The field experiments were conducted on typical WW and SM crop rotations at LAEES Chinese Ecological Research Network, Luancheng County, Shijiazhuang, Hebei Province, China as shown in Figure 1. Shen et al. (2002) note that it is a prominent productive area yielding about 13,500 kg ha⁻¹/y under a WW–SM rotation. The WW crop period from early October to the following mid-June is about 247 days, while the SM crop period from mid-June to late September is about 107 days. The area has a moderate dry monsoon climate, with a mean annual global radiation of 524 kJ/cm² and a mean annual temperature of 12.2°C. Most (60–80%) of the mean annual rainfall of 481 mm occurs during the SM season (June–September), while relatively little falls during March–May at a time when WW grows rapidly, as shown in Table 1. Rainfall does not fulfill the water requirement for WW growth, particularly throughout the windy, dry spring.

The experimental plots (5 × 10 m) were created and divided by concrete walls. According to the FAO (Food and Agricultural Organization) the walls extended 1.5 m under the soil surface and 24.5 cm thick. The fields had been mulched with straw from the WW and SM for ~20 years and rotary tillage was practiced (with a tilling depth of about 10 cm) to mix the broken straw with the top soil; it is termed a reduce tillage practice because the tilling depth is shallower than that of normal traditional tillage at around 25–30 cm. The WW variety Kenong No.199 was sown at a rate of 187 kg ha⁻¹ in 25 cm wide rows. The SM variety Zhengdan No.958 was sown by hand at a rate of 53 kg ha⁻¹ into 50 cm wide rows. Fertilization schedules included di-ammonium phosphate, applied before wheat sowing at a rate of ~600 kg/ha; then, supplementary urea was applied twice with irrigation at the beginning of April (August) 300 kg/ha (600 kg/ha) during the jointing stages of WW (SM), respectively. Generally, farmers apply manure every 2–3 years at a rate of around 100 kg N/ha. The soil type is silt loam, with a field capacity of ~35% (Sun et al., 2007; Shen et al., 2011). Soil organic matter content was 1.8% in the top 20 cm layer at the experimental site. Soil parameters and features are shown in Table 2.

A low pressure water transportation system outlet connected to a plastic pipe was used to irrigate each plot and water use was recorded by a water meter. Normally, 3–4 irrigation applications of 60–80 mm each were given during the WW growing season and 1–2 irrigations were applied for SM, depending on rainfall. Levels of irrigation and their timings are shown in Table 3. Leaf area index (LAI) was estimated by randomly measuring the leaf area of plant samples. Each sample included 10 wheat plants or 3 maize plants. The sum of the leaf area was scaled to a unit area according to plant density and then used to calculate the leaf area index. Plant density was observed simultaneously. Grain yield measurement was taken in the middle of each plot (area of 3 × 8 m), and a 1,000-kernel weight was determined from the harvested grains. For dry matter, 10 plants were measured after 48 h of oven drying at 65°C.

ET measured by an eddy covariance (EC) system consisted of a CSAT3 sonic-anemometer and a LI7500 H2O/CO2 gas analyzer (Campbell Scientific, Inc. USA) at 3 m above ground level. Latent heat flux was measured every 30 min. The water table declined rapidly from 1975 to 2016 with a drawdown rate of about 0.8 m per year as shown in Figure 2.

The water table increased a little after rainfall, but a dramatic decrease was noted at the start of each successive irrigation period. Weather data was recorded on a daily basis, including rainfall, maximum/minimum temperature, relative humidity, dew point, and wind speed throughout the experimental period; this was obtained from the meteorological station at LAEES. The Angstrom equation was used for the calculation of daily solar radiation from the number of sunshine hours (Jones, 1992). Field data was collected from 2010–2011 to 2012–2013 at LAEES and thereafter used to calibrate and validate the CropSyst Model. Management measures during WW and SM growing seasons from 2010–2011 to 2012–2013 are shown in Table 3.

