The experiment was conducted in six locations: Salina, Kansas (KS-TLI; The Land Institute); Mead, Nebraska (NE; University of Nebraska-Lincoln); Rosemount, Minnesota (MN-ROS; University of Minnesota-Twin Cities); Arlington, Wisconsin (WI-UW; University of Wisconsin-Madison); Troy Center, Wisconsin (WI-MFAI; Michael Fields Agricultural Institute), and Wooster, Ohio (OH; The Ohio State University) ( Figure 1 ; Table 1 ). The soil orders presented across these locations are Mollisols and Alfisols (Soil Survey Staff, 2023; Table 1 ). The climates are classified as follows (Kottek et al., 2007): NE and OH have hot and humid continental summers (Dfa); MN-ROS, WI-UW, and WI-MFAI have mild and humid continental summers (Dfb); KS-TLI is between a hot humid subtropical summer (Cfa) and a hot humid continental summer (Dfa).

All locations were planted to oats (Avena sativa) in spring 2021 and harvested in July of that year before establishing experimental treatments. Previous management at each location had been a continuous annual crop rotation typical of the region of each location (a corn-soybean rotation WI-UW, WI-MFAI, OH, MN-ROS, and NE, and a soybean-sorghum rotation at KS-TLI). We established the same random block design with twelve treatments ( Table 2 ) replicated once per each of four blocks. We provide a detailed description of the site characteristics and agronomic management in Supplementary Tables S1 and S2 . Plots were approximately 15.24 m x 6 m at each location. We collected baseline soil samples in a subset of the treatments referred to as the ‘soil health treatments’ with a hydraulic soil probe (5.08 cm diameter and 0-20, 20-50, and 50–90 cm deep) in the summer of 2021 following oat harvest as described below. There were 12 treatments total, including 10 IWG planted treatments with various fertilization strategies described below, and two “business as usual” (BAU) treatments that tested annual crops chosen to reflect typical crops of the given region. These BAU treatments were a soybean-corn rotation with both phases (labeled BAU1 and BAU2) included at most locations and a sorghum-soybean rotation at KS-TLI only ( Table 2 , Supplementary Table S2 ). Data from BAU treatments is not included in the analysis presented in this paper, but will be compared with IWG responses in the coming years. Prior to planting IWG, we applied fertilizer appropriate to each treatment ( Table 2 ) and prepared the seedbed by tillage ( Supplementary Table S2 ). Fertilizer rates were optimized so that at least one treatment could both meet the nutritional needs of IWG and to create a useful comparison between the IWG and BAU crops, as well as balance the amount of fertilizer all crops received over two-year periods (Jungers et al., 2017, Table 2 ). N application rates in this study were selected to encompass the range of applications currently recommended as well as higher rates to assess potential N-limitation at current recommended rates with the newest released variety, MN Clearwater (Jungers et al., 2017; Bajgain et al., 2020). N rate treatments included 0, 45, 90, 135 and 180 kg N ha^-1 applied as urea in the spring. All N addition treatments also included 56 kg P ha^-1 and 168 kg K ha^-1, and two additional fertility treatments tested the 90 kg N ha^-1 rate either without the added P or without the added K. P and K fertilizer rates were selected to approximate crop removal rates through harvest based on unpublished Kernza grain and forage data, as well as published corn and soybean removal data (e.g., Culman et al., 2020; Table 2 ). The K rate likely exceeded the BAU removal rates, but is within range of what would be considered typical farmer practice in Midwest corn and soybean cropping systems.

In addition to examining N rates, we examined IWG responses to N source and N timing. The N source treatment assigned 180 kg N ha^-1 added as chicken manure, with the same product used across all experimental sites. N timing treatments were implemented by applying N (at 90 kg N ha^-1) in the fall or as a spring-fall split (45 kg N ha^-1 at each application). Tillage and seedbed preparation occurred as described in Supplementary Table S2 throughout the late summer of 2021 in all plots. In September 2021 the IWG variety ‘MN-Clearwater’ was planted at a density of approximately 7.8 (MN-ROS, TLI), 12 (NE, WI-UW), or 16 (MFAI, OH) pure live seeds ha^-1, with either 30 (MN-ROS, KS-TLI) or 38 (OH, NE, WI-MFAI, WI-UW) cm between rows, depending on available equipment. We fertilized relevant IWG treatments in April-May 2022 as described in Table 2 . Urea fertilizer was mixed with a urease inhibitor (ANVOL^®) before application, at a rate of 1.5 quarts per ton of urea, per product recommendations (Koch Agronomic Services, LLC). Grain and summer forage harvest took place in August 2022. Fall fertilizer for relevant IWG treatments took place in September 2022. Fall forage harvest for IWG took place after the first frost In October 2022.

