Journal of Hydrology (2008) 354, 90– 101
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/jhydrol
Doomed reservoirs in Kansas, USA? Climate change and groundwater mining on the Great Plains lead to unsustainable surface water storage T.H. Brikowski
*
Department of Geosciences, The University of Texas at Dallas, P.O. Box 830688, Richardson, TX 75083-0688, United States Received 3 August 2007; received in revised form 20 February 2008; accepted 27 February 2008
KEYWORDS Global climate change; Arid zones; Water supply; Groundwater mining
Streamflow declines on the Great Plains of the US are causing many Federal reservoirs to become profoundly inefficient, and will eventually drive them into unsustainability as negative annual reservoir water budgets become more common. The streamflow declines are historically related to groundwater mining, but since the mid-1980s correlate increasingly with climate. This study highlights that progression toward unsustainability, and shows that future climate change will continue streamflow declines at historical rates, with severe consequences for surface water supply. An object lesson is Optima Lake in the Oklahoma Panhandle, where streamflows have declined 99% since the 1960s and the reservoir has never been more than 5% full. Water balances for the four westernmost Federal reservoirs in Kansas (Cedar Bluff, Keith Sebelius, Webster and Kirwin) show similar tendencies. For these four, reservoir inflow has declined by 92%, 73%, 81% and 64% respectively since the 1950s. Since 1990 total evaporated volumes relative to total inflows amounted to 68%, 83%, 24% and 44% respectively. Predictions of streamflow and reservoir performance based on climate change models indicate 70% chance of steady decline after 2007, with a 50% chance of failure (releases by gravity flow impossible) of Cedar Bluff Reservoir between 2007 and 2050. Paradoxically, a 30% chance of storage increase prior 2020 is indicated, followed by steady declines through 2100. Within 95% confidence the models predict >50% decline in surface water resources between 2007 and 2050. Ultimately, surface storage of water resources may prove unsustainable in this region, forcing conversion to subsurface storage. ª 2008 Elsevier B.V. All rights reserved.
Summary
* Fax: +011 972 883 2537. E-mail address:
[email protected]. 0022-1694/$ - see front matter ª 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2008.02.020
Doomed reservoirs in Kansas, USA? Climate change and groundwater mining on the Great Plains
Introduction
ervoir Optima Lake in Oklahoma are emphasized in particular. Finally, the consequences of projected climate change are evaluated quantitatively for the largest of the four reservoirs, and qualitatively for the other three.
The Great Plains are likely to be the first area in the US to experience severe hydrologic disruption as a consequence of long-term climate change (e.g. Hurd et al., 1999). Dramatic declines in streamflow have been evident there since 1970, primarily as a consequence of groundwater mining and changing land use practices (Sophocleous, 2000). While changes in water regulations may have moderated these trends, the streamflow reductions have left the region increasingly sensitive to changes in temperature and precipitation. Such changes appear likely given increasingly certain predictions of global climate change over this century (IPCC4, 2007). An immediate consequence of the streamflow reductions has been to cause a number of reservoirs to become quite inefficient, evaporating over half of the water flowing into them. Climate change will only exacerbate this trend, and may truly doom these reservoirs. In many cases the most feasible alternative will be to convert from surface to subsurface storage of water in this region. This paper examines the situation of four example at-risk sites, the westernmost Federal reservoirs in Kansas: Cedar Bluff, Keith Sebelius, Webster and Kirwin. Similarities to the hydrologic warning signs for the now-empty Federal res-
Keith Sebelius
91
Streamflow changes Widespread groundwater overdraft in the High Plains Aquifer (HPA) has long been recognized (Gutentag et al., 1984; McGuire, 2007), as well as its role in streamflow reductions (Sophocleous, 2000; Sophocleous, 2005; Wahl and Wahl, 1988; Wen and Chen, 2006; Whittemore, 2002). Regional analysis of the spatial distribution of these streamflow changes can be used to determine those locations at greatest risk of adverse impacts. Using non-parametric statistics (Kendall Tau, Conover-1980) the strength of trends vs. time at US Geological Survey, National Water Information System (USGS NWIS) monitoring locations in the Great Plains was assessed. Stream gauges with periods of record greater than 25 years were analyzed, and trend was determined as the Kendall slope multiplied by period of record (i.e. net change) divided by mean daily discharge value for the period of record. Values are compared to the mean to allow for those few stations that do not show monotonic trends with
Legend
Webster Kirwin
Streamflow Changes %of Mean
76
< -150
225 70
-150 - -100 -100 - -50
Colorado
-50 - 0
135
Cedar Bluff
0 - 50 235
Optima
50 - 100 100 - 150
Kansas
Certainty Scale Oklahoma
Certainty 60%
25
75% 35 40
95% 99% 99.5% Freeway
27
New Mexico
Rivers
44
Texas
State Boundaries HPA >25% Drawdown 820 20
0
60
120
240
360
480
Kilometers 600
Figure 1 Long-term trends in streamflow in the central Great Plains. The most severe declines (dark circles) occur in or downgradient from large drawdown region in High-Plains Aquifer (McGuire, 2007). Increases (shown by squares) are primarily related to discharge of imported water by municipalities. Symbol size proportional to certainty of trend as determined by Kendall Tau.
