The study focused on Halfway (34° 10′ N, 101° 56′ W; elevation 1,075 m) in Hale County, TX in the THP where the Texas A&M AgriLife Research Station is located. Measured data from a cotton irrigation water use efficiency (IWUE) experiment conducted at this site with a center pivot irrigation system over four growing seasons from 2010 to 2013 (Bordovsky et al., 2015) were used for model evaluation. The soil at the study site is classified in the Pullman clay loam series (fine, mixed, super active, thermic Torrertic Paleustolls). The average (1977–2018) annual and growing season (from 1 May to 31 October) rainfall was 463 and 344 mm, respectively (Himanshu et al., 2019). A total of 27 irrigation treatments were implemented in the IWUE field experiment with combinations of three levels of daily irrigation [0 mm day⁻¹ (Low), 3.2 mm day⁻¹ (Medium), and 6.4 mm day⁻¹ (High)] applied during the vegetative, reproductive, and maturation growth stages. Conventional tillage was adopted at the study site and pre-plant irrigations, if needed, to elevate the top 0.6 m of the soil profile to field capacity, were applied uniformly on all plots. Additional details about irrigation, fertilizer, and chemical applications in the IWUE field experiment are given by Bordovsky et al. (2015) and used for model parameterization. Measured seed-cotton yield data from the IWUE experiment were used for model evaluation. Daily weather data for the period from 2005 to 2019 were obtained from the on-site weather station for scenario analysis.

The DSSAT is a platform that integrates soil, weather, crop, and experimental data management programs with crop models and different application programs (Jones et al., 2003; Hoogenboom et al., 2019). The DSSAT CSM can calculate crop growth, development, water use and yield, soil water, and C and N processes under combined effects of weather patterns, soil properties, cultivar characteristics, and management practices on a daily interval. The latest 4.7.5 version of the DSSAT model has 42 crop modules including CROPGRO-Cotton (Hoogenboom et al., 2019). The minimum weather data required to run the model include daily precipitation, maximum and minimum air temperature, and solar irradiance, while wind speed and relative humidity are optional input parameters. The model requires several soil parameters including texture, color, bulk density, slope, albedo, soil water constants, hydraulic conductivity, drainage, organic carbon content, and total soil nitrogen. The management data required by the model include dates of planting, harvest, and tillage; planting depth and density; row spacing; cultivar; and details about crop residue management, and fertilizer, chemical and irrigation applications, as applicable. More details about the DSSAT model inputs used in this study can be found in Adhikari et al. (2016).

A tipping bucket approach is used for soil water balance simulation in DSSAT with three key soil moisture variables: Saturated Water Content (SAT), Drained Upper Limit (DUL), and the Lower Limit (LL) of plant extractable water (Hoogenboom et al., 2019). The DUL and LL are equivalent to field capacity and wilting point soil water contents, respectively. In DSSAT, soil water balance is calculated on a daily basis by adding irrigation and rainfall and subtracting surface runoff, drainage, and ET. Rainfall is partitioned to infiltration and surface runoff using the Soil Conservation Service (SCS) curve number approach (Ritchie, 1998). Downward movement of water within the soil depends on a soil drainage factor (fraction day⁻¹), which is limited by saturated hydraulic conductivity of soil layers (Hoogenboom et al., 2019). There are two options for estimating ET; the Priestley-Taylor method (Priestley and Taylor, 1972), and the FAO-56 method (Allen et al., 1998). For computation of SOM dynamics, either the original CERES-based module (Godwin and Singh, 1998) or the CENTURY-based module (Gijsman et al., 2002) can be used. The main difference between these two modules is that the surface fresh organic matter and three pools of SOM are included in the CENTURY model, and it allows more control over initialization of stable carbon pools and hence overall decomposition dynamics (Hoogenboom et al., 2019).

The DSSAT CSM CROPGRO-Cotton model was previously evaluated for Halfway, TX by Adhikari et al. (2016) using measured data from Bordovsky et al. (2015) IWUE experiment by creating a DSSAT experiment project. In an experiment project, the initial conditions (e.g., soil water, crop residue) are initialized at the beginning of each simulation year and this was appropriate for the Adhikari et al. (2016) study, which followed the field experiment in which pre-plant irrigation was applied when necessary based on the soil water status. However, as the goal of the current study was to evaluate the long-term effects of changes in drainage, surface runoff and soil properties on dryland crop production, a re-evaluation of the Adhikari et al. (2016) model by creating a DSSAT sequential project was necessary. In a sequential project, long-term simulations are conducted continuously (without initializing the initial conditions each year) and hence it represents a cotton-fallow cropping system more realistically and ensures that the water and nutrient balances during the fallow periods between cotton growing seasons are simulated accurately.

