Evaluating Optimum Limited Irrigation Management Strategies for Corn Production in the Ogallala Aquifer Region

The experiment was carried out at the Kansas State University Southwest Research-Extension Center (SWREC) Finnup farm (38°01′20.87″N, 100°49″26.95W, elevation of 887 m above mean sea level) near Garden City, Kansas. The site has a semiarid climate with long-term mean annual precipitation and reference evapotranspiration (ETo) based on the Food and Agricultural Organization of the United Nations (FAO) Penman–Monteith equation of 477 and 1,810 mm, respectively (Klocke et al. 2011). The growing season (April to September) average minimum and maximum temperatures are approximately 15.5 and 31.5°C, respectively. The seasonal average precipitation and ETo are 295 and 870 mm, respectively. Soils at the experimental site are Ulysses silt loams with organic matter content 1.5% and a pH of 8.1 (Klocke et al. 2011). The soil water characteristics of the silt loam soils such as the wilting point (13–16 vol.%), field capacity (31–35 vol.%), and saturation (44–48 vol.%) were estimated based on pedo-transfer functions (Saxton et al. 1986). Since the majority of the soil around southwest Kansas are classified as silt loam with relatively small patches of areas dominated by sandy clay loam/sandy loam, two soil types (silt loam and sandy clay loam) were used for scenario analysis, whereas the soils of the experimental site (silt loam soils) were used for model calibration and validation. The physical properties of the sandy clay loam soils were also estimated based on pedo-transfer functions following Saxton et al. (1986) as 15, 27, and 38% for volumetric water content at wilting point, field capacity, and saturation, respectively.

Experiments were carried out during the cropping seasons of 2010, 2011, and 2012. In all years, corn was grown under full and deficit irrigation with six irrigation treatments, all replicated four times and arranged in a randomized complete block design. Each plot had a size of $13.7\times 27.4\mathrm{m}$. The irrigation treatments were designed to represent 100, 80, 70, 50, 40, and 25% of full irrigation (Klocke et al. 2011). The irrigation amount for any irrigation event was the same (25 mm), but the number of irrigations and the interval were varied among the treatments to represent percentages of the full irrigation treatment (Table 1). The wettest treatment was irrigated before 50% of the available soil water was depleted. Irrigation intervals for the wettest treatment were determined based on daily ETo and precipitation received considering the biweekly soil water measurements. Soil water depletions were estimated using neutron probe measurements, which were taken every 2 weeks throughout the growing season. The other irrigation treatments were rescheduled based on the wettest treatment (Klocke et al. 2011). Irrigation per season decreased as the time between irrigation intervals increased (Klocke et al. 2011). Also, irrigation intervals of the same treatment varied slightly from season to season due to variability in seasonal precipitation. Even the full irrigation treatment also varied from growing season to growing season as the irrigation amount was scheduled for nonlimiting conditions (Klocke et al. 2011). Irrigation was applied using a four-span linear move sprinkler system (Model 8000, Valmont Corp., Valley, Nebraska) (Kisekka et al. 2016). Center pivot sprinkler irrigation is the most common irrigation method in the Southern and Central High Plains of the Ogallala Aquifer Region. These treatments were designed to mimic the pumping capacity limitations of center pivot irrigation systems in the Great Plains (Klocke et al. 2011). However, our interest in this study was to utilize the available experimental information to calibrate and validate a crop model (Table 1).

Fertilizer application and other agronomic practices were applied uniformly to all treatments to meet potential yield according to the recommendations for the region. Phosphorus and nitrogen were applied at the time of planting at a rate of 99 and $14\mathrm{kg}/\mathrm{ha}$, respectively. Also, side-dress nitrogen was applied 48 days after planting at a rate of $306\mathrm{kg}/\mathrm{ha}$. Corn was planted in a no-till field. The recommended planting density for the study area varies with irrigation water management such that seeding rates decreased with reduction in irrigation water application, as described in Klocke et al. (2011). In this study, the seeding rates and irrigation amounts were 7.9, 7.3, 6.7, 6.0, 5.4, and $4.8\mathrm{plants}/{\mathrm{m}}^2$ for 100, 80, 70, 50, 40, and 25% of full irrigation, respectively (Klocke et al. 2011). This shows that each irrigation treatment had a matching specific recommended seeding rate based on local practices, which were aimed at matching available water for optimal yield (Klocke et al. 2011). The practice of matching seeding rate with available water was considered one of the agronomic practices for optimizing irrigation water productivity (Kirda 2002).

