Figure 3-2 Effect of land-atmosphere coupling on soil moisture and evapotranspiration (SM-ET) correlations for ensemble mean in JJA from 1981-2014.
Figure 3-3 Effect of land-atmosphere coupling on SM-ET distribution for ensemble mean in JJA over the Great Plains (indicated in Figure 3-2), using (a) land-standalone (LAND) simulations, and (b) LA-coupled ATM simulations only. The darker dots of (a) indicate LA-uncoupled runs compared to LA-coupled runs. Gray dots in (b) shows the individual month of JJA for each year from 1981 to 2014. The green dots of (b) are the top and bottom 7% soil moisture in the probability distribution, which have equally the same average soil moisture values as the blue dots (see also Table 3-1).
Table 3-1 Comparison between JJA ET from dry- and wet-side soil moisture and ET from moderate soil moisture in LA-coupled ATM simulations averaged over the Great Plains. Units are mm/month.
Category Moderate Both-sides Left 6% (Dry) Right 6% (Wet)
Mean SM 247.23 247.25 209.38 285.12
Mean ET 103.84 91.59 67.72 115.46
Figure 3-4 Effect of land-atmosphere coupling on distribution of monthly JJA (a) root zone soil moisture, (b) ET and (d) precipitation for the ensemble mean over the Great Plains. The y axis shows histogram densities. The legend indicates the coupled or uncoupled, atmosphere coupled or land-only set of simulations used for the corresponding variables. ATM simulations were used for atmospheric variables such as precipitation whereas LAND simulations for land surface variables such as soil moisture and ET. Here we present soil moisture distribution from ATM simulations to show the JJA soil moisture values from LA-uncoupled ATM were held fixed from the climatology of JJA LA-coupled ATM.
Figure 3-5 Effect of different irrigation prescriptions on distribution of monthly JJA root zone soil moisture, ET and precipitation in the Great Plains over the 20-year simulations: SMsat, SMrain, SMsat_more, SMrain_more. The y axis is histogram densities. The legend compares the control run with the irrigation run for each experiment. The vertical dashed line indicates the mean of control run, and the solid line indicates the mean of irrigation run. The units are mm/month.
Figure 3-6 Effect of different irrigation prescriptions on SM-ET relationship using JJA monthly values over the 20-year simulations over the Great Plains. The blue dots indicate experiments using soil saturation (mimic flood irrigation) while the green dots indicate experiments using effective rain rate (mimic sprinkler irrigation). The gray line represents the Budyko curve and distinguishes the three hydrological regimes.
Figure 3-7 Distribution of SM-ET, SM-P and SM-ET for their JJA interannual standard deviation (a-c) and JJA climatology mean (d-f) over the Great Plains. The darker dots indicate data points from GLACE while the lighter dots indicate data points from irrigation experiments. The regression lines are computed only by data from irrigation experiments. * indicate significance level greater than 95%
and ** indicate significance level greater than 99%.
Table 3-2 Mean and standard deviation of JJA monthly soil moisture, ET and precipitation for GLACE and irrigation experiments, respectively. The standard deviation indicates interannual variability where JJA are averaged to remove intraseasonal variation. LAND simulations were used for soil moisture and ET in the GLACE experiment. The grids are averaged over the smaller box. Units: mm/month.
Casename
Top 1m Soil Moisture (mm) ET (mm/mon) Precipitation (mm/mon) Mean Standard deviation Mean Standard deviation Mean Standard deviation
LA-coupled 264.68 15.26 106.07 5.08 87.56 7.77
LA-uncoupled 267.22 5.68e-14 118.66 1.97 115.82 6.40
CTL 412.53 23.40 92.53 9.92 72.12 16.18
SMsat 445.90 13.69 98.31 6.05 73.21 13.27
SMsat_more 489.28 5.83 115.22 2.66 77.21 12.19
SMrain 435.93 19.14 113.50 5.26 75.31 16.00
SMrain_more 437.79 20.67 121.41 4.66 70.85 14.96
Figure 3-8 Interannual variation of irrigation-induced precipitation (mm/month) among the irrigation experiments over the Great Plains in JJA. The blue-tone dots are flood irrigation experiments, whereas the orange-tone dots are sprinkler irrigation experiments.
Figure 3-9 Vertical profile of divergence field over the Great Plains in JJA for each set of experiments.