During the sequential evaluation, as was done in Adhikari et al. (2016) study, the DSSAT CSM CROPGRO-Cotton model was calibrated using the measured data over four growing seasons (2010–2013) from four high irrigation treatments with negligible/no water stress (HHH, HHM, MHH, and MHM) in the Bordovsky et al. (2015) experiment. The calibrated model was then evaluated using measured data from the remaining 23 treatments over the four growing seasons. More details about the model setup for the study area and model inputs can be found in Adhikari et al. (2016). Model calibration was carried out manually by slightly adjusting Adhikari et al. (2016) parameters until simulated and measured seed-cotton yields matched. The model performance during the calibration and evaluation was assessed both graphically (by constructing scatter plots) and statistically. For consistency, four model performance statistics used by Adhikari et al. (2016), including the coefficient of determination (r²) (Legates and McCabe, 1999), index of agreement (d) (Willmott et al., 1985), root mean square error (RMSE), and the percent error (PE), were used in this study. The r² and d-values range between 0 (indicating no fit between simulated and measured values) and 1 (perfect fit). The RMSE values closer to 0 indicate a better agreement between the simulated and measured values. The PE varies between −100 and ∞, with smaller absolute values closer to 0 indicating a better agreement. The model calibration was continued until the performance statistics were comparable to those achieved in the Adhikari et al. (2016) study.

Long-term simulations were run with the evaluated model for the period from 2005 to 2019 and this scenario was considered as the “baseline” scenario (hereafter referred to as the S0 scenario). In the long-term simulations, a common planting date of May 9 was assumed each year based on the actual dates implemented in the field experiment. A conventional tillage (with a field cultivator and rotary hoe tillage) was implemented in each year of simulation as per the standard practice followed at the study site. Potential dryland cotton yield increases from changes in runoff, drainage, and soil properties associated with soil health were then assessed by running five hypothetical scenarios: (1) increasing DUL by 25 mm to reflect the potential increase in soil water holding capacity due to adoption of soil health improving practices such as cover crops (S1 scenario); (2) decreasing curve number from 73 to 60 to reduce surface runoff and thereby simulate potential increase in infiltration/soil water holding capacity due to adoption of soil health promoting practices such as no-tillage (S2); (3) increasing soil albedo from 0.2 to 0.4 to reflect potential increase in reflected solar radiation due to placement of crop residue on the surface (S3); (4) increasing the soil drainage factor from 0.2 to 0.4 to reflect potential increase in infiltration and downward movement of soil water from practices such as cover crops and deep tillage (S4); and (5) increasing SOC by 1% through adoption of practices such as no-tillage and cover crops (S5). Additionally, we ran a sixth scenario (S6) with a 50 mm pre-plant irrigation (25 mm each, applied on 28 April and 2 May in each year) to reflect a practice that some of the producers in the THP region adopt. These producers convert a portion of the center pivot area into dryland production due to reduced irrigation capacities, and apply pre-plant irrigations on dryland cotton in late April/early May when groundwater levels are relatively shallower (compared to summer) and crop water demand is negligible (unless cover crops are grown).

A statistical analysis was done using a student's t-test (Gosset, 1908) to determine if the seed-cotton yields under simulated scenarios were significantly different from those under the baseline scenario. A paired two-tailed t-test (considering the same population twice, with- and without changes to soil properties/pre-plant irrigation) was performed using the T.TEST command in Microsoft Excel to test statistical significance at 95% level (p = 0.05). A “combination” scenario (Sc1) was then run by simultaneously implementing all statistically significant scenarios to predict the “potential maximum dryland seed-cotton yield” with the adoption of multiple soil health promoting practices. Additionally, in view of mixed reports on the effect of increase in SOC on plant available water (Cano et al., 2018; Minasny and McBratney, 2018), an additional combination scenario (Sc2) was run by excluding S1 scenario from the Sc1 scenario. The effect of climate variability on simulated results was finally assessed by dividing the simulation years into dry, normal, and wet years based on the growing season precipitation from 1 April to 31 October (Figure 1). After sorting the years based on the growing season precipitation, the 5 years that received the lowest precipitation were classified as the dry years, and the 5 years that received the highest precipitation were classified as the wet years. The remaining 5 years were classified as the normal years.

