This study was conducted on commercial farms in the eastern and central South Dakota from the fall of 2017 to the fall of 2020 on 8 site-years ( Figure 1 ; Table 1 ). The study sites were selected to represent diverse agroclimatic regions in SD. For this experiment, we included the farms that follow a specific rotation: small grains such as wheat (Triticum aestivum) or oats (Avena sativa)), followed by a fall CC mix, and then a corn or soybean crop is grown in the next summer (cash crop season). Therefore, impacts of CCs planted in 2017 fall was investigated for the cash crop in 2018 summer, and so on. At all selected sites, the farmer-cooperators practiced no-till and had previous experiences of cover cropping. Therefore, the farmers selected and planted the cover crop mixes that they preferred, and generally the CC mixes had several species of grasses like sorghum (Sorghum vulgare) and millet (Setaria italica), legumes like vetch (Vicia sativa) and clovers (Melilotus alba) and brassicas like radish (Raphanus sativus) and turnips (Brassica rapa) that were winter-terminated, after the first or second frost event.

A randomized complete block design (RCBD) was used to compare CC plots with no CC plots, where each treatment was replicated four times. Each whole plot size was 9.2 m in length and 4.6 m in width. CC mixes were planted by the farmer cooperator after the harvest of a small grain in rotation, majorly winter wheat or oats. The project team then established the replicated trial by spraying a broad spectrum herbicide (Glyphosate 0.84 kg a.e. ha^-1) to designate no cover control plots, as seen in the picture ( Figure 2 ). No fertilizer was applied to the CC plots. For the following cash crop, the farmer cooperators followed the South Dakota State University recommended fertilization protocol (45). Farmer-cooperators also chose the crop cultivars to be planted on the research plots at populations and row spacings (40, 50, 57, or 75 cm apart) the same as the rest of the field ( Table 1 ).

Twelve core soil samples were collected from each replicated experimental plots in the fall before the CCs were winter-terminated and in the spring, before cash crop planting. Soil samples were collected from a depth of 0 to 15 cm using a soil probe with an inside diameter of 1.9 cm. The samples were then air-dried and ground to pass through a 2 mm sieve. These soil samples were analyzed for general soil fertility measurements (nitrate-N, available P, extractable K, soil organic matter, electrical conductivity, and pH) following the recommended chemical soil test procedures for the North Central Region (46) ( Table 1 ).

Furthermore, the selected SHIs, potentially mineralizable N (PMN), permanganate oxidizable C (POXC) also known as ‘active carbon’, soil respiration (SR), and soil microbial biomass (SMB) were determined following standardized protocols. The PMN test was done using the protocol adopted for the Cornell Soil Health Laboratory (47) based on (48). Briefly, the PMN measurement was calculated by subtracting the zero-day soil ammonium (NH₄)–N measurement from the NH₄–N measurement determined after the soil was incubated in an anaerobic environment (submerged in water) for 7 days. The POXC values were calculated using the protocol adopted by the Cornell Soil Health Laboratory (49), based on methods from (50), with modifications described in (51). Briefly, 2.5g soil was shaken with a potassium permanganate solution to oxidize the ‘active carbon’ fraction in soil, which uses some KMnO4, and reduces the strength of the solution, and then absorption of a further diluted solution is measured to calculate POXC. Soil respiration (SR) was measured using the protocol adopted by the Cornell Soil Health Laboratory (49) that followed methods described by Weaver et al. (52). For SR, 20g soil was incubated for 4 days with a CO₂-trap solution (0.5M potassium hydroxide, KOH); change in electrical conductivity of the trap solution during the incubation is used to estimate SR (53). For assessment of microbial biomass under experimental plots, Phospholipid Fatty Acid (PLFA) analyses was conducted on soil samples following methods described in Chowdhury and Dick (54). PLFA is a universally recognized biomarker that provides a representation of living microbial communities in the soil.