The ability to simulate crop rotations is important in the analysis of cropping systems. Models used in rotation configuration belong to the CERES and CROPGRO families; these are all situated under the umbrella of DSSAT (Jones et al., 1998). Jones et al. (2001) report that DSSAT access has been slow to adopt a more generic simulation platform, hence users tend to merge these models to simulate crop rotation. Efficient and simple multi-crop simulation takes place in the EPIC model for the analysis of cropping systems and rotations (Willmott, 1982). CropSyst was designed to depict the EPIC concept for the analysis of crop growth, including a more process-oriented approach to simulate crop rotation and its interaction with the surrounding environment and management (Stockle et al., 1994; Stöckle et al., 2003). CropSyst is a cropping system model that is designed to simulate a range of management and weather scenarios. This model is suitable for use in the study of soil, water budget, nitrogen budget, and weather variables. The predictive capability of a model determines its reliability, application, and performance evaluation for its use as a research tool. Therefore, in this study, we used the CropSyst model to simulate growth yield and water use of a widely grown rotation of SM and WW in NCP. The results of this study should offer valuable information about agricultural water management and help assess approaches that involve reducing water usage while maintaining crop productivity.

CropSyst (Suite-4) was used to model crop rotations and crop productivity in reaction to soil, management, and weather. Crop growth stages were simulated based on thermal time demands to different developmental stages. CropSyst used five input files for simulation: (i) simulation control (a combination of different input files such as start and ending days, crop rotation, initialization parameters, simulation of soil salinity, erosion, CO₂, and nitrogen effectuates on crop development, and a choice of soil runoff and water distribution models) based on field experiments; (ii) position file [latitude, weather file, rainfall volume parameters, ET model selection Penman-Monteith (PM) or Priestley-Taylor (PT) on automatic mode] obtained from the weather station of the experimental site; (iii) soil file (soil type, pH, wilting point, bulk density, field capacity, cation exchange capacity, and hydraulic conductivity) based on existing field soil properties; (iv) management file (irrigation, fertilization, tillage, residue, conservation, and harvesting) based on field management operation; and (v) crop file (emergence, thermal time accumulation, transpiration, attain growth, canopy growth, phenology, vernalization, photo-period, senescence, dormancy, and harvest). Detailed data of weekly leaf area, biomass, and grain yield were measured, and harvest index was determined during the experiment, that used for crop file preparation. The parameters used for the CropSyst model were measured or estimated using experimental data; some of this information was based on field experience, and other parameters used the default values given in the model, regardless of the year (Table 4).

The CropSyst model allows calibration of cultivar sensitive parameters for crops. A daily weather database, experimental phenologic stages, weekly biomass, weekly green leaf area indices, and biomass and grain yield at maturity were input for each growing season to calibrate phenology, morphology, and growth parameters (Table 4) for the study location.

Calibration of the CropSyst model was performed using values observed from the experiment during 2012–2013 of WW and SM on full irrigation, for ET_a, biomass, and grain yield. Differences between the simulated model and the measured data were minimized by applying a trial and error approach. A detailed data set of weekly leaf area and biomass was available for 2012–2013, so this was selected for calibration.

Validation was performed by equating field observed data with simulated results and it is a significant measure in model assessment (Power, 1993; Addiscott et al., 1995). ET_a, biomass, and grain yield were considered as the verification parameters for the CropSyst model. Crop parameters obtained from the model calibration were used in the validation. The independent data set was used to test the model's performance from 2010 to 2012.