Baseline soil samples were collected from a subset of treatments including 1, 3, 5, 7, 8, and 10 ( Table 2 ). This subset is hereafter referred to as the ‘soil health plots’. Multiple soil cores were randomly extracted from each plot to control for field heterogeneity and divided into the following depth increments: 0-20, 20-50, and 50–90 cm. Samples within each plot were combined by depth into one composite sample per plot (~450g), sieved to 8 mm, and stored appropriately for each subsequent analysis. All soil analysis followed standardized methods ( Supplementary Table S3 ). Briefly, samples were refrigerated at 4˚ C until sieved to 2 mm and weighed and subsampled for gravimetric soil moisture (~10–20 g subsample dried to 105˚ C for 48 hrs). Bulk density was estimated by dividing the dry weight equivalent of each composite sample core by the volume of the core. Soil texture analysis was conducted on subsamples air dried (~100 g) using the hydrometer method (Carter and Gregorich, 2008). Subsamples were analyzed for soil organic matter (SOM), pH, cation exchange capacity, ammonium-N, nitrate-N, available P, exchangeable K, at A&L Great Lakes Laboratory (Ft. Wayne, IN) using standard procedures (Brown, 1998). Total soil carbon and total N were analyzed by combustion analysis using (varioPYRO cube, Elementer Inc., New Jersey, US). For aggregate stability analysis, we collected three bulk soil samples with a 2.54 cm wide (0–20 cm deep) soil probe which were then combined into one composite ~100g sample per plot. Soil was then sieved to 8 mm and placed in a 4 mm sieve at the top of a stack containing 4 mm, 2 mm, 0.250 mm, and 0.053 mm sieves. The aggregates were first wetted with capillary action by setting the sieves just at water level, followed by oscillation for 10 min in a water tank using a mechanical sieving device ((Angers et al., 2008); with a modification that no sand correction was performed). After drying and weighing the soil remaining in each sieve, we calculated the mean weight diameter (MWD) of water stable aggregates (Rakkar et al., 2023). Soil subsamples were analyzed for nitrate and ammonium using extractions with 1M KCl and subsequent colorimetric analysis (Doane and Horwath, 2003; Sinsabaugh et al., 2000).

To assess establishment, we conducted plant counts within three weeks after IWG planting by counting seedlings within a 50 cm section in two rows of each plot. We assessed IWG growth stages using Zadoks growth at stem elongation, anthesis, and harvest (Zadoks et al., 1974). Lodging was assessed following procedures described by Frahm et al. (2018) at harvest and was not a concern in 2022. We measured aboveground production (grain and summer forage yield) by harvesting 1 m sections from three adjacent rows in two representative areas of each plot (see quadrats size for each location in ST2). IWG grain yield was measured by first clipping the seedheads from stems just below the bottom spikelet. Seedheads from both sampled areas were placed into the same bag and then all remaining vegetative biomass above 7.6 cm from both sampled areas was harvested, combined, and placed in a separate bag. The remaining vegetative biomass is referred to as summer forage. In all field sample collection and measurements, the outermost plot rows and 50 cm border of the plot edges were excluded from sampling to avoid edge effects. Both seedhead and summer forage biomass were weighed fresh before being dried for 7 days at 35°C and re-weighed again for determining moisture content and dry matter yield. Grain was threshed from seed heads using a laboratory thresher (Wintersteiger LD 350, Ried im Innkreis, Austria) followed by sieving and aspirating to fully separate the grain from the chaff. All plots were mechanically combine-harvested after the manual harvest such that both grain and remaining straw were cut and removed with only approximately 7.6 cm of residue remaining on the plots. Fall forage samples were collected from treatments 3, 5, and 7 and processed using the same methods described above for the summer forage harvest. Yields of all crops were expressed in kg ha^-1 by multiplying harvested kg per m^-1 by row spacing (m^-1), then multiplying by 10000 (m² ha^-1).

