Recent Acceleration of the Terrestrial Hydrologic Cycle in the U.S. Midwest
Abstract
Most hydroclimatic trend studies considered only a subset of water budget variables; hence, the trend consistency and a holistic assessment of hydrologic changes across the entire water cycle cannot be evaluated. Here we use a unique 31 year (1983–2013) observed data set in Illinois (a representative region of the U.S. Midwest), including temperature (T), precipitation (P), evaporation (E), streamflow (R), soil moisture, and groundwater level (GWL), to estimate the trends and their sensitivity to different data periods and lengths. Both the Mann-Kendall trend test and the least squares linear method identify trends in close agreement. Despite no clear trends during 1983–2013, increasing trends are found in P (8.73–9.05 mm/year), E (6.87–7.47 mm/year), and R (1.57–3.54 mm/year) during 1992–2013, concurrently with a pronounced warming trend of 0.029–0.037 °C/year. However, terrestrial water storageis decreased by −2.0 mm/year (mainly due to declining GWL), suggesting that the increased R is caused by increased surface runoff rather than baseflow. Monthly analyses identify warming trends for all months except winter. In summer, P (E) exhibits an increasing (decreasing) trend, leading to increasing R, soil moisture, GWL, and terrestrial water storage. Most trends estimated for different subperiods are found to be sensitive to data lengths and periods. Overall, this study provides an internally consistent observed evidence on the intensification of the hydrologic cycle in response to recent climate warming in U.S. Midwest, in agreement with and well supported by several recent studies consistently reporting the increased P, R and E over the Midwest and Mississippi River basin.
Key Points
- Yearly and monthly trends for temperature and water budget variables in Illinois are estimated by two trend estimation methods
- Upward trends are identified in all P, E, and R, concurrently with a pronounced warming trend (+0.03–0.04 °C/year)
- This study provides an internally consistent observed evidence on the intensification of hydrologic cycle over the last two decades (1992–2013) in U.S. Midwest
1 Introduction
The potential consequences of anthropogenic climate change in water resources have been widely investigated over the last two decades (Barnett et al., 2008; Intergovernmental Panel on Climate Change, 2013; Labat et al., 2004; McBean & Motiee, 2008; Wang et al., 2012). Numerous studies suggested that climate warming is likely leading to the alteration and intensification of the hydrologic cycle (Huntington, 2006; Jung et al., 2010; Milly et al., 2005; Oki & Kanae, 2006; Peterson et al., 2006), resulting in changes in water resource availability and the frequency and intensity of floods and droughts as well as amplification of warming through the water vapor feedback (Huntington, 2006). Future precipitation is more likely to arrive in the form of heavy rains accompanied an increase in flood risks (Goswami et al., 2006; Hirabayashi et al., 2013; Milly et al., 2005; Trenberth, 1998; Trenberth et al., 2003; Wu et al., 2014). A decrease in the accumulation and earlier melt of seasonal snowpack may cause a shift in the timing and amount of runoff and affect future water resources availability (Barnett et al., 2005; Barnett et al., 2008). Furthermore, the increase of evaporation associated with rising temperature can potentially result in larger depletion of lake level (Walter et al., 2004), soil moisture (SM) (Dai et al., 2004), and groundwater aquifer (Rodell et al., 2009; Rosenberg et al., 1999).
On the global scale, previous studies have indicated the linkage between increasing runoff trends and anthropogenic climate changes associated with the increasing concentration of greenhouse gases (Betts et al., 2007; Gedney et al., 2006; Labat et al., 2004; Piao et al., 2007). The observed changes in the continental river discharges in high latitudes can potentially disturb ocean circulation and affect climate (Curry et al., 2003; Peterson et al., 2006). At the regional scale, the signatures of runoff trends have been found associated with climatic variations of temperature and precipitation in the western United States (Barnett et al., 2008; McBean & Motiee, 2008), Europe (Stahl et al., 2010), Canada (Burn, 2008; Whitfield, 2001), and China (Chen et al., 2007; Wang et al., 2012; Wu et al., 2014). Furthermore, climate variation is also responsible for the detected trends of SM (Dai et al., 2004; Wu et al., 2015), evaporation (Hamlet et al., 2007; Miralles et al., 2014; Qian et al., 2007), and groundwater (Brutsaert, 2008; Taylor et al., 2013; Weider & Boutt, 2010).
Due to the paucity of long-term observed hydrologic data sets, previous trend studies either considered only a subset of water budget variables (often only precipitation and runoff) or relied on model-based data sets produced by poorly constrained large-scale simulations, including the widely used global or regional reanalysis data. In addition, the commonly recognized sensitivity of estimated trends to the lengths and periods of the data is rarely taken into consideration; hence, the stability and robustness of the identified trends cannot be assessed. Owing to these limitations, a holistic and internally consistent evaluation on the regional hydrologic changes is difficult to be obtained from previous trend studies.
