Possible link between irrigation in the U.S. High Plains and increased summer streamflow in the Midwest

Authors


Abstract

[1] We have previously presented evidence that higher rates of evapotranspiration (ET) associated with irrigation in the U.S. High Plains has likely caused an increased downwind precipitation (P). July P over the Midwest increased by 20%–30% from the preirrigation period (1900–1950) to the postirrigation (1950–2000) period. In this study, we test the hypothesis that the increased July P has had hydrologic consequences, possibly increasing groundwater storage and streamflow. Seasonal analyses of hydrologic variables over Illinois suggest that the water table and streamflow response lags P − ET by 1–2 months, indicating August and September as the months when the increased July P may be detected. We analyzed long-term observations of water table depth at 10 wells in Illinois and streamflow at 46 gauges in Illinois-Ohio basins. The Mann-Kendal test for trends suggests field significant increases in groundwater storage and streamflow in August–September over the period of irrigation expansion. Examination of soil moisture response to present-day above-normal July P suggests that the increased July P can reach the water table in normal to wet years. Mann-Kendall tests suggest that there has been no change in pan evaporation and atmospheric vapor pressure deficit. This implies that soil water availability is the driver of changes in ET, and the increased P may have possibly increased ET. Other studies in the literature give further evidence of increased ET due to increased P. By ruling out a reduction in ET, we suggest that the observed increase in groundwater storage and streamflow in the Midwest is linked to the increased July precipitation attributed to High Plains irrigation. We note that the increases in late summer streamflow are rather small when placed in the context of seasonal dynamics, but they are conceptually important in that they point to a different cause of change.

1. Introduction

[2] Groundwater pumping for irrigation in the U.S. High Plains began to accelerate in the 1940s (Figure 1a), and by the mid-1980s, groundwater levels had declined by >30 m over much of the High Plains (Figure 1b) [McGuire, 2009]. The decades-long and regional-scale water transfer, from the groundwater reservoir to the soil moisture reservoir in the warm season, has likely influenced the region's hydrology and climate. Moreover, this influence may not have been confined to the High Plains itself but may have propagated downwind through the atmospheric vapor transport pathways and down gradient through the river and groundwater pathways. We hypothesize the following: first, the large groundwater decline has led to reduced streamflow in the High Plains region wide, particularly where groundwater is a source for streamflow; second, because irrigation drastically increases warm season evapotranspiration (ET), it has increased vapor export and possibly downwind precipitation (P); third, such increased downwind P has altered the land hydrology over the receiving region, far away from the High Plains where the change originated. Figure 1c schematically illustrates these hypotheses: (1) reduced streamflow in the High Plains, (2) increased downwind P, and (3) increased downwind ET and streamflow.

Figure 1.

(a) Volume of groundwater pumped for irrigation from the U.S. High Plains aquifer for selected years [from McGuire et al., 2003], (b) the resulting water table decline [from McGuire et al., 2009], and (c) possible effects of High Plains irrigation on the regional water cycle.

[3] In an earlier paper [Kustu et al., 2010], we tested the first hypothesis. That groundwater pumping in the High Plains reduced streamflow was not a new idea; there had been many reports in the literature on the subject (see detailed review by Kustu et al. [2010]), but they focused on specific areas and applied different methods of analysis, leaving large spatial gaps and making a regional comparison and synthesis difficult. For example, strong climatic and hydrologic gradients from northern to southern High Plains are well documented. In the north (e.g., Nebraska), the cooler and moister climate, the sandy soil and river beds, and the naturally high water table point to groundwater as a main source for streamflow and that changes in the former can directly influence the latter. In the south (e.g., Texas), the warm and dry climate, the dominance of summer thunderstorms and surface runoff, and the naturally deep water table indicate that the water table may lie below local river beds; hence, pumping may have little effect on local streamflow (but it may affect regional streams fed by regional groundwater convergence further down gradient). To achieve this regional synthesis, we analyzed the entire groundwater and streamflow records of the U.S. Geologic Survey (USGS). Our results filled large spatial gaps between previously studied areas and suggest that indeed, decreases in annual and dry season streamflow and increases in the frequency of low-flows are more pronounced in the northwestern part of the High Plains.

[4] In our second study [DeAngelis et al., 2010], we tested the second hypothesis that irrigation in the High Plains, through increased ET and vapor export, may influence downwind precipitation. The idea that irrigation can affect rainfall is not new either (see detailed review by DeAngelis et al. [2010]), but earlier studies focused on local P recycling and were based on short-term observations or model experiments. It is now well recognized that land surface wetness has a large impact on downwind precipitation [e.g., Dominguez et al., 2009; van der Ent et al., 2010], and a recent global modeling study [Puma and Cook, 2010] reports larger downwind than local increases in precipitation due to irrigation. In the United States, it is well understood that the strong winds of the Great Plains Low Level Jet [Weaver et al., 2009] (see wind vectors in Figure 2a), peaking in the warm season, connect the High Plains (region 1 in Figure 2a) to its downwind regions to the northeast. Meanwhile, hundreds of station precipitation records exist in the central United States, dating back to at least the early 1930s. A study based on these long-term precipitation observations with an emphasis on downwind climatic impacts had been lacking. To fill in this knowledge gap, we analyzed 865 long-term station records, over and downwind of the High Plains (the three regions in Figure 2a), for signals of change. The observations, combined with a Lagrangian vapor tracking analysis to trace the fate of the High Plains ET, revealed evidence that irrigation in the High Plains has led to increased downwind precipitation, particularly over the Midwest (region 3 in Figure 2a) in the month of July, the peak month of irrigation and the peak month of wind speed in the Great Plains Low Level Jet [Weaver et al., 2009].

Figure 2.

(a) Spatial pattern of July precipitation change (%) between the periods of 1900–1950 and 1950–2000 and mean July 850 mbar wind fields (m/s) over 1979–2001, obtained from North America Regional Reanalysis (for details, see DeAngelis et al. [2010]). (b) Time series of July precipitation (mm) averaged over 316 station records within region 3 (green box in Figure 2a), shown as a 5 year moving average and with the means (blue) of the first and second halves of the century (84 and 102 mm, respectively, tested statistically significant in the work of DeAngelis et al. [2010]). The green box is the area of focus in this study.