The blue solid line is the ensemble mean of LA-coupled simulations from GLACE, while the green dashed line is the ensemble mean of LA-uncoupled simulations. The remainder are from irrigation experiments.
Summary and Discussion
Since the original land-atmosphere coupling index by GLACE (Koster 2004; Koster et al. 2006) is derived from the difference between intra-ensemble variance and ensemble mean variance, it is the variability driven by land-atmosphere coupling that determines whether a region is a coupling hotspot.
By showing that different irrigation approaches also change the soil moisture variability, we partially reconcile the disparity among the modeling studies: for those who adopted surface irrigation (Harding and Snyder 2012) or directly prescribed ET rate (Segal et al. 1998) to guarantee sufficient moisture flux in the atmosphere, precipitation response can be significant. Conversely, for those who adopted sprinkler method in the model (Pei et al. 2016; Lu et al. 2017), insignificant or reduced rainfall is observed. Note that irrigated fraction also matters, and is probably more important than soil moisture variability changes. For instance, the difference between SMsat and SMsat_more is that the latter irrigated for the entire C3 crop whereas the former only irrigated for around half of the C3 crop fraction, which leads to significant differences in both ET and precipitation response. This might explain why Adegoke et al. (2003) irrigated by saturating the soil columns once per day but obtained insignificant precipitation changes. Contrarily, while Puma and Cook (2010) chose to irrigate by adding water to the topsoil, their choice to irrigate for the entire vegetated fraction is likely at play in determining the significant irrigation signal.
We also show that the precipitation variability is more susceptible to change when soil moisture variability changes, which explains why central North America is identified as a coupling hotspot. As coupling hotspots are originally defined as the locations that are more capable to improve seasonal forecasts, variability matters more than mean changes because a robust response with little variation is hard to provide a strong signal for longer-term forecasts. For instance, even if the evaporation response to soil moisture is robust, the atmosphere is unlikely to show a strong signal at the surface when the evaporation variability is low. The reduced precipitation variability followed by reduced soil moisture variability suggests a negative feedback from the soil moisture to the precipitation. That
is, irrigation tends to mediate precipitation by decreasing rainfall in wet years and increasing in dry years (Figure 3-8). While this finding is contradictory to a handful of studies suggesting that irrigation tends to amplify precipitation but may not be able to trigger convection (Schickedanz 1976; Segal et al. 1998; Harding and Snyder 2012; Huber et al. 2014; Welty and Zeng 2018), we further examined the divergence field with interannual variation. Figure 4-1 shows the divergence changes in dry years and wet years. The wet (dry) years is determined by whether the mean JJA precipitation of that year is higher (lower) than the average of the 20-year simulations. The results show that irrigation tends to induce low-level divergence in dry years and low-level convergence in wet years, whereas wet years have stronger low-level divergence than dry years (the gray lines in Figure 4-1). The relation between moisture divergence and precipitation remains obscure, while the irrigation effect on precipitation on interannual scale in this study requires more vigorous evidence.
Table 4-1 further divides the variance of precipitation into three components: the variance of ET, the variance of vertically-integrated divergence and the covariance between ET and divergence.
Assume precipitation can be separated into the contribution of ET and divergence, then
var(P) = var(ET) + var(D) + 2 cov(ET, D) (4.1) Note that we computed the divergence using the concept of moisture budget conservation, i.e. D = P − ET − , where D’ denotes the anomaly of divergence, P’ denotes the precipitation anomaly,
ET’ denotes the evapotranspiration anomaly, and denotes the rate of change in humidity. We chose this approach because we do not have outputs for each timestep. In addition, the divergence term calculated by monthly outputs would not be conserved over lands, since the monthly wind fields have smoothed out the transient variation. Table 4-1 shows that the reduced variance in P after irrigation is primarily contributed by the changes in ET variance and covariance term despite the marked increase in divergence variance. Since the covariance term is purely a mathematical outcome and might not correspond to certain physical processes, we treat it as a nonlinear term in this variance analysis. While the changes in ET may represent the changes in local moisture sources, the changes
in divergence can indicate the changes in outer moisture sources. The dominant role of ET changes in reducing P variability suggests that the local moisture contribution is more important over this region. Table 4-2 further shows the coefficient of determination (R2) of the P-ET and P-D relationship using a simple linear regression model. Since R2 can be regarded as the proportion of P variance explained by ET or divergence, the weak P-ET relationship in SMsat, for example, indicates the strong contribution of ET on the significant reduction in P variability.
