Predicting regime shifts in flow of the Gunnison River under changing climate conditions

Authors

  • W. Paul Miller,

    Corresponding author
    1. Colorado Basin River Forecast Center, National Oceanic and Atmospheric Administration, Salt Lake City, Utah, USA
    • Corresponding author: W. P. Miller, Colorado Basin River Forecast Center, National Oceanic and Atmospheric Administration, 2242 West North Temple, Salt Lake City, UT 84116, USA. (paul.miller@noaa.gov)

    Search for more papers by this author
  • Gina M. DeRosa,

    1. U.S. Bureau of Reclamation, Lower Colorado Region, Water Operations Control Center, Boulder City, Nevada, USA
    2. Department of Hydrology and Water Resources, University of Arizona, Tucson, Arizona, USA
    Search for more papers by this author
  • Subhrendu Gangopadhyay,

    1. U.S. Bureau of Reclamation, Technical Service Center, Water Resources Planning and Operations Support, Denver, Colorado, USA
    Search for more papers by this author
  • Juan B. Valdés

    1. Department of Hydrology and Water Resources, University of Arizona, Tucson, Arizona, USA
    Search for more papers by this author

Abstract

[1] Water resource management agencies have traditionally relied upon past observations of historical hydrologic records for long-term planning. This assumption of stationarity, that the past is representative of the future, may no longer be valid under changing climate conditions. The Gunnison River Basin contributes approximately 16% of the annual natural streamflow within the Upper Colorado River Basin, affecting water supply availability over the entire Colorado River Basin. Recent studies indicate that streamflow over the Gunnison River Basin, a subbasin within the Colorado River Basin, may decrease on the order of 15% through 2099. Further study has developed a methodology to statistically characterize the risk of regime shifts using observations of past streamflow through the use of a two-parameter gamma distribution. In this study, regime characteristics derived using a paleoreconstruction of streamflow over the Gunnison River Basin are compared regime characteristics developed using 112 projections of future hydrology to better understand how the frequency and duration of persistent dry and wet periods may change as the impacts of climate change are realized over the subbasin. Results indicate that under changing climate conditions, similar regime characteristics may be expected through 2039. However, between 2040 and 2099, more frequent and persistent dry regimes increase on the order of 50%. Conversely, wet regimes are expected to be shorter and less frequent than observed over the paleoclimatic record, decreasing in frequency by as much as 50%.

1. Introduction

[2] The Colorado River Basin spans much of the American West, providing water for over 30 million people and irrigating over 16,000 km2 of land in the upper basin states of Wyoming, Utah, Colorado, New Mexico, and the lower basin states of California, Nevada, Arizona, and the country of Mexico. The U.S. Department of the Interior, Bureau of Reclamation (Reclamation) manages water resources over the Colorado River Basin to provide water for municipal and agricultural use, as well as hydropower facilities with a generating capacity in excess of 4200 MW [U.S. Department of the Interior, Bureau of Reclamation, Lower Colorado Region, 2012]. Most of the water stored over the Colorado River Basin is the result of spring (April–July) snowmelt driven by accumulated winter snowpack throughout the Upper Colorado River Basin [e.g., Timilsena and Piechota, 2008]; numerous studies have indicated that as the impacts of climate change are realized, the character of precipitation and snowmelt may change [e.g., Clow, 2010; Kalra and Ahmad, 2011; McCabe et al., 2007; Painter et al., 2010]. Approximately 92%, or just over 18,500 million cubic meters (mcm), of the annual natural flow realized over the Colorado River Basin is contributed from the upper basin [Prairie and Callejo, 2005]. Reclamation, and other resource management agencies, have traditionally assumed that past realizations of hydrology are representative of future conditions; however, as the impacts of climate change are realized, resource managers now realize that the past may no longer be representative of the future [e.g., U.S. Department of Interior, Bureau of Reclamation, 2009; Milly et al., 2008].

[3] The Gunnison River Basin is an approximately 20,720 km2 drainage area located in the southwest portion of Colorado and contributes approximately 16% (2960 mcm) of the annual natural flow within the Upper Colorado River Basin (Figure 1) as calculated at the outlet of the Gunnison River watershed at Grand Junction, Colorado [Prairie and Callejo, 2005]. Reclamation manages a collection of reservoirs collectively known as the Aspinall Unit to meet downstream flow requirements, hydroelectric power needs, and to provide for endangered fish and their habitat in addition to other approved uses [U.S. Department of the Interior, Bureau of Reclamation, Upper Colorado Region, 2009]. Miller et al. [2011] developed 112 projections of streamflow through 2099 over the Gunnison River Basin under changing climate conditions using bias-corrected, statistically downscaled (BCSD) projections of future climate [Maurer et al., 2007] and the National Weather Service (NWS) River Forecasting System (RFS) [National Oceanic and Atmospheric Administration, National Weather Service, 2005] provided by the Colorado Basin River Forecasting Center (CBRFC). It was found that streamflow over the Gunnison River Basin may decrease by approximately 15% (approximately 490 mcm) by 2099.