In dryland systems, increasing average yield is important. However, producers are risk averse (Sandmo, 1971) and would therefore be willing to give up some revenue to reduce exposure to risk. The impacts of different soil health improvements must also consider the impacts on variance (risk) of returns as well. Farm revenue was simulated using a Monte Carlo method combining the simulated yields in this analysis with a sample of actual cotton market prices drawn from the US Department of Agriculture (USDA) database on farm prices (USDA-ERS, 2019). A beta distribution was fit to the simulated yields because it allows for the skewness inherent in yields censored at zero and has been used in many applications of evaluating yields for insurance modeling purposes (Luitel et al., 2018). A log-normal distribution, which is the most commonly assumed distribution for prices, was fit to cotton prices. Prices and yields are often (at least minimally) negatively correlated, thus Monte Carlo simulations must account for that correlation when modeling farm revenue (price × yield). The procedure outlined by Anderson et al. (2009) was used to generate a simulated distribution of farm revenue assuming a price/yield correlation of r = −0.092 as was used by Luitel et al. (2018) for Texas cotton farms.

In our analysis, only farm revenue was considered (no costs of production and no costs of obtaining the soil health improvements). The analysis aimed to identify two key outcomes. First, a risk averse producer would be willing to accept some level of revenue lower than the expected (mean) revenue to avoid the risk of producing. This value is called the “certainty equivalent” (CE) and represents the amount of revenue required for the producer to avoid taking the risk. Again, following Luitel et al. (2018), this analysis used a constant relative risk aversion (CRRA) coefficient of two, which is associated with a moderately risk averse producer. This value was arbitrary but was held constant across all scenarios for comparability. The CE was interpreted in relation to the base value (S0). Higher CE values indicated that the producer would prefer that scenario to the base and vice versa. Second, an important outcome of soil health improvements must be to reduce the risk of low revenue years if the producer is to be at all interested in pursuing those outcomes. If the soil health improvement does not reduce risk of catastrophic losses relative to the status quo, improvements in average revenue may be ignored in preference for “safer” alternatives. To evaluate this risk, a value of 40% of the expected (mean) revenue was arbitrarily considered to be a catastrophic loss. Again, the value chosen was arbitrary but was held fixed across alternatives to insure comparability. Revenue distributions were found to be best fit by log-normal distributions, so this comparison was conducted by calculating the probability of observing 40% of the mean revenue in the base case, or $69 ha⁻¹ in each of the scenarios.

Several of the calibrated parameters of Adhikari et al. (2016) were adjusted and some of their initial conditions were modified during the sequential calibration of the DSSAT CSM CROPGRO-Cotton model (Table 1). The values of previous crop (cotton) residue and root weight, which were not considered by Adhikari et al. (2016), were estimated in this study, and input to the model. The N and P contents of residue were obtained from the literature (Gemtos and Tsiricoglou, 1999; Wanjura et al., 2014). In addition, the initial soil water content was changed from 0.18 to 0.24 m³ m⁻³ during the sequential calibration (Table 1). Some of the methods used by Adhikari et al. (2016) were also modified to get a good match between the simulated and measured seed-cotton yields (Table 1). Most importantly, the FAO-56 ET method (Allen et al., 1998), which was used by Adhikari et al. (2016), calculated consistently lower seed-cotton yields for the sequential project and hence Priestly-Taylor method was used in this study. A majority of other DSSAT cotton modeling studies in the literature have also used the Priestley-Taylor method (e.g., Guerra et al., 2007; Garcia et al., 2008; Gérardeaux et al., 2018). The SOM simulation method was also changed from the CERES (Godwin and Singh, 1998) to the CENTURY (Gijsman et al., 2002) method as the latter method simulates soil carbon and nitrogen dynamics more efficiently than the former method.

The simulated seed-cotton yield matched well with the measured yield during both calibration and evaluation as shown in Figure 2, and also as indicated by model performance statistics (Table 2). The performance statistics achieved during calibration (r² = 0.82, d = 0.9, RMSE = 18%, and PE = 10%) and evaluation (r² = 0.84, d = 0.95, RMSE = 29%, and PE = −0.3%) were comparable to those obtained by Adhikari et al. (2016). In general, the model predictions of seed-cotton yield were better in case of high irrigation (calibration) treatments as compared to those for low/medium irrigation (validation) treatments as indicated by lower % RMSE values during the model calibration (18 vs. 29%). These trends are in agreement with previous studies that reported unsatisfactory performance of DSSAT CSM under dry conditions (Thorp et al., 2014; Modala et al., 2015; Adhikari et al., 2016; Kothari et al., 2019). The model performance during the year 2012 was not satisfactory, mainly due to the influence of extreme dry conditions in the preceding year. Overall, the evaluated DSSAT CSM CROPGRO-Cotton sequential model demonstrated potential in effectively simulating seed-cotton yield under 27 treatments in Bordovsky et al. (2015) experiment with widely variable irrigation applications over four growing seasons (2010 to 2013) with contrasting weather conditions.