All above-ground plant biomass, including CCs and previous crop residue samples were collected within two 30.5 cm² areas from each replicated CC treatment plot in the fall before CC winter termination. Fall sampling dates occurred between late September and early November, depending on the first freezing event, while spring sampling occurred in late April and May, before the cash crop planting. The surface residue biomass (RB) samples included the total previous crop residue left above-ground in the no-till plots and the CC residue planted in the fall. Previous crop residues are included in the analytical RB samples to include carbon, nutrient and other soil health contribution by these previous crop residues with comparatively higher C:N along with new cover crop residues with comparatively lower C:N to represent nutrient cycling in a traditional no-till system. RB samples were dried in a laminar-flow oven at 60°C until constant mass was reached and weighed to determine dry matter yield, which was converted to mass per area (kg ha^-1) basis. Corn grain was harvested mechanically with a small plot combine in the fall from the middle two rows of each 75 cm row spacing plot and the center three rows from each 40, 50, and 57 cm row spacing plots, while soybean grain (Garretson, 2018-2019) was harvested from the center 1.52 m crop row in each plot. All grain yields were calculated on a Mg ha^-1 basis. Grain moisture was used to adjust grain yield to 15.5% moisture for corn and 13% moisture for soybean.

Monthly precipitation (P) (1991-2020) were obtained from the TerraClimate dataset (https://climatetoolbox.org/) (55). This dataset provides data on a monthly basis for global terrestrial surfaces with ~ 4 Km or 2.5 miles (1/24 degree) spatial resolution. Farm-specific weather data would have been useful if the weather stations’ (South Dakota Mesonet) locations (https://climate.sdstate.edu/) were closer to our experimental sites. The weather stations are miles away from our fields, which might not represent actual field conditions. We also checked the Automated Weather Data Network (AWDN) website (https://hprcc.unl.edu/awdn/) for data availability and found there is only one ‘inactive’ station in South Dakota far from our selected locations. Therefore, we used TerraClimate data to deliver more accurate information.

Standardized precipitation-evapotranspiration index (SPEI) data has been collected from WestWide Drought Tracker (WWDT) (https://wrcc.dri.edu/wwdt/about.php), which uses monthly data from PRISM Climate Mapping Program. Another index, namely the Standardized Precipitation Index (SPI), has been considered a meteorological drought index (56) on a range of timescales. This index is widely accepted for the quantification of drought based on precipitation deficit/surplus and is closely related to soil moisture. SPI is available in different timescales (1,3,6, 12, 24, and 48 months) and the values range from +2 to –2 (57, 58). Smaller values indicate drier condition while >0 values are assigned to non-drought conditions. SPEI accounts for climatic water balance and integrates temperature/potential evapotranspiration using the Thornthwaite equation (59) or Penman-Monteith method. The dataset has been collected from TerraClimate (http://www.climatologylab.org/gridmet.html) (60) which has a spatial resolution of ~ 4 Km or 2.5 mile (1/24 degree). The SPEI value always ranges from –5 to +5. The smaller values indicate higher severity of drought whereas greater values indicate a higher degree of moisture or wetter conditions (61).

