In bad waters: Water year classification in nonstationary climates


Corresponding author: S. E. Null, Department of Watershed Sciences, Utah State University, 5210 Old Main Hill, Logan, UT 84322, USA. (


[1] Water year indices and drought indices are helpful for categorizing water years into similar types, allowing water managers and policymakers to quantify years, visualize variability, and guide water operations. Many water management decisions, such as environmental flow requirements and water supply allocations, are based on water year type designations. They vary by region and index, but are defined by runoff in the current water year compared to average historical runoff, with numerical thresholds categorizing year types. California's Sacramento Valley and San Joaquin Valley Indices are used as case studies to examine how climate change affects indices. Modeled streamflow for 1951–2099 from the climate-forced Variable Infiltration Capacity hydrologic model estimate potential changes in runoff and water year type frequency. We show that the frequency of water year types changes significantly with climate change and different strategies to adapt water year classification indices to climate change affect water allocations as much as the impacts from changing hydroclimatic conditions.

1. Introduction

[2] Water management frameworks that were designed assuming stationary climate conditions will be increasingly difficult to implement in nonstationary climates and present a barrier to climate change adaptation. Given that many water allocation decisions are based on arbitrary thresholds of historical data that are stationary and limited in duration, we must ask: will status quo approaches, which use fixed management prescriptions based on hydrologic stationarity, be used in a future with changing hydroclimatic conditions? Will such approaches result in ad hoc and reactionary management solutions poorly suited to meet complex water demands? Or will adaptive, anticipatory solutions be enacted to more equitably sustain water resources in a nonstationary environment?

[3] These questions are particularly pressing for water resources management of large rivers. Existing research has shown that climate warming is expected to change the timing, magnitude, frequency, and form of precipitation, impacting water supply [Milly et al., 2008; Barnett et al., 2008; Rajagopalan et al., 2009], hydropower generation [Viers, 2011; Madani and Lund, 2009], and flood protection [Zhu et al., 2007; Kundzewicz et al., 2008], as well as species, ecosystems, and ecosystem services [Dudgeon et al., 2006; Shaw et al., 2011]. Water allocation frameworks and environmental regulatory laws were designed assuming climatic stationarity, and more research is needed for adapting them for climate warming [Miller et al., 1997; Cohen et al., 2000]. This is a notable information gap as water managers must respond to changing hydroclimatic conditions to sustain human and environmental water demands.

[4] Water year classification systems and hydrologic indices are common for water planning and management because they simplify complex hydrology into a single, numerical metric that can be used in rule-based decision making [Dracup et al., 1980; Heim, 2002; Quiring, 2009]. Estimated unimpaired runoff for a water year is categorized by year type, such as wet, dry, or normal, compared to historical averages. Year type classification is tied to water resources planning, helping to answer the question of whether there is “enough” water [Redmond, 2002], and allocations for various water uses are adjusted based on water year type (WYT) designation. WYTs inform water allocation decisions for water supply, hydroelectric power generation, reservoir storage, and environmental protection [Simpson et al., 2004]. Many drought and water year indices exist, including the Palmer Drought Severity Index [McKee et al., 1993], Standard Precipitation Index [McKee et al., 1993], Surface Water Supply Index [Shafer and Dezman, 1982], Reclamation Drought Index [Weghorst, 1996], and deciles [Gibbs and Maher, 1967]. All indices compare current year streamflow to the historical average and may not capture changing trends in nonstationary climates.

[5] In this paper, we illustrate this problem using California's Sacramento and San Joaquin Valley Indices as case studies to evaluate: (1) how water year classification indices may function with climate change, (2) how indices can be updated for changing hydroclimatic conditions, and (3) how anticipated changes affect allocations to competing water users.

[6] Previous research indicates that the distribution of WYTs is not stationary through time [Booth et al., 2006; VanRheenen et al., 2004; Vicuna, 2006]. Existing studies have suggested changing WYT thresholds to maintain historical distribution of WYTs for human water uses [Miller et al., 1997; VanRheenen et al., 2004] or changing the seasonal weight coefficients in water year indices to reflect climate-driven changes in inflow timing [Vicuna, 2006]. Other research has focused on improved understanding of the effects of El Nino-Southern Oscillation events or including the paleoclimatic record in hydrologic indices [Anderson et al., 2001; Verdon-Kidd and Kiem, 2010]. There has been little research on climate change impacts to environmental flows and adaptation strategies, except for general agreement that competition could increase for minimum instream flow allocations [VanRheenen et al., 2004; Meyer et al., 1999], increasing economic costs of environmental requirements [Harou et al., 2010].