Among the simulated scenarios, the soil drainage rate increase scenario (S4) showed the greatest increase in simulated seed-cotton yield followed by DUL increase (S1) and soil albedo increase (S3) scenarios as compared to the baseline scenario (Figure 3). While the mean seed-cotton yield increased by 27, 7, 16, 58, and 12% under S1, S2, S3, S4, and S6 scenarios, respectively, when compared to the baseline scenario, it decreased slightly by about 1% under S5 scenario (increased SOC) (Figure 3). There was high variability in simulated seed-cotton yield under S4 and S1 scenarios. The calculated p-values from the student's t-test for S1, S2, S3, S4, S5, and S6 scenarios were 0.022, 0.026, 0.002, 0.0008, 0.162, and 0.090, respectively. These p-values indicated that the simulated seed-cotton yield changed significantly (p < 0.05) under S1 to S4 scenarios, while it did not change significantly (p > 0.05) under the remaining two (S5 and S6) scenarios (p > 0.05). Although the differences in mean seed-cotton yield between the base scenario and the alternate scenarios were substantial, the differences in median seed-cotton yield were small, except in the case of S4 scenario. This was because simulated seed-cotton yield values were low in majority of simulation years. However, long-term simulations were run for only 15 years in this study, and in general, median values present more realistic trends when the simulation period is much longer (~40 years or when the sample size is larger). The effect of changes in selected factors, soil properties, and pre-plant irrigation on seed-cotton yield is discussed in more detail in the subsequent sections.

The increase in DUL by 25 mm (S1 scenario) increased plant available water (the difference between field capacity and permanent wilting point soil water contents), which contributed to a 27% increase in mean seed-cotton yield (Figure 3). The effect of increase in DUL on mean seed-cotton yield was the highest in normal years (34%) and lowest in the dry years (4%) (Figure 4). In the dry years, the average growing season precipitation was about 31% lower than that in the normal years, and hence the benefits of increase in DUL could not be fully realized due to insufficient rainfall. These results are consistent with published studies (Wang et al., 2009; He and Wang, 2019). Using an agricultural production systems model (APSIM), Wang et al. (2009) found that the soils with higher soil plant available water holding capacity (PAWC) enabled more rainfall to be used by crops. In a follow-up study, He and Wang (2019) noted that the long-term averaged dryland wheat biomass and yield increased with an increase in PAWC until a critical value was reached. The increase in DUL also increased soil water evaporation from the surface layer and decreased deep drainage, and therefore the increase in mean seed-cotton yield with the increase in DUL was lower compared to the increase in soil drainage rate [S4 scenario; section Effect of Increasing Soil Drainage Rate (S4 Scenario) on Seed-Cotton Yield].

The reduction of 18% in curve number from 73 to 60 decreased surface runoff and increased infiltration, and thereby plant available soil water, which contributed to a 7% increase in mean seed-cotton yield (Figure 3). As expected, the effect of decreased surface runoff was the highest in wet years (10%; Figure 4), which received, on an average, about 49 and 115% higher growing season precipitation compared to normal and dry years, respectively (Figure 1). Adimassu et al. (2019) and Gharibdousti et al. (2019) have also reported that conservation practices such as contour farming, zero/no-tillage, conservation tillage with sufficient crop residue increased water infiltration to the soil, which resulted in a reduction in surface runoff and enhanced wheat crop yield. In another field study in Croatia, Bogunovic et al. (2018) observed a significantly lower average annual surface runoff and soil loss under no-tillage and deep tillage than under conventional tillage. However, no-tillage had lower grain yields compared to conventional and deep tillage, but higher yields in dry years due to high capacity for water retention.