The values of SPI and SPEI were considered for the month of July each year just before the planting of CCs to indicate the pre-planting hydroclimatic conditions. Our intention was to explore hydroclimatic parameters that are easily accessible and may guide the farmers with information to decide on whether or not they should plant cover crops. All it comes to the fact if they have enough moisture to budget a portion of the water received through precipitation to grow a cover crop. Because, when a farmer needs to decide if they can afford growing a cover crop, they cannot be reliant on the amount of precipitation they might receive precipitation they might receive during the cover crop season or during winter and early spring season next year to be able to grow a ‘profitable’ corn crop. As these farmers practice no-till and can have sufficient surface residue from the previous small grain crop to preserve moisture, they might decide not to grow a cover crop if they did not receive enough precipitation to budget water for a cover crop and save this water for the subsequent cash crop, which might be corn, and growing corn requires a lot of water (62, 63). Otherwise, CCs might adversely affect cash crop yields (15, 23). Therefore, we only considered the hydroclimatic conditions before cover crop planting, because once cover crop is planted, it is already too late for a farmer to decide otherwise. As the water year officially kicks off October each year, the trends of accumulated precipitation and SPEI have been plotted from October (the previous year) to July (before the planting of the CC) for the study sites. Monthly precipitation with the long-term climatology has been plotted ( Figures 3A, B ) using TerraClimate precipitation data for Salem, SD. Hydroclimatic data of similar timelines were used for the other experimental sites, where 'climatology' refers to the long-term (1991-2020) average of monthly precipitation which is considered normal for a thirty-year period, and provides a threshold to identify the months with below or above-normal precipitation.

The effects of CC treatments on all parameters measured were analyzed with RStudio statistical software version 4.0.2 and interpreted using a two-way ANOVA and a linear mixed model for all independent variables (64). Site-year, CC treatments (CC vs. NC), and their interaction was considered a fixed-effect, while block within each site-year was considered a random effect, ‘lmertest’ package was used in R tool for analyses (65). Normality and constant variance assumptions were tested and shown to be met using the Shapiro-Wilk normality test and examining the residuals plots using the ‘ggResidpanel’ package (66, 67). Differences among variables under study as affected by CC treatments and site-year were determined using Fishers Least Significant Difference at P < 0.05 significance level for mean separation using the agricolae package (68) within R statistical software. To decide differences among means, the level of significance was declared at P < 0.05. When variables were tested separately under each site-year, a one-way ANOVA was used, but other considerations remained same. Principal component analysis (PCA) on variables was performed using ‘FactoMineR’ package (69) for which all replicated values were used for all the variables under study. All graphical representation of data were prepared in Microsoft Excel or R 4.0.2.

Cover crops did not improve the selected soil health indicators (SHI) as statistically non-significant (p<0.05) effects were found when data from all site-years under study was analyzed, except for soil nitrate-N ( Table 2 ). Soil and residue biomass (RB) data from spring sampling was used in this analyses as the cash crop season began after this sampling event, therefore, the effect of cover cropping was expected to impact these values the most. High variability in soil properties and agronomic management, existing soil health and climatic conditions are the possible reasons we did not find any statistically significant difference between CC treatments, but for different site-years ( Table 1 ). Furthermore, these farmer-cooperators have been practicing conservation practices like no tillage, cover cropping, suitable crop rotations for several years that helped in building soil health in their farms. Therefore, beneficial impacts of cover cropping on SHIs were absent, which is similar to the scenarios when no effect of fertilization is observed on highly fertile fields (70, 71). The only significant difference in nitrate-N values prove that CC potentially saved labile nitrates-N from environmental losses by incorporating in plant biomass as organic N forms.

Although, the values for the SHIs such PMN, POX-C, and SMB are not significantly different, but 10%, 4%, and 33% higher values were recorded under CC plots compared to no cover plots for PMN, POX-C, and SMB respectively ( Table 2 ). Residual biomass before the cash crop planting was found 9% higher under CC plots compared to the no cover plots, however, we did not find any statistical difference ( Table 2 ). As we sprayed the no cover plots with a broad-spectrum herbicide only during the CC plot establishment, regrowth from previous crops (small grains) and the undecomposed dead plant biomass contributed to the plant residue biomass data under NC treatment. Therefore, under no cover control plots, a combination of volunteer plants (from previous crops) and undecomposed dead plant residues acted as soil cover or soil armor, ‘mimicking’ the CC plots that also had previous crop residues (no-till system) and living plant species, the cover crop species. As the volunteer plant species in the no cover plots are mostly high-biomass producing small grain crops, wheat or oats, therefore most no cover plots had similar amounts of total plant biomass (previous crop residue plus growing plant residue) and hence, provided with similar benefits of CC. Also, we noted that no-till systems leave a significant amount of crop residues, and when plots were laid out randomized, the total biomass therefore often were similar for no cover and CC plots.