When it comes to the inconsistent or insignificant irrigation impacts over the Great Plains, a recent study (Tuttle and Salvucci 2016) used Granger causality to estimate the relationship between soil moisture and subsequent precipitation across the United States, and suggested that ET may not strongly influence precipitation in this region. Given that much of the precipitation over the Great Plains occurs nocturnally because of eastward-propagating convective systems originating in the upstream of the irrigated areas, it is likely that local land-atmosphere interactions are not important, at least not dominant, for predicting rainfall in this region. The propagation of the mesoscale convective systems (MCS) from Rocky Mountains can thus provide a favorable environment for the enhanced moisture from irrigation to take effect. Although our irrigation experiments also show a nighttime rainfall increase from irrigation (Figure 4-2), the model nocturnal rainfall is much less than the afternoon rainfall in terms of total rainfall, making the irrigation-induced changes comparatively negligible. Similarly, Pan et al. (1996) stressed the importance of sensible heating in the convective initiation and showed that the increase in soil moisture can decrease local rainfall when the atmosphere was humid and lack sufficient thermal forcing to initiate convection. The general increasing trend suggested by observational studies might therefore be associated with large-scale oceanic forcings. For instance, Hu et al. (2011) revealed that Atlantic multidecadal oscillation (AMO) provides a fundamental control on JJA precipitation over North America, especially during the cold phase, which incidentally matches the period of irrigation expansion and might overshadow the irrigation effect.
It is worth noting that the empirical relationship between predictors of land and atmosphere is
not linear and may subject to change over time, for instance, due to climate change (Seneviratne et al. 2013) and/or land use and land cover change (LULCC, Hirsch et al. 2014; McDermid et al. 2019).
The potential implication is twofold: (1) one may not expect the precipitation response to be proportional to the land surface changes, for the relationship is nonstationary; (2) it is probable that the coupling hotspot regions identified by GLACE is not always valid, and that irrigation might weaken the coupling hotspot over the Great Plains (Lu et al. 2017). However, one cannot separate the irrigation effect on land-atmosphere coupling strength and the outcome of land-atmosphere interactions such as precipitation, which is why we do not prefer to apply the postulate of “irrigation weakens the coupling strengths” to explain the insignificant precipitation changes, and the other way around (Lu et al. 2017; McDermid et al. 2019). It is also not justifiable to say irrigation cannot take effect over the Great Plains because the surface cooling surpasses the surface moistening, as the irrigation-induced cooling can also act as the cold pool for the MCS to develop. The increased stability of the atmosphere caused by surface cooling also might not be as dominant as the low-level divergence while the latter does not always occur in irrigation scenarios. One should be mindful when interpreting the mutual-causal relationship.
Figure 4-1 Vertical profile of divergence field over the Great Plains in JJA for two irrigation prescriptions in dry and wet years. The darker lines indicate the dry years while lighter lines are the wet years. The blue (orange)-tone lines are flood (sprinkler) irrigation experiments. The gray lines indicate control runs.
Table 4-1 Variance decomposition of mean JJA precipitation for each year averaged over the Great Plains for the irrigation experiments. P indicates precipitation, and D indicates vertical-integrated divergence. The units for the variance are mm2.
Var (P) Var (ET) + Var (D)
+ 2*Cov (ET,D) Var (ET) Var (Q) Var (D) 2*Cov (ET, D)
CTL 261.82 262.52 98.33 0.04 73.01 91.18
SMsat 176 143.4 7.08 0.54 146.58 -10.26
SMsat_more 148.48 165.39 36.63 0.69 90.96 37.8
SMrain 255.9 255.78 27.7 0.08 187.62 40.46
SMrain_more 223.94 209.98 21.72 1.05 153.06 35.2
Table 4-2 Coefficient of determination (R2) for JJA mean precipitation-evapotranspiration (P-ET) and precipitation-divergence (P-D) relationship in the Great Plains over the 20-year simulations using the simple linear regression model.
R2 P-ET P-D
CTL 0.79 0.72
SMsat 0.005 0.95
SMsat_more 0.53 0.77
SMrain 0.32 0.89
SMrain_more 0.35 0.89
Figure 4-2 Diurnal cycle of total precipitation over the Great Plains for control simulations and irrigation runs in unit of mm/hr. The black solid line indicates control run while the blue dashed shows the irrigation-induced rainfall changes. SMonce is similar to SMsat with the top1m soil moisture set to field capacity once per day.