Soil water measurements were carried out using neutron attenuation to a depth of 2.4 m. Soil water measurements were taken at depth increments of 0.3 m twice a month during the growing season and at physiological maturity. Seasonal ET was estimated based on soil-water-balance method (Klocke et al. 2011). In 2011, the leaf area index (LAI) was estimated from leaf area measurements taken during the mid and late season growth stages of the crop using a CI-203 Laser Leaf Area Meter (CID Bio-Science, Camas, Washington). End-of-season biomass was obtained from 3 m in the middle row of a plot. Two 3 m rows were used for grain yield. Harvesting was done by hand, and the dry weight samples of biomass were recorded after oven drying. Crop phenology (emergence, tasseling, silking, senescence, and maturity) was recorded considering the developmental stages of the crop during the growing season.

Two sets of climate data were used in this study. One data set containing 51 years (1963–2013) of daily weather observations (precipitation, maximum and minimum temperatures) for Garden City was obtained from the High Plains Regional Climate Center (HPRCC). A second data set, comprising 14 years (2000–2013) of daily climate observations that include precipitation, minimum and maximum temperatures, wind, solar radiation, and humidity, was obtained from the SWREC meteorological station. The observations for both data sets were taken at the same site. However, it was believed that the accuracy of model calibration and validation could be improved if the full set of measured daily weather data were used. Thus, the SWREC meteorological data were used for model calibration and validation, whereas the data from HPRC were used for scenario analysis. Assessment of irrigation management strategies requires an understanding of the long-term precipitation characteristics of a given area. Thus, the long-term monthly precipitation trend was analyzed for Garden City. Monthly, seasonal, and annual precipitation means and coefficients of variance were calculated.

Analysis of irrigation water management strategies was done by sorting the long-term seasonal average precipitation data in descending order and calculating the probability of exceedance for wet (1–20%), normal (41–59%), and dry (80–100%) years. The years whose probability of exceedance fell in the respective ranges were noted and the respective seasonal precipitation ranges were calculated (444–579 mm = wet, 334–380 mm = normal, and 201–254 mm = dry). After grouping of the years into the three precipitation scenarios, daily precipitation data from the each season that fell under the three different categories (wet, normal, and dry) were sorted and averaged for use in the AquaCrop model. ETo for the experimental season (April–September) was estimated using the short-term Garden City climate data based on FAO-Penman-Monteith (FAO-PM), whereas the ETo for the long-term growing season was estimated using the Hargreaves method because of the limited availability of climate data (Allen et al. 1998; Hargreaves and Allen 2003). To understand the relationship between the two methods (Hargreaves and FAO-PM), short-term (2000–2013) growing season data estimated using the two methods were regressed and compared.

Climate, soil, crop characteristics, and management data (irrigation and agronomic practices) of the site were supplied as input to AquaCrop. In AquaCrop, crop phenology depends on the nature of the cultivar and air temperature (Raes et al. 2009; Steduto et al. 2009). The calendar day option in the model was used because corn was grown in favorable temperatures in the study area. Calibration for fertility stress was not considered in the model because the corn was assumed to be grown under optimal fertility. Model default calibration data sets were used as the starting point. However, most of the cultivar and site-specific crop parameters (none conservative) were entered in the model as per the collected information from the corn experiment. Table 2 shows some important crop data used for calibrating the AquaCrop. Canopy cover is one of the most important crop parameters in AquaCrop because it limits the amount of water transpired, which indirectly determines the biomass (Steduto et al. 2009). Canopy cover at emergence was adopted from Heng et al. (2009) and Hsiao et al. (2009). Since plant density in the study area varied with irrigation treatments, adjustments to default maximum canopy cover to match actual field conditions was made as presented in Steduto et al. (2009). The canopy cover was estimated from LAI and extinction coefficient ($k$), as presented in Hsiao et al. (2009).

However, owing to variation in planting density, a typical maximum canopy cover (estimated from LAI) that is representative of the treatments was selected to simplify complexities. The maximum canopy cover was then set to 80%.

Estimated ranges of normalized crop water productivity values for C4 crops ($30–35\mathrm{g}/{\mathrm{m}}^2$) are available in the literature (Raes et al. 2012). For this analysis, the normalized crop water productivity values that were adopted in Hsiao et al. (2009), Heng et al. (2009), and Paredes et al. (2014) ($33.7\mathrm{g}/{\mathrm{m}}^2$) were used. Similarly, crop evapotranspiration (ETc) at maximum canopy cover for no stress was also taken from Hsiao et al. (2009). The cultivar grown in the experimental site had a maximum harvest index (HI) of up to 59%. However, 52% was taken as the most common HI value for good growing conditions, which is higher than the value (48%) used in Hsiao et al. (2009). This shows that the experimental cultivar was different from those presented in Hsiao et al. (2009). Water stress may affect HI depending on the timing, duration, and severity of the stress (Raes et al. 2009). Model default values for corn were adopted for the effects of water stress on HI. Reports indicated that water stress could negatively affect pollination, fruit setting or abortion, or under filling of young fruits depending on the timing and extent of the stress (Steduto et al. 2012).