In U.S. Midwest, also known colloquially as the Corn Belt, the near-surface air temperature has been rising in particular during recent decades (U.S. Global Change Research Program, 2009) consistent with anthropogenic climate change (Intergovernmental Panel on Climate Change, 2013). Climate change may have contributed to the record-breaking floods in 1993 and 2008 and the severe droughts in 1988, 2005, and 2012 (Budikova et al., 2010; Coleman & Budikova, 2010; Holmes et al., 2010; Illinois Department of Natural Resources, 2013; Kunkel et al., 2006) as well as the reduction in the Great Lake level (Assel et al., 2004). Therefore, one escalating concern for this region is the potential change in the hydrologic cycle and the associated uncertainty in response to ongoing and future climate change.
To enhance our understanding of this critical issue in an agriculturally important region, we assess whether any hydrologic changes have indeed occurred, with a focus on the most recent two decades when temperature rising was most pronounced.
Several recent studies have focused on identifying hydroclimatic trends over the Mississippi River basin (Baker et al., 2012; Gupta et al., 2015; Karl & Knight, 1998; Lins & Slack, 1999; Milly & Dunne, 2001; Qian et al., 2007; Schilling, 2016; Winter et al., 2015; Winter & Eltahir, 2012a, 2012b; Zhang & Schilling, 2006). The national-wide trend studies by Karl and Knight (1998) and Lins and Slack (1999) found that precipitation and runoff have increased substantially during the second half of last century over much of the North America including the Mississippi River basin. Combining various sources of data, Milly and Dunne (2001) analyzed the 1949–1997 trends of surface water and energy balances in the Mississippi basin and found an upward trend in runoff (+0.85 mm/year), as well as in evaporation (+0.95 mm/year) primarily driven by increased precipitation (+1.78 mm/year) and increased irrigation (+0.26 mm/year). Also, large negative trends were found in net radiation and sensible heat flux due to the increased cloudiness. Walter et al. (2004, Table 1) estimated the 1950–2000 trend over the six large river basins in the United States by using the U.S. Climate Division data (Guttman & Quayle, 1996) and U.S. Geological Survey (USGS) streamflow data and found for the Mississippi River basin an upward trend in precipitation (+1.10 mm/year), streamflow discharge (+0.65 mm/year), and evaporation (+0.95 mm/year), rather consistent with the estimates by Milly and Dunne (2001). Qian et al. (2007) analyzed the trends of water budget components by combining hydroclimatic records and model simulations and found an increase in precipitation balanced by increases in both runoff and evaporation during 1948–2004 over the Mississippi River basin.
For the U.S. Midwest, Baker et al. (2012, Figure 1) estimated the 1960–2009 hydrologic trends over the U.S. Midwestern region. Their results show that the mean precipitation has increased 1 mm/year over the majority of the Midwest region and >4 mm/year in portions of Indiana and Ohio, and this precipitation increase has generally been matched or exceeded by streamflow increase, which is estimated to be >1 mm/year for most of the Midwestern regions and >4 mm/year for some watersheds. The corresponding evaporation trends are not pronounced (within ±1 mm/year or less) for most of regions. Winter et al. (2015) compared the results from the ensembles of regional climate model simulations with the trend identified from observed precipitation and SM over the 1985–2009 period and concluded that the impacts of climate change in the Midwest regional hydrology are highly uncertain. Gupta et al. (2015) evaluated the relative importance of changes in precipitation and land use/land cover (LULC) on streamflow in 29 upper Midwestern watersheds for both periods prior to and after 1975. The LULC changes in the agricultural Midwest are often quantified by the large increase in soybean production that occurred from the mid-20th century on. Their results revealed that the increased streamflow after 1975 was mainly due to increased precipitation, and there is a lack of evidence on the impact of LULC change on streamflow; however, an earlier work by Schilling and Libra (2003) concluded that about 30% of the increased streamflow in twentieth century over many Midwestern rivers was caused by increased baseflow due to LULC change, while precipitation changes contributed about 70% (Schilling, 2016).
A unique long-term (1983–2013, 31 year) observed hydroclimatic data set of Illinois is used in this study to identify the trends for the 1983–2013 and 1992–2013 periods. The data set collected from several independent sources includes mainly temperature (T), precipitation (P), evaporation (E), streamflow (R), SM, and groundwater depth (GWL), among others. Illinois is a representative region of the U.S. Midwest. Only few previous studies used observed data sets covering all or most water budget components to estimate trends across the entire hydrologic cycle. The budget closure in the estimated trends need to be checked; therefore, it is of interest to examine whether the data sets of water budget components from different sources can yield internally consistent trends across the water cycle.
The objectives of this study are (1) to identify the trends of water budget variables of Illinois based on an observational 31 year monthly data set, (2) to examine the sensitivity of estimated trends to varying data periods and different data lengths, and (3) to provide a holistic evaluation on the intensification of the hydrological cycle in response to climate warming in U.S. Midwest. In the following, section 2 summarizes the past trend studies in U.S. Midwest and Mississippi River basin and introduces the data set and trend estimation methods used in this study. In section 3, the estimated yearly and monthly trends are summarized and discussed, and their sensitivity to varying data periods and lengths are examined. The correlations among the estimated trends of all hydroclimatic variables for varying data periods are also evaluated. Finally, the main findings drawn from this study and future research directions are summarized in section 4.