Figure 1.

The Gunnison River Basin is a 20,720 km2 area located in the southeast portion of the Upper Colorado River Basin. The Colorado River Basin spans seven states across the western United States and is divided into an upper and lower region (inset).

[4] Gangopadhyay and McCabe [2010] developed a methodology to compute the risk of shifting streamflow regimes (i.e., wet or dry flow periods) over the Upper Colorado River Basin given the length and time of past observations of regime intervals. By using paleohydrologic reconstructions of streamflow over the Upper Colorado River Basin and statistical methods developed by Enfield and Cid-Serrano [2006], robust information regarding the hydrologic regime over the basin could be developed and applied to resource management. The methodology is applicable to data sets with substantial decadal to multidecadal variability, such as streamflow over the Colorado River Basin [e.g., Aziz et al., 2010; Matter et al., 2010; Piechota et al., 1997; Tingstad and MacDonald, 2010; Tootle and Piechota, 2006].

[5] As the impacts of climate change are realized, it is important to investigate how shifting hydrologic regimes may change in the future. Here, the methodology presented in Gangopadhyay and McCabe [2010] is applied to a paleoreconstruction of streamflow developed over the Gunnison River Basin [Woodhouse et al., 2006] and compared to an ensemble of future streamflow projections presented in Miller et al. [2011]. An ensemble of 112 future streamflow projections allows for the development of a broad range of streamflow scenarios that encompass various emissions scenarios, model assumptions, and methodologies. In this study, the probability of regime change over the reconstructed paleohydrologic record is compared to the probability of regime change over an ensemble of projected streamflow in an effort to investigate how the impacts of climate change may affect hydrologic variability over the Gunnison River Basin. These results provide additional information for water resource managers to consider as climate change presents new resource management challenges.

2. Streamflow Data

[6] An ensemble of 112 projections of water year natural streamflow spanning 2010–2099 at the outlet of the Gunnison River Basin located at Grand Junction, Colorado, was developed by Miller et al. [2011]. Each projection was developed using the NWS RFS provided by the CBRFC and forced using 112 BCSD projections of future climate (i.e., precipitation and temperature) derived from global climate projections from the World Climate Research Programme's (WCRP) Coupled Model Intercomparison Project phase 3 (CMIP3) data set [Maurer et al., 2007]. Climate data from the WCRP CMIP3 was statistically downscaled through the bias correction and spatial disaggregation method described in Wood et al. [2002, 2004] in which the distribution of temperature and precipitation data from general circulation models (GCMs) is mapped and scaled to the distribution of regional temperature and precipitation data (i.e., the data set first presented in Maurer et al. [2002]). Distributed data from the WCRP CMIP3 data set was then aggregated to develop inputs for the lumped NWS RFS over three headwater basins over the Colorado River Basin, including the Gunnison River Basin. Streamflow projections are derived using projections developed from 16 global circulation models and 3 emissions scenarios (A2, B1, and A1B) [Nakićenović and Intergovernmental Panel on Climate Change, 2000; Solomon and Intergovernmental Panel on Climate Change, Working Group I, 2007]. Through the use of multiple GCMs spanning a broad range of potential emissions scenarios, a range of potential streamflow conditions under changing climate conditions over the Gunnison River Basin may be developed. The methodology for developing these streamflow projections is described in detail in Miller et al. [2011].

[7] A paleoreconstruction of natural streamflow of the Gunnison River near Grand Junction, Colorado spanning 1569–1997 was developed by Woodhouse et al. [2006] and is utilized for historical comparison to future projections of streamflow. This paleoreconstruction of streamflow is calibrated to natural flow published and updated annually by Reclamation [Prairie and Callejo, 2005]; as such, this historical reconstruction was updated through 2008 using the latest natural streamflow information available from Reclamation. As described in Gangopadhyay and McCabe [2010], each of the 112 projections of future streamflow, and the paleoreconstruction of streamflow developed by Woodhouse et al. [2006], was filtered using a low-pass Kaylor filter [Kaylor, 1977] to retain frequencies in the time series that were equal to and slower than 0.1 cycles per year. Figure 2 illustrates the raw historical and projected time series used as well as the subsequently filtered data derived here.

