver Island). Truncating the NCEP data to match the NARR and observation time period do not change the PAR values by more than 1°. The PAR values from the coarser NCEP data have a particularly smooth transition between locations compared to those generated from the NARR and observational wind data. The NARR and observation locations are closer to shore, and the more abrupt transitions could be partially due to topographic effects on wind. The PAR values estimated using NARR data differ from the nearest available observations by an absolute average of 7°, with a range from 0° to 12°. The PAR values from NCEP data differ from observations by an absolute average of 15°, ranging from 1° to 29°.

Figure 3.

Comparison between observation (circles), NARR (squares), and NCEP (triangles) derived estimates of (a) PAR and (b) variance ratio.

[28] The VR values for NCEP data are generally much lower than the VR values calculated for either NARR or observational data sets (Figure 3b). The differences between NARR and nearby observational VR values are negligible between 40°N and 48°N, except at NARR location I (42.26°N), which has a much higher VR than the nearest observational location (V, 41.95°N). Winds are most isotropic (weaker principal axis) for all locations near 46°N, off the north Oregon coast. In contrast, northern and southern ends of the study area have strong principal axes. South of 39°N, VR values from NARR data sets show a threefold increase relative to VR values derived from observations (Figure 3b). The coastline topography could be affecting the southernmost three observation locations, which are nearer to the shore than NARR locations A–C.

[29] A comparison between observed absolute wind speed, wind direction, and AWS for NARR and NCEP values shows stronger correlations between NARR and observations than between NCEP and observations (Table 2). The weaker correlation between observations and NCEP data is likely due to the distance between the NCEP locations and the observation data and also the coarser grid cells of NCEP data. The NARR and NCEP wind directions are well correlated to nearby observations for all locations and always more correlated than are wind speeds at NCEP and observational locations. One noticeable and unexplained exception to the overall strong AWS correlation performance (R ∼ 0.9) occurs between observational location VIII (46.14°N) and NARR location P (46.33°N) (for AWS R = 0.46). Correlating observation location VIII with the next closest NARR station, O (45.75°N) gave similarly low results (for AWS R = 0.45) (Table 2).

Table 2. Correlation (Pearson R Coefficients) of Winds Between Observation and Nearby NARR and NCEP Locationsa
 NARRNCEP
Observation LocationsLocationWind SpeedWind DirectionAWSLocationWind SpeedWind DirectionAWS
  1. a

    AWS, alongshore wind stress.

XIIW0.830.900.90    
XIU0.830.830.88ζ0.720.820.82
XR0.820.800.89    
IXQ0.860.900.92ε0.700.830.83
VIIIP0.780.900.46    
VIIM0.900.920.94δ0.760.860.86
VIJ0.770.860.85γ0.610.690.69
VH0.810.810.88    
IVG0.860.880.92β0.730.830.83
IIIC0.770.750.89    
IIA0.900.830.91    
I    α0.740.840.84

3.2. Demarcating Seasons

[30] Demarcating upwelling and downwelling seasons based on criteria outlined in section 2 captures the bimodal behavior of coastal winds over an annual cycle in our study area (e.g., Figure 2). The performance of the demarcation criteria was evaluated by comparing our results with two other methods from three different studies: an upwelling ocean current index [Thomson and Ware, 1996, hereinafter referred to as TW96] and a cumulative upwelling index (CUI) derived from winds [Bograd et al., 2009; Foreman et al., 2011, hereinafter referred to as B09 and F11, respectively]. We compare results, specifically season start and duration, from our station U (48.86°N) with those of TW96 from nearby La Perouse Bank (station XI at 48.83°N), between 1987 and 1995 using a Student's t test (Table 3). A direct comparison with TW96 upwelling index must include a small caveat, since their upwelling index is based on actual current measurements. Some upwelling events in this part of the study area are generated via coastal trapped waves originating as far south as southern California and cannot be captured by examining local wind behavior alone [Hickey, 1979]. We therefore expect some discrepancies (e.g., a shorter summer upwelling season using our criteria) between our criteria and the index developed in TW96.

Table 3. Results for Season Onset and Duration Between 1987 and 1995 Compared to Thomson and Ware [1996]a
  Summer OnsetSummer Length
 SeasonMean (Julian Day)SDpMean (Days)SDp
  1. a

    TW96, Thomson and Ware [1996]. TW96 results are based on an ocean current index and thus provide a more direct estimate of upwelling. Stars denote significant differences in p values of the means at the 0.05 level. SD, standard deviation (days); SP, spring; SU, summer; FA, fall; WI, winter.