92
T.H. Brikowski iod (1980–2000) had a significant but transient effect on streamflows (Garbrecht and Rossel, 2002). In most cases the long-term streamflow declines correlate temporally with declines in groundwater levels in the HPA, which has led to the conclusion that the phenomena are related (Whittemore, 2002). The correlation is not as clear for the Smoky Hill River, suggesting that increased farmland terracing in addition to groundwater mining may have had a significant impact in this hydrologic basin (Ratzlaff, 1993). This region is environmentally quite sensitive. A steep spatial gradient in average rainfall is present, ranging from 760 mm (30 in.) per year in Eastern Kansas to 380 mm (15 in.) in the West. Shifts in this pattern are common, and the western half of Kansas has experienced severe to extreme summertime drought for more than 25% of the 20th century (interpreted from gridded instrumental data of Cook and Meko, 1999; Dai et al., 2004.) Similar or worse conditions have been inferred for the past 2000 years (Woodhouse and Overpeck, 1998). This climate sensitivity will likely magnify the effects of any long term climate trends in the Great Plains.
time. In general, stations with declines P100% of flow relative to the mean (dark circles, Fig. 1) have lost at least 50% of initial streamflow over time. This assessment does not distinguish effects of control structures such as reservoirs, and must be interpreted with caution. The region of greatest streamflow decline, including the western half of Kansas and Oklahoma/Texas Panhandles lies immediately downstream from the area of greatest reduction in saturated thickness of the HPA over time (hatchured area, Fig. 1). To simplify further analysis, this paper will focus on the four westernmost Federal reservoirs in Kansas (Cedar Bluff, Keith Sebelius, Webster and Kirwin; Fig. 1), all of which have experienced difficulty maintaining expected water levels. All are the uppermost instream control structures in their respective watersheds. Streamflow above each of these reservoirs has exhibited severe long-term decline (‘‘Inflow change’’, Table 1). Short term trend reversals are evident in the annual records (Fig. 2), which have primarily been attributed to sub-decadal-scale cyclicity in the El Nin ˜o Southern Oscillation (ENSO, Woodhouse and Overpeck, 1998). An unprecedented multi-decadal wet per-
Table 1
Streamflow changes above Cedar Bluff, Keith Sebelius, Webster and Kirwin Reservoirs, Kansas (Kirwin has two inflows)
Reservoir
Upstream gauge #
Record begins year
Kendall Tau
Confidence %
Inflow change %
Mean inflow m3 =year
Cedar Bluff Keith Sebelius Webster Kirwin
6861000 6847900 6873000 6871000 6871500
0.4273 0.1741 0.3665 0.2718 0.2299
1950 1962 1945 1953 1951
88 56 77 50 70
99.95 99.5 99.95 99.50 99.0
ac-ft/year 7
3:55 10 7:30 106 4:49 107 2:26 107 1:14 107
28,814 5,918 36,398 18,320 9,202
Gauge is nearest upstream USGS-NWIS streamgauge (see Fig. 1). Kendall Tau listed for time correlation (negative indicates steady decline), and statistical confidence of that correlation. ‘‘Inflow change’’ is percent change of annual inflow from first to last decades of record, ‘‘Mean inflow’’ is mean annual inflow for period of record.
9
Cedar Bluff Webster Keith Sebelius Kirwin 105
3
Annual Discharge (m )
108
7
10
4
10
6
10
3
10
5
10
2
10
1940
1950
1960
1970
1980
1990
2000
Annual Discharge (ac-ft)
10
2010
Year
Figure 2 USGS reservoir inflow vs. time for the four Kansas reservoirs. Similar trends are evident for all reservoirs, indicating 90% decline (1–2 log units) in streamflow since the 1950s and unprecedented increase from 1992–1998.
Doomed reservoirs in Kansas, USA? Climate change and groundwater mining on the Great Plains
Reservoir changes
93
more of reservoir inflows are lost to evaporation than to any downstream use. In fact a useful criterion will be to consider a time-averaged ratio of evaporation to inflow greater than 50% to be inefficient, and some unspecified higher value to be intolerable. Optima Lake represents the extreme of this ratio (100%), where what little water flows into the reservoir leaves only by evaporation. For comparison, annual evaporative loss rates are 10–20% at Lake Mead, NV and Lake Nasser, Egypt, two lakes often used as examples of high evaporative losses. Examining trends vs. time for inflow and evaporation for the four Kansas reservoirs using Kendall Tau, a strong negative correlation is evident for inflow (i.e. decreasing with time, Table 1), and slightly weaker negative correlation in outflow (Table 2), based on USGS NWIS stream gauge information (USGS, 2007). Monthly evaporation volumes are available from the US Bureau of Reclamation for these reservoirs (pan-based, USBR, 2007,) and the ratio of annual evaporation to inflow exhibits strong positive (increasing) correlation with time. The USBRreported evaporation volumes exhibit strongly linear correlation with mean annual storage and water level
Not surprisingly, decreases of 50–90% in reservoir inflow (Table 1) have profound impacts on reservoir performance. Easily the most striking example in this region is Optima Lake in the Oklahoma Panhandle (Fig. 1). This Federal reservoir is remarkable in never having been more than 5% full (Wahl and Tortorelli, 1996), and has been effectively empty since the early 1990s. Rapid declines in reservoir inflow occurred as reservoir construction began, coincident with declines in groundwater levels (Fig. 3). Groundwater levels continue to decline at this time, and recovery of Optima Lake is highly unlikely. The inflow declines in Kansas have not been as steep, but are approaching similar magnitude.