Future work
In our preliminary analysis, some important issues have not been resolved and might affect the credibility of our results. The author feels it is necessary to address these issues for the readers to interpret these data. First and foremost, we found that the coupling hotspot spatial pattern is susceptible to change when we considered fewer ensemble members and simulation years. Since we can only obtain 9 ensemble members for the LAND simulations, we also computed the coupling hotspot indices using the corresponding 9 ensemble members in ATM simulations, causing the coupling hotspot over central North America to disappear. In addition, to compare with the irrigation experiments, we also have considered using the same length of the years for analysis, but the 20-year simulations from GLACE cannot reproduce a similar spatial pattern of coupling strengths shown in the literature. Secondly, the results of our irrigation experiments are also sensitive to the choice of timespan and areal size for spatial averaging. Since we only have 25-year (and a few 30-year) simulations for each experiment, we initially considered removing the first 10 years as a spin-up period and analyzed the last 20 years for the simulations having 30-year long. We eventually removed the first 5 years to incorporate more experiments, and found that the same 20-year (year 6 to year 25 vs. year 11 to year 30) shows a stark difference in precipitation variability: the stronger reduction in precipitation variability in flood irrigation experiment no longer exists. In addition, since the irrigated areas are not of the equal size for the four selected irrigation experiments, we chose to perform the areal mean using the smaller grid box. However, after we corrected the irrigated areas in SMrain to be consistent with others, the irrigation response in ET and precipitation behaves much differently (Table 5-1, Table 5-2), making the aforementioned conclusions not hold anymore. As we cannot be sure whether the problem results from this irrigation run or the areal size for averaging, we did not show the outcomes using this version of SMrain.
We also found the conventional 10-year spin-up period for land-atmosphere coupled runs might be insufficient for central North America, where the groundwater table depth has not reached a steady
state in the 30-year cold-start control simulations (Figure 5-1). This dry bias in soil moisture and hydrology is likely to affect the results, as previous studies have shown a higher land-driven predictability and land-atmosphere coupling over North America during wet years (Schubert et al.
2008; Guo and Dirmeyer 2013; Kumar et al. 2014). However, this does not mean the above results are untenable, as land-atmosphere coupled simulations in general can reach equilibrium in one year.
In addition, since the average groundwater table depth is about 5-10 meters deep, the top 1m soil moisture should not be drastically different if using longer simulations. On the other hand, as Ho (2017) indicated the potential impact of irrigated area size on the atmospheric response, our irrigation experiments were performed under a medium size of irrigated areas, which can show little changes in precipitation in their sensitivity tests. Current experiments can be amended using larger irrigated areas as Ho (2017) indicated to address the relative importance of irrigated areas, soil moisture variability and soil moisture mean changes. However, we do not expect a drastic increase in precipitation even using larger irrigated areas. The magnitude of precipitation increase in larger irrigated areas is also much weaker than the decrease in smaller irrigated areas in Ho (2017).
A few interesting topics associated with GLACE and irrigation are also worth exploring. While GLACE suggested that land-atmosphere coupling can enhance subseasonal-to-seasonal predictability, the impact of irrigation on potential land-driven predictability has not been well established. The analogy between irrigation and decoupled land-atmosphere feedbacks is also elusive. Current results also do not show consistent signs in changes of precipitation mean and variability for land-atmosphere coupling hotspots, but the implication and physical mechanism have not been unveiled either. In addition, while previous findings suggested that irrigation can only amplify precipitation (Schickedanz 1976; Segal et al. 1998; Harding and Snyder 2012; Huber et al. 2014), some studies such as D’Odorico and Porporato (2004) indicated a positive soil moisture-precipitation feedback through rainfall-triggering mechanisms. Whether the difference comes from the nature of simple soil wetting and irrigation or spatial disparity between the state of Illinois and the Great Plains can be explored to have deeper understanding of the nature of irrigation processes.
Table 5-1 Same as Table 3-2, but included another SMrain experiment which irrigated over a larger extent (denoted as SMrain’).