The water stress coefficients (for leaf expansion, stomatal conductance, and leaf senescence) are considered to be conservative (globally applicable once calibrated), meant for use under various climate and soil conditions (Hisiao et al. 2009). However, slight adjustments were necessary for the model to reproduce our measured data. Adjustment for these parameters was carried out after repeatedly comparing the measured against the simulated data, taking into account the available documented information. After adjusting the water stress coefficients, simulated yield, biomass, and ET were comparable to measured values. However, the upper level for leaf senescence obtained based on iteration in this study slightly deviated from the information for maize presented in Hsiao et al. (2009) and Steduto et al. (2012). Some of the possible reasons for this include (1) cultivar differences; (2) soil heterogeneity; (3) method used for estimation of the soil water characteristics [pedo-transfer function (e.g., Saxton et al. 1986) based on soil texture]; (4) limited LAI measurements together with planting density differences among the treatments; and (5) model version differences. For example, Hsiao et al. (2009) hinted that version differences might contribute to the disparity in some of the documented conservative parameters. Overall, since our interest was to develop indicative optimal irrigation strategies that suit corn in the study area, measured biophysical data with the documented information in the literature were adequately used to characterize the actual crop in the model.

ET was estimated using the soil-water-balance method [Eq. (1)]:where $I$ = irrigation (mm); $P$ = precipitation during sampling period (mm); $R$ = runoff or runon during sampling period (assumed negligible, mm); SW1 and SW2 = total profile soil water (mm) between first and last neutron probe readings, respectively; and $D$ = drainage during sampling period. Drainage was estimated as presented in Miller and Aarstad (1972) and Klocke et al. (2011).

The sampling period refers to the period between the first and last neutron probe readings during the growing season. Data from full irrigation treatment for the 2011 cropping season were used for model calibration. Four treatments each in 2010 and 2012 were used to validate the AquaCrop model (Table 3). Treatments that were dry (not representative of the actual corn growing conditions) were not included.

Goodness-of-fit statistics based on the percent of deviation (d) [Eq. (2)], normalized root-mean square error (NRMSE) [Eq. (3)], index of agreement (IAg) based on Wilmott (1982), and coefficient of determination ($R^2$) were used when testing model performance:where $S{m}_i$ = simulated and $M_i$ = measured values. $M_M$ = mean of the measured value; a $d$ value close to zero indicates better agreement between the measured and simulated values. The root-mean square error (RMSE) is a statistical indicator of model uncertainty; a NRMSE value close to zero indicates excellent agreement and, hence, good performance of the model. IAg varies from below zero to 1. Values closer to 1 indicate better agreement between simulated and measured values.

The 5-year moving average of monthly precipitation illustrates the inter- and intraseasonal climate variability at the research site. In particular, May precipitation showed a decreasing trend over the decade from 2003 to 2013 (not shown). The coefficient of variation for monthly, seasonal, and annual precipitation fell within ranges of 46–53%, 18–59%, and 10–37%, respectively, which indicates considerable precipitation variability. Mean seasonal (April–September) and annual precipitation was as low as 182 and 254 mm in a dry year and 444 mm and 579 mm in a wet year, respectively. The seasonal and annual ETo were 870 and 1,810 mm, respectively. This indicates precipitation is variable and not adequate to meet the maximum crop water demand (ETc). Irrigation must be applied to attain the production potential of crops like corn, especially during below-normal-precipitation seasons.

The FAO-PM method of calculating ETo was reported to have better accuracy compared to other methods (Allen et al. 1998). The growing season ETo calculated based on the Hargreaves approach was slightly higher than that based on the FAO-PM approach (Fig. 1), but there was a strong linear relationship between the daily ETo values estimated based on the two approaches with $R^2$ of 0.86 (Fig. 1). The two approaches simulated the growing season (April–September) short term (2000–2013) daily ETo values with a slight difference considering the difference in their fundamental equations.