Reservoir inefficiency The primary impact of declining streamflow on reservoirs is that water replenishment often decreases more rapidly than do evaporation and other losses. Eventually these trends lead to relative inefficiency, where over extended periods
57 10 8
58
60 10 7
61 62 63
10
6
Depth to Groundwater (m)
3
Annual Stream Discharge (m )
59
64 65 Reservoir
10 5 1934
1940
1946
66
Construction
Groundwater Surface Water 1952
1958
1964
1970
1976
1982
1988
1994
2000
67 2006
Year
Figure 3 Inflow and groundwater level history for Optima Lake, OK. Both measures began precipitous declines as reservoir construction began, making the reservoir impossible to fill by the time construction ended. Reservoir history from Oklahoma Water Resources Board. Discharge from USGS-NWIS gauge (7232500) immediately above Optima Lake, groundwater data from nearby USGS Monitoring Well 01N-12E-35 BDD 1, Texas Co. OK. Adapted from Wahl and Tortorelli (1996); Wahl and Wahl (1988).
Table 2
Time-correlations for reservoir performance at the four Kansas reservoirs
Parameter
Outflow (USGS) Evap/inflow
Cedar Bluff
Keith Sebelius
Webster
Kirwin
Slope
Conf
Slope
Conf
Slope
Conf
Slope
Conf
0.351 1.14
N/A 99.5
1.37 0.917
99.95 N/A
0.896 0.682
97.5 99.0
0.317 0.383
99.95 90.0
‘‘Slope’’ is scaled Kendall slope, (slope times period of record divided by mean value), and gives the fractional change over the period of record relative to the mean value. Confidence is given in percent. Where fewer than 40 years of record are available confidence cannot be determined (‘N/A’).
94
T.H. Brikowski
ðR2 ¼ 0:99Þ, indicating that reservoir surface area is the primary control on reported evaporation volume. Recent negative monthly inflow volumes reported by USBR indicate either overestimated storage losses (probably a small effect, e.g. Ferrari, 2001) or underestimated evaporation; hence USGS discharge from nearby upstream gauges were used to represent reservoir inflow this study, and reservoir inefficiency may be similarly underestimated below. The history of Cedar Bluff Reservoir demonstrates how these parameters interact (Fig. 4). The time-averaged ratio of evaporation vs. inflow has been steadily increasing for the entire period of record (see also trend in Table 2). Some periods such as the mid-1990s exhibit inflow greater than evaporation, and reservoir storage (heavy black line) increased. Note that expressing loss/gain as the ratio of evaporation/inflow means that storage changes relative to ratio change will be similar in sign, but generally not in magnitude. Since 1979 the reservoir lost storage two-thirds of the years, storage has averaged 44% of the active conservation pool (Fig. 4), and since 1990 (USGS) annual inflows average 14% of storage. A clear trend toward annual net loss for all four reservoirs is visible in the historical record (Fig. 5). Perhaps the clearest assessment of reservoir efficiency is to evaluate the net evaporation divided by inflow for an extended period. For the period since 1990 Cedar Bluff, Keith Sebelius, Webster and Kirwin reservoirs have evaporated 68%, 83%, 24% and 44% of their total inflows, respectively (Fig. 6). The first two can certainly be labeled as inefficient, and if not now, will eventually reach unsustainable levels of evaporative losses. The latter two share the same daunting trend toward inefficiency, but since 1990 at least temporarily reversed the trend during an unprecedented wet period. Hopefully some proactive course of action can be developed
for these sites, and action is probably past due for Cedar Bluff and Keith Sebelius reservoirs.
Future changes Previous rates of streamflow decline have been expected to moderate in Kansas, since minimum streamflow and antigroundwater mining regulations were established in the 1980s (Sophocleous, 2005). Consequently other influences on streamflow are likely to become dominant, particularly those related to long-term climate change. Temperature and precipitation predictions are available from global climate models (GCMs, IPCC4, 2007), and have been employed in a variety of approaches to estimate future runoff and streamflow worldwide. An early analysis of a single GCM for a moderate emissions scenario predicted decreasing water yields in the future for the central US (Rosenberg et al., 2003), primarily a result of little change in precipitation away from the coast in mid-latitude US, and greatly increased evapotranspiration driven by significant warming. Milly et al. (2005) utilized IPCC4-modeled surface runoff from GCM’s assuming a relatively conservative emissions scenario (SRESa1b, IPCC, 2000), considering 12 out of 25 GCM’s exhibiting the lowest error between predicted and observed 20th century basin discharge. They predict 10– 20% decline in runoff in the Great Plains by 2050, and up to 30% for the US Colorado River Basin. Several authors have noted strong correlation between Palmer Drought Severity Index (PDSI) and river discharge for large basins (e.g Dai et al., 2004). Hoerling and Eischeid (2007) use such a correlation and 42 IPCC4 high emissions scenario (A2) GCM results to predict 45% reduction in discharge in the western US Colorado River Basin by 2050. A more detailed approach
1000
2.5
Annual Evap/Inflow Moving Avg Evap/Inflow Storage
Full 2
Evaporation / Inflow
8
3
Mean Annual Storage (10 m )
100
1.5
10
1
Loss 1
Gain
0.5
0.1 1950
0 1960
1970
1980
1990
2000
2010
Year
Figure 4 Cedar Bluff Reservoir annual evaporation/inflow and storage vs. time. Ratios >1 indicate net loss, i.e. evaporation of stored water from previous years. Evaporation data from USBR, inflow from upstream USGS gauges. Top of active conservation pool indicated by ‘‘Full’’ arrow.