Comparable soil health values under CCs and no cover control plots are not rare; many previous experiments reported similar observations (23). To investigate further, we analyzed the site-years separately, but did not find statistically significant difference between CC plots and NC plots for selected SHIs, residue biomass, or cash crop yield ( Table 3 ). Therefore, our data indicates that CC did not provide any statistically significant (p<0.05) soil health or cash crop yield benefit in the site-years investigated, which supports previous findings where no yield benefit was reported (72–74). Some studies even mentioned yield reduction when cover cropping was practiced (23, 75). This finding brought up essential questions against CC as a tool to improve soil health and in providing economic benefits from increased cash crop productivity in the following season. Furthermore, reports of cash crop yield drags and economic losses indicates concerns around CC as a ‘sustainable’ practice (72, 75).

The principal component analyses (PCA) for selected soil variables and soil health indicators along with corn yield revealed that SOM status of the field is the major factor affecting crop yields, while the other SHIs do not majorly influence the crop productivity ( Figure 4A ). One probable reason maybe existing soil health status in the study sites as depicted by high SOM status in the Figure 4B . Due to existing conservation management practices adopted by the farmer-cooperators, the mean SOM levels of the research sites were already high (> 28 g kg^-1; highest value 50 g kg^-1) and these sites can be considered to have ‘healthy soils’. This may be the reason the impact of cover cropping was not significant in these sites.

When plotted against cash crop yield for all site-years, SOM found to be well correlated with the cash crop yields ( Figure 4B ). After further investigations, we found that the SOM levels in the study sites showed different trends when categorized year-wise ( Figures 4C, D ). Cash crop yields followed SOM trends in the study sites during 2019 and 2020 cash crop season, but not during the 2017 season. Furthermore, the SOM levels in all the study sites were high as indicated earlier, therefore, we can speculate that more than soil health status, hydroclimatology of those site-years might be influencing the crop yields following CCs. This finding raised the question if CCs are not contributing to soil health benefits or yield benefits ( Tables 2 , 3 ), should it still be considered as economically feasible? Especially, if we consider the Beresford1718 and Henry1920 site-years, where CCs actually reduced cash crop yield ( Figures 3C, D ). Therefore, it was necessary to study hydro climatological data for these site-years.

For a rainfed system, climate regulates major yield attributes, especially the hydroclimate as the crop is completely dependent on climate for its water requirement throughout the season. Therefore, we have explored a few indices that indicate the availability of water for the cover crop, and the subsequent cash crop as they share the available water. Figure 3A represents a long-term (1981-2010) average of precipitation received in different parts of the state of SD. The state has shown an increasing trend of precipitation from west to east. From Figure 3B it is evident that the 2017-18 season was drier than, 2018-19 and 2019-20 seasons, but it was hard to predict that before CC planting in the previous season. So, we investigated other hydroclimatological parameters: (i) accumulated precipitation during 10 months before CC planting i.e., October through July, (ii) standardized precipitation index or SPI, and (iii) standardized precipitation-evapotranspiration index or SPEI for the month of July as CCs are generally planted in August.

