SEARCH

SEARCH BY CITATION

Keywords:

  • Pacific Northwest;
  • climate change;
  • water resources;
  • evapotranspiration;
  • runoff ratio

Abstract

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Water resources in the Pacific Northwest (PNW) are very sensitive to climate change. There are still big knowledge gaps on how evapotranspiration (ET) varies in responding to changing temperature (T) and precipitation (P) over different zones in terms of supply and demand regime for ET. Here, we employ the Variable Infiltration Capacity (VIC) hydrologic model and a high-resolution meteorological data set to quantify spatial and seasonal variations of ET and the runoff (R)/precipitation (P) ratio over the PNW and attribute effects of T and P. We evaluate modeled ET and R with eddy covariance measurements, upscaled regional ET, and reconstructed natural streamflow. Simulation results indicate that water-limited (annual potential ET (PET) ≥ P) and energy-limited zones (annual PET < P) have different responses to changing climate. In general, water-limited zones tend to be more associated with increasing ET and decreasing R/P than do energy-limited regions. With controlled simulation experiments, we document that trends in annual and warm-season ET and R/P are dominantly controlled by P, while in the cool season they are mainly controlled by T. During an entire cycle of the Pacific Decadal Oscillation (1947-2006), the PNW experienced a substantial increase in ET and a decrease in R and R/P under the trends of warming and drying. In this snowmelt-dominated region where warming-induced changes to the snowpack are impacting seasonal freshwater availability, decreases in R/P could further aggravate water scarcity. To reduce uncertainties, high-resolution meteorological data sets and intensive model calibrations and evaluations against ET are required.

1 Introduction

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] Recent hydrologic studies show that the global water cycle has been intensely modified by the changing climate, atmospheric compositions, and human activities during recent decades [Hutjes et al., 1998; Vorosmarty and Sahagian, 2000; Jackson et al., 2001; Huntington, 2006]. An earlier snowmelt peak and a decrease of snow-water equivalent (SWE) have been detected in the northern snowmelt-dominated basins across western North America [Mote, 2003; Hamlet et al., 2005; Mote et al., 2005; Stewart et al., 2005]. Modeling studies have also predicted that a projected increase in global temperature (T) and changes in precipitation (P) will decrease mountain snowpack and cause an earlier snowmelt, reducing warm-season runoff (R) in many mid- to high-latitude snowmelt-dominated areas [Barnett et al., 2005; Adam et al., 2009].

[3] Here we apply a well-developed macroscale hydrological model to examine how climate change during 1921-2006 affects evapotranspiration (ET) and R in basins across the Pacific Northwest (PNW), which covers the Columbia River Basin (CRB) and the coastal drainage area of Washington and Oregon (Figure 1). In earlier modeling studies, Hamlet and Lettenmaier [2007] and Hamlet et al. [2007] reported long-term trends in ET, R, SWE, soil moisture (SM), and the timing of ET and R in the mountainous areas of the western U.S. during the last century by using the Variable Infiltration Capacity (VIC) model at resolutions from one-sixteenth to one-eighteenth degree resolution [Elsner et al., 2010; Hamlet et al., 2012]. Extending previous studies, there is a need to 1) more thoroughly quantify the historical trends of these variables by focusing on how P and T changes (and related changes in other surface-water and energy-balance components) are translated into changes in water resources availability through changes in ET and R/P, 2) identify and characterize the spatial and temporal patterns (i.e., long-term trends and seasonal variations) of these relationships, and 3) evaluate VIC-simulated ET and streamflow more comprehensively than has been done previously. Here, by employing the VIC driving data and parameters developed by Elsner et al. [2010] and Hamlet et al. [2012], respectively, we address three major questions. First, how do ET and R/P change in the PNW under the context of climate change over the long term (1921-2006) and during the recent Pacific Decadal Oscillation (PDO) cycle (1947-2006). Second, what are the seasonal patterns of change in water-limited (Potential ET (PET) > P) and energy-limited (PET < P) zones. Third, what is the relative contribution from T and P (and related surface-energy and water-balance variables) to these variations of ET and R/P, and how do their relative effects vary geographically and seasonally?

image

Figure 1. PNW study area and distributions of sites for model evaluations. The land cover map is from the MODIS product MOD12Q, Friedl et al. [2002]. Crosses represent locations of eddy flux tower sites (Table 1), and rectangles represent selected gauges for model evaluations.

Download figure to PowerPoint

Table 1. List of Eddy-Flux Tower Stations Used for Evaluating VIC-simulated ET
AbbreviationsFull NameLatitudeLongitudeVegetation and Disturbance HistoryVIC Vegetation TypesObservation PeriodReferences
US-WrcWind River Crane Site45.8205–121.952Old growth evergreen needle forest (ENF)ENF: 87% Woodland: 13%21 Mar 1998 to 2006 without 2003[Parker et al., 2004; Paw et al., 2004; Shaw et al., 2004]
US-Me1Metolius Eyerly Burn44.5794–121.5Severely burned region in 2002; 100% stand replacementWoodland: 87% ENF: 13%13 Apr 2004 to 14 Jul 2005[Irvine et al., 2007]
US-Me2Metolius: intermediate aged ponderosa pine44.4523–121.5574Logging early 20th century; regenerated naturallyWoodland: 100%2002-2007 without 2003[Thomas et al., 2009]
US-Me3Metolius: second young aged pine44.3154–121.6078Timber are removed in 1987; regenerated naturally after thenWoodland: 81% Wooded grasslands: 19%2004-2005[Vickers et al., 2009]
US-Me5Metolius: first young aged pine44.4372–121.567Completely clearcut in 1978; regenerated naturally after thenWoodland: 94% Wooded grasslands: 6%2000-2002[Law et al., 2001a, 2001b, 2001c]