Doomed reservoirs in Kansas, USA? Climate change and groundwater mining on the Great Plains
95
10
Annual Evaporation/Inflow
Cedar Bluff Keith Sebelius Webster Kirwin
Loss 1
Gain
0.1 1950
1960
1970
1980
1990
2000
2010
Year
Figure 5 Smoothed evaporation/inflow vs. time for 4 western Kansas federal reservoirs (compare to Fig. 4). Cedar Bluff and Keith Sebelius are currently in serious water deficit conditions. Low values at beginning of curves represent effect of initial reservoir infilling (minimum evaporation). All show improvement during the unusually wet period 1990–2000.
Legend
Lovewell
Reservoir Efficiency
Kirwin
Kieth Sebelius
Waconda Webster
Evap Outflow
Wilson
Freeway
Cedar Bluff
Streams
Kanopolis
State Boundaries 135
Kansas Cheney 235
Optima
Oklahoma
0
20 40
80
120
160
Kilometers 200
Figure 6 Spatial distribution of post-1990 reservoir efficiency, indicated by evaporative loss as percentage of annual inflow. The westernmost reservoirs are at greatest risk, with losses are 100% at Optima Lake, OK; 68% at Cedar Bluff, 44% at Kirwin, and 24% at Webster reservoirs.
computing runoff using Variable Infiltration Capacity (VIC) macroscale hydrology models and a conservative emissions scenario predicts 18% decline for that basin (Christensen et al., 2004). When coupled with accurate demand and reservoir management models, VIC models driven by an aver-
age of 12 IPCC4 GCM’s for high (A2) and low (B1) emissions predict an increase of 20% in years of shortage in Lower Colorado Basin reservoirs (Christensen and Lettenmaier, 2007). All but the PDSI-based method are calibrated using comparison between models of 20th century condi-
96
T.H. Brikowski
tions (20c3m) and observed basin stream discharge. Some studies indicate that extended periods of extreme conditions are likely in areas like the Great Plains and Western US, and are not well-modeled in GCMs (Pelletier and Turcotte, 1997). One study of the effect of extended droughts indicates a 50% probability failure of Lakes Mead and Powell in the Colorado River basin by 2021 (Barnett and Pierce, 2008).
Correlation of Kansas stream discharge and PDSI For the Great Plains, the large historical declines in streamflow related to baseflow and landuse changes greatly complicate calibration of streamflow prediction by comparison to IPCC4-20c3m results, since those forcings are not included in the GCM models. For this reason the PDSI-based method is adopted here. An advantage of this choice is that temporal changes in PDSI-discharge correlations directly indicate effects of declining baseflow, and can be treated by restricting calibration to post-decline times. Earlier studies have examined historical correlations between PDSI and streamflow declines and soil moisture deficit in this region, finding strong positive correlations (Kansas: Young and Buddemeier, 2002; Nebraska: Wen and Chen, 2006). Temporal trends in baseflow and runoff decline in the Great Plains are evident in a plot of annual discharge vs. PDSI (Fig. 7), in this case for the USGS Cedar Bluff Reservoir inflow gauge listed in Table 1. Declining streamflow, primarily as a result of declining baseflow contribution, leads to progressively smaller discharge at a given PDSI. With minimal baseflow contribution, zero discharge is now possible at normal conditions (PDSI = 0) at this location. In general the discharge vs. PDSI data can be divided into two parts, a relatively high discharge-steeply sloped group prior to
1970, and a low discharge-gently sloped group post-1980 (Fig. 7). Data for the 1970s occupy a transitional zone between these two populations. An independent indication of baseflow changes with time can be obtained using the Baseflow Index (BFI, Wahl and Wahl, 1988), an automated technique for baseflow separation of hydrographs. Annual BFI declined 13% for the Smoky Hill River at Arnold since 1950, and 20% for the Beaver River above Optima Lake since 1940 (both calculations using storm hydrograph duration n ¼ 5 days). For the purposes of prediction, a linear fit to the post1980 data will be used, implicitly assuming that post-1980 changes in baseflow and other non-climate effects are minimal. Kendall Tau analysis of this dataset indicates a strong positive correlation between annual PDSI and log discharge, with 99.95% certainty (Table 3), and similarly for the full record (1951-present). A linear fit to post-1980 data gives an R ¼ 0:72, quite similar to variances reported for such fits (Dai et al., 2004; Colorado River, R ¼ 0:79, Hoerling and Eischeid, 2007). This indicates that the linear PDSI relationship explains slightly more than half the observed variation in stream discharge in most of these cases. Note limitation to post-1980 times reduces the coefficient of determination in this case. Linear fits to the historical variation of stream discharge above each of the reservoirs vs. PDSI are summarized in Table 3. Comparison of R2 to drainage area indicates that in the three smaller watersheds local effects overwhelm PDSI correlation with streamflow. Relatively high Kendall Tau and resulting certainty indicate strong positive rank correlation between PDSI and log(Q), but low R2 indicates this relationship is not particularly linear. Uncertainty in predictions based on these fits can be quantified by computing a prediction interval (Chatfield, 1993), but in practice only the Cedar Bluff inflow provides a useful interval. The remaining quantitative discussion will be limited to that case.