[5] In this paper, we test our third hypothesis that the irrigation-enhanced July precipitation over the Midwest has had hydrologic consequences. Precipitation is a key driver of land hydrology, and changes in P will propagate through the various hydrologic pathways: canopy interception, surface runoff, infiltration, soil and plant ET, water table recharge, and groundwater discharge into streams. The noticeable increase in July P from the first to the second half of the century (Figure 2b) likely manifested itself in one or more of these hydrologic variables. In this paper, we analyze available observations of these hydrologic variables to search for signals of change that may be attributable to the increased July precipitation.

[6] In section 2, we discuss the dominant hydrologic pathways in the region and the associated time scales whereby precipitation propagates through land hydrology. In section 3, we analyze long-term water table, streamflow, soil moisture, air temperature, and pan evaporation time series. In section 4, we discuss the implications of the work and future research to improve our understanding of hydrologic-climatic interactions in the context of climate variability and change and land use or water use changes.

2. Hydrologic Features of the Study Area

[7] The study area is centered over the states of Illinois and Indiana (Figure 2a, green box), where July precipitation increased by 10%–30% from the first half to the second half of the twentieth century. The change occurred near the middle of the century (Figure 2b), after which the means and lows all increased and extreme dry periods have been absent. The questions are, how does such a change in July P (hereafter referred to as July equation image) propagate through land hydrology, and can we detect its signal in historically observed hydrologic variables such as water table depth and streamflow?

[8] As precipitation increases, the vegetation and the near-surface soils are the first to sense it, and if ET is water limited, equation image will likely engender increased ET, leaving no trace in groundwater and streamflow (historically observed). However, if ET is energy limited, then equation image may infiltrate deeper and recharge the groundwater, leaving a signature in groundwater and river flow. It may also take the route of increased surface runoff if equation image is in the form of higher storm intensity.

[9] To explore the possible partition of equation image into increased ET (not observed) versus groundwater storage and streamflow (observed), we examine the seasonality of land hydrology in the region. Illinois has one of the best hydrologic monitoring networks in the world, including soil moisture beginning in 1981 [Hollinger and Isard, 1994] and shallow water table in the late 1950s (Illinois State Water Survey (ISWS)). Although they began after the initial irrigation expansion, the water table records cover a good portion of the period. In addition, soil moisture and water table observations provide essential insight into the cascading of P signals through the hydrologic stores and the associated time scales.

[10] Figure 3a plots the seasonal cycle of observed P, estimated ET, and observed streamflow (the fluxes), and Figure 3b plots the seasonal cycle of the observed top 2 m soil moisture (SM) and water table depth (WTD) (the states), averaged over the state of Illinois and the period of 1983–1995. The data in Figure 3 are directly taken from the work by Eltahir and Yeh [1999], which is a seminal study on the hydrologic linkages in the region, where the ET is the mean of two independent estimates, one based on atmospheric vapor convergence and the other on soil water budget analysis [Yeh et al., 1998]. We note the following.

Figure 3.

Seasonal cycle in (a) hydrologic fluxes: precipitation (P), evapotranspiration (ET), soil water surplus (P − ET), and streamflow (Qr). (b) Hydrologic states: soil moisture (SM) and water table depth (WTD) (data are from Eltahir and Yeh [1999]).

[11] First, ET flux, with its large seasonal swings, dominates the seasonal dynamics, exceeding P in May through August. In July, P accounts for 80% of ET, suggesting a net soil water deficit (P − ET < 0). Long-term mean July pan evaporation in central Illinois is 227 mm (as shown later in Figure 10b and Table 6), suggesting that the 122 mm ET here is below potential and that a July equation image of ∼20% may directly translate into increased July ET. However, if higher storm intensity accounts for equation image, it would lead to increased infiltration excess surface runoff, leaving a signature in streamflow. Since surface runoff responds to rainfall quickly, the signature in streamflow would be found in the same month (July). If equation image represents longer periods of rainfall, it would increase infiltration into the soil.

[12] Second, the top 2 m soil moisture closely follows the P-ET cycle, with the best correlation obtained at the 1 month lag (Figure 4a). That is, the top 2 m soil moisture as a whole responds to climate forcing 1 month later, although the shallow soils may respond in the same month. Therefore, the signal of equation image is likely found in July (at shallow depths) and August (at deeper depths) in the soil moisture records.

Figure 4.

Phase relations between (a) soil moisture and P − ET, (b) water table depth and soil moisture, and (c) streamflow and water table depth, with the Pearson correlation coefficient (r) given for the different lags. (d) Summary of the lag time of response of the hydrologic variables, where black lettering indicates variables that are observed over the period of interest (1940–1980).

[13] Third, the water table cycle closely follows the soil moisture cycle, with the best correlation obtained at the 1 month lag (Figure 4b). That is, the groundwater on average is recharged 1 month after the soil moisture is replenished. This suggests that the signal of equation image is likely found in the groundwater records (if at all) in August and September.

[14] Fourth, groundwater fluctuations are closely linked to streamflow. The streamflow seasonal cycle is mostly in phase (no lag) with that of the water table depth (Figure 4c). Eltahir and Yeh [1999] estimated that surface runoff explains <10% of streamflow variations and accounts for <25% of streamflow, leaving groundwater as the main source and driver of monthly and seasonal dynamics. This suggests that the July equation image signal would find expressions (if at all) in August and September streamflow.

[15] The above analysis is summarized in Figure 4d with the expected lag times of the relevant hydrologic variables indicated. The above discussion helps us focus our subsequent analysis on relevant hydrologic variables and at relevant time scales.

3. Signals of Increased July P in the Observed Hydrologic Variables

[16] Figure 5 gives the mean region 3 (Figure 2a) precipitation time series (shown as a 5 year moving average to highlight long-term variabilities) for May–September based on 316 station monthly data from the National Climate Data Center (NCDC, http://www.ncdc.noaa.gov/oa/ncdc.html). They are shown here because the signal of May and June P may be present in July and August water table level and streamflow. Over the period of irrigation expansion (1940–1980, shaded gray in Figure 5), there is a slight decline in May and June, a step-like increase midcentury in July, a rise in late 1970s in August, and no apparent trend in September. Of the hydrologic variables in Figure 4d, only the groundwater level and streamflow are observed over the period of interest (1940–1980), and we start our analysis with these observations.

Figure 5.