2 Model, Input Data, and Model Evaluations

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1 Model Description and Input Data

[4] In this study, we use the three-layer VIC model with improved algorithms for snow accumulation and ablation [Liang et al., 1994, 1996b; Cherkauer and Lettenmaier, 1999, 2003; Cherkauer et al., 2003]. We use the one-sixteenth degree VIC implementation over the PNW as developed by Hamlet et al. [2012]. This implementation uses three SM layers in which the thin upper layer (~10 cm) simulates “fast” R processes and soil evaporation, and the intermediate (~30 cm) and lower layers (~0.5–2.5 m) represent “slower” storage and drainage processes contributing to base flow [Liang et al., 1996b; Hamlet et al., 2007]. All soil layers potentially contribute to plant transpiration, weighted by root distributions and SM within each layer. In VIC, total ET includes three parts: evaporation from the canopy layer; transpiration from vegetation leaves through stomata; and evaporation from bare soil [Liang et al., 1994, 1996b]. The PET is calculated in the VIC model by using the Penman-Monteith approach [Liang et al., 1994, 1996b]. Transpiration is estimated by the formulation of Blondin [1991] and Ducoudré et al., [1993], which depends on canopy-intercepted water storage, PET, canopy resistance, architectural resistance, and aerodynamic resistance [Liang et al., 1994]. The bare soil evaporation is calculated using PET, the fraction of the surface soil layer that is saturated, and the infiltration capacity of the unsaturated surface layer [Liang et al., 1994, 1996a]. Elsner et al. [2010] implemented VIC to output PET in several alternative forms. The model parameters were originally calibrated for the one-eighteenth degree VIC model by Matheussen et al. [2000] using nine naturalized flow locations in the CRB. These calibrations were used as a starting place by Elsner et al. [2010] to construct a one-sixteenth degree implementation of the model. Additional calibration was performed by Hamlet et al. [2012], and the simulations were validated at 80 streamflow gauging stations in the CRB. In this study, we performed further evaluations of ET and streamflow simulations over the PNW.

[5] The daily meteorological data include daily total P, maximum T (Tmax), and minimum T (Tmin), cover the period from 1915 to 2006 and are gridded at a resolution of one-sixteenth degree spatial resolution. They were developed from daily station observations by Elsner et al. [2010] using techniques developed by Maurer et al. [2002] and Hamlet and Lettenmaier [2005]. The primary station data sources include the National Climatic Data Center Cooperative Observer Network and Environment Canada, and the major sources for data correction and topographic adjustment include the U.S. Historical Climatology Network, the Adjusted Historical Canadian Climate Database, and the Parameter-elevation Regressions on Independent Slopes Model. More detailed descriptions on the climate data set can be found in Elsner et al. [2010] and Hamlet et al. [2012]. Daily wind-speed data from 1949 to 2006 were downscaled from the National Centers for Environmental Prediction–National Center for Atmospheric Research reanalysis products [Kalnay et al., 1996]. For years prior to 1949, wind-speed data were derived from a daily climatology of the 1949-2006 data [Hamlet and Lettenmaier, 2005; Elsner et al., 2010]. We perform our current analysis of simulation results for the period of 1921-2006, allowing for 6 years of model spin-up. We also analyze the results for the period of 1947-2006 to capture the period of one full PDO cycle, as the decadal climate variability associated with the PDO has been shown to have strong impacts on the hydrologic cycle in the region [Hamlet et al., 2007].

2.2 Model Evaluations on Streamflow Simulations

[6] In general, the VIC model accurately simulates monthly streamflow over moderate to large river basins across the western U.S. [Maurer et al., 2002; Hamlet et al., 2005; Mote et al., 2005]. Hamlet et al. [2007], for example, demonstrated the VIC model's capacity in reconstructing hydrologic variability associated with P and T as the primary drivers and capturing long-term trends in the seasonality of streamflow for two rivers: Colorado River at Lee's Ferry and the Arizona and the Sacramento River at Shasta Dam, California.

[7] Here we compared VIC simulations and naturalized observations of long-term average annual discharge and long-term trends in annual discharge over 86 streamflow gauging sites in the PNW (Figure 2). The VIC-simulated gridded surface R and base flow over each watershed were accumulated and assumed to be equivalent to the river discharge from its gauge (outlet) at the annual time scale. Reconstructed natural streamflow observations were downloaded from the Columbia Basin Climate Change Scenarios Project (CBCCSP) website [http://www.hydro.washington.edu/2860/] [Hamlet et al., 2012]. Many of these naturalized data sets for large rivers are based on Bonneville Power Administration (BPA) naturalization efforts [Crook, 1993]. However, some come from observed streamflow records at sites with minimal human influence. Although some natural streamflow records extend back to the early 1900s, as noted above, we only used post-1921 data for model comparisons. Among the naturalized streamflow gauges, we selected 86 that were within ±10% differences between U.S. Geological Survey (USGS) surveyed drainage area and the delineation basin area from the one-sixteenth degree flow-direction data. On average, these observed records of naturalized flow are 60 years long, and the shortest record is 29 years long, which is generally sufficient for meaningful trend analysis.

image

Figure 2. Comparisons of simulated river discharge and R with reconstructed observations over the PNW. The lower right plot box shows the estimated long-term trends versus reconstructed records during observation periods of each site. The upper left small plot box is the R; The reconstructed R is calculated with river discharge divided by drainage area; the simulated R is the average over simulated drainage basin.