8 80% Prediction Interval Pre-1970 1970-1980 Post-1980
6 4 2
PDSI
0
1980
Post-
-2 -4 -6
70
-8
-19
Pre
-10 -12 10
5
10
6
10
7
10
8
10
9
3
Annual Discharge (m )
Figure 7 Correlation between PDSI and stream discharge, Smoky Hill River gauge at Arnold (Cedar Bluff inflow). Pre-1970 and post-1980 trend lines demonstrate that zero discharge is increasingly likely at normal conditions (PDSI = 0) since 1980. Post-1980 fit line ðR2 ¼ 0:52Þ and 80% prediction interval (shaded zone) described in Table 3.
Doomed reservoirs in Kansas, USA? Climate change and groundwater mining on the Great Plains Table 3
97
Linear fit of annual log inflow (Q) vs. PDSI for the four reservoirs, post-1980, listed in decreasing order of drainage area
Reservoir
Cedar Bluff Kirwin Webster Keith Sebelius
Linear equation logðQm3 Þ ¼ a PDSI þ b a
b
0.1742 [0.170–0.175] 0.1146 0.1355 0.04484
6.681 [6.228–7.188] 7.194 7.066 6.659
R2
s
0.519
0.4758
0.4102 0.3663 0.2188
0.4701 0.4017 0.3267
% Certainty
99.95 75 <60 60
Drainage area ðkm2 Þ
13,515 3,081 2,693 1528
Total of two inflows used for Kirwin Reservoir. Coefficient ranges for 80% certainty prediction interval for linear fit given in square brackets for Cedar Bluff. Certainty of a rank correlation between PDSI and inflow Q is listed (based on Kendall Tau).
Future climate, PDSI and streamflow Future PDSI is required for this approach, and was determined from temperature and precipitation for each of 19 IPCC4 models of the SRESa1b scenario. This scenario assumes intermediate levels of greenhouse gas emissions over the next century. Current levels of emissions (2.5% increase per year in atmospheric CO2 since 2000, Canadell et al., 2007) are consistent with the SRESa1b scenario, although they would exceed scenario emissions if continued beyond 2025. To minimize GCM bias, model increments (model departure from 20c3m results) of surface air temperatures (TAS) and precipitation for each of the IPCC4 SRESa1b scenario models were determined. These were downscaled to US climate divisions (nine of these in Kansas) using intersection-area weighting, and future TAS and precipitation values were computed by adding the increment to observed normals for each division. Monthly and annual average PDSI was computed from these for each division for each model. The mean, 90th and 10th percentile values over the 19 models were computed for each division for each year, and were the basis for the mean and uncertainty ranges shown in Fig. 8a. Mean predicted streamflow was computed using the linear fit for Cedar Bluff inflow from Table 3. Uncertainty related to the climate models was determined by applying the same equation to the 90th and 10th percentile values of PDSI. Uncertainty related to the PDSI vs. discharge fit was determined by applying the prediction interval range to the 90th and 10th percentile values of PDSI; consequently the lighter shaded region in Fig. 8b represents the joint 80% certainty range arising from future climate and historical PDSI vs. discharge uncertainty. Future PDSI and therefore streamflow rise slightly until 2020, buoyed by a 10–20 year period of increased precipitation in the averaged model results (see discussion), then decline steadily thereafter (Fig. 8). Although only Kansas climate division 4 is discussed here, projected conditions are similar for the three Kansas climate divisions of interest (1: Keith Sebelius, 2: Webster-Kirwin, 4: Cedar Bluff). In general only small changes in precipitation are predicted by the IPCC4 models, with the averaged values declining by 2–5% by 2050. In contrast temperature is predicted to increase by 2.1 C. PDSI correlates negatively with positive departures from normal temperature, and changes little for < 5% departures in precipitation; therefore PDSI can be expected to become significantly more negative by 2050, chiefly as a function of temperature. After 2050 mean
PDSI values approach those experienced during the major historical droughts in this region (left side, Fig. 8a). Uncertainty in the above projections is large, since IPCC4 model results, and therefore PDSI values derived from them, exhibit considerable variance in time and space. Variance in climatic and hydrologic conditions is expected to increase as a result of global warming (Held and Soden, 2006), and in this region the anomalous conditions from 1980–2000 may represent just such an effect. Although large, variance in the projected discharge (as represented by the joint 80% confidence interval Fig. 8b) is of similar magnitude as post-1980 observed variability.
Future reservoir performance As shown above, historical declines in inflows have severely impacted reservoirs in Kansas, and certainly elsewhere in the Great Plains. Climate change related continuation of these declines will extend that impact. Given projected inflow to the reservoirs as calculated above, mass balance calculations can be used to predict future performance, provided evaporation and outflows can be determined. For these shallow reservoirs, reported evaporation can be closely approximated by a linear function of annual storage ðR2 ¼ 0:99Þ, and this function used to predict future evaporation. In this region other outflows will be dominated by deliberate releases in response to demand, but future demand models have yet to be completed (Streeter, 2007). In lieu of such information, the least restrictive case of zero future demand (and no flood releases except from emergency spillway) can be assumed. There is historical precedent for this assumption in the case of Cedar Bluff since poor reservoir performance led to elimination of demand in 1979 (irrigation district was disbanded), and no intentional releases were made for the next 26 years. Only minimal releases have been made since 2004. An iterative approach must be taken for the mass balance, since future evaporation and storage are interdependent. A method which best matches historical performance is to first estimate the next year’s evaporation from an initial estimate of next year’s storage (current year storage minus current evaporation). Then next year’s storage is recalculated as the sum of this year’s storage plus next year’s inflow minus evaporation. Finally next year’s evaporation is recalculated from next year’s storage. Uncertainty bounds for these projections are estimated by assuming year-to-year inflows at various percentile levels (see Fig. 9).