Region 3 mean monthly rainfall (5 year moving average) for May through September based on 316 station records, with the irrigation development period (1940–1980) shaded gray.

3.1. Changes in Water Table Depth

[17] Water table observations, dating back to the 1950s, are obtained from USGS at one site (all others began in the late 1980s) at ∼10 day steps and at nine sites from the ISWS Water and Atmosphere Resource Monitoring network (WARM, http://www.isws.illinois.edu/warm/sgwdata/wells.aspx) at monthly steps. All these long-term observations are in the state of Illinois; no historic groundwater data could be found for Indiana, where the largest July equation image was observed (Figure 2a). The well locations are shown in Figure 6a (orange and green), with site information given in Table 1 (see the wells with early data; the rest have shorter records and are used for later analyses). Monthly water table depths at these 10 sites are plotted in Figure 7 for July–September.

Figure 6.

Maps showing sites of observations used in this study: (a) groundwater wells and soil moisture sites and (b) streamflow gauges (including considered, selected, and dam locations), pan evaporation, and air humidity sites. Bottom color bar gives percent increase in July P.

Figure 7.

Observed July, August, and September water table depth (m below land surface) at 10 long-term monitoring sites, with a linear regression line fitted to data over 1940–1980.

Table 1. Information on Groundwater Observation Wells Used in This Studya
SiteSite NameLatitudeLongitudeLand Elevation (m)Well Depth (m)Begin YearEnd YearObservation FrequencyMean Water Table Depth (m)
  • a

    Wells with early data are shown in Figure 7 and Table 2. USGS, U.S. Geological Survey; ISWS, Illinois State Water Survey; WARM, Water and Atmosphere Resource Monitoring; ICN, Illinois Climate Network.

Wells With Early Data
USGS 412220089290301USGS 141.37−89.47212.758.841942199010 days3.34
ISWS-WARM W11Cambridge  246.9812.819612004monthly2.91
ISWS-WARM W21Galena  222.697.6219632007monthly6.50
ISWS-WARM W31Mount Morris  282.0316.7619602007monthly5.85
ISWS-WARM W41Crystal Lake  273.045.4919502007monthly1.57
ISWS-WARM W61Barry  190.88.5319562007monthly3.60
ISWS-WARM W91Snicarte  148.2912.819582007monthly11.29
ISWS-WARM W171Sparta/Eden  156.068.2319602007monthly2.27
ISWS-WARM W181SWS 2  128.3524.3819522004monthly4.52
ISWS-WARM W191Dixon Springs  131.672.7419552007monthly0.96
Mean        4.28
 
Other WARM Wells
ISWS-WARM W53Fermi  233.574.5719882007monthly2.04
ISWS-WARM W72Good Hope  233.179.1419802007monthly2.44
ISWS-WARM W132Greenfield  185.936.7119652007monthly3.52
ISWS-WARM W143Janesville  220.523.3519692007monthly1.66
ISWS-WARM W153St. Peter  182.274.5719652007monthly0.94
ISWS-WARM W202Harrisberg  116.133.3519842007monthly1.39
ISWS-WARM W221Boyleston  123.607.0119842007monthly1.44
ISWS-WARM W1120Bondville  213.916.4019822007monthly1.27
Mean        1.84
 
ICN Wells at SM Sites
ISWS-ICN W10Bellville38.52−89.88133.006.1020002010daily1.65
ISWS-ICN W1Bondville40.05−88.37213.006.1020012010daily1.44
ISWS-ICN W3Brownstown38.95−88.95177.004.5719972010daily0.98
ISWS-ICN W11Carbondale37.70−89.23137.007.6220012010daily1.53
ISWS-ICN W5DeKalb41.85−88.85265.007.6219972010daily1.02
ISWS-ICN W2Dixon Springs37.45−88.67165.002.7420082010daily10.28
ISWS-ICN W34Fairfield38.38−88.80136.006.4020032010daily0.98
ISWS-ICN W13Freeport42.28−89.67265.007.6220042010daily5.28
ISWS-ICN W6Monmouth40.92−90.73229.007.6219972010daily4.62
ISWS-ICN W12Olney38.73−88.10134.005.7920032010daily1.17
ISWS-ICN W8Peoria40.70−89.52207.0012.1920072010daily1.56
ISWS-ICN W4Perry39.80−90.83206.006.1020012010daily2.48
ISWS-ICN W14Rend Lake38.13−88.92130.006.1020042010daily1.46
ISWS-ICN W9Springfield39.68−89.62177.008.5320042010daily1.73
ISWS-ICN W15Stelle40.95−88.17213.004.5720012010daily0.89
Mean        2.47
 
Other ICN Wells
ISWS-ICN W3Kilbourne40.17−90.08152.00 20022010daily9.14
ISWS-ICN W20St. Charles41.90−88.37226.00 20002010daily5.99
ISWS-ICN W22Big Bend41.64−90.04182.00 20052009daily4.52
Mean        6.55

[18] In July, more sites showed an upward trend despite the flat or downward trend in May and June P. In August and September, the upward trend is more apparent. Table 2 gives the result of the statistical test for water table trends in July to September over the period of 1940–1980, using the nonparametric Mann-Kendall test [Mann, 1945; Kendall, 1975]. Eight of the 10 sites show a rising trend in the July water table, but it is statistically significant (at 5% level, or p < 0.05, shown in bold) at only two sites, and one site (W191) has a significant falling trend. In August and September, the number of sites with significant rising trends increased, consistent with our expectation that if the signal of equation image is to be detected in the groundwater, it would be in August and September. The decreasing trends at W61 (August) and W191 (August–September) are unexplained.

Table 2. Results of Water Table Trend Analysis Over 1940–1980 Using the Mann-Kendall Testa
SiteRecord PeriodJuly Water TableAugust Water TableSeptember Water Table
Mann-Kendall Test Statistic SZ Statisticp ValueTrendMann-Kendall Test Statistic SZ Statisticp ValueTrendMann-Kendall Test Statistic SZ Statisticp ValueTrend
  • a

    Bold indicates statistically significant at the 5% level.