We compiled a suite of climate variables from 1979–2022 for each site so that we could characterize the 2022 data of this study within the context of historical trends. Actual evapotranspiration (Eta), precipitation (Prcp), net solar radiation, and temperature were derived by the European Centre for Medium-Range Weather Forecasts (ECMWF) and Reanalysis v5 (ERA5) (Hersbach et al., 2020; Supplementary Tables S4, S5 , Supplementary Tables metadata). ECMWF consists of a dynamically consistent combination of numerical weather prediction forecasts, satellite, and in-situ observations of climatic variables. Data for 2021-2022, as the more recent trends for each location, were downloaded from National Centers for Environmental Information (National Oceanic and Atmospheric Administration (NOAA), 2023), the CFAES Weather System (Ohio State University, 2013), and Davis weather station at TLI ( Supplementary Tables S6, S7 ).

To assess the correlation of climate factors with IWG grain and biomass yields, we selected a number of variables that would balance comprehensiveness and minimization of covariance. The selected variables included precipitation (Pp; mm, the sum of rainfall, snow and hail), actual evapotranspiration (Eta; mm), temperature (˚C, annual, monthly and daily averages), frequency occurrence Prcp (25%, 50% and 75%; Irigoin, 2011) and evaporative fraction (Ef). Ef is defined as the ratio of latent heat flux to the available energy, i.e. how much of the radiation received can be used for Eta. Ef is controlled by water in the root zone (Bastiaanssen et al., 1997).

In addition, the cumulative actual evapotranspiration from May to October was calculated as an estimate of evapotranspiration over the crop growing season (EtaCrop). We also calculated cumulative Prcp over several different time intervals associated with various critical physiological periods to determine if any were correlated with yield. The cumulative growing season precipitation (PrcpCrop) was calculated as the sum Prcp from May to October 2022 (de Oliveira et al., 2018). Cumulative precipitation was also calculated for the 60 days before anthesis (Prcp60Anthesis, Figure 2 ), for the 20 days before + 10 days after anthesis (Prcp20Anthesis10; the critical period of the wheat based on Fischer, 1985), for the entire calendar year before anthesis (PrcpAnthesis) (beginning January 1^st and ending on the date of observed anthesis at each location), and for the entire calendar year (PrcpAnnual). The mean temperature of crop establishment through harvest (TempCrop) was calculated as well (September 2021 to August 2022 for grain and summer forage; September 2021 to October 2022 for fall forage).

Additionally, growing degree days (GDD) were calculated according to Jungers et al. (2018) and Barribal et al. (2022) (Equation 1), where Tmax and Tmin are the maximum and minimum temperatures and Tbase (0 ˚C) is the minimum temperature threshold for IWG plant growth (Jungers et al., 2018; Barribal et al., 2022). Growing degree day accumulation began after the average daily temperature exceeded the base temperature for 5 consecutive days (Jungers et al., 2018) after January 1^st 2022 (Barriball et al., 2022).

To assess how fertility treatment, climate and soil factors could be related to differences in IWG establishment and growth across locations, we used several data analysis approaches. We analyzed data using R Software (version 2023.06.1; R Core Team, 2021) and tools in the dplyr package to structure data frames appropriately (Wickham et al., 2021). For all analyses, we first assessed whether measured soil and IWG variables conformed to model assumptions of homogeneity of variance and normality, and applied a square root transformation as needed. We tested for the fixed effect of location on IWG plant counts and baseline soil parameters with blocks specified as random variables, since there were no fertility treatments implemented for these response variables. Linear mixed effects models were used to test the effects of the fertility treatments and locations for IWG grain and forage yield. We used both one-way ANOVA and mixed effects models using the following R packages: stats (R Core Team, 2021), scales (Wickham and Seidel, 2020), and lme4 (Bates et al., 2015). For mixed effects models, we specified fixed effects for treatment, location, and the interaction of treatment and location and assigned block as a random effect. When fixed effects were deemed significant (α ≤ 0.05), mean comparisons of levels within each fixed effect were conducted using Tukey’s HSD posthoc pairwise comparison with the emmeans package (Lenth, 2024).