 SP5115 7636 
TW96SU14030 14432 
 FA28930 6527 
 WI35628 5229 
        
 SP77220.0762140.44
Location USU138200.92118220.04*
 FA256200.0659240.50
 WI315170.06127270.00*
        
 SP72230.1563160.48
Location MSU135210.69131180.44
 FA266180.2654150.34
 WI320110.07118260.00*

[31] While the average start dates of summer upwelling are almost identical for the two stations, the average summer durations estimated using our demarcation method are significantly shorter than found by TW96 (significant at the p = 0.05 level). On average, the whole fall season is shifted earlier, with the end of summer upwelling occurring 36 days earlier and the beginning of winter downwelling starting 42 days earlier. Our demarcation criteria also result in a later end to the winter downwelling season and lead to a significantly (74 days) longer winter season, on average.

[32] Results for season onset and duration from TW96 were also compared to location M off the coast of Oregon. Interestingly, the results at location M more closely matched those from TW96 than location U, as indicated by the larger p values for six out of eight metrics (Table 3). The duration of winter was still significantly longer than TW96 estimates (although shorter than at station U), and the summer onset was slightly earlier than at location U. A key difference between results for location U and M is the later fall onset at location M, which effectively stretches out the duration of its summer. The summer duration at location M is in much closer agreement with TW96, and is likely the result of the coastal trapped waves propagating from the Oregon coast to the TW96 location [Hickey, 1979].

[33] The summer start and end were also compared to results from B09 and F11, who studied the majority of the CCS (up to 48°N) and the northern limit of the CCS, respectively, using the CUI approach (Figure 4). Using our method, NARR and NCEP start and end dates are consistent with each other, disagreeing by no more than a few days along the entire coast. Likewise, the results from B09 and F11 appear to be consistent with each other, although there is no spatial overlap (Figure 4). It is clear, however, that the CUI methodology yields different results in the start and end date of the summer season. The methodology presented herein predicts an average delay in the summer onset and an earlier end of the summer season of roughly 1 month, when compared to B09 and F11. The difference between the two methodologies appears greatest in the southern CCS (Figure 4). All approaches show a later onset and earlier end date with increasing latitude, except for a zone around the Washington state coast (∼47°N −48°N) in which the onset time roughly stays the same (Figure 4).

Figure 4.

(a) The mean summer onset dates (in Julian days) and (b) the mean summer end dates, for TW96 (gray star), NCEP (empty triangles), NARR locations (filled triangles), F11 (empty squares), and B09 (filled squares), plotted by latitude. TW96 results are averaged over 1987–1995; B09 results are averaged from 1967 to 2008; NARR results cover 1979–2010; F11 results span 1958–2008; and NCEP results span 1950–2010.

3.3. Cluster Analysis of NARR Data

[34] Hierarchical clustering was initially performed on data from all NARR locations for season timing, duration, and upwelling and downwelling intensity. The clustering results for season timing and duration at the second branching stage were identical, except for location J, which switched from group 2 to 3 for the duration of all four seasons, respectively. The clustering results of season start were therefore also used to group the season duration time series (Figure 5a). The grouping of NARR stations by upwelling and downwelling intensity was dramatically different than the cluster results based on upwelling and downwelling onset and duration, respectively, and therefore required their own clusters (Figures 5b and 5c).

Figure 5.

Hierarchical clustering results for all four seasons for (a) season onset and duration, (b) upwelling intensity, and (c) downwelling intensity. Season onset and season duration produced almost identical results in the clustering analysis (Station J was included in cluster 2 for all four season durations).

[35] Clusters were generally grouped by latitude (dendrogram plots for clusters available in Figures S7–S10, supporting information). Locations north of 50°N (Y-Z) were consistently grouped together, and experienced longer, more intense downwelling seasons, and shorter, lower intensity upwelling seasons (Figure 6). The main branches of the clusters were divided at roughly 43°N, except for downwelling intensity (Figure S10, supporting information). While downwelling intensity generally increases in the more northern locations (Figure 6b), the relatively large cluster extending from 40°N to 50°N in downwelling suggests that downwelling exhibits more consistent behavior than upwelling throughout the region. This inference is supported by clustering results which show that locations 40°N–42°N (E to G) and near 47°N (Q and T) are more similar to each other than locations H to P between 41°N and 47°N (Figure 10, supporting information). Downwelling becomes infrequent enough by cluster 4 that downwelling events are no longer registered, leading to gaps in the time series (Figures 6b, 6d, and 6f). Except for upwelling intensity, locations south of 40°N (A–D) were always clustered together (Figures 6a and 6c). The length of summer (winter) generally decreases (increases) with increasing latitude (Figures 6e and 6f).

Figure 6.

Time series of NARR clustered data, including onset and duration clusters 1 (dashed gray line), 2 (dashed black line), 3 (solid gray line), and 4 (solid black line) for summer (a) upwelling intensity, (c) onset, and (e) duration, and winter (b) downwelling intensity, (d) onset, and (f) duration.