USGS11943–1990−237−2.96700.00rising−343−4.29960.00rising−353−4.42530.00rising
W111962–2004−45−1.66660.10rising−75−2.58890.01rising−39−1.32950.18rising
W211963–2006−44−1.77130.08rising−51−2.06140.04rising−34−1.35940.17rising
W311961–2006−52−1.65470.10rising−54−1.71950.09rising−62−1.97910.05rising
W411950–2006−203−3.60390.00rising−219−3.88930.00rising−246−4.16470.00rising
W611956–2006541.60040.11falling391.00360.32falling−37−0.95080.34rising
W911958–2006−45−1.16210.25rising−35−0.89800.37rising−53−1.46630.14rising
W1711961–2006−5−0.13990.89rising−17−0.55980.58rising−33−1.03880.30rising
W1811955–2006−63−1.29250.20rising−81−1.76330.08rising−55−1.12570.26rising
W1911952–20061633.78450.00falling1262.75590.01falling701.52120.13falling

[19] We also evaluate the field significance of the trend test results, which is necessary when assessing regional trends at multiple sites [e.g., Livezey and Chen, 1983; Lettenmaier et al., 1994; Douglas et al., 2000; Yue and Wang, 2002; Renard et al., 2008; and Khaliq et al., 2009]. Field significance (equation image) is the combined significance of N tests; if the percentage of significant results is greater than equation image, then the results are said to be field significant. Two methods can be used. If the sites are spatially independent, equation image follows the binomial distribution. The wells used for the trend analysis (orange and green in Figure 6a) are isolated from one another by several streams, and we consider them hydrologically independent (e.g., land use change or pumping near one well will not affect another). The binomial test [Livezey and Chen, 1983] indicates that the water table trends are field significant at the 5% level in August and September but not in July (they might have occurred by chance). If the multiple sites are not independent, then the regional Kendall's S test [Douglas et al., 2000] is appropriate, results of which suggest that the water table trends are not field significant in any of the months.

3.2. Changes in Streamflow

[20] Streamflow records were obtained from the USGS National Water Information System (NWIS) database (http://nwis.waterdata.usgs.gov/nwis/sw) for a total of 1428 gauges in the Ohio River basin and 343 in the Illinois River basin. We selected 46 gauges (24 in the Ohio basin and 22 in the Illinois basin) for this study according to the following criteria. First, they are located in areas where more than 10% of July equation image is detected (see Figure 2a). Second, their records cover at least 30 years, starting no later than 1941 and ending no earlier than 1970. Third, the streams are not affected by reservoirs which cause significant changes in streamflow, especially during summer months, making attributions of change difficult [Yang et al., 2004; Haddeland et al., 2007]; however, those gauges where regulation began after 1970 are retained with the data after the changes removed. Fourth, these streams do not drain into one another, so each gauge represents an independent measurement; if one drains into another, the larger basin is retained. Figure 6b gives the location of the 46 gauges selected (yellow), as well as all the gauges considered (pink) and the dams (light blue) that rendered many stations unusable. More information on the gauges is in Table 3.

Table 3. Information on the 46 Stream Gauges Used in This Study
Gauge SiteUSGS GaugeLatitudeLongitudeStateRecord PeriodRiver NameDrainage Area (km2)
1334550038.9364−88.0225Illinois1915–2009Embarras River3,926
2338050038.3583−88.5847Illinois1909–2008Skillet Fork1,202
3334600039.0100−87.9456Illinois1941–2009North Fork Embarras River824
4337800038.3864−87.9756Illinois1941–2009Bonpas Creek591
5541900042.25278−90.28583Illinois1935–1977Apple River640
6542000042.11417−90.09278Illinois1941–1977Plum River596
7543550042.30278−89.61944Illinois1915–2009Pecatonica River3,434
8544000042.19444−88.99889Illinois1940–2009Kishwaukee River2,846
9544050042.09889−89.05194Illinois1940–1971Killbuck Creek303
10544400041.90278−89.69611Illinois1940–2009Elkhorn Creek378
11544800041.44222−90.55583Illinois1940–2008Mill Creek162
12546650041.18694−90.96722Illinois1935–1972Edwards River1,153
13546700041.12889−90.91917Illinois1935–2009Pope Creek451
14546900041.00139−90.85417Illinois1935–2009Henderson Creek1,119
15546950040.85694−90.86389Illinois1940–1971South Henderson Creek215
16550204039.69306−91.14861Illinois1940–1986Hadley Creek188
17551300039.44306−90.79583Illinois1940–1986Bay Creek383
18552750041.34667−88.18639Illinois1915–2009Kankakee River13,338
19552900042.08167−87.89056Illinois1941–2009Des Plaines River932
20554050041.52222−88.19250Illinois1941–2008Du page River839
21554200041.28611−88.35972Illinois1940–2009Mazon River1,178
22555550041.25528−89.01222Illinois1932–1971Vermilion River3,310
23555650041.36583−89.49833Illinois1936–2009Big Bureau Creek508
24558300040.12417−89.98500Illinois1940–2009Sangamon River15,289
25559200039.40722−88.78139Illinois1941–1970Kaskaskia River2,730
26559700037.90139−89.01389Illinois1908–1971Big Muddy River2,056
27332650040.5764−85.6600Indiana1924–2009Mississinewa River1,766
28334000039.9311−87.1258Indiana1941–1971Sugar Creek1,735
29327500039.5794−85.1581Indiana1929–2009Whitewater River1,352
30332800040.9939−85.7814Indiana1930–2009Eel River1,080
31336350039.4175−85.6342Indiana1931–2009Flatrock River785
32325250038.3908−84.3031Kentucky1938–2009South Fork Licking River1,608
33340650037.1711−84.2961Kentucky1936–2009Rockcastle River1,564
34331400036.8953−86.3806Kentucky1941–1971Drakes Creek1,238
35343800036.7778−87.7217Kentucky1940–2009Little River632
36321700038.5642−82.9522Kentucky1941–2009Tygarts Creek627
37329900037.4972−85.3239Kentucky1938–1992Rolling Fork619
38321950040.4194−83.1972Ohio1925–2010Scioto River1,469
39323050039.7006−83.1103Ohio1922–2009Big Darby Creek1,383
40326500040.0578−84.3561Ohio1917–2009Stillwater River1,303
41323750038.8036−83.4211Ohio1926–2010Ohio Brush Creek1,002
42323200039.3792−83.3756Ohio1927–2009Paint Creek645
43323850038.8581−83.9286Ohio1925–2009White Oak Creek565
44326700040.1075−83.7992Ohio1926–2009Mad River420
45343450036.1219−87.0989Tennessee1926–2009Harpeth River1,764
46343600036.5153−87.0589Tennessee1939–1991Sulphur Fork Red River482

[21] Monthly flow at these 46 gauges is plotted in Figure 8, with the 5 year moving average shown in blue and the period of interest shaded gray. Casual inspection suggests that many sites experienced increasing streamflow. A trend analysis was performed using the Mann-Kendall test, with the results given in Table 4 (statistically significant trends, at the 5% level, are in bold). Over the month of July, 34 of the 46 sites show an upward trend, but only four are significant; for August, 42 of the 46 sites show an upward trend, with eight being significant; for September, 40 of the 46 sites have an upward trend, with 12 being significant. This result is consistent with our expectations that if the signal of July equation image can be detected in streamflow records, it could be in July from increased surface runoff but would be more likely in August and September from increased groundwater base flow because the latter accounts for >75% of streamflow in the region.