2. Background

2.1. California's Sacramento–San Joaquin Bay Delta

[7] The Sacramento and San Joaquin Rivers drain the west slope of the Sierra Nevada mountain range, then merge and flow through the Sacramento–San Joaquin Bay Delta (Bay Delta) to the Pacific Ocean (Figure 1). Measured historical (1906–2009) average annual flow for both basins is 29,600 million cubic meters (mcm), with the Sacramento River contributing approximately three fourths of the flow. No observed long-term trend of increasing or decreasing discharge is evident for the Sacramento River using measured data from 1906 to 2009 (p = 0.773) or for the San Joaquin River using measured data from 1901 to 2009 (p = 0.285) (Figure 2). These rivers provide approximately 43% of California's total average annual surface runoff and provide drinking water for about 24 million residents. This region has a Mediterranean-montane climate and a notably variable hydrology, where interannual variability is less predictable than seasonal or geographic variability.

Figure 1.

Sacramento and San Joaquin watersheds with gage locations.

Figure 2.

Observed discharge for the SVI (1906–2009) and SJI (1901–2009). p values reflect the probability of the linear trend is nonstationary.

[8] The Bay Delta serves as the primary water wheel for California, and is ground zero for extreme sociopolitical conflict over water transfer schemes [Connick and Innes, 2003], habitat degradation [Light and Marchetti, 2007], and failure to proactively manage California's water resources [Null et al., 2012]. The state relies on pumping facilities in the Bay Delta to divert water south to urban and agricultural users. This necessitates continuous freshwater purging of the seasonally brackish delta. The Bay Delta is an environmentally sensitive area, which provides habitat for fish and wildlife (some species are protected under the state and federal Endangered Species Acts), and holds public trust value for common use (State Water Resources Control Board (SWRCB),, 2000). Adequate environmental water allocations are needed to protect habitat and biodiversity, and regulatory requirements sometimes drive water management in this water scarce region. Thus, it is not surprising that for decades the Bay Delta has been the source of lawsuits, headaches, and hand-wringing [Connick and Innes, 2003].

2.2. Water Year Indexing

[9] The Sacramento Valley Index (SVI) and San Joaquin Valley Index (SJI) were designed with estimated unimpaired historical hydrology and are used in a complex and evolving water delivery allocation scheme shaped by operational constraints, regulatory restrictions, and objective demands (SWRCB, online, 2000). Numerical thresholds separate WYTs, set by weighted winter and spring runoff volume for major rivers, as well as the previous year's index (Table 1). The indices were intended to divide runoff into wet, above normal, below normal, dry, and critical categories (originally weighted approximately 30%, 20%, 20%, 15%, and 15%), respectively, of the historic record [California Department of Water Resources (CDWR), 1989, 1991] (Figure 3). The SVI was developed by the SWRCB in 1989 from a previously existing Sacramento River classification scheme, and the SJI was developed in 1991 [CDWR, 1989, 1991].

Figure 3.

Percentile of WYT with observed data (1906–2000).

Table 1. Sacramento Valley Index and San Joaquin Valley Index Year Type Classification Thresholds in Millions of Cubic Meters (mcm) and Millions of Acre-Feet (maf)
Water Year TypeSacramento Valley Index mcm (maf)San Joaquin Valley Index mcm (maf)
Wet≥11,348 (≥9.2)≥4,687 (≥3.8)
Above normal>9,621 to <11,348 (>7.8 to <9.2)>3,824 to <4,687 (>3.1 to <3.8)
Below normal>8,018 to ≤9,621 (>6.5 to ≤7.8)>3,084 to ≤3,824 (>2.5 to ≤3.1)
Dry>6,661 to ≤8,018 (>5.4 to ≤6.5)>2,590 to ≤3,084 (>2.1 to ≤2.5)
Critical≤6,661 (≤5.4)≤2,590 (≤2.1)

2.2.1. Sacramento Valley Index

[10] The SVI (also known as the “4 River Index” and the “40-30-30 Index”) uses the sum of calculated monthly unimpaired runoff from the following gages: Sacramento River above Bend Bridge, Feather River at Oroville, Yuba River near Smartsville, and American River below Folsom Lake (California Data Exchange Center (CDEC),, 2010) (Figure 1). It is calculated using equation (1), and year type classification is based on the thresholds in Table 1. The term for the previous year's index is a proxy for carryover reservoir storage on system capability [CDWR, 1989].