Figure 2.

The plot shows (a) the paleoreconstruction of water year streamflow at Grand Junction, CO, as defined by Woodhouse et al. [2006], (b) the filtered values derived after subjecting the paleoreconstructed data to a Kaylor Filter; positive values are indicative of a wet regime, whereas negative values are indicative of a dry regime, (c) gray lines illustrating an ensemble of 112 traces of projected natural streamflow at Grand Junction, from Miller et al. [2011] and resultant mean indicated by a black line, and (d) the filtered values derived after subjecting the projected natural streamflow data to a Kaylor Filter; a zero line, shown in black, is provided for reference.

3. Methodology

[8] To determine wet and dry regimes over the Gunnison River Basin, this study follows the methodology described in Gangopadhyay and McCabe [2010]. Both the paleoreconstruction and each of the 112 projections of streamflow time series over the Gunnison River Basin were filtered using a low-pass Kaylor filter [Kaylor, 1977] to retain frequencies that were equal to and slower than 10 years. Regimes are then defined by counting the number of years between successive zero crossings of the filtered time series. Figure 2b illustrates wet regimes in gray and dry regimes in over the paleohydrologic record. Figure 2c illustrates each of the 112 filtered time series derived from the ensemble of streamflow projections. Both the paleoreconstruction and ensemble of projected streamflow provide sufficiently long records with numerous wet and dry intervals of varying lengths that are useful for statistical analysis.

[9] The time interval (T) between wet and dry regimes is assumed here to be a stochastic process that can be fitted to a given probability distribution [Enfield and Cid-Serrano, 2006; Gangopadhyay and McCabe, 2010]. As in Gangopadhyay and McCabe [2010], a two-parameter gamma distribution is used to study the shift between wet and dry regimes. Gamma distributions have been previously used to model the distribution of regime intervals [Gangopadhyay and McCabe, 2010; Salas et al., 2005]. From the defined probability distribution function (P), a cumulative distribution function may be derived. Using an assumption of timepass, water resource managers may use the cumulative distribution to determine the conditional probability of a regime shift over an assumed subsequent period of time [Enfield and Cid-Serrano, 2006; Gangopadhyay and McCabe, 2010]. As such, given the time elapsed since a previous regime shift, t1, the conditional probability that a regime shift will occur within a future time horizon, t2, can be expressed as equation (1) from Gangopadhyay and McCabe [2010] and described prior by Enfield and Cid-Serrano [2006].

display math

where t = t1 + t2 is the current climate regime interval and Γ[t] is the two-parameter gamma cumulative distribution function.

[10] A comparison of shifting regimes within the paleohydrologic streamflow record to those projected in an ensemble of future streamflows is presented in the next section. Anomalies within the projected data set are calculated with respect to trends present within the original climate projections used to derive the streamflow projections used here; in other words, the data were detrended to remove bias from model simulations of future climate [U.S. Department of Interior, Bureau of Reclamation, 2010, 2011; Miller et al., 2011].

4. Comparison Between Paleohydrologic and Projected Streamflow

[11] Wet and dry regime durations developed from filtered paleohydrologic and projected ensemble streamflow data were compared. Regime characteristics from the paleohydrologic data set spanning water years 1608–1997 were compared to regime characteristics derived from the ensemble of projected streamflows over the Gunnison River Basin. The ensemble period of record (water years 2010–2099) is further divided into three, projected 30 year periods for comparison. As such, results presented here compare regime characteristics derived from the paleohydrologic streamflow record to regime characteristics projected between 2010 and 2039, 2040 and 2069, and 2070–2099. For each 30-year period, 5-, 10-, and 15-year regimes (both wet and dry) durations were recorded for each ensemble trace. Overlapping regime intervals are considered; for example, covering a 6 year time frame over which each annual value is characterized as “dry,” two 5-year dry regimes are recorded.

[12] The frequency and duration of wet and dry regime intervals were further compared. Wet and dry regime durations over the projected ensemble traces were separated and compared to wet and dry regime characteristics from the paleohydrologic record. Five-, 10-, and 15-year dry and wet regimes for each data set were compared by fitting a two-parameter gamma exceedance distribution function (GEDF). The GEDF is considered to be a Poisson process to determine the likelihood of time to the kth event [Montgomery and Runger, 2011]. Here, the event referenced is the hydrologic regime shift.