Figure 8.

Observed July–September streamflow at 46 gauges. Blue curves are 5 year moving averages to show the long-term variability.

Figure 8.
Figure 8.
Table 4. Results of Streamflow Trend Analysis Over 1940–1980 Using the Mann-Kendall Testa
Stream SitesStateRecord PeriodBasin Area (km2)July StreamflowAugust StreamflowSeptember Streamflow
Mann-Kendall Test Statistic SZ Statisticp ValueTrendMann-Kendall Test Statistic SZ Statisticp ValueTrendMann-Kendall Test Statistic SZ Statisticp ValueTrend
  • a

    Bold indicates statistically significant at the 5% level.

1Illinois1915–20093926120.12360.900increasing1781.98810.047increasing1661.85330.064Increasing
2Illinois1909–20081202420.46050.650increasing1021.13440.257increasing−4−0.03370.739Decreasing
3Illinois1941–2009824981.13020.258increasing1561.80590.071increasing180.19810.842Increasing
4Illinois1941–20095911001.15350.248increasing981.13020.258increasing750.86220.389Increasing
5Illinois1935–197764030.02510.979increasing−47−0.57830.563decreasing−27−0.32690.744Decreasing
6Illinois1941–1977596100.11770.906increasing180.22230.824increasing100.11770.906Increasing
7Illinois1915–2009343480.07860.937increasing−12−0.12360.902decreasing340.37070.711Increasing
8Illinois1940–200928461902.12280.033increasing1822.03300.042increasing1501.67360.094Increasing
9Illinois1940–1971303901.44330.149increasing180.27570.783increasing−10−0.14590.884Decreasing
10Illinois1940–20093782002.23520.025increasing1461.62860.103increasing1441.60620.108Increasing
11Illinois1940–2008162−4−0.03370.973decreasing350.38190.703increasing620.68510.493Increasing
12Illinois1935–19721153−36−0.54230.587decreasing40.04650.963increasing00.00001.000Increasing
13Illinois1935–2009451−48−0.52790.597decreasing1081.20180.229increasing840.93230.351Increasing
14Illinois1935–20091119−8−0.07860.937decreasing1241.38150.167increasing1321.47140.141Increasing
15Illinois1940–1971215−94−1.50810.131decreasing20.01620.987increasing−6−0.08110.935Decreasing
16Illinois1940–1986188−28−0.30330.762decreasing−9−0.08990.928decreasing1061.17940.238Increasing
17Illinois1940–1986383−32−0.34820.727decreasing1401.56120.119increasing1641.83080.067Increasing
18Illinois1915–200913,338500.55040.582increasing1301.44890.147increasing1852.06680.039Increasing
19Illinois1941–20099322022.34190.019increasing3243.76330.000increasing3143.64680.000Increasing
20Illinois1941–20088392763.20400.001increasing3423.97300.000increasing3193.70530.000Increasing
21Illinois1940–20091178740.81990.412increasing1081.20180.229increasing3083.44820.001Increasing
22Illinois1932–19713310240.37300.709increasing−50−0.79460.427decreasing−14−0.21080.833Decreasing
23Illinois1936–2009508−38−0.41560.677decreasing100.10110.920increasing1281.42650.154Increasing
24Illinois1940–200915,289−8−0.07860.937decreasing1681.87570.061increasing1862.07790.038Increasing
25Illinois1941–19702730−55−0.96340.335decreasing310.53520.593increasing70.10700.915Increasing
26Illinois1908–19712056180.27570.782increasing20.01620.987increasing−26−0.40540.685Decreasing
27Indiana1924–20091766520.57280.566increasing1581.76340.078increasing1962.19020.029Increasing
28Indiana1941–19711735590.98580.324increasing430.71380.475increasing370.61190.541Increasing
29Indiana1929–200913521201.33660.181increasing1321.47140.141increasing1321.47140.141Increasing
30Indiana1930–2009108080.07860.937increasing890.98850.323increasing1441.60620.108Increasing
31Indiana1931–20097851321.47140.141increasing1942.16780.030increasing1161.29170.196Increasing
32Kentucky1938–200916081081.20180.229increasing2402.68440.007increasing2803.13370.002Increasing
33Kentucky1936–20091564−42−0.46050.645decreasing80.07860.937increasing1681.87570.061Increasing
34Kentucky1941–19711238290.47590.634increasing400.66300.507increasing390.64590.518Increasing
35Kentucky1940–2009632460.50540.613increasing850.94350.345increasing1191.32550.185Increasing
36Kentucky1941–2009627360.40780.683increasing780.89710.370increasing1761.96560.049Increasing
37Kentucky1938–1992619−51−0.56160.574decreasing1441.60620.108increasing1781.98810.046Increasing
38Ohio1925–20101469420.46050.645increasing550.60660.544increasing1481.65110.099Increasing
39Ohio1922–20091383900.99960.317increasing1731.93200.053increasing2402.68440.007Increasing
40Ohio1917–20091303660.73010.465increasing580.64020.522increasing1221.35910.174Increasing
41Ohio1926–201010021101.27000.204increasing2763.20400.001increasing1681.94570.052Increasing
42Ohio1927–2009645991.66560.095increasing851.42770.153increasing1652.78740.005Increasing
43Ohio1925–20095651481.65110.099increasing2682.99890.003increasing1381.53880.124Increasing
44Ohio1926–20094201421.58370.113increasing941.04460.296increasing1481.65110.099Increasing
45Tennessee1926–200917641001.11200.266increasing360.39310.694increasing2022.25760.024Increasing
46Tennessee1939–19914821321.47140.141increasing1701.89820.058increasing1091.21310.225Increasing

[22] We assess the field significance of the streamflow trends. The 46 gauges were chosen to be independent of one another by excluding nested basins. The binomial test indicates that similar to water table trends, the streamflow trends are field significant at the 5% level in August and September but not in July. The regional Kendall's S test, if independence cannot be assumed, suggests the same.