Download figure to PowerPoint

[8] Figure 2 shows comparisons between simulated and observed mean annual discharge, mean annual R, and long-term trends of discharge over the selected watersheds. Overall, the simulated mean annual discharge for each water year (October–September) is close to observations (1:1 line). In particular, 67% of the sites have Nash–Sutcliffe Efficiency (NSE) values larger than zero and 37% of the sites with NSE values larger than 0.5. Simulated and observed time series of annual discharge are generally highly correlated with an average squared correlation coefficient (R2) of 0.84. The simulated R values over these representative watersheds have a root mean squared error of 125.8 mm/year. By comparison, long-term trends in R are less accurately reproduced in the simulations for a substantial number of gauging locations; 27% of the sites examined show simulated trends of incorrect sign, although it is important to note that most of these are associated with relatively small trends (<~0.25% per year) (Figure 2).

[9] To evaluate the interannual variability of river discharge from the CRB, we compared the annual simulated river discharge for the CRB at The Dalles, Oregon (USGS #14105700), with USGS streamflow observations (regulated flow), reconstructed natural flows that were obtained from the Climate Impacts Group at the University of Washington (UW) but originally developed using methods of the BPA [Crook, 1993], and the reconstructed historical streamflows by Dai et al. [2009] (Figure 3). The two reconstructed historical annual streamflows are close to USGS observations and the simulated results have very high correlations with UW-reconstructed streamflows (R2 = 0.92 for the period of 1931-1989) and USGS observations (R2 = 0.90 for the period of 1921-2006). In general, VIC overestimates the annual mean discharge over the CRB (by 5% during 1931-1989), but has consistent interannual variability and long-term trends with observations.

image

Figure 3. Comparisons of simulated river discharge at The Dalles near the outlet of the Columbia River with reconstructed historical streamflow from the Climate Impacts Group at UW, USGS observations, and reconstructed streamflow by Dai et al. [2009].

Download figure to PowerPoint

2.3 Evaluation of VIC-Simulated ET

[10] Although prior studies have extensively evaluated the VIC model's capability to reproduce naturalized streamflow [e.g., Hamlet et al., 2012], very few site-level evaluations of VIC-simulated ET have been conducted, and only regional-scale ET has been indirectly evaluated via analysis of the large-scale water balance [Maurer et al., 2002]. Here we compare grid-scale VIC-simulated ET with eddy flux level 4 data products at five sites in the CRB (ftp://cdiac.ornl.gov/pub/ameriflux/data/Level4/AllSites/; http://daac.ornl.gov/FLUXNET/fluxnet.shtml) [Baldocchi et al., 2001] (Figure 1; Table 1). We also evaluate the regional behavior of VIC-simulated ET with eddy covariance measurements that have been upscaled using the empirical model tree ensemble (MTE) method [Jung et al., 2009]. To convert latent energy to liquid water, we used the method developed by Henderson-Sellers [1984], as suggested by the AmeriFlux data support system (http://ameriflux.ornl.gov/).

[11] Figure 4 demonstrates that the VIC model generally captures the seasonal patterns and total annual ET at these five sites, although considerable seasonal bias is present at four out of five sites. For the semiarid Metolius sites (Figure 4b–e), for example, the VIC model generally underestimates total ET, especially during warm/dry seasons. One possible reason is that VIC-simulated ET is very sensitive to SM in the top thin (10 cm) SM layer, as well as the distribution of roots along the soil profile. The missing mechanisms of deep-root effects on ET in semiarid and arid areas in the VIC model could lead to underestimations of ET in this region when the limited SM storage simulated by the model is exhausted. Another possible explanation could be insufficient water storage due to missing large-scale horizontal redistribution of groundwater in the VIC simulations. So, for example, at the Metolius sites, snowmelt from the higher-elevation areas could sustain higher ET than simulated by the VIC model in the warm/dry season via groundwater recharge and deeply rooted plants. The underestimation of ET in warm/dry seasons could also potentially be caused by an overestimation of evaporation from cool-season canopy interception processes, resulting in less water delivery to the soil column for warm-season ET. To accurately simulate ET, high spatial and temporal resolution data for vegetation cover and root distributions in these semiarid regions are also needed [Liang et al., 1996b]. Mismatches between the actual vegetation at the observation sites and simulated vegetation cover in the associated VIC cells are also a possible source of disagreement between simulations and observations (Table 1).

image

Figure 4. Comparisons of simulated (black lines) and observed ET from flux towers (black dots). Site names and time period: (a) Wrc (Wind River Crane Site) in 2001; (b) Me1 (Metolius Eyerly Burn) during 13 April 2004 to 12 April 2005; (c) Me2 (Metolius intermediate aged ponderosa pine) in 2002; (d) Me3 (Metolius second young aged pine) in 2005; and (e) Me5 (Metolius first young aged pine) in 2001.

Download figure to PowerPoint

[12] Simulated spatial patterns of ET, in comparison to upscaled eddy covariance measurements, generally capture higher ET rates in the eastern PNW and Rocky Mountain areas and lower ET rates in semiarid lowland areas (Figure 5a and b). However, the VIC model underestimates ET over lower-elevation semiarid grasslands and agricultural lands and overestimates ET over high-elevation mountain areas in Idaho (Figure 5c and d). For this study, we did not simulate irrigation over irrigated agricultural lands, which could be the explanation for the underestimation of ET in some portions of the lowland sites. The differences in precipitation data sources for our study versus those used for the upscaled ET products could be another reason for these discrepancies.

image

Figure 5. Comparisons between the VIC simulated ET and upscaled ET. (a) Upscaled mean annual ET during 1983-2006 [Jung et al., 2009]. (b) Simulated mean annual ET during 1983-2006. (c) Relative bias of VIC-simulated ET comparing with upscaled ET. (d) Simulated vs. upscaled ET by each grid cell. In (d), the dotted line is the 1:1 line, the solid black line is the linear regression between the simulated ET and upscaled ET over the energy-limited zone, and the solid red line is the regression over the water-limited zone; the red triangle points are grids in the water-limited zone (annual PET ≥ P), and the square points are grids in the energy-limited zone (annual PET < P).