display math(1)

2.2.2. San Joaquin Valley Index

[11] The SVI and SJI were intentionally given different weights on each segment of the index to account for snowmelt-dominated runoff and occasional large winter floods that provide less water deliveries in the San Joaquin basin [CDWR, 1991]. San Joaquin watersheds are generally higher elevation with geology that limits infiltration of groundwater. The SJI (or the “60-20-20 Index”) uses the sum of unimpaired runoff from Stanislaus River below Goodwin Dam, Tuolumne River below La Grange Dam, Merced River below Merced Falls, and San Joaquin River inflow to Millerton Lake (CDEC, online, 2010) (Figure 1). It is calculated using equation (2), and year type thresholds are based on the values in Table 1.

display math(2)

2.3. Historical Water Year Thresholds

[12] For planning purposes, year types are set by forecasts beginning in February (and updated monthly through May), although for this study we use calculated unimpaired runoff (CDEC, online, 2010) or modeled data. Values of the SVI and SJI account for geographic variation in streamflow, so the SVI has greater thresholds than the SJI (Table 1). The historical relative frequency of year types also varies slightly between the SVI and SJI (Figure 3). For example, the threshold for critically dry year types falls at the 13th percentile of the observed period of record for Sacramento Valley streamflow, but at the 17th percentile for San Joaquin Valley streamflow. Operationally, this means there is a slightly higher chance that any year will be critically dry in the San Joaquin Valley, and more environmental flow is allocated from Sacramento Valley rivers than the San Joaquin rivers. The opposite is true for dry and below-normal year types.

2.4. Water Allocation

[13] Generally, the SVI and SJI determine WYT for California's two largest water projects, the State Water Project (SWP) and the federally funded Central Valley Project (CVP) to allocate water for out-of-stream agricultural users in the Bay Delta, environmental flows, and export limits to water users south of the Bay Delta (SWRCB, online, 2000). Environmental flow objectives for the region include Bay Delta outflow, flow-dependent salinity and water temperature objectives, and instream flows for rivers in the Sacramento and San Joaquin watersheds. The SVI is the most important for managing the Bay Delta, although the SJI impacts environmental flow objectives and the “Eight River Index” uses both Sacramento and San Joaquin system runoff to determine salinity in Suisun Bay. The SVI and SJI were designed with the understanding that water shortages would occur in critical year types (M. Roos, personal communication, 2011). In reality, water scarcity exists in critical, dry, and below normal year types during August, September, and October (SWRCB, online, 2000). The SVI and SJI directly influence water policy in the state through regulatory restrictions and directly affect dozens of federal, state, and local agencies [Simpson et al., 2004].

3. Methods

3.1. Climate-Forced Hydrology

[14] Downscaled, climate-forced streamflow estimates are from the Variable Infiltration Capacity (VIC) model, a large-scale, distributed, physically based hydrologic model that balances surface energy and water over a grid [Maurer et al., 2002; Liang et al., 1994]. VIC uses subgrid representation for vegetation, soils, and topography to retain local variability for partitioning precipitation into runoff and infiltration and uses nonlinear representation for simulating baseflow. Data were downscaled using bias correction and spatial downscaling (BCSD), a statistical downscaling method that preserves monthly climate patterns between course and fine resolutions [Maurer and Hidalgo, 2008]. Water routing was postprocessed to estimate streamflow at river outlets using an algorithm developed by Lohmann et al. [1996] [as cited in Cayan et al., 2008]. Parameterization for deriving streamflow is identical to that used by VanRheenen et al. [2004] for the Sacramento–San Joaquin basin. VIC is commonly used to assess hydrologic effects of climate change in the western United States [VanRheenen et al., 2004; Maurer et al., 2002; Cayan et al., 2008; Vicuna et al., 2007].

[15] This application of VIC uses a 1/8° spatial grid and a daily timestep (later aggregated to a monthly timestep) for 1951–2099 water years. Twelve VIC runs were analyzed, with climate input data from six general circulation models (GCMs) for the A2 and B1 emissions scenarios (Figure 4). These models and emissions scenarios are consistent with research from California Energy Commission's climate change research center [Cayan et al., 2012; California Energy Commission (CEC),, 2012]. Modeled water years are separated into three time periods and GCMs simulated historical conditions from 1951 to 2000 to represent interannual variability [Cayan et al., 2008]. Here 2001–2050 represents the near-term future, and 2051–2099 are far-term estimates of runoff conditions. Water years (October–September) are used throughout this paper.