2 Data and Methodology
2.1 Data
The 1983–2013 (31 year) monthly hydroclimatic data set of Illinois used in this study is summarized in Table 1, including in total 12 hydroclimatic variables. The locations of measurement networks are shown in Figure 1, including 19 Illinois Climate Network weather stations maintained by Illinois State Water Survey (ISWS), Water and Atmospheric Resources Monitoring (WARM) Program (2018) and 11 USGS stream gauges. The 1983–2013 monthly data of mean temperature (Tmean), mean daily maximum and minimum temperature (Tmax and Tmin), precipitation (P), and snow water equivalent are acquired from 19 Illinois Climate Network stations based on the criterion of missing data less than 5%. The daily temperature data are averaged to obtain monthly data. The 11 USGS streamflow (R) stations encompassing the largest river basins collectively cover 85% of the entire Illinois areas. The 1983–2013 R data are weighted by the corresponding drainage areas to yield the state-average. Weekly to biweekly SM measurements have been collected using the neutron probe technology at totally 19 ISWS sites since 1981, 11 of which with the complete 1983–2013 record are used in this study.
| Variable | Unit | Annual mean | Yearly trend (change per year) | Data source | |
|---|---|---|---|---|---|
| (1983–2013) | 1983–2013 | 1992–2013 | |||
| Mean temperature (Tmean) | °C | 11.7 | 0.006 | 0.037 | |
| −0.011 | −0.029 | ||||
| Minimum temperature (Tmin) | °C | 5.9 | 0.015 | 0.029 | |
| −0.014 | −0.019 | ||||
| Precipitation (P) | mm/year | 962.4 | −0.787 | 8.732* | |
| −0.423 | (9.049*) | ||||
| Snow water equivalent | mm/year | 50.2 | 0.053 | 0.532 | |
| −0.103 | −0.456 | ||||
| E (mm/year) | mm/year | 640.2 | −0.949 | 6.872** | Estimated from terrestrial water budget |
| (−0.570) | (7.474**) | ||||
| Precipitable water (PW) | mm | 19.6 | −0.007 | 0.017 | ERA-Interim Reanalysis data |
| (−0.006) | −0.004 | ||||
| Streamflow (R) | mm/year | 323.3 | 0.969 | 3.543 | 10 U.S. Geological Survey (USGS) stations weighted by the drainage areas |
| −1.276 | −1.571 | ||||
| Soil moisture (SM) | mm | 719.2 | 0.359 | −0.069 | ISWS, the mean value of 11 stations |
| −0.381 | (−0.446) | ||||
| Groundwater depth (GWL) | m | −3.71 | −0.009 | −0.021 | ISWS, the mean value of 10 wells |
| (−0.004) | (−0.012) | ||||
| Terrestrial water storage anomaly (TWS) | mm | 0 | 0.021 | −1.999 | Estimated by combining the anomalies of soil moisture and groundwater |
| −0.057 | (−1.423) | ||||
| Terrestrial water storage change (TWSC) | mm/year | −1.06 | −0.05 | 0.073 | Derived by backward differencing |
| (−0.283) | −0.004 | ||||
- ** Statistical significance level p < 0.05.
- * Statistical significance level p < 0.10.
Monthly GWL data are also obtained from ISWS. Ten of the total 19 wells measuring the depth from surface to the water table with the complete 1983–2013 records are used in this study. The GWL data are converted into groundwater storage by multiplying a constant specific yield of 0.08 representative of the predominant soil type of silt loam in Illinois, following the values used by Yeh et al. (1998, 2006), Eltahir and Yeh (1999), and Yeh and Famiglietti (2008, 2009). These wells measuring the response of underlying unconfined aquifers to climate forcing are all located far away from pumping centers and streams (Yeh et al., 1998); hence, the measured fluctuations in GWL primarily reflects climatic variability and not human influences (e.g., pumping, water withdrawal and urbanization). The same GWL data set was used in previous studies (Brutsaert, 2008; Yeh & Famiglietti, 2008, 2009; and Kustu et al., 2011). The 31 year (1983–2013) monthly anomalies (i.e., deviation from the long-term mean seasonal cycle) of terrestrial water storage (TWS) are estimated by combining SM and GWL anomalies since other components such as snow water equivalent and surface storage were found to be insignificant for the monthly and annual water balances in Illinois (Yeh et al., 1998; Yeh et al., 2006; Yeh & Famiglietti, 2008, 2009). Most of snow accumulations melt within one week to 10 days, so the contribution to monthly water storage change and balance is negligible.
The 1983–2013 monthly evaporation (E) is estimated from the water balance based on observed P, R, SM, and GWL, and this estimate is within 10% of accuracy from the comparison with another estimation from atmospheric water balance (Yeh & Famiglietti, 2008). Finally, the precipitable water (PW) data are obtained from a subset of European Centre for Medium-Range Weather Forecasts Reanalysis (ERA-Interim) data (Simmons et al., 2007). Figure 2 plots 1983–2013 annual and monthly time series of all the hydroclimatic variables used in this study (snow in this figure is the water equivalent of snowfall rate in millimeter).