98
T.H. Brikowski
a Annual Mean Palmer Drought Severity Index
8
6
4
2
0
-2
-4
-6 Projected
Observed 90-10 range Projected Mean -8 1900
1950
2000
2050
210
Year
10
8
10
7
10
10
10
5
10
4
6
10
3
5
10
2
Annual Discharge (ac-ft)
80% Prediction Interval Climate 80% Certainty Mean Projected Observed
3
Annual Discharge (m )
b
Projected 1940
1960
1980
2000
2020
2040
2060
2080
2100
Year
Figure 8 (a) Observed and projected PDSI for Kansas climate division 4, and (b) streamflow for Smoky Hill River at Arnold (central western Kansas, inflow to Cedar Bluff Reservoir). Streamflow projections based on PDSI vs. Q fit, Table 3. Future climate-related streamflow declines continue trend of historical groundwater-mining related declines.
The most general aspect of these projections is that reservoir levels decline with time for virtually all estimates, and that these declines will begin no later than 2020 (Fig. 9). If the models adequately describe the full range of potential variability in discharge, these percentiles can be interpreted as an indication of risk. Taking this approach, the results indicate a 70% chance reservoir levels will decline steadily from the present. Once levels drop into the inactive pool, the reservoir must be considered an unreliable resource, and indeed reservoir releases were halted in 1979 when this first occurred. A 50% risk this will occur by 2035 is indicated, and by 2100 the risk increases to 75%. By 2050 there is a 50% chance levels will drop into
the dead pool and reservoir water will be inaccessible for gravity release (reservoir failure). There is a 25% chance Cedar Bluff reservoir will have failed as early as 2026. Similarly quantitative projections for the remaining three reservoirs of interest are not currently possible using the methods employed above. Keith Sebelius has demonstrated historical performance limitations quite similar to Cedar Bluff, and the data examined here offer no indication that it is any less threatened than Cedar Bluff (Fig. 5). Kirwin and Webster reservoirs exhibited notable increases in efficiency from 1980–2000, but unfortunately this was a historically unprecedented period of increased rainfall, and their efficiency seems likely to decline rapidly in the future.
Doomed reservoirs in Kansas, USA? Climate change and groundwater mining on the Great Plains Observed
99
Projected
3.5
2.5
3.0
2.0
3
8
Annual Storage (x10 ac-ft)
2.5
3
Annual Storage (x10 m )
Flood Control Pool
2.0
Conservation
1.5
Pool 1.5
90% 1.0
1.0 75% 5.0 5.0 10%
Inactive Pool 1960
1980
2000
25% 2020
50% 2040
2060
2080
0.0 2100
Figure 9 Projected performance of Cedar Bluff Reservoir, based on projected inflow from Fig. 8. Shading indicates historical storage and 50th percentile (median) projected storage. Other percentiles indicated by labeled lines. Reservoir allocation levels indicated by horizontal dashed lines. By 2055 levels will drop to near the minimum accessible by gravity flow (dead pool).
Discussion Although regional streamflow declines in the Great Plains have been recognized for some time, the effect of these changes on reservoir performance has received little attention. Certainly the primary issues will be efficiency and sustainability of the reservoirs. In the development above it is clear that reservoir inefficiency has reached potentially intolerable levels and sustainability is threatened for at least two (Cedar Bluff, Keith Sebelius) of the four westernmost Kansas reservoirs considered here. In the case of Cedar Bluff, these evaporative losses are by far the largest consumptive ‘‘use’’ of surface water in the hydrologic basin. This inefficiency has arisen almost exclusively from declining inflows resulting from groundwater mining and farm terracing. The remaining question of future trends in reservoir sustainability is more difficult to answer accurately. Climate projections universally indicate significant warming by 2050, accompanied by small decrease in precipitation. This will certainly result in drier conditions because of enhanced evapotranspiration, as indicated by projected declines PDSI (Fig. 8a). Future trends in runoff and reservoir input are much more uncertain, but at least in the case of Cedar Bluff Reservoir historical correlation between PDSI and streamflow is good enough to allow approximation of future climate-related streamflow declines. These declines are projected to continue at half the rate of historical baseflow-related declines, resulting in about a 99% decrease in streamflow from 1955 levels by 2050 (Fig. 8b). This degree of decline caused the failure of Optima Lake, and future reservoir mass balance for Cedar Bluff indicates a nearly 50% chance of it failing (reaching dead pool) by 2050.