Figure 4.

Climate scenarios, GCMs, and modeled time periods.

[16] Differences between emissions scenarios are due to uncertainty in human actions such as population growth and greenhouse gas (GHG) emissions, while differences in GCMs are from uncertainty in climate models such as representation of physical processes and sensitivity to GHG forcings. The A2 scenario has more severe climate change, assuming maximum carbon dioxide (CO2) emissions of 850 parts per million (ppm), continuously increasing global population, and slow economic growth. The B1 scenario is more moderate, assuming maximum CO2 emissions of 550 ppm, global population that peaks mid-century and later declines, and global sustainability solutions that introduce resource-efficient technology [Intergovernmental Panel on Climate Change, 2000]. Other potential future changes (such as land cover change) are ignored here.

3.2. Statistical Analysis

[17] Nonparametric Kruskal–Wallis and Wilcoxon tests were used to determine whether differences in mean runoff between modeled and observed data or different modeled time periods were statistically significant. Distributions of the modeled historical 1951–2000 datasets (modeled A2 and B1 simulations) were tested against observed historical data for the same time period. Modeled distributions from GCM-driven VIC simulations for each emissions scenario were pooled in statistical analyses and some figures to capture uncertainty associated with individual climate models. Pairwise comparisons were used to determine whether simulated SVI and SJI indices changed through time for simulated 1951–2000, 2001–2050, and 2051–2099 time periods. Lastly, we examined frequency of WYTs for different time periods with chi-square tests using observed values for expected frequency. Statistical analyses were conducted using R [R Core Team, 2012].

4. Results

4.1. Water Year Index Response to Climate Change

[18] We evaluated the response of water year indices to climate change using a multiple model, multiple emissions scenario approach (Figure 4). To determine whether model run distributions (n = 12, 6 GCMs and 2 emissions scenarios) were different than historical observations for the 1951–2000 modeled historical period for each basin, we used nonparametric Kruskal–Wallis tests with adjustments for multiple comparisons where modeled water year index values were considered treatments and observed historical data were considered controls (kruskalmc procedure) [Giraudoux, 2012]. Using a two-tailed test at the p = 0.001 significance level, we found no significant differences between modeled and observed index values and conclude that the modeled index values represent measured conditions.

[19] To assess whether GCM-driven, VIC-modeled future discharge and corresponding WYT index values were statistically different from the modeled historical equivalent, we compared modeled results for near-term (2001–2050) and far-term (2051–2099) time periods against each other and to modeled historical conditions (1951–2000) for each basin and emissions scenarios, aggregating GCMs as ensembles. Nonparametric Wilcoxon rank sum tests, adjusted for multiple comparisons and false discovery [Benjamini and Hochberg, 1995], were applied to each pairwise comparison (Table 2). Index values for both basins and both emission scenarios were statistically different between historical and far term, as well as near term and far term (Table 2). Only the A2 scenario for SJI was significantly different in the historical versus near term (p = 0.0002).

Table 2. Pairwise Comparisons of GCM-Driven, VIC-Simulated Seasonal Discharge and Index Values for Sacramento River and San Joaquin River Basins for Two Emissions Scenarios Using a Wilcoxon Rank Sum Test, Corrected for Multiple Comparisons and False Discovery Rates [Benjamini and Hochberg, 1995]a
BasinEmissions ScenarioOctober–March DischargeApril–July DischargeANN Index
1951–2000 vs. 2001–20501951–2000 vs. 2051–21002001–2050 vs. 2051–21001951–2000 vs. 2001–20501951–2000 vs. 2051–21002001–2050 vs. 2051–21001951–2000 vs. 2001–20501951–2000 vs. 2051–21002001–2050 vs. 2051–2100
  1. a

    p values given in bold indicate <0.005.