AquaCrop was satisfactorily calibrated and validated in this study. The moderate NRMSE (13.8%) and the strong IAg (0.80) values indicated good agreement between simulated and measured total seasonal ET. Similarly, there was moderate $d$ between the measured and simulated total seasonal ET values ($-20$ to 17.9%). However, there was a weak correlation between the measured and simulated total seasonal ET ($R^2=0.42$), which might be attributable to inaccuracy in measurements of the soil physical properties, as mentioned previously. Also, the limited LAI measurements that were used to estimate maximum canopy cover might not have perfectly represented the actual conditions because planting density varied across irrigation treatments. Besides the inaccuracy caused by variation in planting density, the limited number of canopy cover measurements might also have contributed to the weak relationship between the simulated and measured ET. However, the goodness of fit based on the percent of deviations, NRMSE, and IAg were satisfactory (Table 3) and were similar to those reported in Katerji et al. (2013).

There was a very strong relationship between measured yield and seasonal ET with an $R^2$ of 0.83 [Fig. 2(a)]. This result mirrors previous reports on crop yield—ET relationships (Doorenbos and Kassam 1979; Molden et al. 2007; Stewart and Peterson 2015; Klocke et al. 2015). Similarly, there was a strong relationship between measured biomass and seasonal ET with an $R^2$ of 0.74 [Fig. 2(b)]. AquaCrop simulated yield and biomass of corn well with a NRMSE of 5.1 and 15.5%, respectively. The relationship between the measured and simulated yield ($R^2=0.99$) [Fig. 3(a)] and biomass ($R^2=0.96$) [Fig. 3(b)] was strong. Similarly, IAg values for yield and biomass were 0.99 and 0.94, respectively (Table 3). These indices of agreement values were similar to those reported in Hsiao et al. (2009) and Heng et al. (2009). Despite the simplification of the complex processes, simulation outputs satisfactorily agreed with the measured data. Overall, the goodness-of-fit indicators showed satisfactory performance of the model in simulating the measured yield, biomass, and ET. Hence, the model can be used for assessing alternative irrigation management strategies based on water productivities for various irrigation water allocation amounts, long-term seasonal precipitation, and planting dates.

Water productivities for yield (WPy) and biomass (WPb) values were higher in 2010 compared to 2012 (Table 4), which may be due to the better scheduling of the irrigation and synchronization of growing season precipitation to the sensitive growth stages. The growing seasons in 2010 and 2012 were wet and dry, respectively, which might have contributed to the variability in WPy and WPb. The WPb values for unstressed treatments in 2010 were similar to the WPb values documented in Keller and Seckler (2005) for California areas. In 2010, a point was reached at which WPy and WPb did not increase with the increase in ET (Table 4). This reduction in water productivity at high ET levels during 2010 might be partly caused by loss of water via evaporation and in some cases may be due to increase in nutrient leaching. It is also worth noting that 2011 and 2012 were severe drought years at the study site.

The highest WPy values for the April 20 planting date were 1.7, 1.8, and $1.8\mathrm{kg}/{\mathrm{m}}^3$ for dry, normal, and wet seasons, respectively (Table 5). The WPy values were comparable to the May 5 planting date for the corresponding growing season climatic scenarios (1.8, 1.8, and $1.8\mathrm{kg}/{\mathrm{m}}^3$) under the respective irrigation water allocation limits (Table 5). Our simulation results were somewhat similar to those values reported in Mengu and Ozgurel (2008) for semiarid regions of Turkey. They also observed a 10–27% yield reduction when 30% less water was applied compared to their full irrigation treatment (Mengu and Ozgurel 2008). Similarly, the WPy obtained for the May 15 planting date under the dry season scenarios was $1.7\mathrm{kg}/{\mathrm{m}}^3$, obtained from a similar irrigation water allocation (450 mm) limit (Table 5). However, simulated yield was slightly lower under late planting compared to early and normal planting dates. This indicates that late planting was not the best strategy since the yield and biomass levels were substantially lower compared to those simulated for early and normal planting. One of the reasons for lower yield under late planting during a dry climate was attributed to poor matching of the rainfall with the peak demand of the crop during the growing cycle.

The biomass water productivity (WPb) for dry growing season precipitation scenarios were $3.4\mathrm{kg}/{\mathrm{m}}^3$ for early, $3.5\mathrm{kg}/{\mathrm{m}}^3$ for normal, and $3.3\mathrm{kg}/{\mathrm{m}}^3$ for late planting dates, while the corresponding WPb values for wet growing season precipitation scenarios were 3.5, 3.5, and $3.7\mathrm{kg}/{\mathrm{m}}^3$, respectively (Table 5). This shows that WPb values slightly increased from dry to wet growing season precipitation scenarios.