The increase in soil albedo from 0.2 to 0.4 reduced heat flux into the soil, and water vapor flux from the soil (Lascano and Baumhardt, 1996; Bogunovic et al., 2018), resulting in an overall reduction in soil water evaporation and an increase in plant available water, and consequently a 16% increase in mean seed-cotton yield (Figure 3). The effect of increasing soil albedo was relatively higher in the normal years (21%) as compared to dry years (16%) and wet years (14%) (Figure 4). However, the effect of changing soil albedo depends on the average growing season temperature in addition to precipitation. Using a numerical model, Lascano and Baumhardt (1996) also found that the cumulative soil water evaporation was 50% of the seasonal ET in conventional cotton and 31% of the seasonal ET in the wheat-stubble cotton while the seasonable ET was similar in both systems.

The DSSAT crop models simulate only a one-dimensional water balance with vertical flow, and the downward movement of water within the soil depends on the soil drainage rate (Hoogenboom et al., 2019). Soil drainage rate directly affects the amount of plant available water, and hence accurate calculation of crop yield (Suleiman and Ritchie, 2004). While a higher soil drainage rate may result in a significant deep percolation loss and thereby reduce the plant available water, lower soil drainage rate can cause excess water stress in the cases of extended saturation periods. Saturation resulting from slow soil drainage rate can also cause surface ponding/runoff (Vanderborght et al., 1997; Suleiman and Ritchie, 2004). A soil drainage rate of 0.4 (fraction day⁻¹) is recommended for moderately drained soils (Romero et al., 2012). In this study, an increase in soil drainage factor from 0.2 (baseline) to 0.4 (S4 scenario) enhanced downward movement of water to deeper soil layers and resulted in a reduced soil water evaporation and increased availability of soil water for plant transpiration requirements, and thereby contributing to the highest increase (58%) in simulated mean seed-cotton yield (Figure 3). The effect of increasing soil drainage rate on seed-cotton yield was the highest in normal years (83%) followed by wet years (52%) and dry years (26%) (Figure 4) indicating that the increase in soil drainage rate led to higher percolation losses and hence reduction in plant available water in wet years, compared to normal years.

A statistically insignificant and slightly declining trend (1% decrease) in simulated mean seed-cotton yield was found with a 1% increase in SOC (Figure 3) and this was contrary to many published results in the literature. For example, Lal (2006, 2020), Qiu et al. (2009), and Baumhardt et al. (2013) report a substantial improvement in crop yield with an increase in SOC. However, large productivity gains with the increase in SOC are associated with the implementation of a combination of practices including judicious application of fertilizers, irrigation, and other amendments. Baumhardt et al. (2013) reported that cotton irrigated at 2.5- and 5.0-mm day⁻¹ averaged 55 kg ha⁻¹ lint for every 25 mm of water used. Increasing soil available water content (AWC) was attributed to greater SOM associated with improved soil health (Cano et al., 2018); that is, a 1% SOM increase in the surface 150 mm of soil expanded AWC by >252.5 m³ ha⁻¹ (NRCS, 2013a). However, increasing the SOC by 1% did not result in any statistically significant increase in our simulated cotton yields compared with simulations for unadjusted soil conditions. The matched increases of 1% SOM and 252.5 m³ ha⁻¹ AWC appearing in an NRCS graphic media (NRCS, 2013b) identified a source article by Mengel (2012) that summarizes results from Australia (Emerson, 1995). In that work, the AWC of loams and fine sandy loams determined between 10-and 1,500-kPa increased 3% by volume with every 1% increase in SOC and was also reported by Hudson (1994) using 1,500-kPa for permanent wilt and 10–33 kPa for field capacity. The NRCS (2019) stipulates AWC is determined at pressures of 33 and 1,500 kPa. By contrast, the near saturated 10 kPa condition may exaggerate the effect of SOM on AWC due to gravity drainage to 33 kPa that in turn reduces potential water for crop use.

Application of 50 mm of pre-plant irrigation increased mean seed-cotton yield by 12% (Figure 3). Interestingly, the effect of pre-plant irrigation was the highest (21%) under wet years followed by dry years (4%) (Figure 4). In normal years, pre-plant irrigation reduced seed-cotton yield, on an average, by 4%. Water requirement of cotton varies with the growth stages (Himanshu et al., 2019) and hence, in addition to the total amount of rain received during the growing season, its distribution during the growing season is equally important and affects seed-cotton yield. These percent changes in seed-cotton yield with pre-plant irrigation application were statistically not significant (p > 0.05). These trends are consistent with those reported by Bordovsky (2020) from a field experiment near the study site. Bordovsky (2020) found that the commonly applied pre-plant irrigations and larger irrigation applications during the vegetative period did not significantly increase seed-cotton yield due to high evaporative losses during the late spring and early summer in the semi-arid environment prevalent at the experimental site. Evett et al. (2019) have also reported high evaporative losses from sprinkler-irrigated corn fields at Bushland, TX in the THP region.