San JoaquinA20.980.980.986.2e-05<2e-166.2e-110.0002<2e-162.5e-09

[20] Seasonal discharge simulations for winter (October–March) runoff showed no significant differences, regardless of time period, basin, or emissions scenario (Table 2). This reflects variable, but generally stable, hydroclimatic conditions during the winter season. Discharge during the spring snowmelt period (April–June) shows statistically significant differences between all time periods across all basins and under all emissions scenarios (p < 0.005) (Table 2). This suggests that index calculations are sensitive to spring discharge and this season may drive hydrological nonstationarity through time. For future projections, modeled ensembles represent uncertain estimations of future conditions and may not be truth-centered [Annan and Hargreaves, 2011]; however, significant differences in far-term streamflow, driven by spring season runoff, correspond to existing climate-induced volume and timing runoff research for this region [VanRheenen et al., 2004; Stewart et al., 2004; Null et al., 2010]. Table 3 gives modeled average seasonal and annual runoff for the three modeled time periods.

Table 3. Modeled Average Annual Flow by Time Period
Index and DataAverage Annual Flow (mcm)
Annual runoffA224,78123,90522,560
October–March runoffA214,43214,90115,419
April–July runoffA29,0547,7836,056
San Joaquin
Annual runoffA27,4386,7845,908
October–March runoffA22,7632,8993,084
April–July runoffA24,2933,5892,566

4.2. Climate Change Affects the Frequency of Water Year Types

[21] We used chi-square tests of simulated and observed WYT frequencies to determine if the abundance of each WYT is expected to change through time. In the near term (2001–2050), three or fewer simulations were significantly different than the historical period (p < 0.01) (Table 4). However, the frequencies of four or more of the six GCMs were significantly different from historical frequencies (p < 0.01) (Table 4) when far-term (2051-2100) simulations for each basin index were compared with the historical time period. The relative frequency that water years are classified as each WYT is illustrated with histograms by modeled time period for SVI (Figure 5) and SJI (Figure 6) (note scale change between figures). Observed data are included for the 1951–2000 historical period (Figures 5a and 6a) for visual corroboration of modeled and observed data. Differences between emissions scenarios (warm hues versus cool hues) are due to uncertainty in human actions such as population growth and GHG emissions, while differences in GCMs (variability within the warm hues or cool hues) are due to uncertainty in climate models, such as representation of physical processes and sensitivity to radiative forcings.

Figure 5.

SVI relative frequency histograms for (a) 1951–2000, (b) 2001–2050, and (c) 2051–2099.

Figure 6.

SJI relative frequency histograms for (a) 1951–2000, (b) 2001–2050, and (c) 2051–2099.

Table 4. Chi-Square Results Test Simulated Future Frequency of WYTs Against Simulated Historical Frequency for all SVI and SJI GCMs and Emission Scenariosa
ESGCM1951–2000 vs. 2001–20501951–2000 vs. 2051–2100
  1. a

    p values given in bold indicate <0.005.

A2CNRMCM3χ2 = 1.958χ2 = 5.124
p = 0.743p = 0.275
A2GFDLCM21χ2 = 5.867χ2 = 16.648
p = 0.209p = 0.002
A2MIROC32MEDχ2 = 29.506χ2 = 49.333
p = 0p = 0
A2MPIECHAM5χ2 = 11.89χ2 = 30.861
p = 0.018p = 0
A2NCARCSM3χ2 = 13.97χ2 = 29.321
p = 0.007p = 0
A2NCARPCM1χ2 = 11.325χ2 = 32.171
p = 0.023p = 0
B1CNRMCM3χ2 = 9.397χ2 = 3.876
p = 0.052p = 0.423
B1GFDLCM21χ2 = 8.711χ2 = 15.486
p = 0.069p = 0.004
B1MIROC32MEDχ2 = 22.6χ2 = 44.23
p = 0p = 0
B1MPIECHAM5χ2 = 14.653χ2 = 53.067
p = 0.005p = 0
B1NCARCSM3χ2 = 15.487χ2 = 12.17
p = 0.004p = 0.016
B1NCARPCM1χ2 = 7.638χ2 = 4.479
p = 0.106p = 0.345
A2CNRMCM3χ2 = 4.785χ2 = 29.335
p = 0.31p = 0
A2GFDLCM21χ2 = 7.554χ2 = 37.96
p = 0.109p = 0
A2MIROC32MEDχ2 = 26.837χ2 = 116.72
p = 0p = 0
A2MPIECHAM5χ2 = 31.025χ2 = 28.424
p = 0p = 0
A2NCARCSM3χ2 = 7.789χ2 = 34.268
p = 0.1p = 0
A2NCARPCM1χ2 = 2.122χ2 = 2.661
p = 0.713p = 0.616
B1CNRMCM3χ2 = 26.492χ2 = 22.089
p = 0p = 0
B1GFDLCM21χ2 = 3.421χ2 = 28.547
p = 0.49p = 0
B1MIROC32MEDχ2 = 50.267χ2 = 125.524
p = 0p = 0
B1MPIECHAM5χ2 = 7.594χ2 = 46.249
p = 0.108p = 0
B1NCARCSM3χ2 = 17.003χ2 = 6.748
p = 0.002p = 0.15
B1NCARPCM1χ2 = 3.956χ2 = 13.506
p = 0.412p = 0.009