3.4. Temporal Trends in Onset, Duration, and Intensity of Alongshore Winds

[36] There were no statistically significant (p < 0.05) linear temporal trends for the onset or duration of the summer or winter season at any NARR location (regressions not shown; time series in Figures S11–S18, supporting information). Five NARR locations showed significant trends toward higher upwelling intensity with time during the summer season (Figure 7), including two sites off northern California (F and G, p < 0.01) and three sites off northern Washington and southern Vancouver Island (R and U, p < 0.01; T, p < 0.05). Comparing the time series of upwelling intensity between locations E, F, G, and H off northern California shows that the significant trend in F and G is mostly due to enhanced wind-driven upwelling observed between 2003 and 2010 (Figure 7a). Location H appears to follow locations F and G closely until 2003; thereafter, locations F and G show enhanced upwelling.

Figure 7.

(a) Time series of summer upwelling intensity for locations E (black line), F (red line), G (blue line), and H (gray line); (b) time series of upwelling intensity for locations R (red line), S (gray line), T (blue line), and U (green line). Stars denote time series with a significant trend of more intense upwelling with time at the p = 0.01 level (**) and p = 0.05 level (*). Note the different scale for the vertical axis for Figure 7b.

[37] With substantially longer time series, trends in NCEP data tell a different story. In contrast to the NARR onset data, simple linear regression of data from locations α, near central California, and δ, ε and ζ, extending from central Oregon to northern Washington, all indicate a trend toward a significantly later start to the upwelling season (p < 0.05) (Table 4). Also unlike the NARR results, no significant linear trends in upwelling intensity were found at the NCEP locations. Visual inspection of location β shows it to follow NARR locations F, G, and H closely (Figure 8a). After 2003, location β shows a similar level of upwelling intensity to location H, but not locations F and G. A similar phenomenon occurs for location ζ, which closely matches location S (the nearest NARR location) until 2010 (Figure 8b).

Table 4. Linear Trends in Summer Onset and Intensity Estimate From NCEP Data Using the Full (1948–2010) and Also the Truncated (1979–2010) Time Seriesa
 NCEP Summer OnsetNCEP Summer Total Upwelling
 1948–20101979–20101948–20101979–2010
StationDays/yearpDays/yearpN days m−2/yearpN days m−2/yearp
  1. a

    Bold print denote significant trends (p < 0.05).

α0.350.020.130.75−0.020.14−0.040.22
β0.300.060.370.39−0.0180.08−0.050.07
γ0.300.090.300.49−0.0120.18−0.050.08
δ0.300.04−0.060.88−0.010.36−0.030.14
ε0.440.01−0.070.87−0.010.26−0.020.17
ζ0.620.01−0.450.3100.760.020.03
Figure 8.

(a) Time series of upwelling intensity between NCEP location β (dashed line) and nearby NARR locations F (green line), G (blue line), and H (orange line). Stars in legend denote significance of the linear trend (p < 0.01). The gray line is the simple linear regression with time at NARR location G. (b) time series of upwelling intensity for NCEP location ζ (dashed line) and NARR locations R (green line), S (blue line), T (orange line), and U (purple line). Gray line is simple linear regression with time at NARR location T.

[38] The NCEP time series for upwelling start date and intensity were truncated to 1979–2010 and linearly regressed again to test for the impact of time series length on the results (Table 4). Similar to the NARR data, there were no statistically significant trends in upwelling onset dates when using the truncated NCEP data. The only statistically significant increasing trend (p < 0.05) in upwelling intensity for the truncated NCEP data was found at location ζ. This result is in contrast to the NARR results for which five stations show statistically significant increasing trends (Figure 7).

3.5. Correlations to Climate Oscillations

3.5.1 Correlation Between NARR Clusters and Climate Indices

[39] The clustering approach (section 3.3) was ultimately a convenient tool to summarize results from NARR locations, which would have otherwise been intractable. It was verified that correlations with climate indices to the NARR clusters accurately represent correlations to individual NARR locations, and therefore the clusters are suitable in summarizing our results. Based on a t distribution, a few notable, significant correlations were found between the NARR time series for onset, duration, and intensity derived from the cluster time series and the averages of the NPGO, ENSO, or PDO climate indices for the 4 months prior to season onset (Table 5).

Table 5. Pearson's R Correlation Coefficients Between 4 Month averaged NPGO, PDO, and ENSO Indices and Season Onset, Duration, and Intensity Derived From NARR (1979–2010) Cluster Time Seriesa
 NPGOPDOENSO
ClusterSUWISUWISUWI
  1. a

    SU, summer; WI, winter. Bold denotes significance at the p = 0.05 level. Note that the significance of the correlation is based on a t distribution, and is dependent on the number of pairs. The NARR data have 32 pairs of observations (1979–2010), while NCEP data (Table 6) have 63 pairs of observations (1948–2010). Therefore, significant (p < 0.05) correlations between NARR and a given climate index are achieved with a higher r score than the same score with NCEP data.