5. Results and Discussion

[13] The distribution of wet and dry regimes derived from the paleohydrologic streamflow record over the Gunnison River Basin was fitted using maximum likelihood estimation to a two-parameter gamma distribution model with shape and scale sample parameters equal to 2.57 and 3.63, respectively. This distribution is fitted to the paleohydrologic record, indicating that the duration of most regimes over the Gunnison River Basin is between 6 and 10 years and only rarely exceeds 15 years (Figure 3a). The median distribution of regime intervals over the projected ensemble is presented in Figure 3b. In comparison with the distribution of regime intervals throughout the paleohydrologic record, there is a marked decrease in regime intervals of lengths between 11 and 15 years and an increase in the occurrence of regime intervals of lengths less than or equal to 5 years. A fitted gamma distribution model developed using median-scale and shape parameters derived from the ensemble of projected streamflows indicates a tendency similar to the paleohydrologic record (Figure 3c). It is important to note that the median characteristics were selected because the pairing of the median shape and median-scale parameter is representative of the pairings derived over the ensemble; the pairing of the average shape and scale parameters is not.

Figure 3.

(a) The distribution of regime intervals and gamma distribution model derived using the paleohydrologic streamflow record over the Gunnison River Basin, (b) the median distribution of regime intervals and standard error at the 95% confidence interval derived from an ensemble of projected flows over the Gunnison River Basin in conjunction with the gamma distribution derived using the median scale and shape parameter, (c) a comparison of the 112 gamma distribution models developed from each trace of the projected ensemble of streamflow compared to the gamma distribution models presented in Figures 3a and 3b, and (d) the distribution of scale and shape parameters developed during the course of this study.

[14] Figure 3d describes the distribution of scale and shape parameters derived from the ensemble of projected streamflows over the Gunnison River Basin compared to the same parameters derived from the paleohydrologic record. The average shape parameter is approximately the same as the shape parameter derived from the paleohydrologic record; however, the average-scale parameter is nearly double that of the value derived over the paleohydrologic record. Median shape and -scale values derived from the projected ensemble are similar to those derived from the paleohydrologic record (Table 1). The gamma distribution derived using the paleologic streamflow record was compared to the gamma distribution derived from the ensemble of projected flow values using the Kolmogorov-Smirnov (KS) test. The KS test indicated statistically significant differences (at P ≤ 0.05) between the two distributions. Comparison of the gamma distribution derived using the paleohydrologic streamflow record and the gamma distribution derived from each member of the projected streamflow ensemble also yielded statistically significant differences for all 112 projected traces.

Table 1. A Summary of the Distributions of Shape and Scale Parameters Derived by Fitting a Two-Parameter Gamma Distribution to an Ensemble of 112 Projections Over the Gunnison River Basina
 Scale and Shape Parameter Characteristicsb
ShapeScale
  1. a

    Using a single trace of paleohydrologic streamflow, shape and scale parameters equal to 2.57 and 3.63, respectively, were developed.

  2. b

    Derived from 112 Trace Ensemble of Projected Streamflow.

Maximum11.5950.72
Median2.433.65
Minimum0.490.52
Average2.676.60

[15] The risk of future regime shift given the number of years since the prior regime shift and a specified number of years into the future may be derived. Figure 4 compares the risk of a future regime shift derived using parameters of a gamma distribution model derived using the paleohydrologic streamflow record and the risk derived using median gamma distribution model parameters derived using the ensemble of projected streamflow. As Figure 4 illustrates, regime intervals are expected to be more persistent than those observed within the paleohydrologic record. As an example, given a current 10-year regime, the risk of a regime shift within the upcoming 5 years is approximately 70% given the paleohydrologic record; however, the risk of a regime shift within 5 years decreases to approximately 60%.

Figure 4.

Compared distribution of the probability of a regime shift risk (%) occurring within a future time frame (ordinate) given that some time (abscissa) has elapsed since the last regime period. Paleohydrologic record distribution is based on the two-parameter gamma distribution with scale and shape parameters X and Y, respectively. Projected ensemble median record distribution is based on the median scale (X2) and shape (Y2) parameter developed using the two-parameter gamma distribution.

[16] The distribution of regime shift risk over the entire ensemble of projected traces and its comparison to risk observed over the paleohydrologic record is described in Figure 5. As Figure 5 illustrates, the risk observed over the paleohydrologic record is slightly lower than the median value for regime intervals of less than 10 years when projecting forward 5–10 years. The median risk from the projected ensemble is similar to that observed over the paleohydrologic record for longer (greater than 10 years) regime intervals. For longer regime intervals, the risk of regime shift ranges approximately between 20% and 95%.

Figure 5.