[22] Figures 5 and 6 demonstrate that the relative frequency of WYTs is expected to shift throughout the 21st century. For the SVI, modeling suggests a more even distribution of WYTs in each category by the end of the century (Figure 5). Projections from both the A2 and B1 emissions scenarios indicate the Sacramento Basin will likely have more dry and critical years, and less normal and wet years throughout the current century (Figure 5, Table 5), although some categories—such as wet years in 2001–2050—suggest considerable uncertainty between individual models. By the latter half of the 21st century (2051–2099), 6–10% more critical years and 10–12% more dry years could occur if water year thresholds remain the same. The more drastic changes occur if the higher CO2 emissions and increasing population assumptions of the A2 emissions scenarios are realized. For the SJI, many more years fall into the critical category with fewer years in all other year types, particularly toward the end of this century (Figure 6). Results indicate a 28–35% increase in critical water years by the last half of this century, with the larger changes under A2 assumptions (Figure 6, Table 5).

Table 5. Percentage of Years in Each Water Year Type by Modeled Time Period and Emissions Scenarioa
 1951–2000 (%)2001–2050 (%)2051–2099 (%)
  1. a

    Values given in italics are percent change from historical period.

Critical8.78.311.3 (2.7)6.7 (−1.7)18.4 (9.7)14.0 (5.6)
Dry7.710.012.0 (4.3)15.7 (5.7)19.4 (11.7)20.1 (10.1)
Below normal23.321.323.3 (0.0)17.3 (−4.0)18.7 (−4.6)19.4 (−1.9)
Above normal21.022.716.7 (−4.3)20.7 (−2.0)12.9 (−8.1)18.4 (−4.3)
Wet39.337.736.7 (−2.7)39.7 (2.0)30.6 (−8.7)28.2 (−9.4)
Critical26.026.041.3 (15.3)35.3 (9.3)60.9 (34.9)54.1 (28.1)
Dry13.012.311.0 (−2.0)12.7 (0.3)8.2 (−4.8)11.9 (−0.4)
Below normal19.319.715.7 (−3.7)14.0 (−5.7)10.5 (−8.8)10.9 (−8.8)
Above normal13.713.39.3 (−4.3)12.0 (−1.3)8.5 (−5.2)10.9 (−2.5)
Wet28.028.722.7 (−5.3)26.0 (−2.7)11.9 (−16.1)12.2 (−16.4)

[23] The SVI and SJI are numerical indices, so they can continue to be used with severe climate change without altering how indices are calculated. However, WYT classifications and threshold definitions will likely become less representative with climate change, and in fact, may lose meaning if most years fall into the same year type classification. By the end of this century, the distribution of particular year types is anticipated to be significantly different from the historical record.

5. Discussion

5.1. Adapting Water Year Type Indices for Climate Change

[24] Water allocations are affected by the climate change impacts discussed above, as well as strategies to adapt WYT indices to hydroclimatic changes. Maintaining the status quo of fixed WYT thresholds under nonstationary hydroclimate regimes is likely to result in system-wide inequities over the far term and could have severe consequences for water management. While it is easy to acknowledge that some level of adaptation is inevitable, and is likely to include a variety of approaches, the scope, direction, and timeliness of adaptation are less clear. This section presents two extreme adaptation methods: (1) maintaining historical WYT thresholds with changing flow regimes for a new distribution of WYT, or (2) maintaining the historical distribution of WYTs by modifying WYT thresholds for changing flow regimes. Other adaptation options exist, such as changing the seasonally weighted coefficients of each index in response to climate-driven runoff timing but are not discussed here.