[23] Although it may be concluded on the basis of the previous analyses that groundwater storage and streamflow in the study region has increased in August and September since the onset of High Plains irrigation development, we have not yet established a link to the increased July P. Evidence of such a link may be found in the soil moisture, the filter between the climatic forcing and the groundwater-river system.

3.3. Changes in Soil Moisture

[24] SM at 11 levels down to 2 m depth is observed over 1981–2004 at 18 sites across the state of Illinois (Figure 6a, brown symbols). The observations began after the period of irrigation expansion (1940–1980), but a close examination of how, in the postirrigation era, July rainfall propagates through the shallow to the deeper soils in years with above-normal July P may shed some light on whether the July equation image signal can reach the deeper soil depths and recharge the water table.

[25] Table 5 gives the P anomaly in May–September covering the period of SM observations (1981–2004) based on 316 long-term precipitation station data obtained from the NCDC and averaged over region 3 (Figure 2a). It is calculated as the deviation of monthly P from the 1980–2004 mean and divided by the mean (i.e., (P − mean)/mean). We examine three years, 1986, 1992, and 2003, when a wet July is sandwiched between a normal or dry June and a dry August. Here any positive anomaly in the soil moisture may be attributable to the above-normal July P, allowing us to see whether a positive July P anomaly alone can reach the deep soil.

Table 5. Warm Season Precipitation Anomaly (%) Based on the Mean of 316 Station Records in Region 3 Over the Period of 1980–2004 When Soil Moisture Observations Are Availablea
YearMayJuneJulyAugustSeptember
  • a

    The anomaly is calculated as monthly P deviation from the 1980–2004 mean divided by the mean. The years 1986, 1992, and 2003 are examined. Region 3 is shown as a green box in Figure 2a.

1980−0.3460.035−0.1540.4510.276
19810.1410.2060.3190.290−0.094
1982−0.064−0.0940.2260.137−0.174
19830.329−0.264−0.392−0.300−0.159
19840.091−0.273−0.125−0.4070.237
1985−0.242−0.039−0.1930.338−0.132
19860.0050.0000.243−0.2280.977
1987−0.365−0.1980.1610.266−0.263
1988−0.630−0.748−0.157−0.1410.063
1989−0.088−0.084−0.0440.0470.094
19900.4420.335−0.0450.165−0.277
1991−0.016−0.451−0.247−0.224−0.072
1992−0.565−0.4440.607−0.2830.481
1993−0.1470.5650.3250.2950.776
1994−0.5130.047−0.0610.011−0.186
19950.515−0.155−0.1760.190−0.497
19960.3350.2230.049−0.4500.182
19970.0100.104−0.2880.147−0.249
1998−0.1110.7640.010−0.013−0.393
1999−0.1970.087−0.014−0.338−0.463
20000.1090.548−0.013−0.0250.147
20010.083−0.005−0.0630.0900.159
20020.363−0.022−0.2570.032−0.053
20030.223−0.1130.214−0.2380.323
20040.638−0.0240.0760.189−0.702

[26] Biweekly soil moisture observations in Illinois are obtained from the Global Soil Moisture Databank (http://climate.envsci.rutgers.edu/soil_moisture/illinois.html) at three depths: 0.1–0.3, 0.9–1.1, and 1.7–1.9 m. The topmost (0–0.1 m) and bottommost (1.9–2.0 m) layers have many missing data; hence, the next shallowest and deepest layers are used. Soil moisture anomaly is calculated for each site and month as the deviation from the mean divided by the mean, the latter being obtained from the entire record (1981–2004) for each site for the respective layer and month. The regional anomaly is then calculated as the mean anomaly of the 18 sites. Figure 9 plots the P and SM anomalies at three depths over the warm season of the three years.

Figure 9.

Anomalies in regional mean precipitation (based on 316 station records) and soil moisture (based on 18 site observations) at three depths, May–September of (a) 1986, (b) 1992, and (c) 2003. (d) The long-term mean water table depth distribution based on 34 wells in Illinois (data source is USGS and Water and Atmosphere Resource Monitoring and Illinois Climate Network groundwater monitoring networks, both run by Illinois State Water Survey (data in Table 1)).

[27] In 1986 (Figure 9a), the entire soil moisture profile is near normal in June because of the near-normal P in both May and June. The above-normal July P not only wetted the shallow soil but also elevated the deeper soils to above normal. This positive anomaly in the deeper soils persisted into August despite the below-normal August P. The July P anomaly here (24.3% increase) is at a similar magnitude to the July equation image signal (Figure 2a). In 1992 (Figure 9b), despite the large precipitation and soil water deficit in May and June, the above-normal July P wetted the deepest soil layer to above normal, which persisted into August despite the large deficit in August P. In 2003, the below-normal soil moisture in the deep layers in June is elevated to above-normal values by the above-normal July P. These cases suggest that a positive July P can reach the deeper soils (1.7–1.9 m) despite the normal to dry antecedent soil moisture conditions and high ET rates in July and August.

[28] Water table observations are available at 15 Illinois Climate Network (ICN) wells collocated with 14 of the 18 soil moisture sites used in the above analyses (Figure 6a and the ICN wells at SM sites in Table 1). The temporal (over 1998–2009) and spatial (over 15 sites) mean water table depth at these ICN wells is 2.47 m, not far from the 1.7–1.9 m soil layer analyzed above. To further characterize the groundwater conditions in Illinois, we compiled observations from a total of 34 wells, including the 10 historic wells used in the trend analyses earlier (wells with early data in Table 1), the shorter ISWS-WARM well records (other WARM wells), the 15 ICN wells collocated with soil moisture sites (ICN wells at SM sites), and the rest of the ICN wells (other ICN wells). All data are maintained by ISWS (see http://www.isws.illinois.edu/warm/sgwdata/wells.aspx) except for the USGS well, and their locations are shown in Figure 6a. The temporal mean at the 34 wells gives the frequency distribution of water table depth in space shown in Figure 9d. It suggests that the water table in Illinois clusters around the 1–2 m depth, with 53% of the sites <2 m and 68% <3 m deep. If the above-normal July P in 1986, 1992, and 2003 could reach the soils at the 1.7–1.9 m depth with normal to dry antecedent soil moisture conditions, then the July equation image signal might have also reached the shallow water table, at least in the years with normal to wet antecedent soil moisture conditions.