Figure 3 plots the correlation coefficients among all the 31 year (1983–2013) annual and monthly hydroclimatic variables. At the annual time series (Figure 3a), Tmax and Tmean show weak positive correlation with E and PW and weak negative correlation with any other variables (p < 0.01). In contrast, Tmin is strongly correlated with PW (p < 0.01). P shows strong positive correlations with R, SM, GWL, and TWS (p < 0.01). However, no significant correlation can be identified between P and E (p < 0.01). Annual snowfall shows modest negative correlation with all temperature variables and weak positive correlation with annual P and R. As anticipated, E has higher positive correlation (the correlation coefficient Corr = 0.41) with Tmax than other two temperature variables. In terms of water fluxes, E only shows significant negative correlation with TWSC (Corr = −0.61). A high positive correlation (Corr > 0.7) can be found among annual R, SM, GWL, and TWS, indicating high dependence among all of these hydrologic variables.
At the monthly time series (Figure 3b), temperature (Tmax, Tmean, and Tmin) shows much stronger positive (negative) correlation with E and PW (Snow and TWSC) (p < 0.01) than that at the annual scale. In contrast to that found at the annual scale, monthly P is only weakly correlated with R, indicating that P is not the only factor affecting R at the monthly scale. Strong positive correlation can be found between R, SM, GWL, and TWS for both monthly and annual time series (p < 0.01), but unlike that at the annual scale, all these four hydrologic variables have only rather weak positive correlation with monthly P. Finally, E shows strong positive (negative) correlation with PW (TWSC) (p < 0.01), but no significant correlation with monthly R, SM, GWL, and TWS.
2.2 Methodology
The trends of hydroclimatic variables are evaluated by using (1) the standard nonparametric Mann-Kendall trend test (M-K test) (Kendall, 1975) and (2) the least squares linear trend method (L-S method) in this study. The M-K trend test is a rank-based method widely applied in previous hydrologic literature for identifying the monotonic trends of hydroclimatic variables such as T, P, and R (Burn, 2008; Chen et al., 2007; Kustu et al., 2011; Rice et al., 2015; Stahl et al., 2010; Whitfield, 2001). Trend magnitude is estimated using the Kendall's rank correlation tau (Sen, 1968). In addition to check the consistency in estimated trends from different methods, the L-S method can ensure closure in the trends of water budget components, thus providing more internally-consistent hydrologic changes across the entire water cycle in response to environmental drivers. The significance levels of p = 0.1 and 0.05 are used in both trend test methods to judge the statistical significance. Notice that due to the differences in their algorithms, the trends identified from the L-S linear trend method can ensure the exact closure in the trends of water budget components (i.e., TWSC = P-E-R), while the trends estimated from the M-K test cannot.
Notice that in some previous trend studies, the procedures of prewhitening and effective sample size were used to reduce the influence of the presence of serial correlation with respect to the significance of M-K test (Hamed & Rao, 1998; Yue & Wang, 2002; Yue & Wang, 2004). By using the lag-one serial correlation method (Yue & Wang, 2002), it is found that more than 90% of annual and monthly time series of the total 12 hydroclimatic variables analyzed in this study (as listed in Table 1) do not show any significant autocorrelation at the 95% confidence level. Therefore, the prewhitening procedures are considered unnecessary and hence not performed in the study.
3 Results
The trend analysis conducted in this study is divided into yearly and monthly trend analysis. For yearly analysis, the trends are identified from annual time series of hydroclimatic variables, while for the monthly analysis, they are estimated for each of the 12 months based on 31 year data. The purpose of monthly trend analysis is to investigate seasonal differences in the estimated trends in order to explore the mechanisms controlling the interactions among water budget variables at seasonal time scale.
3.1 Yearly Trend Analysis: 1983–2013
Table 1 summarizes the 1983–2013 yearly trends of 12 hydroclimatic variables based on the M-K test and L-S method. The statistically significant trends with the levels of p < 0.05 and p < 0.10 are marked by the asterisks. Overall, the consistent trends in terms of magnitude and sign are estimated from both methods for most variables; hence, the trends estimated from two methods can be taken as the reasonable range of trends. For the 1983–2013 period, despite no statistically significant trends identified, Tmean exhibits a warming trend of 0.006–0.011 °C/year, close to the trend of global mean temperature. Tmin shows a larger trend (0.014–0.015 °C/year) than Tmean and Tmax (0.008–0.010 °C/year). For P, two methods estimate the opposite signs but both with a small trend magnitude. Both E and PW show a weak downward trend. Snow, R, SM, and TWS all show weakly increasing trends, while TWSC shows a weakly deceasing trend. Note that only the L-S method can ensure the closure in the estimated trend of water budget components (P, E, R, and TWSC).
3.2 Trend Sensitivity to Data Periods and Lengths
It is well recognized that the trend magnitudes may vary with different data lengths and initial and ending years (Brutsaert, 2008; Kite, 1989; McCabe & Wolock, 2002; Radziejewski & Kundzewicz, 2004; Yue et al., 2002). To investigate this sensitivity, trend tests are performed by varying the initial and ending year of the data periods. A total of 153 subperiods are selected with the minimum length of 15 years within the 1983–2013 period (17 + 16 + 15 + … + 3 + 2 + 1 = 153). Figure 4 plots the contours of estimated trends from the M-K test corresponding to the 153 subperiods for each variable. As seen, both trend magnitude and sign are extremely sensitive to varying data periods, and no consistent contour patterns among variables can be observed in Figure 4. For Tmax, Tmean, and Tmin, the pronounced trends are found only for the subperiods starting from around 1992, and the trends for the subperiods beginning earlier than 1990 are all weak. For Tmax and Tmean, the increasing trends are the strongest from 1992 to 2007, while a weakly decreasing trend can be found for the subperiods starting from 1996. For Tmin, the trends are all significant for any subperiods starting from 1992 or after. For example, the trend of Tmax during 1992–2007 (+0.1 °C/year) is about 12 times larger than that during 1983–2013 (+0.008 °C/year; Table 1).