[25] If current WYT thresholds are maintained (proactive changes are rarely made in the Bay Delta's challenging political and regulatory climate, effectively maintaining the status quo), substantially more dry and critically dry years are anticipated to occur as explained above and further illustrated with the modeled distribution of WYTs using historical thresholds (Figures 7a and 7b) (black bars show thresholds, and wider bars quantify uncertainty between the A2 and B1 runoff estimates). This would disproportionately impact environmental uses (i.e., Bay Delta outflows are reduced by approximately 36% between wet and dry years, whereas exports and out-of-stream agricultural uses have relatively constant deliveries among year types (Bay Delta Conservation Plan,, 2012)). With persistent dry conditions under this scenario, California risks failing to provide adequate baseflow and hydrologic variability to support various ecosystems, and failing to protect species and habitat as required by the state and federal ESA and the Clean Water Act. Additional regulatory drivers include oversight by the SWRCB to uphold public trust values, hydropower relicensing through Federal Energy Regulatory Commission [Viers, 2011], and the emergence of state interest in safeguarding public trust values through the California Fish and Game Code [Baiocchi, 1980].

Figure 7.

Modeled distribution of water year types using historical thresholds, where black bands show uncertainty between A2 and B1 projections for (a) SVI and (b) SJI (C is critically dry, D is dry, BN is below normal, AN is above normal, and W is wet).

[26] Conversely, WYT thresholds could be redefined to reflect changes in climate, recognizing that the normal years of the future may resemble the critical or dry years of the past century (Figures 8a and 8b). The thresholds determining year types could be lowered to maintain the historical distribution of water years with climate-driven modeled data [CDWR, 1989, 1991]. For example, for modeled SJI 1951–2000 data, the threshold for critically dry year types should be set at about 2097 mcm for the historical distribution of 17% of years to be in the critically dry year type. The threshold would have to be reset between 1110 and 1357 mcm for 17% of years to be in the critically dry category by 2051–2099. If volumetric environmental flow requirements tied to each WYT remain the same, much of the burden of climate change would fall on human water uses under this scenario because environmental requirements for wet years would remain unchanged, but future wet years would be drier than historical wet years. In this case, regulatory restrictions could increasingly drive water policy in California. If environmental flow allocations were also altered to reflect overall drier conditions, water scarcity from climate change could be shared more equitably among water uses. Wider black bars in Figure 8 represent uncertainty and perhaps model bias in this sort of threshold change, which could be explored in greater detail with climate models, downscaling methods, or stochastic parameterization of VIC.

Figure 8.

Modeled change to water year classification thresholds using historical percentages of years per category, where black bands show uncertainty between A2 and B1 projections for (a) SVI and (b) SJI (C is critically dry, D is dry, BN is below normal, AN is above normal, and W is wet) (note scale change between figures).

5.2. Climate Change Adaptation Strategies Affect Water Allocations

[27] Water year classification forms the framework for water exports through SWP and CVP pumping facilities, out-of-stream agricultural uses in the Bay Delta, and environmental flow objectives in the Bay Delta and Sierra Nevada rivers. We estimated water deliveries as percentages of historical annual flow for wet, above normal, and dry years using values from Bay Delta Conservation Plan (online, 2012), with the above normal percentage allocations also used for below normal years and dry year percentage allocations applied for critical years. This illustrates how adaptation strategies affect water allocations differently using simplified deliveries as a percentage of runoff for water exports, out-of-stream users, and Bay Delta outflow. Adapting water allocation frameworks to climate change is as important as anticipating climate change impacts because adaptation strategies directly affect water allocations by shifting climate-driven water scarcity and associated economic costs between water users. Yet, most research focuses on climate change impacts to water resources, while adaptations to climate change are underrepresented in the scientific literature.

[28] Modeled data indicate a reduction of approximately 3700 mcm in unregulated inflow to the Bay Delta by the end of the 21st century (ignoring direct precipitation and contributions from eastside tributaries). Unimpaired estimates are used in indices, which also ignore upstream diversions and all other water regulation. If WYT thresholds are redefined so that the historical distribution of water years is preserved (Figure 8), then water reductions are split between exports to central and southern California (−3%), out-of-stream agricultural Bay Delta uses (−2%), and Bay Delta environmental outflows (−14%) (Bay Delta outflows lump undeveloped—or uncontrolled—water with environmental flow requirements) (Figure 9). However, if current WYT thresholds are maintained (Figure 7), then the burden of climate-driven water scarcity falls entirely on environmental outflow through the Bay Delta (−16%), while the percentage of average annual flow to exports (+2%) and out-of-stream uses (+4%) increase somewhat to preserve relatively constant deliveries in drier years (Figure 9). The WYT framework, and how it could be altered to reflect climate change, directly affects water winners and losers in the state.

Figure 9.