3.4. Changes in ET

[29] Last, we address the role of possible changes in ET. The increased groundwater storage and streamflow in August–September could have been caused by the increased July P, but it also could have been caused by reduced July ET because it is P − ET, the net soil water surplus, that reaches the water table. Since actual ET is not routinely and historically observed, we infer changes in ET from changes in those variables that are historically observed and indicative of ET, such as maximum air temperature, pan evaporation, air relative humidity, and atmospheric vapor density deficit computed from the latter two.

[30] July mean daily maximum air temperature (Tmax) averaged over 104 station records in the states of Illinois and Indiana (data from the NCDC) is plotted in Figure 10a. A notable cooling began in the mid-1950s and continued to the late 1970s. This is consistent with the observed U.S. [e.g., Liepert, 2002] and global-scale cooling due to reduced solar radiation over the period of 1950–1980 (i.e., solar dimming; see recent review by Wild [2009]) caused by changes in anthropogenic aerosols and their interaction with changes in clouds. In the central United States, the cooling has also been linked to large-scale land use changes such as converting forest to crops and, particularly, irrigation [e.g., Bonan, 2001; Govindasamy et al., 2001; Milly and Dunne, 2001; Baidya Roy et al., 2003; Boucher et al., 2004; Feddema et al., 2005; Lobell et al., 2006, 2008; Adegoke et al., 2007; Kueppers et al., 2007; Diffenbaugh, 2009]. The mechanisms include the higher albedo (reflecting more solar radiation) of croplands, increased latent versus sensible heat due to irrigation, and, indirectly, increased cloud cover caused by higher ET. A recent global model simulation study [Puma and Cook, 2010] forced by observed sea surface temperature and reconstructed global irrigation development history shed much light on the cause of the cooling by illustrating that cooling occurred in both the irrigation and nonirrigation ensemble simulations but more so in the ensemble with irrigation. Although the causes might have been multiple, the cooling is certain.

Figure 10.

(a) July mean maximum daily temperature (°C) averaged over 104 stations, (b) station pan evaporation (mm) at five sites, and (c) surface air temperature, humidity, and vapor pressure deficit at three sites (5 year moving average is in blue; period of interest is shaded gray).

[31] The relevance of this cooling to the present study is that ET might have been reduced since the 1950s, allowing deeper infiltration and water table recharge, without additional precipitation. Records of observed pan evaporation, a direct indicator of atmospheric ET demand, are available from six stations in Illinois and Indiana (from the NCDC) dating back to at least the 1950s and continuing to the 1980s. Figure 10b plots the July total pan evaporation at these six sites, and Table 6 gives the site information and the result of the Mann-Kendall trend analysis. There are no statistically significant (at the 5% level) trends in any of these records, suggesting that the cooling alone may not have caused a change in the atmospheric ET demand.

Table 6. July Pan Evaporation Site Information and Mann-Kendall Test Results for Trends Over 1940–1980a
Coop IDSite NameStateLatitudeLongitudeElevation (m)PeriodMean (mmol)July Pan Evaporation
Mann-Kendall Test Statistic SZ Statisticp ValueTrend
  • a

    No significant trends (at the 5% level) are found at the six sites. Coop ID, cooperative station ID.

118179Springfield Capital APIllinois39.83−89.681811948–1990227360.69200.49increasing
122309Dubois S in Forage FMIndiana38.45−86.702101957–1999180130.29800.77increasing
122738Evansville Regional APIndiana38.03−87.521221948–1987205580.88360.38increasing
126506Oaklandon Geist RSVRIndiana39.90−85.982421937–1998159−119−1.37550.17decreasing
128999Valparaiso WTR WKSIndiana41.50−87.032441948–1999150−62−0.94570.34decreasing
129430West Lafayette 6 NWIndiana40.47−86.982181957–1999192330.79470.43increasing

[32] To supplement these few pan evaporation records, we assessed changes in ET demand by computing the atmospheric vapor pressure deficit (VPD) from relative humidity (RH) and air temperature (Ta) records. We found long-term observations of air humidity at only three stations from the NCDC archive; most of the long-term records have a data gap (1948–1973) over our period of interest (1940–1980), unfortunately. The VPD is computed as

equation image

where e* (kPa) is the saturation vapor pressure at air temperature Ta (°C). Figure 10c plots the July Ta, RH, and VPD (with 5 year moving averages). At first glance, the two short records suggest an upward VPD trend over 1940–1980, though the lack of data in the 1940s and 1950s makes them difficult to judge. On the basis of a trend analysis (Table 7), no significant trends (at the 5% level) are found at the three sites.

Table 7. July Relative Humidity and Temperature Site Information and Mann-Kendall Test Results for Trends in the Atmosphere Vapor Pressure Deficit (VPD) Over 1940–1980a
U.S. Air Force SiteSite NameStateLatitudeLongitudeElevation (m)PeriodJuly Atmosphere VPD
Mann-Kendall Test Statistic SZ Statisticp ValueTrend
  • a

    No significant trends (at the 5% level) are found at the three sites.

725300Chicago/O'Hare ARPTIllinois41.986−87.914631946–1977901.75830.08increasing
724338Scott AFB MidAmericIllinois38.545−89.835431938–1998−48−0.52790.60decreasing
725335Grissom ARBIndiana40.650−86.150751955–1993671.74310.08increasing

[33] However, neither pan evaporation nor VPD are sufficient to infer the actual ET because they only tell half of the story (the atmospheric demand side); soil water availability (or the land supply side) can be the dominate control where ET is water limited. The relationships among pan evaporation, VPD, and actual ET are complex and multidimensional, involving land-atmosphere feedbacks, vegetation and land cover, and changes in the dominant forcing [e.g., Brutsaert and Parlange, 1998; Lawrimore and Peterson, 2000; Teuling et al., 2009; van Heerwaarden et al., 2010]. Nevertheless, actual ET in the region has likely increased, rather than decreased, over the period of 1940–1980 for the following reasons.