Combined implementation of changes in four factors, i.e., increase in DUL, soil albedo, and soil drainage rate, and decrease in runoff curve number, significantly affected seed-cotton yield (scenario Sc1) and resulted in a 93% increase in mean (2005–2019) seed-cotton yield (Figure 5). When only soil albedo, soil drainage rate, and runoff curve number were changed (scenario Sc2), the simulated mean seed-cotton yield increased by 86%. Letey (1958) also suggested considering multiple soil properties while assessing productivity potential of a soil because soil properties affect water use potential, oxygen diffusion rate, and mechanical resistance, which have a direct relationship with production.

Changing a single soil physical property that is linked with other properties and ignoring their relation may lead to results that do not reflect the interaction that takes place in the soil system. For example, an increase in SOM may change the soil structure, which may lead to changes in bulk density and the porosity of the soil. More importantly, what is ultimately affected is the soil pore distribution, which in turn affects the soil moisture release curve and thus transport of water in the profile. The complexity of capturing the interaction of basic processes with a modification of the state of the system is a key challenge in modeling soil processes (Vereecken et al., 2016). This is the subject of a forthcoming paper where we explore the combined effect of changing soil properties and how they affect the simulated cotton lint yield. This exploration is crucial to provide realistic and practical recommendations on management options to dryland producers in the THP.

The effect of changing multiple soil properties together was the highest under normal years (157 and 132% increase under Sc1 and Sc2, respectively) followed by wet years (75% increase under Sc1 and 73% increase under Sc2). Potential reason for a significantly lower increase in mean seed-cotton yield under wet years as compared to normal years could be development of excess water stress due to excessive precipitation in critical growth stages in some years (Himanshu et al., 2019). Overall, based on these simulations, if a combination of soil health promoting practices could be successfully implemented in dryland cotton production systems in the semi-arid regions, and if these practices resulted in the modeled soil characteristics, then seed-cotton yield could be improved substantially by as high as about 86–93% as compared to the baseline scenario. However, successful implementation of soil health promoting practices in dryland cotton production in the THP region would be challenging. For example, no-till practice results in blowing soil (sand) following a rain event, and hence could potentially destroy an early plant stand. Cover crops could potentially reduce soil water availability for establishing dryland cotton in dry years (Adhikari et al., 2017; Lewis et al., 2018). Moreover, there is no empirical evidence, nor is it suggested, that a simple increase of SOC leads to DUL increases on the order of those modeled herein. It should be borne in mind that the purpose of using a 1% increase in SOC or of an additional 25 mm DUL was to examine the effects of two separate parameters associated with soil health upon potential crop yields.

Results of the CE, expected (mean) revenue, the probability that a value of $68.81/ha (40% of the mean revenue under the base scenario) or less would be observed in each scenario, and a Kruskal–Wallis test of equivalence of median revenues (compared to the base) are shown in Table 3. In the base case (S0), the CE is equal to $156/ha, the mean revenue is $172/ha, and the probability of observing a revenue ≤$68.81 is 0.34. By comparison, the CE for S1 (DUL) is $242/ha, indicating that producers would prefer the risk adjusted returns for S1 compared to S0, which is reflected in the higher expected revenue per hectare. The increase in DUL does reduce the risk of lower returns to 0.30, or a 4-percentage point decrease in probability. Note that across scenarios, S2 and S5 generate higher mean returns ($237 and $223, respectively) but the Kruskal–Wallis test [low Chi-squared statistic values of 1.29 (S2) and 2.73 (S5)] indicates that the median returns from those scenarios is not statistically different from the base case.

The single factors that we used in our analysis and that had the largest impact on producer returns is soil drainage rate (S4), with larger CE ($285) and mean returns ($317) while reducing overall risk of revenue shortfall to 0.23 vs. 0.34 in the base case. The combined soil health improvements (Sc1 and Sc2 scenarios) do reduce risk and raise both CE ($339 and $326, respectively) and mean returns [$376 (Sc1) and $360 (Sc2)] relative to all other scenarios and the base (both CE and mean return are over twice the base case). Thus, the results indicated that a combined suite of soil health improvements does reduce revenue risk for producers with most of that impact coming from soil drainage rate improvements. This result, however, does not include the cost (or feasibility) of those improvements and thus impacts profitability.