(a) Example Bay Delta water balance, where historical average annual water allocation for Delta Outflow, Exports, and Consumptive Use are given (mcm), and average annual change in allocation for 2051–2099 are given in bold values (mcm) for maintaining historical WYT thresholds (orange columns) and maintaining historical WYT distribution (red columns). Black bars quantify uncertainty between A2 and B1 emissions scenarios, and all values use the average of A2 and B1 emissions scenarios. Arrow and bar size represents flow percentage. (b) Approximate water allocations as a percentage of annual flow estimated from the Bay Delta Conservation Plan (Bay Delta Conservation Plan, online, 2012), which ignores upstream diversions, direct precipitation, and eastside tributaries. Bay Delta outflow lumps undeveloped water with some environmental flows.

6. Major Findings and Implications

[29] Water year typing has inherent limitations. Water year indices typically focus on runoff volume, with less emphasis on timing [Vicuna, 2006], although the SVI and SJI have seasonally weighted coefficients that incorporate runoff timing. More research is needed to update coefficients to reflect climate-driven shifts in runoff timing. Accurately describing the ecological differences between year types is another current information gap. Likewise, the volume and timing of undeveloped water will likely change and merits further research. Separating undeveloped water from managed environmental flow allocations would also help quantify anticipated flow changes [Null, 2008].

[30] The modeled data used here suggest that 34–38% of years may be classified as dry and critically dry years in the SVI, and 66–69% may be classified as dry and critically dry years in the SJI by the end of the 21st century without changing thresholds values. These findings indicate that hydrologic and drought indices should be adapted for changing hydroclimate conditions, particularly in large rivers which are relied upon for water supply, hydropower generation, flood protection, and ecosystem function. Where water allocations are reliant on WYT classifications, deliveries are sensitive to both climate change impacts and adaptations. Hydroclimate changes may not be sufficient to prompt adaptation in the short term, in part, because bimodalities in wet and critically dry frequencies provide little evidence to a public focused on “normal” water years. In the long term, maintaining the status quo (historical WYT threshold values) versus redefining thresholds to reflect the historical distribution of WYTs affects which water users face climate-driven water scarcity. Climate, WYTs, and water allocation decision making are interrelated.

[31] Extensive studies have been conducted to develop adequate environmental flows [Acreman and Dunbar, 2004; Arthington et al., 2006] and to determine how much water ecosystems need [Richter et al., 1997]. But the quality, accuracy, and utility of WYT indices, like the SVI and SJI, which in practice determine how much water ecosystems receive, have yet to be extensively studied with projected hydroclimatic changes. Failing to recognize how probabilities of year types may shift with climate change introduces error and uncertainty into long-term regulatory stability and may not preserve the hydrologic variability needed to maintain ecosystem health.

[32] Aquatic ecosystems depend on hydrologic variability to preserve function and integrity [Richter et al., 1997]. In developed systems, water managers have some responsibility to maintain ecological function and health of aquatic and riparian systems with anticipated hydroclimate changes. In a future where more than half of all years are designated as critically dry, larger instream flows may be warranted to manage hydrologic variability if we are to maintain existing ecosystems. However, preserving historical assemblages of species may not be the ecosystems we choose to manage for in the future [Lund et al., 2010]. As a society, we tend to preserve ecosystems that we are accustomed to, although that may not be realistic in a future with severe climate change [Hanak et al., 2011]. Future conditions, as well as unanticipated events such as exotic species invasions, food webs collapse, or changing migration barriers, could alter the historical range of species. Changing frequencies of WYTs may present an opportunity to openly recognize that ecosystems are already heavily managed and to more explicitly decide what complement of ecosystems, functions, and species we opt to manage for. We recommend an adaptive WYT framework, with a formal process to review year type thresholds periodically to ensure that water is allocated according to water rights, regulatory requirements, and desired ecosystems.


[33] This work was funded by the California Energy Commission and a white paper describing research and results was submitted to the funding agency. We would like to thank Maury Roos and Jay Lund for critical feedback, Guido Franco for oversight with the CEC PIER program, and past UC Davis student Danielle Salt for processing data. We also thank Mary Tyree and the California Nevada Applications Program/California Climate Change Center (directed by the Climate Research Division of Scripps Institute for Oceanography) for sharing VIC climate data, as well as Ed Maurer and Dan Cayan for their advice. The majority of this research was completed while Sarah Null was a postdoctoral scholar at UC Davis.