  2. b

    All SU correlations are with upwelling intensity and WI correlations are with downwelling intensity.

 Season Onset
1−0.13−0.200.35−0.20−0.130.14
2−0.180.390.62−0.060.170.16
30.35−0.190.50−0.110.530.18
4−0.30−0.030.47−0.250.480.08
 Season Duration
20.19−0.250.540.15−0.12−0.07
30.31−0.230.560.090.420.04
40.14−0.06−0.570.01−0.34−0.10
 Upwelling/Downwelling Intensityb
20.15−0.31−0.330.41−0.060.50
30.35−0.340.410.260.600.50
40.49−0.130.48−0.14−0.550.22

[40] The NPGO has been most strongly related to oceanographic changes south of 38°N [Di Lorenzo et al., 2008], and significant correlations further north in our study region were not expected. We found a weak negative correlation between the NPGO and summer onset (between 39°N and 43°N) and upwelling intensity (south of 43°N), suggesting earlier and more intense summer upwelling seasons occur during the positive phase of the NPGO (Table 5). However, these earlier onsets and more intense summer upwelling seasons do not correspond to any significant correlations between the NPGO and season duration (Table 5). A significant negative correlation was also found between the NPGO and winter onset between 43°N and 50.5°N.

[41] The PDO is positively correlated with summer onset at clusters 2–4 (south of 50.5°N), and negatively correlated to summer duration and upwelling intensity throughout the study area, save cluster 2 (Table 5). Together, these results strongly suggest that the positive (negative) phase PDO is associated with later (earlier) and shorter (longer) upwelling seasons with decreased (increased) upwelling intensity. While no significant correlations were found between PDO and winter onset or duration, locations north of 49°N were positively correlated to enhanced winter downwelling during positive PDO phases (Table 5).

[42] ENSO is positively correlated to summer onset and negatively correlated to summer duration and summer upwelling intensity south of 43°N (Table 5). The El Niño phase of ENSO occurring during the summer is therefore associated with later and shorter summer seasons of less intense upwelling. Similar to PDO, the El Niño phase of ENSO was significantly correlated to enhanced winter downwelling at all but the most southerly NARR locations (Table 5).

[43] In summary, when considering the NARR data in our study area, we found that the NPGO was generally uncorrelated to seasonality in coastal winds. The ENSO has significant correlations during the summer season for all three variables south of 43°N and is positively correlated to enhanced winter downwelling north of 39°N. The PDO appears to be most correlated to coastal wind dynamics, and is significantly correlated to all three variables during the summer throughout the study area and correlated to enhanced winter downwelling north of 49°N.

3.5.2. Correlation Between NCEP Locations and Climate Indices

[44] As with the NARR data, significance in the correlations between NCEP locations and climate indices was based on a t distribution; however, while NARR and NCEP data might have the same correlation value for a given climate index, they would have two different p values due to the difference in their temporal extent.

[45] Correlations between NCEP season onset, duration, and intensity and the climate indices are generally similar to the NARR results, with some notable exceptions. Significant negative correlations were found between the NCEP data and the NPGO index for the winter downwelling intensity at all locations other than α (Table 6). Otherwise, the NPGO is sparsely correlated to the duration or onset of either the winter or summer seasons, much like the NARR data results.

Table 6. Pearson's R Correlation Coefficients Between 4 Month Averaged NPGO, PDO, and ENSO Indices and Season Onset, Duration, and Intensity Derived From NCEP (1948–2010) Time Seriesa
 NPGOPDOENSO
LocationSUWISUWISUWI
  1. a

    SU, summer; WI, winter. Bold denotes correlations significant at the p = 0.05 level. Note that the significance of the correlation is based on a t distribution, and is dependent on the number of pairs. The NARR data (Table 5) have 32 pairs of observations (1979–2010), while NCEP data have 63 pairs of observations (1948–2010). Therefore, significant (p < 0.05) correlations between NCEP and a given climate index can be achieved with a lower r score than the same score with NARR data.

  2. b

    All SU correlations are with upwelling intensity and WI correlations are with downwelling intensity.