To compare the potential productivity between locations, we estimated the crop productivity index (CPI) for each location using the National Commodity Crop Productivity Index (NCCPI) model developed by the NRCS-USDA. The CPI ranks soils according to their inherent capacity to produce dryland commodity crops. Ratings are based on soil properties and climate, and range from 0 to 100, with higher ratings indicating greater yield potential (Natural Resources Conservation Service (NRCS), 2022; Supplementary Figure S1 ).

Separate multiple linear regression models were developed to estimate the predictive value of soil baseline (0–90 cm depth average) and climate (2021–2022 series) data on grain, summer forage, or fall forage yields. Before constructing and fitting the models, Pearson correlation analysis (R Core Team, 2021) was performed on all possible predictors, and variables with a correlation of 0.6 or greater were considered significantly correlated and thus excluded from consideration ( Supplementary Table S8 ). Predictors were also centered by subtracting the mean. Based on this preliminary analysis, we used the following predictors in each model: SOM, CEC, P, K, Mg, Na, S, the MWD of water stable soil aggregates; sand; Prcp60Anthesis; PrcpAnthesis, and EtaCrop. For the fall forage model, we used PrcpCrop as the climate predictor instead of Prcp60Anthesis or PrcpAnthesis. We then used the dredge function of MuMln package (Barton, 2013) for each of the three full models to explore all possible combinations of predictors and selected the model with the lowest AIC score. After a first run of the models, outliers were identified by calculating the standardized residuals for the models, identifying and deleting observations where the absolute value of the standardized residual was greater than 1.5. The variance inflation factor (VIF) analysis was performed to identify multicolinearity between predictors (Fox and Weisberg, 2019). Predictors with VIF values greater than 5 were deleted for fitting the models. Lastly, we performed an ANOVA, and tested the assumptions for linearity, independence, homoscedasticity, normality and non-collinearity to each estimated model using the lmtest package (R Core Team, 2021).

Once the full models were finalized, we created bivariate linear regression models for each remaining predictor to better understand how each factor alone predicted variability in IWG grain yields, summer forage yields, and fall forage yields.

Over the past 40 years (1979-2022), the month with the lowest average temperature (Feb) and highest average temperature (Aug) were consistent across locations; while the lowest mean temperatures (-0.16 C) occurred at MN-ROS and the highest (25.3 ˚C) at KS-TLI ( Table 1 ).

The 1979–2022 the minimum precipitation (Prcp) occurred in January or February at all locations. Maximum Prcp occurred in spring for KS-TLI, NE, and OH, and spring-summer for WI-MFAI, WI-UW and MN-ROS ( Figure 3 ). The maximum actual evapotranspiration (Eta) occurred during late spring and summer in all locations. Precipitation varied seasonally for all locations, as was the case for Eta at KS-TLI, NE and MN-ROS ( Supplementary Figures S2, S3 ). The evaporative fraction (Ef) was highest during late spring and summer months, and NE, MN and WI-UW had the highest values, while WI-MFAI and OH had the lowest ( Supplementary Figure S4 ).

Temperature trends during our study period of 2021–2022 were similar to the long-term averages from 1979-2022, with the minimum occurring in February and maximum in August. MN-ROS was an exception with maximum temperatures in June ( Figure 3 ). The years 2021 and 2022 had notably less precipitation than 1979–2022 means at all locations (by 188 and 153 mm in KS-TLI; 301 and 517 mm in NE, 186 and 198 mm in MN-ROS; 341 and 19 mm in WI-UW; 314 and 71 mm in WI-MFAI; 274 and 69 mm in OH, lower annual and accumulated Prcp, respectively; Figures 2 , 3 ). Annual Prcp at NE was exceptionally low relative to its long-term average and was the lowest of all locations (367 mm). May and June 2022 were the only months at KS-TLI and NE that were not drier than average during our study. In this two-month window, Prcp in KS-TLI was higher (by 142 (May) and 79 (June) mm) than the 1979–2022 averages. Higher than average (by 73 mm) Prcp also occurred in MN-ROS in August 2022 ( Figure 3 ; Table 1 , Supplementary Table S7 ).