A revalidation of the model depends upon its successful calibration based on field experimental data, and the accurate estimation of the specific model's coefficients in a given environment. Eitzinger et al. (2003) revalidated the CERES-Wheat model using the grain yield observed over 9 years (1985–1993) at the experimental site, and compared it with model outputs. Iqbal et al. (2014) revalidated the AquaCrop model for wheat on the NCP under deficit irrigation from field experiments to regional yield simulation. In the current study for the CropSyst evaluation, we used ET_a, biomass, grain yield, and ET_a partition data measured at the Luancheng experimental site for WW (2008–2009) using well-water irrigation. Small experiment plots (5 × 10 m) were constructed with concrete walls to avoid soil water recharge, according to stipulations adjusted by the FAO. Plot experiments have been in operation for approximately 15 years. The well-watered plots were irrigated to maintain ≥85% of field capacity. Three irrigation applications totaling 230 mm were given in April to May 2009. Soil volumetric moisture content was measured using a neutron probe (IH-II, Institute of Hydrology, Wallingford, UK) at 10 cm intervals between a depth of 0 and 180 cm. Depth of the water table was measured with a water level logger (Hobo U20-001-01, Onset Corp., Bourne, MA). Vapor pressure deficit was calculated from temperature and relative humidity data that was measured with a temperature-relative humidity data logger at 0.1, 3, and 10 m. Soil temperatures at 0.05 m were measured with a digital thermometer. ET was measured by an EC system composed of a CSAT3 sonic-anemometer and a LI7500 H₂O/CO₂ gas analyzer (Campbell Scientific, Inc.) at 3 m above ground level. Latent heat flux was measured every 30 min. Evaporation (E) data was measured using two micro-lysimeter systems (MLS) (Shen et al., 2002; Sun et al., 2006). Stable isotopes are excellent tracers of soil water cycling because of the isotopic fractionation imparted by E. Isotope composition of soil water, groundwater, plant-stem water, rainfall, and atmospheric water vapor were analyzed over one growing season. ET_a and its partitioning was performed by combined isotopic and micro-meteorologic approaches in an irrigated WW field; a more detail description of this experiment can be found in a related paper, Zhang et al. (2011). This season was chosen for revalidation because transpiration (T) and E data, collected by a micro-lysimeter and isotopes, was available.

Since no individual criterion can accurately gauge the accuracy of a simulation model, combine statistical indexes are broadly applied to judge model performance (Caton et al., 1999; Kobayashi and Salam, 2000; Gauch et al., 2003). Correspondence between the observed and simulated values were measured by adopting six statistics: RMSE (root mean square error), NRMSE (normalized root mean square error), MBE (mean bias error), MAE (mean absolute error), and IoA (index of agreement) (Willmott, 1982). Percentage differences were determined using the following equations, where Si indicates simulated values and Mi indicates measured values of all statistical indices.

The simulation results are viewed as excellent when an NRMSE of <10% is achieved, good if the NRMSE is >10% <20%, fair if the NRMSE is >20% <30%, and poor if the NRMSE is >30% (Jamieson et al., 1991). For IoA, a measure result of 1.0 indicates excellent correspondence between measured and simulated values. For RMSE, a result close to zero indicates better model performance. MBE reveals the duration of execution of the model. A positive MBE result gives the mean sum of overvaluation in the estimated values and vice versa. MAE measures the weighted average magnitude of the absolute errors.

The CropSyst calibrated data set of WW (2012–2013) and SM (2013) is shown in Table 5 for ET_a, biomass, and grain yield. In WW, the minimum (maximum) deviation in grain yield (ET_a) was 4.76% (9.51%), respectively, while the deviation in biomass was 5.03%. For the SM crop, the minimum (maximum) deviation in ET_a (grain yield) was −0.11% (−0.51%), respectively, while biomass deviation was 0.27%. The calibration results show a reasonably close match between the measured values and those simulated by the model (Table 5).

Considering the same parameters used in the calibration procedure, CropSyst was validated for 2010–2011 to 2011–2012, as shown in the second part of Table 5.

The results of revalidation in Table 7 show no significant deviation in grain yield (−2.8%), biomass (−1.81), and ET_a (4.12%); the measured and simulated variables are a close match. The results of this study suggested that the CropSyst model can be used with a considerable degree of accuracy to simulate grain yield, biomass, and ET_a of WW in the NCP.

After successful revalidation of the CropSyst model for WW, simulated ET_a partitions were performed into E and T. The temporal trends in rainfall, irrigation, and simulated ET_a illustrate that ET_a increased substantially when rainfall or irrigation occurred (Figure 3). During growth periods for WW from planting to harvesting, simulated ET_a partitions of WW (2008–2009) into E and T show that the total ET_a was 430 mm, and E and T were 129 mm and 301 mm, respectively.