Most surprising in these results is the indication of an approximately 30% chance that reservoir water levels will rise until 2020, and 20% chance the reservoir will reach ‘‘full’’ levels (top of conservation) at least once more before beginning a steady decline. Indeed such an event has already occurred, related to the unprecedented 20 year increase in central Great Plains precipitation (of 10–15%, Garbrecht and Rossel, 2002). These authors have suggested this wet period may lead to an erroneous view of sustainability of some water resources on the Great Plains, and the projections made here buttress that concern. The increase in storage indicated by the higher percentile results (upper lines Fig. 9) represents the averaged influence of shorter sub-decadal increases in precipitation present in a number of the SRESa1b models. These spikes are concentrated between 2000 and 2025, but decline at later times. The process of averaging the models mimics the ‘‘long memory’’ or extended climate extreme periods noted by Garbrecht and Rossel (2002) and Pelletier and Turcotte (1997). Although streamflow in the western Great Plains responded only moderately to the 1980–2000 departure the Great Plains (Fig. 2, and Garbrecht et al., 2004), it was apparently enough to result in the extended period of storage recovery that began in 1990 for many reservoirs (e.g. Fig. 4 or Garbrecht and Schneider, 2008). While the forcing mechanism of this unprecedented period is uncertain, it coincides with the onset of recognizable global warming effects in the US, and similar phenomena are evident in IPCC SRESa1b model results. In light of the model results, it seems that the much-needed recent recovery in reservoir performance in this area is to be short-lived. Fortunately, alternative means of water storage are available along these valleys, namely underground storage
100 in the alluvial valley aquifers. These aquifers represent a more efficient storage medium of similar volume in this context, and preliminary studies indicate evapotranspiration losses from the Smoky Hill River aquifer system are <40% (Brikowski and Anderson, 2006). While higher than surface reservoirs in most settings, in the Great Plains such losses are notably better than other alternatives. In fact just such a transfer from surface to groundwater storage is underway on a small scale at Cedar Bluff in response to downstream water right demands. The projections made here include undesirably high uncertainty. Inherent uncertainty in predictions of future climate are by far the largest contribution, and at this time are irreducible. The hydrologic contribution to uncertainty in these results may be reduced most directly by carrying out detailed runoff models in each basin of interest, and using these to parameterize GCM-driven runoff projection models coupled with reservoir demand and management models (e.g. Christensen and Lettenmaier, 2007). For larger basins, PDSI-based hydrologic models such as the ones developed here may yield higher estimates of climate impact on streamflow, but appear to more accurately simulate extended drought/wet period effects (compare Barnett and Pierce, 2008; Christensen and Lettenmaier, 2007; Hoerling and Eischeid, 2007). Finally, the warming scenario chosen here may be conservative in view of steadily increasing observed CO2 emissions rates. Even given these uncertainties, it appears that two and perhaps all of the reservoirs examined here can expect to see long term declines, possibly following near term increases (perhaps these have already finished), and greater chance than not of failure by 2050.
Conclusions In the Great Plains, adverse streamflow trends historically attributable to groundwater mining appear likely to continue into the future, now attributable to climate change. These trends have already served to push reservoirs in the Great Plains into inefficient, and potentially unsustainable modes. Water resources plans that rely on these reservoirs are potentially in peril, and water rights appropriations, environmental flow regulations and water law in general will likely require adjustment to accommodate these trends. Many sustainable systems today are likely to evolve into unsustainability with time. As a result, future resource management in this area may be forced to convert from surface to subsurface storage of water.
Acknowledgements Thanks to Dr. L. Balzer of John Brown University for bringing Optima Lake to my attention. Thanks also to Sam Perkins of the Kansas Department of Water Resources and Wayland Anderson of BBA Assoc. for the original efficiency calculation at Cedar Bluff Reservoir, released in public testimony before the Kansas State Engineer, May 2006. IPCC4 datasets were made available to us via the Earth System Grid webserver. Comments by two anonymous reviewers motivated extensive revision and improvements in the methodology and manuscript. U. Texas-Dallas Geosciences contribution # 1131.
T.H. Brikowski