For the hydrologic variables, P exhibits no apparent trends during 1983–2013 or even shows significantly decreasing trends for the subperiods starting from 1983–1985 to 2006–2008 (Figure 4d), in contrast to the significantly increasing trends during 1992–2013 (Figure 4d). Following the upward trends of P from 1992 to 2013, similar upward trends can be observed in E, PW, and R, which become increasingly pronounced from 1996 onward, suggesting that the intensification of the hydrologic cycle and a more humid Illinois could be in response to temperature warming during the 1992–2013 period. In contrast, consistent downward trends can be observed in R, SM, GWL, and TWS during the 1990–2008 period when P shows zero or slightly negative trend, suggesting that decreasing land water storages could be contributing to decreasing R. As mentioned above, R has an increasing trend consistent with the increasing P from around 1995 until 2013, but this increasing trend is not observed in all water storage variables (SM, GWL, and TWS). This is because that both P and E have increasing trends of comparable magnitudes during 1992–2013 (Table 1), the land water storages may increase or decrease, implying that the increasing R is mainly caused by the increased P due to more intense storm events rather than the enhanced baseflow due to water table rise. Finally, TWSC does not show any clear trend, since it is merely the residual term of three much larger terms (P, E, and R).
3.3 Yearly Trend Analysis: 1992–2013
As a clear shift in trend is observed from around 1992 onward for most variables (Figure 4), the trends are also identified for the 1992–2013 period and also summarized in Table 1. When compared with the 1983–2013 period, T has an upward trend of +0.029–0.037 °C/year during 1992–2013 compared to +0.006–0.011 °C/year during 1983–2013, and most hydroclimatic trends are larger (P: +8.73–9.05 mm/year, E: +6.87–7.47 mm/year, and R: +1.57–3.54 mm/year) than the entire 31 year period.
The following results are based on the M-K test; similar results with slightly different trend values are also obtained from the L-S method. For the 1992–2013 period, all Tmean, Tmax, and Tmin show warming trends of +0.037, +0.047, and +0.029 °C/year, respectively, considerably larger than the corresponding +0.006, +0.008, and +0.015 °C /year for the 1983–2013 period. Under such intense warming during the 1992–2013 period, P and E exhibit significant increasing trends of +8.73 mm/year (~9% per decade) and +6.87 mm/year (~11% per decade), respectively; also, R shows a consistent increasing trend of +3.54 mm/year (~11% per decade), although not statistically significant. In contrast, a concurrent weak decreasing trend is found for both SM (−0.07 mm/year) and GWL (−0.02 m/year) as a result of the larger increasing trends of E plus R than that of P. Meanwhile, TWS shows a deceasing trend (−2 mm/year) contributed mainly by decreasing GWL trend, which can be approximated as ~80% assuming that the typical specific yield for converting GWL into groundwater storage is 0.08, following the value used by Yeh et al. (1998) and Yeh and Famiglietti (2009) for Illinois. Annual TWSC exhibits no apparent trend. Overall, the stronger trends for the 1992–2013 period in most water budget variables are consistent with the intensification of the hydrologic cycle as a response to climate warming particularly pronounced from 1990s onward (Huntington, 2006).
3.4 Monthly Trend Analysis: 1992–2013
The monthly trends are identified for each of the 12 months of a year. In Illinois, the four seasons are winter (December–February), spring (March–May), summer (June–August), and autumn (September–November). Figure 5 plots the monthly trends for the 1992–2013 period obtained from both trend methods. The detailed 1992–2013 trends for all variables are summarized in Table 2. The shaded areas in Figure 5 indicate the confidence interval of estimated trend magnitude at the statistical significance level of 95%. For most of the 12 months, a clear seasonal pattern can be discerned despite no statistically significant temperature trends can be identified. All temperature variables show increasing trends during warm seasons (March to September) and decreasing trends during cold seasons (October to February). Based on the M-K test, Tmax exhibits a significant increasing trend of +0.115 °C/year in April and +0.088 °C/year in August. For both Tmean and Tmin, the largest increasing trends are identified in March and September though both are not statistically significant.