Corn response to planting date, growing season climate, and irrigation water allocation scenario in silt loam soils was very similar to that in sand clay loam soils discussed earlier. However, there was slightly lower yield for the dry climate scenario under silt loam soils compared to sand clay loam soils (Table 5). For this reason, the WPy for dry climatic scenario in silt loam ($1.5–1.7\mathrm{kg}/{\mathrm{m}}^3$) was slightly reduced compared to sand clay loam soils ($1.7–1.8\mathrm{kg}/{\mathrm{m}}^3$).

This study showed that it is possible to stabilize (optimize) corn yield at yield levels slightly lower than the maximum. Considering initial plant allowable soil water at 70%, the optimal deficit irrigation water allocation under the various growing season precipitation scenarios in both soils (silt loam and sandy clay loam) were 300, 450 and 150 mm during the normal, dry and wet seasons, respectively. However, growers should also take into account the initial soil water status (at planting), which could vary with soil type and previous crop and climatic conditions. For example, slightly lower yield and biomass per unit of ET was simulated for dry precipitation seasons compared to wet [Figs. 4(a–d)], indicating that water during dry seasons might not be sufficient to meet the crop water need. Also, during the dry season where rainwater is variable, water loss through soil evaporation could increase slightly compared to the wet season.

Deficit irrigation might maximize crop water productivity depending on many factors, including the amount of precipitation received, targeted yield level, planting date, crop and cultivar type, irrigation scheduling, soil type and initial soil water status (at planting), and irrigation capacity of the well. For example, deficit irrigation of wheat was found to maximize crop water productivity without significant reduction in yield (Zhang and Oweis 1999), which might require understating the response of each growth stage of the crop to water stress (Oweis and Hachum 2003; Zhang 2003), precipitation patterns, and economic impacts of yield penalties (English 1990). For corn, some studies have indicated that deficit irrigation may not be appropriate if drought stress occurs during late vegetative and flowering stages (Kirda et al. 1999). Cakir (2004) reported that water stress during vegetative and tasseling growth stages could reduce leaf area index whereas water stress during cob development and early grain filling (milking stage) could reduce biomass and grain yield. Yield can be significantly reduced if water stress occurred during tasseling and ear formation. Some authors have reported better response of corn to deficit irrigation at early vegetative stages (Fabeiro et al. 2001).

In this study, the targeted optimal yield was found to be achievable with an ET of 675–703 mm for early, 664–702 mm for normal, and 623–675 mm for late planting in sandy clay loam soils (Table 5), whereas the corresponding ET values for silt loam soils were 679–709 m, 662–714 m, 625–687 mm (Table 5). However, irrigation water should be carefully applied based on the sensitivity of the crop to water stress, although this is not always possible with limited well capacity.

Table 6 shows the effect of spreading versus concentrating irrigation water on attainable corn yield (as simulated based on AquaCrop model), and crop water productivities assuming corn is grown in a dry season in sandy clay loam and silt loam soils with limited irrigation capacity. A typical irrigated area for a center pivot sprinkler system (46.8 ha) was considered in this analysis in proportions from 100, 75, and 50% of 46.8 ha. By reducing the irrigated area, farmers can increase irrigation capacity (the ratio of flow rate to the land area irrigated), as shown in Table 6. Considering sandy clay loam soils, Farmer A planted 100% of the area and produced $4.8\mathrm{t}/\mathrm{ha}$, which was lower than Farmers B and C, who planted 75 and 50% of the total area and produced 9.2 and $11.8\mathrm{t}/\mathrm{ha}$, respectively. Farmer C obtained the highest yield per unit area with the highest crop water productivity ($1.8\mathrm{kg}/{\mathrm{m}}^3$).

Similarly, under silt loam soils, Farmer A planted 100% of the total area and produced $8\mathrm{t}/\mathrm{ha}$ compared to Farmers B and C, who planted 75 and 50% of the total area and produced 9.8 and $11.3\mathrm{t}/\mathrm{ha}$, respectively. There were substantial differences in yield between the three farmers, as shown in Table 6. However, Farmer C achieved the highest crop water productivity and yield compared to Farmers A and B. These results imply that on well-drained sandy clay loam and silt loam soil, it is possible to plant 50% of the area and optimize crop water productivity under a typical center pivot sprinkler system. However, under commercial agricultural production systems similar to those in semiarid southwestern Kansas, profitability needs to be evaluated in addition to crop water productivity whenever assessments of deficit irrigation strategies are made. Crop model–guided irrigation water optimization and yield simulations as presented in this study could be enriched with information such as net income to guide comprehensive decision making for sustainable water use and profitability. It is also worth noting that targeting irrigation to critical growth stages is only possible if the producer reduces the total irrigated area in order to increase irrigation capacity. However, in wet years, a reduction in the total irrigated area might reduce income potential.