The irrigation experiments were performed for 1971-2000 with the atmosphere/land stand-alone configuration of CESM1.2.2 with a horizontal resolution of roughly 0.9° x 1.25°. We used the version of CAM5 and CLM4.0, which is slightly different from CLM4.5 using in the GLACE-Hydrology experiment. Nonetheless, according to the CLM45 documentation (Oleson et al. 2013), major revised parameterizations are to reduce biases of biogeochemical substances and products, including low carbon stocks and unrealistic low gross primary production (GPP) in the Arctic, excessive GPP and unrealistic high leaf area index (LAI) in the tropics. The updates to hydrological processes are more about frozen soils, which should have little effects on our analysis since we are focusing on midlatitudes and boreal summer. To isolate the irrigation impacts, we used climatology SSTs and aerosol concentrations from a temporal 20-year window surrounding year 2000 (Rasch et al. 2019) instead of the observed ones as GLACE-Hydrology to reduce interannual variability and maximize irrigation effects on climate. A comparison of model configurations between our irrigation experiments and GLACE-Hydrology is briefly summarized in Table 2-1.
Table 2-1 Model configurations for GLACE-Hydrology and irrigation experiments. The section of GLACE is the Table 1 in Kumar et al. (2020).
Exp. Type Experiment ID Soil moisture Climate forcing SST forcing
LA coupled ATM Interactive Interactive
atmosphere model
LA coupled LAND Interactive LA coupled ATM N/A LA uncoupled LAND Interactive LA uncoupled
ATM N/A
To further testify our hypothesis that the irrigation methods cause the diverged response in precipitation, we considered two types of irrigation prescription using the framework of CLM4.0 crop and irrigation model (Levis and Sacks 2011): flood irrigation and sprinkler irrigation. Although the sprinkler method is more dominant than surface irrigation (Maupin et al. 2014), we also considered flood irrigation in which soil moisture variability is more likely to be changed. We mimicked sprinkler irrigation by adding irrigation rate to effective precipitation on the top soil, bypassing canopy interception, over the entire grid cell specified over the Great Plains. The irrigation rate for each timestep is disaggregated from the monthly irrigation water dataset (Wada and Bierkens 2014). We prescribed the same 2001-2014 climatology for each simulation year to remove interannual variability for irrigation water use. On the other hand, we imitated flood irrigation by directly modifying soil moisture values to field capacity. Since this approach is doomed to consume much more water than reality, we adjusted the fraction of irrigated grid cells to ensure the irrigation water amount of the two approaches are comparable. The procedure of flood irrigation is described below.
When irrigation is enabled, the fraction of C3 generic crop of each grid cell is divided into irrigated and non-irrigated portion based on a dataset of areas equipped for irrigation (Siebert et al.
2007, 2015), where the fraction of total cropland area is determined according to the default CLM
dataset (Oleson KW et al. 2010). To remove the irrigation effects from other irrigated regions, we restricted irrigation to be implemented only over the Great Plains for the entire C3 crop fraction, then forced the irrigated C3 crop portion to be zero elsewhere (Figure 2-1). Specified irrigated areas were identified by monthly mean irrigation water over 1 mm/month averaged over 1960 to 2014 (Figure 2-2a), using a 5 arc-second observation-based, hydrological-model-driven global irrigation dataset (Wada and Bierkens 2014) which has been calibrated against national statistics from Food and Agriculture Organization of the United Nations (FAO). The chosen irrigated areas can also correspond to intensive irrigated crop fractions in CLM4.0 (Figure 2-2b). To ensure that the model only irrigates during the growing season, two criteria have to be satisfied: (1) the soils are not frozen;
(2) the transpiration wetness factor falls below 1, that is, when the transpiration rate is less than the evaporation rate of a hypothetical water surface having the same temperature and exposure as the leaf (vanBavel et al. 1965), indicating the vegetation is growing. Once the two conditions are met, irrigation is triggered at the specific timing for each experiment. Note that although by definition flood irrigation approach does not allow the soil to be dried out, the soil moisture was replenished at the beginning of the model and gradually dried out as runoff or evapotranspiration in the simulating process. In this regard, the output soil moisture values would not be held fixed.
To further examine if the effect of soil moisture variability would be altered under different irrigation water amounts, we considered another set of irrigation experiment by increasing the irrigation rate in sprinkler irrigation and increasing the fraction of irrigated grid cell to entire C3 crop in flood irrigation. The seasonal cycle and irrigation spatial pattern for the four irrigation experiments are shown in Figure 2-3, where SMsat and SMsat_more indicate flood irrigation and SMrain and SMrain_more represent sprinkler irrigation. The irrigation experiment settings are briefly summarized in Table 2-2. Note that the irrigated areas for SMrain is much smaller than others for the different research purpose in the first instance. Therefore, the following analysis is based on the areal mean using the smaller grid box to avoid the dilution of irrigation impacts in the SMrain experiment.
Table 2-2 CESM1 irrigation experiment settings.
Casename How to apply When to trigger Irrigation water amount SMsat Saturate top 1m SM
Throughout the day Realistic SMrain Add effective rain rate
SMsat_more Saturate top 1m SM
Throughout the day Hypothetical SMrain_more Add effective rain rate