| Variable | Jan | Feb | Mar | Apr | May | Jun | Jul | Aug | Sep | Oct | Nov | Dec |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Tmax (°C/year) | 0.013 | −0.083 | 0.1 | 0.115* | 0.048 | 0.072 | 0.053 | 0.088* | 0.065 | −0.016 | 0.145 | −0.06 |
| −0.038 | (−0.107) | −0.103 | −0.099 | −0.035 | −0.066 | −0.054 | −0.065 | −0.072 | (−0.035) | −0.146 | (−0.046) | |
| Tmean (°C/year) | −0.004 | −0.072 | 0.078 | 0.074 | 0.053 | 0.068 | 0.027 | 0.046 | 0.083 | −0.018 | 0.051 | −0.056 |
| −0.004 | (−0.104) | −0.103 | −0.082 | −0.042 | (0.060*) | −0.04 | −0.043 | −0.072 | (−0.008) | −0.072 | (−0.061) | |
| Tmin (°C/year) | −0.031 | −0.071 | 0.087 | 0.066 | 0.062 | 0.059 | 0.012 | 0.035 | 0.061 | −0.017 | −0.006 | −0.075 |
| (−0.022) | (−0.099) | −0.105 | −0.066 | −0.05 | −0.064 | −0.026 | −0.021 | (0.073*) | (−0.003) | −0.021 | (−0.075) | |
| P (mm/year) | −0.594 | 0.867 | 1.452 | 1.067 | 2.715 | 0.365 | 0.982 | 0.118 | 0.727 | 1.865 | −1.993 | 2.144** |
| (−0.381) | −0.933 | −1.385 | −1.771 | −1.451 | −0.463 | −0.101 | −0.183 | −0.215 | −2.333 | (−1.651) | (2.247**) | |
| Snow (mm/year) | −0.164 | 0.329 | −0.141 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | −0.032* | 0.435* |
| (−0.314) | −0.501 | −0.084 | (−0.067) | −0.002 | 0 | 0 | 0 | 0 | (−0.028) | (−0.128*) | −0.405 | |
| E (mm/year) | 2.298** | 0.658 | −0.238 | 0.213 | −0.05 | −0.561 | −0.992 | 0.614 | −0.04 | 1.604** | 1.042 | 2.410** |
| (2.093**) | −0.999 | (−0.292) | −0.682 | (−0.003) | (−0.532) | (−1.163) | −0.584 | −0.068 | (2.069**) | −0.673 | (2.299**) | |
| PW (mm/year) | −0.037 | −0.059 | 0.05 | 0.031 | 0.094 | −0.052 | −0.074 | −0.062 | 0.061 | −0.001 | −0.038 | 0.019 |
| (−0.022) | (−0.065) | −0.103 | −0.035 | −0.092 | (−0.017) | (−0.042) | (−0.101) | −0.06 | (−0.001) | (−0.025) | −0.029 | |
| R (mm/year) | 0.29 | 0.181 | 0.593 | 0.277 | 0.451 | 0.683 | −0.06 | −0.171 | 0.03 | 0.033 | −0.269 | 0.096 |
| −0.273 | −0.252 | −0.543 | −0.392 | −0.493 | −0.426 | (−0.018) | (−0.363) | (−0.191) | (−0.124) | (−0.335) | −0.224 | |
| SM (mm/year) | −2.338** | −2.340** | −1.575* | −1.277 | −0.426 | 0.452 | 1.352 | 0.858 | 2.213* | 1.387 | 0.457 | −0.717 |
| (−2.081**) | (−2.602**) | (−1.572**) | (−1.290*) | (−0.680) | (−0.126) | −0.645 | −0.645 | −1.271 | −0.826 | −0.17 | (−0.564) | |
| GWL (m/year) | −0.029 | −0.011 | −0.021 | −0.016 | −0.014 | −0.01 | −0.013 | −0.003 | −0.006 | −0.002 | −0.013 | −0.013 |
| (−0.025) | (−0.022) | (−0.021) | (−0.016) | (−0.011) | (−0.011) | (−0.005) | (−0.005) | (−0.009) | −0.002 | (−0.015) | (−0.009) | |
| TWS (mm/year) | −4.767* | −2.935** | −3.555* | −2.727 | −1.605 | −0.909 | 0.568 | 0.264 | 1.094 | 1.124 | −0.816 | −1.867 |
| (−4.055*) | (−4.372**) | (−3.237*) | (−2.540) | (−1.579) | (−1.009) | −0.274 | −0.235 | −0.573 | −0.962 | (−1.028) | (−1.304) | |
| TWSC (mm/year) | −2.661** | −0.168 | 0.955* | 0.543 | 2.136 | 0.818 | 1.41 | 0.534 | 0.617 | −0.104 | −2.714 | −0.579 |
| (−2.747**) | (−0.317) | −1.135 | −0.697 | −0.961 | −0.57 | −1.283 | (−0.039) | −0.338 | −0.388 | (−1.989) | (−0.276) |
- ** Statistical significance level p < 0.05.
- * Statistical significance level p < 0.10.
For the water budget variables, the trends estimated from both methods indicate that P has an increasing trend for all months except November and January, but the trend of P is only statistically significant in December (+2.14 to +2.25 mm/year). Snow also has a significant upward trend in December (+0.405 to +0.435 mm/year), explaining ~20% of the increasing P trend and also consistent with the large decreasing trend in temperature in December (Table 2). Also, the increasing trend of Snow in February (though not significant) is also consistent with the concurrent decreasing trends in all temperature variables. E shows significant increasing trends in cold months: January (+2.10 to +2.30 mm/year), October (+1.60 to +2.07 mm/year), and December (+2.30 to +2.41 mm/year), but lacks of any clear trend in warm months. The PW shows a weak increasing trend in spring generally consistent with the pattern of P. The R has an upward trend for the first half of the year (particularly in spring months) and weak downward trends for the second half, but all trends are not significant at the level of p = 0.1.