 Season Onset
α0.03−0.240.510.020.180.14
β−0.03−0.300.470.160.170.25
γ0.05−0.080.510.090.190.12
δ−0.04−0.140.360.080.260.13
ε−0.24−0.130.340.000.330.13
ζ−0.240.020.47−0.150.52−0.07
 Season Duration
α0.03−0.06−0.510.27−0.240.13
β0.12−0.24−0.440.32−0.180.22
γ0.09−0.06−0.400.19−0.160.04
δ0.12−0.06−0.250.16−0.170.05
ε0.26−0.01−0.21−0.01−0.24−0.10
ζ0.19−0.05−0.440.13−0.44−0.04
 Upwelling/Downwelling Intensityb
α0.01−0.33−0.230.40−0.130.53
β0.02−0.32−0.090.22−0.020.43
γ0.05−0.29−0.170.14−0.100.34
δ0.11−0.27−0.160.17−0.100.38
ε0.33−0.28−0.120.25−0.190.47
ζ0.34−0.03−0.170.19−0.470.51

[46] The PDO correlation to summer onset and duration likewise mirrors results from the NARR analysis, showing a significant coastwide positive and negative correlation to onset and duration, respectively (Table 6). Unlike the NARR result, however, there are no significant correlations to summer upwelling or winter downwelling intensities at any NCEP locations, save location ζ (Table 6). Also different from NARR are significant positive correlations between winter duration and PDO at locations ζ and ε (Table 6).

[47] The correlations between ENSO and NCEP data also generally match NARR results. The onset of summer upwelling is positively correlated with ENSO at locations α, β, and γ in the southern part of the study area (Table 6), and summer duration and upwelling intensity only show significant positive correlations to ENSO at location α (Table 6). Downwelling intensity is prominently correlated to ENSO throughout the entire study area during the winter (Table 6).

[48] While the majority of correlations between the climate indices and both the NARR and NCEP data are in general agreement, two distinct differences appear: the negative correlation between the NPGO and winter downwelling intensity found at all locations using NCEP wind data, and the negative correlation between PDO and summer upwelling intensity found across much of the CCS using NARR data.

[49] One potential cause for the discrepancies between NARR and NCEP correlations is the differing length of the time series. When the length of the NCEP data set is truncated to match the NARR time period (1979–2010), correlations between the PDO, ENSO, and onset dates from the truncated NCEP data set (Table 7) are in closer agreement with those using the NARR data set (Table 5). Thus, the differences in correlation values between the climate indices and data derived from NCEP and NARR may result in part from the fact that the NARR time series is substantially shorter.

Table 7. Pearson's R Correlation Coefficients Between 4 Month Averaged NPGO, PDO, and ENSO Indices and Season Onset, Duration, and Intensity Derived From NCEP Locations Using a Truncated (1979–2010) Time Seriesa
 NPGOPDOENSO
LocationSUWISUWISUWI
  1. a

    Bold denotes correlations significant at the p = 0.05 level.

  2. b

    Al SU correlations are with upwelling intensity and WI correlations are with downwelling intensity.

 Season Onset
α−0.180.380.550.040.020.32
β−0.160.390.430.01−0.090.21
γ−0.07−0.350.45−0.07−0.010.05
δ−0.11−0.350.41−0.050.290.08
ε−0.25−0.270.43−0.080.470.23
ζ−0.330.000.45−0.250.44−0.14
 Season Duration
α0.21−0.290.560.17−0.04−0.02
β0.19−0.290.420.160.03−0.14
γ0.13−0.13−0.460.05−0.04−0.19
δ0.10−0.180.420.16−0.220.06
ε0.170.100.46−0.070.40−0.09
ζ0.18−0.220.54−0.03−0.35−0.22
 Upwelling/Downwelling Intensityb
α−0.19−0.300.330.360.010.61
β−0.18−0.280.180.290.000.59
γ−0.20−0.240.250.230.110.49
δ−0.32−0.250.390.170.210.45
ε0.46−0.300.420.190.450.44
ζ0.35−0.140.370.030.610.51

4. Discussion

4.1. Seasonal Demarcation Methodology Performance

[50] No one of the many methods for demarcating the summer upwelling season [e.g., Thomson and Ware, 1996; Bilbao, 1999; Lynn et al., 2003; Schwing et al., 2006; Bograd et al., 2009] has been successful at predicting biological response over the entire region of the CCS [e.g. Holt and Mantua, 2009]. Meanwhile, the ability to determine the downwelling season has not been evaluated. Given the changes in the character of upwelling (and downwelling) with latitude (section 3.3), it would not be surprising if multiple metrics are necessary to adequately identify coastwide trends in transitions. While our method is based solely on winds (like B09 and F11), it offers some flexibility in the choice of the COP. It is possible that as the character of winds and stratification changes with latitude, that biological responses may be better predicted by allowing COP to change with latitude.

[51] Since our ultimate goal is to predict the timing of biological response, that response itself (e.g., satellite derived chlorophyll [Henson and Thomas, 2007]) provides the best metric. However, direct observations are not easy to obtain, and satellite observations may not always be available (due either to cloud cover or the finite lifetime of the satellites themselves). In addition, it is difficult to associate downwelling with biological markers. The TW96 index uses ocean currents, one step removed from biological response, and the direct result of wind forcing. However, these data are even more difficult to obtain. Thus, wind metrics (such as the CUI, B09, and our own) have a distinct advantage. These data are readily available, and we expect them to be available in the future.