Download figure to PowerPoint

2.4 Simulation Experiments

[13] To quantify the relative contributions of T and P to changes in hydrologic processes, we carried out three separate model simulations as described by Hamlet et al. [2007]. The first simulation was the base run (BR) that uses unaltered historical climate data (i.e., Tmax, Tmin, P, and wind speed) as model input. The second simulation was the fixed P run (FixP), which uses unaltered historical T (i.e., Tmax and Tmin) while fixing P at monthly climatological mean values; that is, the 1915-2006 monthly average total P at each cell is held constant while preserving the daily variations of the original historical data. In the third simulation (FixT), we followed the same procedure, but fixing monthly T (Tmax and Tmin) and leaving P unaltered. For the FixP and FixT simulations, the daily covariance between T, P, solar radiation, and other derived forcing variables are essentially preserved, while removing the long-term trends and monthly variations in the fixed variable during the simulation [Hamlet and Lettenmaier, 2005; Hamlet et al., 2007].

3 Results

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

3.1 Changes in T and P During 1921-2006 Over the PNW

[14] On average, over the PNW from 1921 to 2006, annual average Tavg, Tmax, and Tmin increase by 0.8°C (P < 0.05), 0.4°C (P = 0.12), and 1.2°C (P < 0.05), respectively. The different trends for Tmax and Tmin cause a substantial, significant decrease in the daily T range (DTR) by 0.75°C during 1921-2006, a phenomenon also reported by Hamlet et al. [2007]. The magnitude of warming is not evenly distributed, although the cells with significant warming trends are widely distributed. For example, the lowland areas of western Washington, the Clearwater Mountains in Idaho, and the upper basin of the Snake River show no significant trend (Figure S1a). The cells with decreases in DTR are mainly located in the mountainous areas of the eastern and southern portions of the PNW, while the lower CRB shows no significant trend and, in some cases, an upward trend in DTR during the last 86 years (Figure S1b).

[15] Annual total P increases by 10% or 88 mm year–1 during 1921-2006 (P = 0.07) in the PNW (Figure 6a). Large areas show increases in P, except for a few areas, mainly in and near Washington State (Figure 7a). Due to warming effects, the estimated partitioning of precipitation between rain and snow changes substantially. For instance, rainfall increases by 17.6% or 100 mm year–1 while snowfall decreases by 5% or 13 mm year–1 during 1921-2006 over the study domain. Grid cells with increasing rainfall are widely distributed across the PNW, except for a small area in the upper Snake River Basin and the Salmon River Mountains in Idaho (Figure S1c). The western PNW regions show substantial decreases in snowfall (Figure S1d).

image

Figure 6. Anomalies of climate factors and estimated water fluxes over the PNW during 1921-1006. (a) Annual mean T (°C), annual P (mm yr–1), and estimated annual total R. (b) Estimated annual ET (mm yr–1), potential ET (mm yr–1), and R/P. The baseline is the average during 1921-2006.

Download figure to PowerPoint

image

Figure 7. Trends (% change in 85 years) in precipitation and estimated water fluxes during 1921-2006. (a) P; (b) ET; (c) R; (d) R/P; (e) SM; and (f) SWE. Trends are calculated as total change (i.e., linear slope multiplied by 85 years) divided by the average condition during 1961-1990.

Download figure to PowerPoint

3.2 Long-Term Trends in ET

[16] Our simulation results demonstrate a linear upward trend in ET of 4.5 mm year–1 decade–1 or an 8.8% total increase during 1921-2006 (P < 0.05) (Figure 6b; Table 2). Most areas of the PNW show a positive trend; the highest relative increases are located in the water-limited (warmer and dryer) areas, such as the Central CRB between the Cascade Ranges and the Rocky Mountains (Figures 7b and 8). In general, the dryer and warmer regions have larger increases in ET than the wetter and colder regions (Figure 8).

Table 2. Trends in P and VIC-Simulated ET, R, and R/P in the PNW During the Periods of 1921-2006 and 1947-2006 From Different Simulation Experiments with VIC
PeriodVariableBRFixTFixP
  • *

    Represents a significant trend with P value <0.05.

1921-2006ET (mm century–1)45.2*44.7*10.5*
R (mm century–1)55.856.3–10.0*
T (°C century–1)0.9*0.9*
P (mm century–1)102.6102.6
R/P (century–1)–0.005–0.005–0.015*
1947-2006ET (mm century–1)23.77.722.3*
R (mm century–1)–78.9–63.8–21.6*
T (°C century–1)1.9*1.9*
P (mm century–1)–59.2–59.2
R/P (century–1)–0.043–0.030–0.024*
image

Figure 8. Trends (% change in 85 years) in simulated ET, R, P, and R/P ratio along the climate gradients during 1921-2006. (a) Along the temperature gradient. (b) Along the precipitation gradient.

Download figure to PowerPoint

3.3 Long-Term Trends in R

[17] The interannual and decadal variations of R are much larger than those of ET. R has a regional-average positive trend of 6.6 mm year–1 decade–1, or a total 11.0% increase in the PNW during 1921-2006, but the trend is not significant (Table 2; Figure 6b). Most places in the PNW show R increases, except for much of the lower elevation areas in and around Washington State (Figure 7c). The trend in R is not apparently a function of temperature zone, except that it is slightly more positive in the cold regions (Figure 8a). Along the P gradient, R has the largest increase in the semihumid regions, with mean annual P from 500 to 800 mm (Figure 8b).