[34] First, the available pan evaporation and VPD records in the region suggest no significant changes in atmospheric ET demand in July (Figures 10 and 11 and Tables 6 and 7). If the atmospheric demand stayed the same, then any changes in the actual ET would have been caused by changes in soil water availability. This is to assume that wind speed and land cover have not changed significantly or they are a weak driver of ET. It has been shown that ET is insensitive to reduction of wind speed (or stilling [see van Heerwaarden et al., 2010]). Since precipitation has increased, soil should be wetter, in general. Hence, if actual ET has changed at all, it is more likely that it has increased rather than decreased.

Figure 11.

Changes in streamflow seasonal cycle at the 46 gauges (as percent annual total).

[35] Second, the idea that the actual ET in the region is driven by precipitation rather than temperature is supported by a simple and elegant study by Teuling et al. [2009], who conclude that changes in actual ET are governed by changes in its key driver (or limiting factor) in a given region. That study shows that annual ET in the central United States, inferred from flux tower and multimodel syntheses, is far more responsive to changes in P than changes in radiation. If this holds true for annual ET, it must hold true for warm season ET because the latter is more water limited than all season ET (Figure 3a, P < ET in warm season).

[36] Third, the annual river basin water balance analyses in the same study [Teuling et al.., 2009] demonstrate that ET has increased over the period of solar dimming in the upper Mississippi (including Illinois) and the Ohio river basins. The upward trend in annual ET is explained by the upward trend in annual P, which is partitioned into both increased ET and increased streamflow, as has been shown by Milly and Dunne [2001] and Qian et al. [2007] over the same region and period. Summer ET dominates annual ET, and if annual ET has increased, then summer ET has likely increased as well.

[37] Thus, it is plausible that the increased July P caused both increased ET and increased streamflow. This is corroborated by our earlier seasonal analysis, which suggested that ET was likely to increase in response to the July equation image signal because ET exceeds P and there is a net soil water deficit in the warm season. It is also consistent with our earlier soil moisture analysis, which shows that present day above-normal July P could reach the deep soil in dry to normal antecedent soil moisture conditions despite high ET demand. There is no evidence that ET has decreased because of cooling. We conclude that the observed increase in late summer groundwater storage and streamflow in the Midwest is caused by the increased July precipitation.

4. Summary and Discussion

[38] In this study, we set out to detect changes in land hydrology in response to the increased July precipitation over the U.S. Midwest attributed to High Plains irrigation in an earlier study [DeAngelis et al., 2010]. Seasonal analysis of hydrologic variables over Illinois suggests that the seasonal cycle of P − ET is followed by the soil moisture cycle 1 month later, which is followed by the water table and streamflow cycles another month later; thus, it is expected that the increased July P may be detected in August–September groundwater and streamflow. We analyzed 30 year and longer time series of water table depth at 10 wells in Illinois and streamflow at 46 gauges in the Illinois and Ohio river basins. The Mann-Kendall test for trends indicates that groundwater storage and streamflow have increased in August–September since the onset of irrigation in the High Plains, and these trends were determined to be field significant. Examination of soil moisture response to above-normal July P in the postirrigation era suggests that the increased July P due to High Plains irrigation can be sufficient to reach the shallow water table, at least in normal to wet years, hence providing a possible link between increases in July P and groundwater storage and streamflow. The Mann-Kendall test for trends in pan evaporation and atmospheric vapor pressure deficit, both indicators of atmospheric ET demand, suggests that the ET demand has remained constant. The latter points to the soil water availability as the driver in changes in ET and the possibility of increased ET because of the increased P. An annual water balance study by Teuling et al. [2009] gives further evidence of increased ET due to increased P. By ruling out the reduction in ET as a cause, we conclude that the observed increase in groundwater storage and streamflow in the Midwest is linked to the increased July precipitation attributed to High Plains irrigation.

[39] We briefly address the effect of land use change. Historic reconstructions [e.g., Ramankutty and Foley, 1999; Bonan, 2001] of the Midwestern states suggest accelerated conversion of forest to cropland over 1850–1900, but it has slowed down significantly since then. In addition, forest to cropland conversion has been shown to increase ET [Bonan, 1999, 2001; Baidya Roy, 2003; Diffenbaugh, 2009], not to decrease ET. Urban expansion can also affect water budget, but greater paved area is known to reduce groundwater recharge, not to increase it. Therefore, land use change cannot explain the observed increase in late summer groundwater storage and streamflow.

[40] Last, to place the observed increase in summer streamflow in the context of seasonal dynamics, we plot in Figure 11 the seasonal cycle difference at the 46 gauges between two periods, 1940–1960 (early irrigation development) and 1960–1980 (late irrigation development). Changes in August–September, the focus of this study, are rather small compared to changes in March–April at many gauges; hence, its signal is buried in the total annual flow, which is often the focus of regional hydrologic change studies [e.g., Groisman et al., 2001; Zhang and Schilling, 2006; Qian et al., 2007; Kalra et al., 2008; Raymond et al., 2008]. We note that it is important to isolate the signals of change in different seasons because they are likely caused by different mechanisms. Although the signal of summer change is small, it is conceptually significant in that it may point to human modification of the water cycle in the far-away High Plains region as a possible source and cause.

[41] Our results and their interpretations are limited by the available observations, particularly the sparse and short records in water table depths, the lack of soil moisture observations in the preirrigation era, and, particularly, the lack of actual ET measurements. A regional climate-hydrology model simulation over the irrigation development era, similar to the approach by Puma and Cook [2010] but with fully integrated hydrologic (including groundwater) and climatic interactions and feedbacks, may help to disentangle the different causes of the observed hydrologic changes in the region.

Acknowledgments

[42] Support comes from the U.S. National Science Foundation (NSF-ATM-0450334), U.S. Environmental Protection Agency (EPA-STAR-R-834190), and Rutgers University Academic Excellence Fund (AEF2006-07). We thank Mathieu Gerbush and David Robinson at Rutgers University Climate Lab for assisting us in obtaining climate data from the archive at the National Climate Data Center. We also thank three anonymous reviewers for their thoughtful comments and the editors for their handling of the manuscript.