Casename Top 1m Soil Moisture (mm) ET (mm/mon) Precipitation (mm/mon)
Mean Standard deviation Mean Standard deviation Mean Standard deviation
LA-coupled 264.68 15.26 106.07 5.08 87.56 7.77
LA-uncoupled 267.22 5.68e-14 118.66 1.97 115.82 6.40
CTL 412.53 23.40 92.53 9.92 72.12 16.18
SMsat 445.90 13.69 98.31 6.05 73.21 13.27
SMsat_more 489.28 5.83 115.22 2.66 77.21 12.19
SMrain 435.93 19.14 113.50 5.26 75.31 16.00
SMrain’ 431.23 16.91 104.47 8.18 80.96 10.39
SMrain_more 437.79 20.67 121.41 4.66 70.85 14.96
Table 5-2 Similar to Table 5-1, but the grids were averaged over the large box (red box indicated in Figure 2-1).
Casename Top 1m Soil Moisture (mm) ET (mm/mon) Precipitation (mm/mon)
Mean Standard deviation Mean Standard deviation Mean Standard deviation
LA-coupled 254.76 13.22 104.69 4.07 88.72 6.94
LA-uncoupled 256.78 5.68e-14 116.11 1.88 115.53 5.44
CTL 394.79 20.03 90.46 9.27 74.22 15.39
SMsat 418.01 12.70 93.85 5.87 74.40 11.98
SMsat_more 458.10 5.29 110.06 2.45 79.25 10.61
SMrain 407.35 16.30 100.37 6.10 75.96 13.35
SMrain’ 410.46 14.83 101.57 7.43 81.81 10.46
SMrain_more 419.09 17.98 119.19 4.16 74.86 12.20
Figure 5-1 Variation of groundwater table depth over time for the irrigation experiments in the Great Plains over the course of simulation years.
REFERENCES
Adegoke, J. O., R. A. Pielke, J. Eastman, R. Mahmood, and K. G. Hubbard, 2003: Impact of Irrigation on Midsummer Surface Fluxes and Temperature under Dry Synoptic Conditions: A Regional Atmospheric Model Study of the U.S. High Plains. Mon. Weather Rev., 131, 556–564, https://doi.org/10.1175/1520-0493(2003)131<0556:IOIOMS>2.0.CO;2.
Alter, R. E., Y. Fan, B. R. Lintner, and C. P. Weaver, 2015a: Observational Evidence that Great Plains Irrigation Has Enhanced Summer Precipitation Intensity and Totals in the Midwestern United States. J. Hydrometeorol., 16, 1717–1735, https://doi.org/10.1175/JHM-D-14-0115.1.
——, E.-S. Im, and E. A. B. Eltahir, 2015b: Rainfall consistently enhanced around the Gezira Scheme in East Africa due to irrigation. Nat. Geosci., 8, 763–767, https://doi.org/10.1038/ngeo2514.
Barnston, A. G., and P. T. Schickedanz, 1984: The Effect of Irrigation on Warm Season Precipitation in the Southern Great Plains. J. Clim. Appl. Meteorol., 23, 865–888, https://doi.org/10.1175/1520-0450(1984)023<0865:TEOIOW>2.0.CO;2.
van Bavel, C. H. M., F. S. Nakayama, and W. L. Ehrler, 1965: Measuring Transpiration Resistance of Leaves. Plant Physiol., 40, 535–540, https://doi.org/10.1104/pp.40.3.535.
Berg, A., B. R. Lintner, K. L. Findell, S. Malyshev, P. C. Loikith, and P. Gentine, 2014: Impact of Soil Moisture–Atmosphere Interactions on Surface Temperature Distribution. J. Clim., 27, 7976–7993, https://doi.org/10.1175/JCLI-D-13-00591.1.
Budyko, M. I., 1956: Тепловой баланс земной поверхности (Heat Balance of the Earth’s Surface).
Gidrometeoizdat, Leningrad, 255pp.
——, 1974: Climate and Life. Academic Press, Inc., 508pp.
Ceglar, A., A. Toreti, C. Prodhomme, M. Zampieri, M. Turco, and F. J. Doblas-Reyes, 2018: Land-surface initialisation improves seasonal climate prediction skill for maize yield forecast. Sci.
Rep., 8, 1322, https://doi.org/10.1038/s41598-018-19586-6.
Cook, B. I., M. J. Puma, and N. Y. Krakauer, 2011: Irrigation induced surface cooling in the context
of modern and increased greenhouse gas forcing. Clim. Dyn., 37, 1587–1600, https://doi.org/10.1007/s00382-010-0932-x.
of modern and increased greenhouse gas forcing. Clim. Dyn., 37, 1587–1600, https://doi.org/10.1007/s00382-010-0932-x.