3.4 Long-Term Trends in R/P

[18] The region-average annual R/P has no significant long-term trend for 1921-2006, but has a significant negative trend since the mid-1970s (Table 2; Figure 6b). Both positive and negative trends occur in different regions of the PNW, but cancel each other out over the region as a whole (Figure 7d). Overall, the warmer and drier areas (e.g., the Central CRB lowlands) show a substantial decrease in R/P, while the colder and wetter areas (e.g., the high-elevation mountainous areas in Idaho) show a substantial increase in R/P (Figures 7d and 8).

3.5 Seasonal Variations of Water Fluxes and the Driving Forces Over Different Regions

[19] The VIC simulations support the hypothesis that P is the dominant factor controlling the geographic patterns of long-term trends in annual ET and R/P over the PNW (i.e., the results from simulation experiment FixT more closely match the overall results from the BR than do the FixP experiments) (Figure 9). The results also demonstrate that the spatial pattern of the long-term ET trend is overwhelmingly regulated by the availability of water (i.e., P), which is consistent with Hamlet et al.'s [2007] findings in this region and Jung et al.'s [2010] findings at the global scale. More specifically, increases of ET in the PNW during the last 86 years are predominantly a result of increasing P, particularly in the semiarid regions. Because semiarid lowland areas of the PNW are water limited in terms of ET, most of the increasing P is lost through ET, which results in little increase in R and negative R/P trends, even without warming effects (Figure 9f). With warming effects, these negative R/P trends are intensified (Figure 9g).

image

Figure 9. Long-term estimated trends (% change in 85 years) in ET and R/P ratio in annual and seasonal time periods under different simulation experiments. First three columns: trends during 1921-2006 under different simulation experiments. First column: BR. Second column: FixT. Third column: FixP. Fourth column: the simulated trends during 1947-2006 with the BR. First two rows: ET and R/P ratio in annual total, respectively. Second two rows: ET and R/P ratio in the warm season (October–March), respectively. Third two rows: ET and R/P ratio in the cool season (April–September), respectively.

Download figure to PowerPoint

[20] As to the seasonal pattern, VIC results show a slightly higher ET increase during the warm (April–September) season (10% increase) than the cool (October–March) season (6.8% increase) during 1921-2006 (Figures 9i, q, and S3). The trend of R/P shows significant seasonal differences with a 4.4% increase during the cool season and a 15.5% decrease during the warm season (Figures 9m and u). Overall, long-term trends of ET and R/P in the warm season generally control their annual patterns (Figure 9). Figure 9 also demonstrates that the spatial distributions of the warm-season ET and R/P trends are mainly controlled by P (FixT experiment results) (Figures 9i–k and m–o), which was also reported by Hamlet et al. [2007], while during the cool season, trends of ET and R/P are mainly controlled by T (FixP experiment result) (Figures 9q–s and u–w).

[21] The long-term trends of water fluxes in responding to a changing climate are different between water-limited (PET ≥ P) and energy-limited (PET < P) zones (delineated in Figure 10). PET from natural vegetation and P during 1961-1990 were used for generating this water- and energy-limited zone map. Generally, water-limited zones have larger annual ET increases and R/P decreases than energy-limited zones during 1921-2006 (Figure 11a). The seasonal response of ET and R to the changes in T and P in water- and energy-limited zones are similar in direction, but the magnitudes have substantial differences (Figure 11b and c; Table 3).

image

Figure 10. Distribution of water-limited (annual PET ≥ P) and energy-limited zones (annual PET < P) in the Pacific Northwest. The zone map is generated with annual mean PET and P over the period of 1961-1990.

Download figure to PowerPoint

image

Figure 11. Estimated trends (% change in 85 years) in seasonal and annual water fluxes in each zone with different simulation experiments. (a) BR (i.e., transient T and P); (b) FixT; and (c) FixP.

Download figure to PowerPoint

Table 3. Long-Term Trends of Annual and Seasonal Total Water Fluxes, Mean Temperature, and Derived Meteorological Variables (Daily Mean Radiation, Daily Mean VPD) During 1921-2006 in Different Zones of the Pacific Northwest
ZonesVariablesWarm SeasonCool SeasonAnnual
  • *

    Represents a significant trend with P value <0.05.

Water-limited zoneP (mm century–1)43.2*15.658.8*
ET (mm century–1)40.4*6.8*47.2*
PET (mm century–1)–13.59.5*–4.1
R (mm century–1)5.94.910.8
T (°C century–1)0.59*1.24*0.92*
Short-wave radiation (W m–2 century–1)–10.8*–1.9*–6.4*
Long-wave radiation (W m–2 century–1)10.7*5.5*8.1*
VPD (kPa century–1)–0.050.01–0.02
Energy-limited zoneP (mm century–1)95.8*45.8141.6
ET (mm century–1)37.2*6.343.5*
PET (mm century–1)–12.59.0*–3.5
R (mm century–1)51.844.195.9
T (°C century–1)0.67*1.24*0.95*
Short-wave radiation (W m–2 century–1)–11.8*-2.0*–6.9*
Long-wave radiation (W m–2 century–1)9.6*4.8*7.2*
VPD (kPa century–1)–0.040.01–0.02

[22] In water-limited zones, the FixT and FixP experiments demonstrate that increasing T in the cool season (+1.2°C during 1921-2006) raises ET by 13% and PET by 20%, while in the warm season the increasing T (+0.6°C during 1921-2006) does not produce large responses in ET or in PET (Figure 11c; Table 3). The increased ET during the cool season leads to decreased R in the coming warm season (Figure 11c). The increase in T in water-limited regions reduces R by 9% and 2% in the warm season and the cool season, respectively (Figure 11c). When considering only P changes, roughly 80% of increased P during the warm season is lost to the atmosphere through ET; increases in R come primarily from added P during the cool season in water-limited zones (Figure 11b; Table 3).