References Barnett, T.P., Pierce, D.W., 2008. When will Lake Mead go dry? Water Resour. Res., in press, doi:10.1029/2007WR006704. Brikowski, T.H., Anderson, W.J., 2006. Storage efficiency: converting from surface to groundwater storage to survive long-term drought. Managing Drought and Water Scarcity in Vulnerable Environments: Creating a Roadmap for Change in the United States, 18 Sep. Canadell, J.G., Que ´re ´, C.L., Raupach, M.R., Field, C.B., Buitenhuis, E.T., Ciais, P., Conway, T.J., Gillett, N.P., Houghton, R.A., Marland, G., 2007. Contributions to accelerating atmospheric CO2 growth from economic activity, carbon intensity, and efficiency of natural sinks. Proc. Nat. Acad. Sci., 104. Chatfield, C., 1993. Calculating interval forecasts. J. Busin. & Econ. Statist. 11 (2), 121–135. Christensen, N.S., Lettenmaier, D.P., 2007. A multimodel ensemble approach to assessment of climate change impacts on the hydrology and water resources of the Colorado River Basin. Hydrol. Earth Syst. Sci. 11, 1417–1434. Christensen, N.S., Wood, A.W., Voisin, N., Lettenmaier, D.P., Palmer, R.N., 2004. The effects of climate change on the hydrology and water resources of the Colorado River basin. Climatic Change 62 (1–3), 337–363. Conover, W.J., 1980. Practical Non-Parametric Statistics, second ed. John Wiley and Sons, New York. Cook, E.R., Meko, D., 1999. Drought reconstructions for the continental United States. J. Climate 12, 1145–1162. Dai, A., Trenberth, K.E., Qian, T., 2004. A global dataset of Palmer Drought Severity Index for 1870–2002: relationship with soil moisture and effects of surface warming. J. Hydrometeorol. 5, 1117–1130. Ferrari, R.L., 2001. Cedar Bluff Reservoir 2000 Reservoir Survey. Tech. Rep., US Bureau of Reclamation, Denver Federal Center, Denver, CO, Mar. Garbrecht, J., Liew, M.V., Brown, G.O., 2004. Trends in precipitation, streamflow, and evapotranspiration in the Great Plains of the United States. J. Hydrologic Eng. 9 (5), 360–367. Garbrecht, J.D., Rossel, F.E., 2002. Decade-scale precipitation increase in great plains at end of 20th century. J. Hydrologic. Eng. 7 (1), 64–75. Garbrecht, J.D., Schneider, J.M., 2008. Case study of multi-year precipitation variations and the hydrology of Fort Cobb Reservoir. J. Hydrol. Eng. 13 (2), 64–70. Gutentag, E.D., Heimes, F.J., Krothe, N.C., Luckey, R.R., Weeks, J.B., 1984. Geohydrology of the high plains aquifer in parts of Colorado, Kansas, Nebraska, New Mexico, Oklahoma, South Dakota, Texas and Wyoming. Professional Paper 1400-B. Held, I.M., Soden, B.J., 2006. Robust responses of the hydrological cycle to global warming. J. Climate 19, 5686–5699. Hoerling, M., Eischeid, J., 2007. Past peak water in the Southwest. Southwest Hydrol. 6 (1), 18–20. Hurd, B., Leafy, N., Jones, R., 1999. Relative regional vulnerability of water resources to climate change. JAWRA 35 (6), 1399– 1409. IPCC, 2000. Special report on emission scenarios. Tech. Rep., UN Intergov. Panel on climate change. IPCC4, 2007. Climate change 2007. The Physical Science Basis, Summary for Policymakers (4th Climate Assessment Report). Tech. Rep., UN Intergov. Panel on Climate Change, 5 Feb, p. 18. McGuire, V.L., 2007. Water-level changes in the high plains aquifer. Predevelopment to 2005 and 2003 to 2005. Scientific Investigations Report SIR 2006-5324, US Geol. Survey, Reston, VA. Milly, P.C.D., Dunne, K.A., Vecchia, A.V., 2005. Global pattern of trends in streamflow and water availability in a changing climate. Nature 438 (7066), 347–350. Pelletier, J.D., Turcotte, D.L., 1997. Long-range persistence in climatological and hydrological time series: analysis, modeling
Doomed reservoirs in Kansas, USA? Climate change and groundwater mining on the Great Plains and application to drought hazard assessment. J. Hydrol. 203 (1–4), 198–208. Ratzlaff, J.R., 1993. The effects of agricultural practices upon the hydrology of Cedar Bluff Reservoir, western Kansas. Compass 70 (3), 92–99. Rosenberg, N.J., Brown, R.A., Izaurralde, R.C., Thomson, A.M., 2003. Integrated assessment of Hadley Centre (HadCM2) climate change projections on agricultural productivity and irrigation water supply in the conterminous United States. Ag. Forest Meteorol. 117 (1–2), 73–96. Sophocleous, M., 2000. From safe yield to sustainable development of water resources – the Kansas experience. J. Hydrol. 235 (1– 2), 27–43. Sophocleous, M., 2005. Groundwater recharge and sustainability in the High Plains aquifer in Kansas, USA. Hydrogeol. J. 13, 351– 365. Streeter, T., 2007. The State of Kansas’ Reservoirs. In: Proceedings Kansas Reservoir Summit. Kansas Water Office, 17 Oct. USBR, 2007. HYDROMET Data System. website, US Bureau of Reclamation, 16 May. URL http://www.usbr.gov/gp/hydromet/index.cfm. USGS, 2007. National Water Information System (NWIS). website, 16 May. URL http://water.usgs.gov/waterwatch/.
101
Wahl, K.L., Tortorelli, R.L., 1996. Changes in flow in the BeaverNorth Canadian River Basin Upstream from Canton Lake, Western Oklahoma. Water Resour. Investig. 96-4304, US Geol. Survey, Reston, VA. Wahl, K.L., Wahl, T.L., 1988. Effects of regional groundwater level declines on streamflow in the Oklahoma Panhandle. In: Proceedings of Symposium on Water-Use Data for Water Resources Management. AWRA, Tucson, AZ, Aug., pp. 239–249. URL http://. Wen, F., Chen, X., 2006. Evaluation of the impact of groundwater irrigation on streamflow in Nebraska. J. Hydrol. 327 (3–4), 603– 617. Whittemore, D.O., 2002. Recharge changes along rivers crossing the High Plains Aquifer. Geol. Soc. Amer. Abstr. w. Programs 34 (6), 98, Oct. Woodhouse, C.A., Overpeck, J.T., 1998. 2000 years of drought variability in the Central United States. Bull. Am. Meteorol. Soc. 79 (12), 2693–2714. Young, D.P., Buddemeier, R.W., 2002. Climate variation: implications of long-term records and recent observations. Open File Report OFR 2002-25E, Kansas Geol. Survey, The University of Kansas, Lawrence, KS.