Furthermore, during the field experiment, T was measured by two methods, (i) E was measured using a micro-lysimeter (E(T)-MLS) and subtracted from ET_a as measured by EC; (ii) T was measured using the stable isotope method (E(T)-ISO). Complete isotopic data was collected on two dates representing the late stage of filling and the stage of wax ripeness [days of year (DOY) 138 and DOY149, respectively] and E data from MLS recordings available on the following days of year, as shown in Table 8. Observed T was 83% of total ET_a on DOY138 and 60% of ET_a on DOY149. The maximum percentage of T occurred during the filling stage; this was expected because this stage corresponds to a higher LAI and increasing biomass.

Figure 4 shows the measured vs. simulated percentage comparison of T/ET_a and statistical assessment of the CropSyst model for the WW season of 2008–2009. The results of RMSE, MAE, MBE, d, and NRMSE were 4.68, 15.78, 3.01, 0.99, and 5.8% respectively, for WW grain yield.

The straight-line equation and coefficient of determination shows that the model simulated T/ET_a with a high degree of reliability, having a regression value of 0.83, as shown in Figure 5.

The simulated ET_a partitions into E and T of WW and SM from 2010 to 2013 are given in Figure 6 and show that the average value of total ET_a, E, and T were 707, 282, and 425 mm, respectively. The temporal trend in rainfall, irrigation, and ET partition illustrates that ET_a increased substantially during rainfall or irrigation, (for example see Figure 6).

When compared to past studies, the CropSyst model calibration of WW and SM showed better results, i.e., Singh et al. (2008) simulated the highest deviation for wheat biomass, at −14.5% (underestimate) and Donatelli et al. (1997) simulated the highest deviation (0.96%) in maize grain yield.

Simulated T closely matched observed T according to two different methods (i) micro-lysimeter subtracted from ET_a measured by EC (E(T)- MLS); (ii) the stable isotope method (E(T)-ISO). The deviation results of simulated vs. measured ratio T/ET_a in Table 8 are similar. The percentage comparison and statistical assessment of measured vs. simulated T/ET_a of the CropSyst model for WW 2008–2009, fell within an acceptable range and IoA was in the excellent range. E was 30% of total ET_a in the WW growing season of 2008–2009 and is comparable with results obtained by Zhang et al. (2011) for WW grown in LAEES, NCP.

E accounts for 40% of total annual ET_a for the growing seasons of both WW and SM, as shown in Figure 6. These results are comparable with Liu et al. (2002) for WW and SM grown in LAEES, NCP. The pre-sowing irrigation method for wheat and maize is practiced in the NCP, in which irrigation is performed before harvesting of the previous crop and sowing is performed at soil field capacity. This leads to huge soil E losses after harvesting of the previous crop to emergence of the new crop up to the stage of canopy cover (see for instance, Figure 6).

Averages of the 3 years growth period for WW and SM were divided into 3 developmental periods; (i) PSI to EP, (ii) EP to CC (end of new leaf) period, and (iii) CC to HP, as shown in Table 9. Daily E accounts for ~10% of E in the pre-irrigation sowing to emergence period, ~60% emergence to canopy cover (end of new leaf) period, and ~30% canopy cover to harvesting period for both WW and SM crops. Most E occurs during the pre-sowing irrigation to canopy cover period. Therefore, results suggest that there is a need to reduce E through the use of precise irrigation (i.e., surface and sub-surface drip irrigation or by mulching) to overcome the huge water losses. For example, if we assume a precise irrigation method (such as drip irrigation), this saves 10% soil E losses (Evett et al., 2005; Abdelraouf and ElHabbasha, 2014; Qin et al., 2016). So, an irrigation reduction of 10% (ET_a reduction) across 1.7 million hectares of agricultural land in Hebei Province would account for approximately 1,700 million m³ year⁻¹ of water saving.

The results of this study clearly indicate that the CropSyst model can be used with a high degree of accuracy for yield simulation and soil water loss analysis of WW and SM crop rotations on the NCP. It can also serve as a useful tool for assessing national food and water security in the agricultural sector.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.