[23] The seasonal patterns of water fluxes affected by T and P in the energy-limited zones are roughly the same as the water-limited zones, such as reductions in warm-season R due to increasing T (Figure 11c). The major difference between water- and energy-limited zones is that increasing T in the cool season could increase R in energy-limited zones even with slight ET rises (due to reductions in the quantity of water stored in the snowpack), while the warming generally decreases R in water-limited zones (Figure 11b); that is, the lower-elevation areas where changes in the snowpack play a lesser role and it cannot generate effective R.

[24] Overall, the positive trend of P in both water- and energy-limited zones increases ET and R, but the magnitudes of these responses are different (Figure 11a; Table 3). In water-limited zones, only 18% of increased P is converted into R, while in energy-limited zones the conversion rate is 68% (Table 3). The impacts of T on ET predominately occur during the winter season in both water- and energy-limited zones because of increases in cool-season PET (Figure 11c). While, in the warm season, PET slightly decreases in both water- and energy-limited zones (Figure 11a) over the period of 1921-2006. However, we should point out that a substantial positive trend in PET is apparent in the period from 1947-2006 over one entire cycle the PDO from the cool to warm phase [Latif and Barnett, 1994] (Figure 6b).

[25] From mapping the seasonal water and energy limitations during warm and cool seasons, we found that most of the PNW is water limited during the warm season and most of the PNW is energy limited during the cool season (Figure S3). Therefore, Figure 11 not only demonstrates the differences of impacts from T and P on water fluxes in different zones, but also reveals a common pattern among these two regions that T controls cool-season fluxes, while P controls warm-season fluxes (Figure 11).

3.6 Effects of SWE and SM on ET and R/P

[26] Simulated results indicate that the spatial patterns of long-term trends in ET and R/P are inversely correlated (Figure 7), and changes in SWE and SM are playing a role in these changes. The increasing SWE in the eastern mountainous areas and the southern Harney Basin are almost counterbalanced by the decrease in the coastal and central lowland areas of the PNW (Figure 7f). Over the higher elevations of the Rocky Mountains, increasing SWE is generally associated with decreasing ET and increasing R/P (Figure 7b, d, and f). This is consistent with albedo/surface energy effects (e.g., more persistent snowpack in the spring results in lower surface energy available for ET).

[27] Overall, the mean annual volumetric SM content of the total soil column shows small positive trends, with an average of 4.4% increase over the PNW during 1921-2006 (P = 0.15). The upper and intermediate SM layers during March–May have the largest increase in SM (Figures S3f and g). The simulations show wetter soils over large areas in the Rocky Mountains, while in the northern central lowland between the Cascade Range and the Rocky Mountains the soil becomes drier due to increasing ET and little change in P. The dryer soil could further decrease R through increasing infiltration and water-holding capacities.

3.7 Comparison of 1921-2006 and 1947-2006 Trends

[28] In long-term trend analysis, decadal variations of water variations due to multidecade climate cycles, such as the PDO, can obscure the effects primarily due to anthropogenic climate change if the period of analysis is not over an entire PDO cycle [Latif and Barnett, 1994]. Following Hamlet et al. [2007], we calculated ET and R/P trends for the complete cycle of PDO from the cold to warm phase during 1947-2006 [Mantua et al., 1997]. Mean annual T increased by 1.9°C century–1, while the annual total P decreased by 59.2 mm year–1 century–1 (Table 2; Figure 6). Simulated results from VIC demonstrate that ET increased considerably due to warming even while P decreased, which resulted in a substantial decrease in R/P (Table 2; Figure 9d and h). While both effects contribute to changes in ET and R, P is still the dominant controlling factor (contributing about 70%) to the decrease of R and R/P during 1947-2006 in the PNW as a whole (Table 2).

4 Discussion

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1 Implications for the Future

[29] Climate change has resulted not only in warming (which is unequivocal at the global scale [IPCC, 2007]), but also in an intensification of the global water cycle during the last century due to a warming-induced increase in the atmosphere's water-holding capacity with impacts on P generation [Huntington, 2006]. Previous studies reported on the importance of warming in affecting the shape of the seasonal hydrograph (with less water availability during the summer of peak water demand) in snowmelt-dominated mountainous areas of the western U.S. [Hamlet et al., 2005, 2007]. However, as indicated by this study, a combination of the effects of changes in T and P vary spatially and seasonally. Even with an increase in P over water-limited regions in the warm season, R may not increase accordingly because of the water demand for ET, not to mention that we are anticipating fairly strong decreases in warm-season P in the future, as estimated by Mote and Salathe [2010]. Only an increase in P during the cool season could generate substantial effective R because of an insignificant loss in ET. Conversely, in energy-limited regions, an increase in P and an earlier shift in the seasonal hydrograph due to early snowmelt results in more R and a rise in R/P, especially during the cool season (Figures 7, 9, and 11). Because of the robust prediction of warming in the near future (i.e., global climate models (GCMs) are consistent in predicting warming across the region), but inconsistent in annual P projections, although with some consistency in projecting decreases in summer precipitation and increases in precipitation in the other seasons [Mote and Salathe, 2010], we can expect that in the water-limited areas of the PNW, R/P will continue to decrease. It is in these water-limited areas that there is a growing concern over dwindling water availability during the growing season; reductions in warm-season R/P and increasing water demand, particularly for irrigation, will exacerbate current water scarcity problems. In the energy-limited areas, the warming effects on annual R/P are less certain because R/P depends more strongly on changes in P; however, it is more certain that R/P will increase during the cool season in this region because of the influence of warming on the snowpack and therefore the seasonal hydrograph. However, there are fewer water scarcity concerns in these energy-limited regions.