These boxplots illustrate the risk outlooks over the 112 projections of streamflow within the Gunnison River Basin. The lower and upper box extents indicate the range between the 25th and 75th percentiles, respectively, and the thick horizontal lines indicate the median values. The outermost lower and upper thin lines indicate the 10th and 90th percentiles, respectively. The “X” indicates the risk as observed over the paleohydrologic record.

[17] The comparison of the paleohydrologic record to the projected records show the probability of dry regimes in the projected data set to have a higher frequency and persistence than those observed over the paleohydrologic record (Figure 6). Wet regimes also indicate extended regime intervals though reduced frequency; wet regimes rarely exceed longer than 15 years. Over the paleohydrologic record, 10 consecutive years of a 5-year dry regime displays a probability of about 33%, whereas the projected data set from 2010 to 2039 shows a similar probability of about 30% (Figure 6). From 2040 to 2069, the projected results show an increase in probability of about 62% of 10 years of a 5-year dry regime, and 2070–2099 displays a greater increase in probability of about 80% likelihood of 10 years of a 5-year dry regime (Figure 6).

Figure 6.

GEDFs consisting of 90 years of projected natural annual streamflow segmented into 30-year periods and 390 years of paleoreconstructed natural annual streamflow. The gray lines indicated the projected data and the black lines indicate the paleohydrologic data. The consecutive exceedance of (a) 5-year dry periods, (b) 10-year dry periods, and (c) 15-year dry periods.

[18] Similarities of regime interval length between paleohydrologic and projected data through 2039 are shown in both regimes, but deviates slightly in the 15-year dry and wet regimes (Figures 6 and 7). Distributions show the most likely regime shifts will maintain a time interval of 5–10 years (Figures 3a and 3b), but the two data sets differ in the density of the 11–15 year regime interval.

Figure 7.

GEDFs consisting of 90 years of projected natural annual streamflow segmented into 30-year periods and 390 years of paleoreconstructed natural annual streamflow. The gray lines indicated the projected data and the black lines indicate the paleohydrologic data. The consecutive exceedance of (a) 5-year dry periods, (b) 10-year dry periods, and (c) 15-year dry periods.

[19] Exceedance plots show a trend toward more persistent and frequent in future years because the likelihood of future wet regimes from 2040 to 2099 is near zero (Figure 7). The exceedance plots illustrated in Figures 6 and 7 display a probabilistic outlook for the Gunnison River Basin to experience drier conditions than what it experienced over the paleohydrologic record. Regime shifts projected over the next 30 years are not expected to be significantly different, but begin to vary markedly after 2040.

[20] It is important to note that this study does not explicitly or directly compare the paleohydrologic streamflow profiles developed by Woodhouse et al. [2006] to the ensemble of projected streamflows developed by Miller et al. [2011]; rather, the duration and frequency of wet and dry regimes within each of these data sets are compared. During the derivation of each of these data sets, calibration and validation methods were applied to relate both the streamflow reconstructions and streamflow projections to derived, historical natural flow over the Colorado River Basin and published by Reclamation (available at: http://www.usbr.gov/lc/region/g4000/NaturalFlow/current.html) [Prairie and Callejo, 2005]. The streamflow data sets utilized in this study provide an opportunity to compare past hydroclimatic variability to possible future conditions from which some insight as to the changing character of this variability may be gained. Additionally, as the spatial and temporal resolution of GCMs continues to improve, projections of regional hydroclimatic events such as drought may be refined with increased accuracy.

[21] As the impacts of climate change are realized, resource managers must adapt to changing hydroclimatic regimes. The results of this analysis indicate higher potential risk for more frequent and persistent dry conditions over the Gunnison River Basin under projected climate conditions. Although this analysis did not investigate the impact of prolonged dry regimes to the operation of the Gunnison River Basin, decreased water supply may reduce the flexibility of reservoir operations within the basin to achieve environmental and water demand needs. Future analysis may investigate the impacts to the entire Colorado River Basin region. This analysis further presents a methodology through which resource managers may assess long-term risk. For instance, based on a current drought regime, a resource manager may calculate the statistical likelihood of a more favorable wet regime within a specified amount of time based on an observational record. Interesting future research could focus on specific basin thresholds to withstand prolonged and changing regime intervals and the risk of exceeding those thresholds in the future. As projections of future climate are further refined and developed, resource managers must attempt to incorporate leading science to continue to make well-informed and broadly beneficial decisions.

Acknowledgments

[22] The authors thank the U.S. Bureau of Reclamation and the National Oceanic Atmospheric Administration, Colorado Basin River Forecast Center, for supporting this study.

Ancillary