For water storage variables, both SM and TWS exhibit significant downward trends from January to April, partially associated with the increased E in January and February, and also with the increased Snow, which can reduce infiltration and the replenishment of subsurface water storage. The GWL also shows a declining trend (not statistically significant) in winter, and no any increasing trends can be found throughout the year. Notice that the declining trend of GWL remains possible to be associated with groundwater withdrawals, despite all the ISWS monitoring wells were originally selected to be far away from pumping centers (Yeh et al., 1998). The attribution of the causes for declining GWL in Illinois is beyond the scope of this study, hence will not be explored further here. For TWSC, the upward (downward) trend is identified from March to September (October to February), suggesting gaining more (or losing less) water in warm months, while losing more (or gaining less) in cold months.
Note that for all data used in this study (Table 1) are from direct observations except for only E (estimated from the water budget) and PW (taken from the Reanalysis data). Given that the E data are more difficult to obtain, estimating annual E from annual water budget by assuming long-term storage change is zero is a commonly approach used in literature (Baker et al., 2012; Milly & Dunne, 2001; Walter et al., 2004). However, when estimating monthly E based on monthly P, R, and storage changes, negative E values can be obtained in winter months when E is typically small (Figure 2). Therefore, the uncertainty in the estimated monthly trends of monthly E and PW should be higher than other variables based on direct observations.
3.5 Monthly Trend Sensitivity to Varying Analysis Periods
The monthly trends (Figure 5) discussed in section 3.4 are based on the 1992–2013 period with significant temperature warming. Recognizing the large sensitivity of trends to data lengths and periods (Figure 4), it is meaningful to examine the range of all trends corresponding to various subperiods. Figure 6 presents the box-plots of the monthly trends estimated from the total 153 M-K tests for each variable. The 153 subperiods are identical to that used to construct Figure 4, with the minimum length of 15 year selected from the 1983–2013 period. As shown, the widely scattered trends in some months indicate the high sensitivity of trend to the data lengths and periods, consistent with annual trend analysis shown in Figure 4. Notice that the mean monthly trends and their ranges in Figure 6 are not the same as that plotted in Figure 5, which are based only on one single 1992–2013 trend test. However, a detailed comparison between Figures 4 and 5 shows that the mean and range (the confidence interval) of trends in these two figures are quite consistent.
For Tmax, Tmean, and Tmin, the ranges of monthly trends in cold months (particularly winter) are larger than in warm months, indicating the larger uncertainty in winter temperature trend than summer. Tmax has the largest uncertainty among three temperature variables. In addition, the contour plot similar to Figure 4, but for the monthly trends (not shown), shows that most months (except winter) have more significant warming trends since early 1990s, consistent with the yearly trend analysis (Figure 4).
The P shows the largest decreasing (increasing) trend in November (December), consistent with the monthly trends for the 1992–2013 period (Figure 5). The range (i.e., uncertainty or trend stability) of P trends in November is remarkably large, following by the trends from April to June. PW exhibits the largest decreasing trend in June and July, and high uncertainty for all months. For Snow, the range of trends in the most snowiest January–February is the largest, consistent with the trends of Tmean and Tmin. The monthly trends of E tends to increase in fall-winter and decrease in summer, and they are rather sensitive to the varying subperiods except for March and April. For R, its monthly trend is extremely unstable from April to June, but more stable from summer to fall (July–November). The highly similar pattern of monthly trend can be found for GWL and TWS, indicating the close correspondence between R and subsurface water storages (in particular GWL) in Illinois, as shown in Figure 3b with a correlation coefficient of 0.79 between them (Eltahir & Yeh, 1999; Yeh et al., 1998). It is interesting to note that the trend magnitudes of SM, GWL, TWS, and R are more sensitive to data periods in spring and summer than in winter, which is opposite to the patterns of temperature variables. For TWSC, the decreasing trend is larger in November than any other months, while the largest increasing trend is in May.
3.6 Correlation Between Hydroclimatic Trend Magnitudes
Figure 7 presents the correlation matrix among the 1992–2013 yearly trends (estimated from the M-K test) of all hydroclimatic variables based on 36 trend test results as summarized in Table 2. The following two main points can be observed from Figure 7. First, a strong positive correlation between the trends in P, E, R, SM, GWL, and TWS can be noted, indicating consistent signals of hydrologic changes among these hydrologic variables for the 1992–2013 period. A larger trend in P for any subperiod is concurrently accompanied with the large trends in E and R, and also large trends in water storage variables SM, GWL, and TWS (despite that the estimated trends of these water storage variables are mostly negative on average; see Table 2). Note that the correlation in trend magnitudes between two variables should not be confused with the correlation between the two variables (as shown in Figure 3); the former measures the correspondence between the long-term change between two variables, while the later measures the correspondence of year-to-year or month-to-month fluctuations.