4.2 Dynamics of Budyko Space over the PNW During 1921-2006

[30] Budyko space, the relationship between the mean annual evaporative index (ET/P) and the mean annual dryness index (PET/P), has been widely used to assess the relative contributions of water supply and energy demand to ET variations [Budyko, 1974; Gerrits et al., 2009; Williams et al., 2012; Zhang et al., 2012]. Figure 12 shows the distributions of annual ET/P and PET/P in water- and energy-limited zones, respectively. The simulated results indicate that, in both the water- and energy-limited zones, the slope of the ET/P vs. PET/P relationship has significantly increased in the most recent 43 years of the simulations (P < 0.05), which means that, under the same dryness condition, the evaporative index has increased (i.e., a greater percentage of P is evaporated) (Figure 12). This is likely due to the combined effects of warmer conditions in the cool season (increasing PET in the energy-limited seasons) and increasing trends in warm-season P. If the trend continues in the future, water-limited zones may experience future reductions in the R/P ratio relative to current conditions.

image

Figure 12. Dynamics of Budyko space. Points represent pairs of annual ET/P and PET/P in each zone during the first and second half of 1921-2006. Solid lines are the regression over the period of 1964-2006; dotted lines are the regression over the period of 1921-1963.

Download figure to PowerPoint

[31] On the other hand, future projections of increasing cool-season precipitation and decreasing summer precipitation for the PNW [Mote and Salathe, 2010] may tend to bring the opposite effect, reducing the ET/P ratio in water-limited areas because P is increasingly distributed to seasons with relatively low surface energy and PET, as discussed above. It is also worth noting that projections of changing summer ET in water-limited areas are essentially opposite to those observed in the historical period. GCM/VIC projections suggest drier summers (and therefore lower summer ET in water-limited areas) [Hamlet et al., 2012], while summer precipitation has been increasing overall historically, resulting in increases in summer ET in water-limited areas (Figure 9). By comparison, for energy-limited areas, summer ET has increased historically, and the GCM/VIC projections show broadly similar effects associated with ongoing regional warming (resulting in increasing ET in energy-limited areas).

4.3 Uncertainties

[32] Land-use practices and vegetation dynamics can also influence ET and R/P [Vorosmarty and Sahagian, 2000; Vorosmarty et al., 2000; Jackson et al., 2001; Foley et al., 2005; Liu et al., 2008]. The interactions between climate change, hydrologic processes, land-use practices, and vegetation dynamics should be explicitly incorporated into future model development to better predict how changes in water quantities and regimes affect terrestrial ecosystems (and vice versa) at regional scales. For example, the warming-induced early onset of plant growth could increase ET and hence intensify soil dryness in late summer if P does not change; this dynamic is modulated by the impacts of P and T changes on snowpack and SM processes. This dynamic is not captured in this study because the monthly vegetation parameters (e.g., leaf area index) are held constant throughout the simulation period; that is, the current vegetation parameters have no interannual variations or long-term trends. As another example, the CO2 fertilization effects on stomatal conductance have been shown, in some studies, to increase water-use efficiency and therefore R [Farquhar and Sharkey, 1982; Field et al., 1995; Gedney et al., 2006]. However, its overall impacts on the water cycle at regional and global scales are still uncertain [Milly et al., 2005; Piao et al., 2007; Dai et al., 2009]. The VIC model used in this study, which does not have a dynamic vegetation component, has no specific consideration of the CO2 fertilization effect on plant growth and therefore on ET.

[33] Another limitation in this modeling study is that other climate variables in these offline simulations, such as surface short- and long-wave radiation and vapor pressure deficit (VPD), are estimated from T and P with empirical algorithms [Liang et al., 1994, 1996b; Hamlet et al., 2005; Adam et al., 2009; Elsner et al., 2010]. VPD has been used as a dominant factor representing water stress in estimating remote-sensing-based ET [Mu et al., 2007]. Our modeling work also showed that annual PET has a very high correlation with VPD in both water- and energy-limited zones, with an R2 of 0.96. PET trends in both water- and energy-limited zones are generally negative in the warm season and positive in the cool season, which corresponds to the decreases in derived radiation and VPD in the warm season and increases in the cool season (Table 3). As discussed by Hamlet et al. [2007], the uncertainties from the derived meteorological variables, just based on Tmax and Tmin with Thornton and Running's [1999] method and the coarse resolution of the reanalysis data used for wind-speed data, could affect our simulation results. Therefore, there is a need to better represent meteorological conditions, as well as to capture land-surface feedback effects through coupled land-surface/atmospheric models that explicitly represent dynamic vegetation processes.

[34] Furthermore, similar to most other hydrological models, VIC was calibrated only with river streamflow data and not with ET observations. According to Zhang et al. [2009], the combination of calibrations against remotely sensed ET and streamflow can improve general R/P models of ungauged catchments. Our studies also indicated considerable bias from ET estimations at seasonal patterns. Further modeling studies and analyses should use available ET observations to calibrate key parameters related to ET processes, such as the maximum stomatal conductance for individual biome types. Likewise, water limitations in late summer in the model simulations are clearly introducing a low bias in the simulation of ET over most of the sites examined.