[52] The CUI method (B09, F11) is elegant, however, in the northern limit of the CCS prediction of transitions from a single year of data can be ambiguous, as multiple minima are often present (M. Foreman, personal communication). As a result, F11 used decadal averaged CUI. For a given COP, our method has the advantage of producing clear results for single years at all latitudes. Here, choosing an appropriate COP value is a potential limitation. The average length of both upwelling and downwelling seasons produced by our method is significantly shorter at all latitudes than those predicted by B09 and F11 (Figure 4). We could bring our method into close agreement with B09 and F11 by reducing the COP (see supporting information Figures S4–S6, for COP sensitivity). A common thread among these studies and the results found herein is high interannual variability of the onset, duration, and intensity of seasons.

[53] In 2005, the winds appeared to transition from poleward to equatorward by late May in the north-central CCS, but biologically relevant upwelling only began in early July [Kosro et al., 2006]. This delayed summer season had significant negative effects on the ecosystem [e.g., Schwing et al., 2006; Sydeman et al., 2006; Barth et al., 2007]. The transition time calculated by our proposed methodology for 2005 shows a much closer linkage to the biologically relevant transition (26 June for NARR location M and 4 July for NCEP location δ (Table S2, supporting information) than the CUI [B09]. While anecdotal, this result supports our choice of COP and our methodology in general. Mechanistically, it is possible that a certain amount of upwelling forcing (more associated with a larger COP) is required to bring the pycnocline, and the denser, nutrient-rich waters beneath it, back up onto the shelf after it has been depressed at the coast by winter downwelling in the central-northern CCS where downwelling occurs.

[54] Both our method and the CUI method integrate wind (or wind-derived) signals (although F11 uses different methods to obtain winds than B09). The main difference between methods is that we “rectify” the signal by integrating only the upwelling-favorable winds to determine the onset and duration of upwelling (and vice versa for the downwelling season, section 2.3). This difference in accounting procedures means that isolated downwelling events that occur prior to the spring transition would not delay our date of onset of upwelling, while such events could cause a significant delay in CUI prediction. Because we expect downwelling to play a significant role in preconditioning nutrient conditions at least at northern latitudes [Ianson and Allen, 2002, Tortell et al. 2012], it is unclear if this difference is an advantage. However, our method does provide a robust measure of the downwelling season itself in the northern and central CCS. The performance of the winter demarcation criteria breaks down in the southernmost locations, where downwelling plays a diminished role.

4.2. Temporal Trends

[55] Results from our analysis of NARR winds are consistent with some previous studies indicating no significant temporal trends in summer onset or upwelling duration north of 40°N from 1979 to 2010 [Schwing and Mendelssohn, 1997; Garcia-Reyes and Largier, 2010]. However, there is disagreement between the current study and B09 and F11, not only in the mean summer start and end date but also in temporal trends of onset and intensity.

[56] F11 find a later onset in the summer season at five out of six locations in the study area off the coast of British Columbia, while B09 found a later onset in the northern section of the CCS. A later onset was not found at any NARR locations off the coast of Vancouver Island; however, the NCEP data did find a statistically significant trend at location ζ. The lack of a significant trend for any NARR location could be a consequence of the relatively short NARR time series compared to NCEP, B09 and F11 data sets. Indeed, when the NCEP upwelling onset data were truncated to match the length of the NARR time series, no significant trends were found at any station.

[57] Summer and winter intensity were found to be increasing over time in F11. Select locations in the NARR data were found to have statistically significant temporal increases in upwelling intensity (F, G, S, T, and U), but not as regionally widespread as in F11. Additionally, no locations were found to have statistically significant increases in downwelling intensity. The discrepancy between the present study and F11 is likely a consequence of the differing regression analyses used and the different definitions of intensity. F11 uses the Total Downwelling Magnitude Index (TDMI) instead of the cumulative total downwelling wind stress employed herein. The TDMI is computed with a wind stress based on a constant drag coefficient and a constant coastline orientation. Variations in these parameters could be changing the upwelling or downwelling magnitude measured. Also, the significant trend found in F11 is based on a linear regression of six decadal average values. The decadal averages were used due to difficulties in employing the CUI methodology at northerly locations, related to difficulty in distinguishing the global (i.e., yearly) minimum for the CUI index (M. Foreman, personal communication, 2012). The regression results found in F11 might therefore be a consequence of the decadal averaging. Conversely, the lack of trends found herein could be either due to a short time series (in the case on NARR data) or large grid size (in NCEP data).