Soil parameters were statistically different across locations and with depth ( Table 3 , Supplementary Tables S9, S10 ). These location differences were reflected in the crop productivity index ( Supplementary Figure S1 ). Soil texture analysis showed that all locations had relatively fine soil, ranging from loam to silty clay loam or silt loam 0–20 cm. At 50–90 cm, the texture was coarser, sandy clay loam in WI-MFAI, and loam for MN-ROS and OH. Soil organic matter (SOM) percentage varied across depths and locations with SOM being lower at depth at a given location. The highest SOM was reported at MN-ROS at 56 g kg^-1, at 0–20 cm, and the lowest values being reported at OH and WI-MFAI at 11.5g kg^-1 at the 50–90 cm depth.

Establishment success, measured by plant counts after germination, varied across locations (p < 0.001). KS-TLI had the lowest plant count at an average of 5 ± 4 plants m^-1 and OH had the highest, at an average of 23 ± 3.0 plants m^-1 ( Table 4 , Supplementary Table S11 ).

IWG reached the anthesis growth stage at a similar timing at the MN-ROS, WI-MFAI, and WI-UW sites both when measured by DOY (day 181, 186 and 186 respectively) and by GDD (1249°Cd, 1391°Cd, 1378°Cd). KS-TLI and NE also reached anthesis at a similar DOY (164 and 174 respectively) and GDD (1536°Cd and 1422°Cd respectively). OH had the earliest DOY of anthesis (day 161) and the least GDD 1153°Cd to anthesis across the locations ( Figures 2A, B ).

IWG yields varied significantly by location (p < 0.001). MN-ROS had the lowest mean grain yields at 556 ± 34 kg ha^-1 and WI-UW had the highest mean grain yields at 1343 ± 24 kg ha^-1 for IWG ( Figure 4 , Table 4 , Supplementary Table S11 ). Summer forage mean yields varied across locations with the lowest being at WI-MFAI at 2054 ± 124 kg ha^-1 and the highest at WI-UW at 8930 ± 159 kg ha^-1.

Various combinations of soil nutrient concentrations, percent sand, aggregate stability, and climate variables including Prcp60Anthesis, EtaCrop, and PrcpCrop explained 74%, 92% and 69% of the variability in grain (Equation 2), summer forage (Equation 3) and fall forage (Equation 4) yields, respectively, across locations (p<0.001) ( Tables 5 , 6 ; Supplementary Table S12 ). Variables that best explained each harvest product varied, though all included some measure of precipitation ( Table 6 ). The final models were as follows:

After model construction, we performed individual regressions for those variables included in the final models ( Table 7 ). Based on this analysis, we found that Prcp60Anthesis (accumulated precipitation 60 days before anthesis) had the greatest influence on grain and summer forage yields and alone explained 21% and 45% of the yield variability in the individual regressions respectively ( Table 7 ). By contrast, PrcpCrop (accumulated precipitation from May to October) alone explained 2% of the variability in fall forage ( Table 7 ). Alone, EtaCrop (accumulated actual evapotranspiration from May to October) alone also had an important influence on summer forage and explained 20% of the variability based on individual regressions ( Table 7 ).

Soil variables tended to explain less of the variation in yields in individual regressions than precipitation with the exception of fall forage. Mg, Na and P explained 28% in grain, 9% summer forage and 47% fall forage when individual regressions were summed ( Table 7 ). K explained 7% in grain and 30% fall forage yields models ( Table 7 ). Agg Stability (soil water stable aggregate stability) explained 15% and 5% in summer and fall forage model in individual regressions ( Table 7 ). Sand explained 32% of grain yields in individual regressions ( Table 7 ). A difference between the individual regressions and the full multiple linear regression models was that although the full multiple linear regressions showed positive relationships with each response variable, some individual regressions demonstrated negative relationships with the response variables ( Tables 6 , 7 ).

Neither fertility treatment alone nor the interaction of treatment by location influenced grain yields (p > 0.05). Treatment and location both influenced IWG summer forage yield (p<0.05 and p<0.001 respectively; Figure 5 , Table 4 ). Among treatments, only the addition of 135 kg ha^-1 spring applied mineral N increased summer forage yields compared to the no N control (p<0.05; Figure 5 , Table 4 ). Fall forage yields were affected by the interaction of treatment and location (p < 0.01; Figure 6 , Table 4 ), where treatments with 90 and 180 kg N ha^-1 of spring mineral applied N had mean yields significantly greater than when no N was applied, only at the NE location or when fall forage was averaged across locations (p<0.05; Figure 6 , Table 4 , Supplementary Table S11 ).