[35] In summary, to assess the impacts of climate change on water resources, an integrated Earth systems model that couples the processes of water, ecosystem dynamics, biogeochemical cycles, atmospheric chemistry, regional climate, and human-natural system interactions is critical. However, this should be done with careful consideration of the scales of each of these processes. For example, the discrepancy between eddy-flux observations and VIC-simulated ET is most likely due to a scale mismatch. As proposed by Wood et al. [2011], hyper-resolution land-surface models and spatial data are needed to monitor the terrestrial water cycle from local to global scales. However, Beven and Cloke [2012] argue that improvements in our understanding of scaling effects on spatial heterogeneities are more important than just improving the simulation resolution. Therefore, Earth systems model development efforts for improving our understanding of the water cycle must be performed in tandem with experimental studies to ensure that the appropriate scales for each of these processes are captured.

5 Conclusions

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[36] In this study, we evaluated and applied a well-calibrated macroscale hydrological model (VIC) with high-resolution meteorological data to investigate how climate has affected long-term and seasonal trends on ET, R, and R/P in the PNW during 1921-2006. Comparisons between simulated ET and (both in situ and upscaled) observed ET indicate that VIC captures the annual magnitude and general seasonal patterns in ET across the study domain, with the exception that it underestimates ET in semiarid regions and during the dry season. However, there are some discrepancies between these independent data sets because of the differences in precipitation data and spatial scales. Evaluation against reconstructed natural streamflow demonstrates that the VIC model is able to simulate the general patterns of interannual variations, magnitude, and long-term trends for selected streamflow gauging locations in the region. However, there are still some limitations in the model regarding reconstructing long-term trends over historical periods, and further development is needed to improve how vegetation processes are treated in the model as well as how the meteorological data are processed.

[37] Simulation results indicate a significant upward trend in ET over the PNW during 1921-2006. However, the trends vary spatially and seasonally. Generally, water-limited (warmer and dryer) regions, such as the Central CRB, have greater ET increases and R/P decreases than the energy-limited (colder and wetter) regions, such as the western PNW and the mountainous areas in the east. Generally, the R/P ratio increases in the cool season and decreases in the warm season due to warming effects on the form of P and the reduction of seasonal snowpack. Simulation results indicate that P is the dominant factor controlling the geographic patterns of trends in annual ET and R/P in the PNW. However, the effects of T and P on water fluxes also vary spatially and seasonally. Over the PNW, most areas are water limited in the warm season and energy limited in the cool season. Therefore, long-term trends of ET and R/P in the warm season are mainly controlled by P, while in the cool season they are mainly controlled by T. With a continued warming trend, R will continue to decrease in the warm season and increase in the cool season.

[38] With Budyko space analyses on the distribution of ET/P and PET/P over energy- and water-limited zones, we find that ET/P is less sensitive to PET/P in water-limited zones than in energy-limited zones. However, in the more recent half of our study period, the slope between ET/P and PET/P increases, with a steeper slope increase in water-limited zones, indicating that under the same dryness condition the water loss through ET is higher.

[39] During the most recent complete cycle of the PDO (1947-2006), increasing ET and decreasing P results in a substantial decrease in R and R/P, suggesting that a long-term warming trend is exacerbating the water shortage that is already being experienced in some areas. Given that continued warming in the future is projected with relatively high certainty, water-limited regions are expected to experience further decreases in R/P because of the increase in ET during the cool season due to warming effects, and the increased warm-season P, if any, will likely be counterbalanced by increases in warm-season ET. Furthermore, most GCMs project that summer precipitation will decrease in the future. Therefore, these water-limited areas are the regions where we anticipate that water stress will be the greatest. As these are also the regions that already experience water shortage problems, current water scarcity in these areas will be exacerbated in the future, which has important implications for irrigated agriculture, ecosystems, hydropower production, and other in- and out-of-stream water needs.

[40] To increase the model's reliability in simulating the terrestrial water cycle under multienvironmental stresses, development of high-resolution data sets and coupling with ecosystem dynamics, soil biogeochemical cycles, and regional climate processes in an integrated regional Earth systems framework are critical.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[41] We would like to thank Marketa Elsner (UW) and Qichun Yang (Auburn University) for providing data. Thanks also go to Dr. Martin Jung (Max Planck Institute for Biogeochemistry) for providing the MTE data. Eddy-flux tower sites are part of the AmeriFlux network, and we gratefully acknowledge the efforts of researchers at these sites. Thanks also go to our anonymous reviewers and to Julian Reyes and Kirti Rajagopalan (Washington State University) for their thoughtful comments. Sites are funded through grants from the U.S. Department of Energy Office of Biological and Environmental Research, unless otherwise noted. The flux level 4 data being used in this study were downloaded from the Carbon Dioxide Information Analysis Center (ftp://cdiac.ornl.gov/pub/ameriflux/data/Level4/AllSites/). Reconstructed naturalized streamflow for model evaluations were downloaded from the CBCCSP website at http://www.hydro.washington.edu/2860/. These materials were produced by the Climate Impacts Group at UW in collaboration with the Washington State Department of Ecology, BPA, Northwest Power and Conservation Council, Oregon Water Resources Department, and the B.C. Ministry of the Environment. This study has been supported by the United States Department of Agriculture (grant no.: 20116700330346 for Earth System Modeling).

References

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Model, Input Data, and Model Evaluations
  5. 3 Results
  6. 4 Discussion
  7. 5 Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

All supporting information may be found in the online version of this article.

FilenameFormatSizeDescription
Appendix_rev1.docxWord document1318KSupporting Information

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.