In this study, six different locations (around South Dakota) were identified based on cropping season and yield data availability. Firstly, the trends of accumulated precipitation, SPI, and SPEI were compared against yield advantage under CC (the difference in cash crop yields between CC plots and NC control plots) for the only site we could investigate for three consecutive seasons, Salem, SD ( Figures 5A–C ); high correlations were found for all the three selected hydroclimatological parameters for all three years. This data provides a unique opportunity to investigate further ( Figures 5A–C ). Therefore, the trends of hydroclimatic parameters, accumulated precipitation, a precipitation index, and a drought index (SPEI in July) were plotted ( Figures 6A, B ) together to provide an overview of the weather variability 'at the selected locations. According to the trend, Beresford, Salem, and Gettysburg have received lower precipitation (Oct-July) during pre-planting season (2016-17) along with negative SPEI values ( Figures 6A, B ). Whereas Blunt, Henry, and Salem received higher precipitation during the pre-planting season (2018-19) along with positive values of SPEI. Garretson and Salem during 2017-18 received higher amount of precipitation and SPEI values than 2016-17 season but less than 2018-19. The variability of these three seasons has indicated three distinct hydroclimate periods i.e., dry, neutral, and wet; positive values of SPEI indicate wetter condition whereas negative values indicate drier condition (76, 77).

To explore further, we plotted yield advantage under CC against location-wise accumulated precipitation (Oct-Jul), SPI, and SPEI values ( Figures 7A–C ). Polynomial regression models were calculated for accumulated precipitation and the selected hydroclimatic indicators (accumulated precipitation, SPI, and SPEI); linear regression models were not found statistically significant. SPEI values were found better correlated (coefficient of determination, R² = 0.53) with CC yield advantage in these locations compared to accumulated precipitation (coefficient of determination, R² = 0.30) or SPI (coefficient of determination, R² = 0.14). A better correlation was found with SPEI as this index is calculated using both precipitation and evapotranspiration data, which are critical hydroclimatological parameters, and possibly, majorly contribute to crop growth and yield. It was necessary to find critical levels for SPEI that might guide a farmer to decide on cover cropping in the fall. Solving the model, we found that SPEI values below -0.21 and above 1.62 might provide yield advantage, but if SPEI value for any location is in between these range (-0.21 to 1.62) then the farmer might not get any yield benefit using CCs. Therefore, the farmer may chose not to plant CCs if the subsequent crop is corn ( Figure 6 ). We speculate that when more moisture is available under a very wet year then cover crops can provide a yield advantage just by improving nutrient cycling and soil health, and also by removing excess moisture so that the cash crop can be planted at the ideal time. Contrastingly, when the soil moisture level is very low then cover crops can conserve moisture providing a ‘live mulch’ against evaporation loss (78). Under low moisture availability, cover crops may not produce a lot of biomass that ties up a lot of nutrients. This information may also guide selection of cover crop species under different hydroclimatic scenarios (79). For example, when plenty of moisture is available before cover crop planting, a farmer might choose to select cover crops that produce large biomass. But if the hydroclimatological indices indicate drought conditions, a farmer might decide to use drought resilient, and ‘specific management-targeted’ cover crops, such as radishes to break compaction (80), legumes to fix nitrogen (81), brassicas to irradicate soil-borne-pathogens (82), etc. A recent article from the same region reported that precipitation was the key factor influencing yield and not CC composition or cover cropping (15); this relationship between cash crop yields and SPEI reaffirms that hydroclimatological parameters might be more efficient in deciding crop rotations when cover cropping is included. Mohammed et al. (83) also indicated SPEI as a predictor for corn and wheat yield across Hungary. As we have not seen any soil health benefits in these study sites where farmer-cooperators have already built ‘good soil health’, practicing no-till, sustainable crop rotations, and cover cropping in the earlier seasons, cover cropping might be a ‘potential financial decision’ for a farmer if yield benefits are not expected. Additionally, Figure 8 indicates that higher cover crop biomass negatively influenced cash crop yields. We can speculate that higher amounts of nutrient (especially nitrogen) were accumulated in the cover crop biomass and more moisture was ‘stolen’ from the cash crop when cover crop biomass was higher. This outcome echoed the concerns of many farmers growing cover crops under rainfed system and experiencing cash crop yield loss. Hence, it might be beneficial to use hydroclimatological information before going into another CC season. Overall, our findings indicated that a multiscalar index like SPEI can guide better CC decisions and should be investigated further to create a robust decision-support tool for cover cropping, especially in the rainfed cropping systems.