[58] In any event, the results from the current study should be treated with caution; most of the increase occurs after 2003 and is based on a simple linear regression that is unable to capture nonlinear relationships. The rate of increase in summer intensity for all three locations is −0.049 to −0.040 N m−2 per year. If one assumes a drag coefficient of 0.0015 (equation (1)), then these increases are equivalent to an average increase in equatorward wind speed of ∼0.05 m s−1 per year, during summer, or an impressive annual increase of roughly 2% per year in mean wind speed. The same estimation yields ∼0.06 m s−1 near central California (locations F and G), which is similar to estimated changes in upwelling intensity previously found between 1982 and 2008 at 38°N [Garcia-Reyes and Largier, 2010]. Other estimates for changes in wind-driven upwelling intensity are as high as 0.15 m s−1 per year between 1946 and 1990 at 40°N, but are based on a static summer period of April-July [Schwing and Mendelssohn, 1997] as opposed to a more realistic, variable summer onset and duration used in our analysis.

[59] Our analysis with the NCEP data illustrates the effect of changes in both temporal and spatial scales on resulting trends and correlations. The enhanced summer upwelling intensity trend at NARR locations F, G, R, T, and U are not present at the nearby NCEP locations β and ζ (Figure 8). On the one hand, the coarse resolution of NCEP data, combined with possible deficiencies in data assimilation in earlier epochs used in developing the NCEP data set, could make these data unable to capture the true magnitude of coastal winds, especially at the extremes. On the other hand, the NCEP time series is twice the length of the NARR time series and provides more information about the long-term behavior of coastal winds in our study area. When the NCEP upwelling intensity data are truncated to match the length of the NARR data, the slope of the linear trends do increase in magnitude, and although still not significant, the p values for these trends decrease for all NCEP locations. This result indicates that finding a significant linear trend for upwelling intensity is more likely for data between 1979 and 2010 than for data between 1948 and 2010.

[60] As previously mentioned, both the NARR and NCEP data illustrate strong interannual variability in AWS throughout the study area. We emphasize that this analysis considers only the wind-driven component of upwelling and downwelling and has not considered the influence of stratification [Lentz and Chapman, 2004] or coastal-trapped waves [Hickey, 1979]. Such metrics are required to fully characterize the behavior of the coastal waters and ultimately determine the true magnitude of upwelling. Therefore, even in the absence of detected trends in seasonal wind behavior found herein, in B09 and F11, the enhanced upwelling measured further south in historical analyses and model predictions [e.g., Auad et al., 2006; Snyder et al., 2003] could still lead to enhanced upwelling through coastal trapped waves north of 40°N.

4.3. Implications of Correlations With Climate Indices

[61] The positive phase of the PDO has previously been associated with a deepening of the Aleutian low pressure system, leading to decreased upwelling north of 38°N [Chhak and Di Lorenzo, 2007]. Correlations between the PDO and the NARR and NCEP data presented here show later, shorter upwelling seasons during the positive phase of the PDO; however, while NARR data show a significant correlation between positive PDO and a more intense summer upwelling, the NCEP data do not. Correlations between positive PDO and upwelling intensity from a truncated NCEP data set from only 1979 to 2010 are significant south of 43°N, and indicates that the correlation between PDO and summer upwelling intensity has increased since 1979.

[62] The positive phases of both ENSO and PDO are generally linked to a delay and shortening of summer upwelling throughout most of the study area; however, only the NARR data suggest that positive phases of ENSO and PDO are linked with a reduction in upwelling intensity, and then only at southern latitudes (Tables 5 and 6). In fact, given that the summer season becomes shorter during positive PDO and ENSO phases, late summer upwelling must be more intense than during a negative phase, to account for little change in upwelling intensity. A possible explanation of this behavior is that positive PDO and ENSO phases lead to a change in wind speed frequency, resulting in fewer but stronger pulses of upwelling in northern latitudes, but more sustained changes to winds in southern latitudes. If PDO and ENSO patterns are expected to change through time (i.e., become positive more often), then one might anticipate delayed onsets of summer upwelling more often, and might cause significant disturbances to plankton assemblages, and a bottom-up effect on pelagic communities in general [Ware and Thomson, 2005; Kosro et al., 2006; Barth et al., 2007; Bakun et al., 2010].

[63] The negligible correlation between NPGO and alongshore winds north of 40°N is supported by previous work [Di Lorenzo et al., 2008]. However, the NCEP data show that the negative (positive) phase of NPGO is correlated to enhanced (weaker) winter downwelling throughout much of the study area (Table 6). This pattern is expected because a negative NPGO phase is indicative of an anomalously strong N-S pressure gradient resulting from a high pressure centered at ∼30°N and 150°W, with a deepening low centered at 55°N and 150°W. The resultant wind stress creates enhanced upwelling through equatorward winds along the Eastern Pacific Coast south of 38°N and enhanced downwelling from poleward winds north of 40°N [Di Lorenzo et al., 2008]. Winter preconditioning may be an influential component to coastal productivity, especially in northern sections of the CCS [e.g., Tortell et al., 2012]. Our results suggest that the negative phase of the NPGO leading into the winter months may lead to lower spring nutrient concentrations in shelf waters.