Second, the trends in temperature variables (particularly Tmean and Tmax) have negative correlation with the trends in all hydrologic variables (P, Snow, E, R, SM, GWL, and TWS). Albeit somewhat counterintuitive, it can be explained as follows. Although Tmean and Tmax show increasing trends consistent with water budget components P, E, and R for the 1992–2013 period (Table 1), the temperature trends over other subperiods are also likely to be opposite to the trends of water budget components. For example, as shown in Figure 4, both Tmean and Tmax have decreasing trends for all subperiods starting from 1996 and ending in 2013, but all hydrologic variables (e.g., P, Snow, E, SM, GWL, and TWS) have significant increasing trends for the same periods. In fact, a close inspection of Table 2 reveals that the strong negative correlation between the trends in T and P is mainly resulted from the smallest or even slightly negative trend of T for all the periods starting from 1997 and the concurrent large positive trend of P, as also can be observed clearly in Figure 4. These several tests starting from 1997 with the estimated strong positive trends in P but negative trends in T are the main reason for the strong negative correlation (−0.81) between the trends of P and T shown in Figure 7.
In summary, the strong correlation (>0.90, Figure 7) between the trends of P and R, SM, GWL, and TWS not only reveal the close linkages between the long-term changes of these hydrologic variables as mainly driven by the trend of the P forcing, while the lack of correlation between temperature variables and P (and hence other hydrologic variables) indicate complex interactions between temperature and hydrologic variables, which prohibit a clear correspondence to be identified. In addition, the consistent high (positive) correlation among hydrologic trends found in Figure 7 also provides extra confidence to the overall findings of this study given that the data of most variables analyzed here are from independent sources.
4 Discussion and Conclusions
In this study, we use a unique 31 year (1983–2013) observed data set in Illinois (a representative region of the U.S. Midwest) covering the (mean, maximum, and minimum) temperatures (T), precipitation (P), evaporation (E), streamflow (R), SM, and GWL, among other variables, to estimate the trends and their sensitivity to different data periods and lengths. Both yearly and monthly trends are estimated using the nonparametric M-K trend test and the L-S linear trend method. Two methods identify consistent trends in close agreement. In addition to checking the consistency of the estimated trends from two methods, the L-S method can ensure the exact closure in estimated trends of water budget components, thus providing an internally consistent picture on the hydrologic changes across the entire terrestrial water cycle.
Results of this study reveal an overall warming trend at both yearly and monthly time scales in Illinois. Despite no statistically significant trends in most hydroclimatic variables found for the entire 31 year period, statistically significant increasing trends are identified from both methods in P (8.73–9.05 mm/year), E (6.87–7.47 mm/year), and R (1.57–3.54 mm/year) for the 1992–2013 period, concurrently with a pronounced warming trend in T (0.029–0.037 °C/year). All of these increasing trends in P, E, and R approximately corresponds to a significant 1% increase per year as shown in Table 1. However, all water storage variables (SM, GWL, and TWS) show consistently decreasing trends for the 1992–2013 period. TWS is decreased by −2.0 mm/year (~80% of which due to GWL decline), suggesting that the increasing R is mainly a result of increasing P rather than baseflow increase. Taking together, an observation-based evidence on the intensification of the hydrologic cycle in Illinois from early 1990s until recently is obtained from this study, in general agreement with and well supported by several recent studies consistently reporting the increased P, R, and E over the U.S. Midwest and the Mississippi River basin (e.g., Baker et al., 2012; Milly & Dunne, 2001; Qian et al., 2007; U.S. Global Change Research Program, 2009; Walter et al., 2004).
One of the underlying assumptions in the L-S method is that the estimated trend is linear. If the trend is nonlinear, it would be reasonable to see that the linear trends calculated for subperiods with different periods and lengths would yield different magnitudes. In this study, it is likely that the underlying change is not linear with a stronger trend in the recent decade. For a linear trend, a longer data period would be preferable to yield more robust and statistically significant trend and also avoid misinterpretation of multidecadal variations as a trend. Regarding the estimated nonlinear trends in this study regions, it will be left as a significant research topic in the near future.
It is recommended that the trend stability as well as the consistency of estimated trends in closing water budget should be investigated for all trend studies. The evidence obtained here shows that statistically significant trends can only be identified for the 1992–2013 period but not for the entire 31 year (1983–2013) period. This study presents an observation-based, internally consistent trend analysis among water budget components and temperature variables and provides a new scientific reference for enhancing understanding on the efficient management of water resources in U.S. Midwest. More efforts are required to investigate the hydrological sensitivity of regional hydroclimatic variables to climate warming and predict how these dependences may shift in the future.
Acknowledgments
This research was supported by the project no. R 302-000-125-112 funded by the Singapore Ministry of Education's (MOE) AcRF Tier-1 project. The hydrologic and meteorological data used in this study are publicly available from the Illinois State Water Survey (ISWS) Water and Atmospheric Resources Monitoring (WARM) Program (http://www.sws.uiuc.edu/warm/) and U.S. Geological Survey (USGS) Water Data for the Nation (https://waterdata.usgs.gov/nwis), for which the authors are greatly thankful. We also thank two anonymous reviewers for their valuable comments that considerably helped to improve this paper.