[64] Similar preconditioning may occur as a result of the significant positive relationships found between the positive (warm) phases of the PDO and ENSO and winter downwelling in the NCEP data throughout most of the CCS. The behavior associated with positive ENSO and PDO phases are similar, whereby anomalously low pressure is created south of Alaska through the warming of tropical surface water in the Pacific, generating more poleward winds through our study area [Rasmusson and Wallace, 1983]. The strength of both of these correlations suggests that this mechanism can lead to increased downwelling during the winter.

[65] The general lack of correlation between upwelling intensity and the NPGO, PDO, and ENSO indices north of 40°N is surprising, given the correlation in summer onset and timing to PDO, and to a lesser extent, ENSO (Tables 5 and 6). Alternative approaches to correlating climate indices to our data, including using different averaging lengths for the climate indices (between 1 and 12 months) and an average of climate index values within the upwelling and downwelling seasons, did not substantially change the outcomes. Another reason for these poor correlations could be our definition of upwelling intensity (see section 2.3). A different metric that can explicitly quantify the magnitude or number of wind stress pulses might better elucidate this dynamic and should be the focus of future research.

[66] Our result indicating that the positive phase of ENSO is related to enhanced winter downwelling intensity throughout the study area supports the work of previous studies [Hsieh et al., 1995; Penland et al., 2010]. The effect of climate change on future ENSO conditions is uncertain, but some research suggests that more El Niño phases will occur in the future [Collins, 2005]. A climate experiencing more El Niño conditions would lead to enhanced downwelling during the winter throughout the study area, and subsequently, less nutrient delivery to the surface along the shelf. The interplay between enhanced preconditioning and the potential weakening of upwelling favorable winds during the summer upwelling season (at least in the southern latitudes) could lead to substantial mismatches in nutrient delivery along the northern CCS during the spring and summer [Ware and Thomson, 2005; Kosro et al., 2006].

5. Conclusion

[67] We present new objective criteria to demarcate the upwelling and downwelling seasons along the North American Pacific coast. These criteria are simple, rely only on readily available wind data, and provide unambiguous results for single years. One parameter (the cutoff percentage, COP) is required (we use a COP of 10% throughout the CCS) that could potentially be tuned to biological data and vary with latitude. Our method shows general agreement with the TW96 index, which is based on ocean currents off the west coast of Vancouver Island (Table 3) but predicts a shorter upwelling season than the B09 and F11 wind-derived method when a COP of 10% is used (Figure 4).

[68] Generally, the onset, duration, and intensity of seasons show high interannual variability; on the order of 50–100 days for onset and duration, and 5–10 N m−2 per season for upwelling and downwelling intensity (Figure 6). Winter onset occurs earlier and increases in duration and intensity with increasing latitude. The same is true for summer onset, duration, and intensity as latitude decreases.

[69] There were few significant temporal trends in the onset, duration, or intensity of seasons in either NCEP or NARR time series. However, linear regression of the time series derived NARR data (1979–2010) did indicate a significant intensification of summer upwelling with time off the coast near the Oregon-California border and also near the mouth of the Strait of Juan de Fuca. However, these trends were not found in the longer (1948–2010) NCEP time series from nearby locations. Regression analysis of NCEP time series did indicate significantly later summer onsets at four of the six locations between 1948 and 2010 (Table 4).

[70] Correlation tests between NARR and NCEP time series and PDO, ENSO, and NPGO indicated that (1) some significant correlations exist for the onset, duration and intensity in our study area; (2) the positive phase of the PDO leads to later and shorter upwelling seasons but not necessarily less total upwelling in the northern CCS; (3) the warm phase of ENSO is associated with a later onset of the summer upwelling season and higher downwelling intensity during the winter throughout the study area; and (4) NPGO has a limited influence on wind regimes north of 40°N.

[71] In general, our research characterizes a system that experiences large interannual to interdecadal variability in the onset, duration, and intensity of upwelling and downwelling. The variability of coastal wind swamps most temporal trends introduced by global climate change but can be linked to large-scale climate oscillations. Time series that indicate changes in alongshore winds along the North American Pacific coast, but are too short to resolve climate oscillations (i.e., less than 50 years), should be treated with caution.

Acknowledgments

[72] We thank NOAA, Environment Canada, and NCAR for the raw wind data used in this study. Special thanks to Ben Cross and Zahid Hossain for NARR data acquisition. We are grateful to M. Foreman and S. Bograd for helpful and stimulating conversations and correspondence concerning this research. K.E.K. was supported by a NSERC Discover Grant and through the NSERC Canada Research Chair Program. D.I. was supported by Fisheries and Oceans Canada. This work is also supported by Simon Fraser University's Climate Change Impacts Research Consortium (CCIRC). We also extend sincere gratitude to our three anonymous reviewers for their insightful and constructive comments.

Advertisement