Establishment of IWG varied significantly across locations with OH having more than double, and KS-TLI with less than half the number of plants per length of planted row than the other locations (MN-ROS, WI-UW, WI-MFAI, and NE ( Table 4 ). The low establishment observed at KS-TLI could have been associated with low precipitation at that location during the months around seeding ( Figure 3 ). Despite variability across locations, plant counts did not predict grain or biomass yields. IWG spreads via tillers and rhizomes that can take up space between plants within sown rows. This ability to vegetatively propagate is an establishment strategy in the first year of growth (Woeltjen et al., 2024; Shoenberger, 2022). Because of this, initial plant counts in the range we observed (5.3–23 plants per m¹) are expected to be able to produce a robust stand in the first year given other growth conditions and agronomic variables such as weed pressure. Research is needed to determine threshold plant populations for IWG after establishment and the first winter period to inform growers on when stands should be terminated as a result of poor establishment. Such information is available for overwintering crops like winter wheat and canola, for example (Paulsen, 1987; Assefa et al., 2018).

First year IWG grain yields (ranging from 586–1343 kg ha^-1) were similar to and sometimes higher than year 1 yields reported in other studies (ranging from 494–1074 kg ha^-1; see Supplementary Table S1 for a summary of publications reporting grain yields). WI-UW and KS-TLI yields were ~30% and ~20 – 30% higher than those described in the literature for Wisconsin and Kansas respectively (Bowden, 2023; Culman et al., 2023; Pinto et al., 2022).

At the same time, MN-ROS yields were ~20 – 25% lower than those reported in other literature from Minnesota, WI-MFAI yields were 35% lower than other reports from Wisconsin, and OH yields were similar to previous findings from Ohio (Bowden, 2023; Culman et al., 2023; Pinto et al., 2022; Bajgain et al., 2020; Fernandez et al., 2020). The highest IWG grain yields are typically observed during the first production year. Although some studies have reported comparable or higher yields in second-year stands (Culman et al., 2013; Bowden, 2023), a decline over time is more commonly observed when IWG is well established in the first year (Culman et al., 2023; Pinto et al., 2024; Bajgain et al., 2020; Mårtensson et al., 2022; Tautges et al., 2018; Zhen et al., 2024). Also, as discussed below, precipitation in the 60 day period before anthesis and local soil characteristics could be related to variance in grain yields across locations. Based on these results, expectations of yields should be established on a regional basis with the knowledge there is likely significant local variation in IWG performance depending on precipitation and establishment in the first growing year.

IWG summer forage yields also differed across locations. Similar to the trend with grain yields, WI-UW summer forage (8930 kg ha^-1) was ~30- 50% higher than that previously reported for Wisconsin (Pinto et al., 2024; Culman et al., 2023) (see Supplementary Table S1 for a summary of publications reporting summer forage yields), and summer forage values at MN-ROS were ~20 – 75% lower than those previously reported for Minnesota (Bowden, 2023; Culman et al., 2023). KS-TLI summer forage yields were roughly the same as those previously reported in Kansas for the MN-Clearwater variety (Bowden, 2023) but ~20% higher than those previously reported for C4 TLI variety (Culman et al., 2023). This makes sense in that higher productivity is a major breeding goal (Bajgain et al., 2020). Although fall forage yields were lower than summer forage yields ( Table S11 ), they were not as variable as summer forage yield by location, with the exception of NE, which had ~50 – 75% higher yield than the other sites. Together, the variance we and others have observed in grain and summer forage yields indicates a need to better understand what could give rise to this large range in growth from one year or location to the next. Fall forage has not been as extensively studied across our study locations, but our range of fall forage biomass, when averaged across locations, were similar to that reported previously (Pinto et al., 2024; Culman et al., 2023), with the average at NE approximately two-three times higher. Below, we discuss how climate and soil variables were also related to this variance in biomass yields across locations.

One objective of our study was to determine the extent to which local climate and soil conditions explained variance of first year grain and forage yields. The context within which we performed this analysis with a first growing year of 2022 was that at our experimental locations, the temperature and precipitation in 2021 and 2022 were respectively colder and drier than the 1979–2022 means for all locations except OH, which had temperature similar to the historic average ( Table 1 ). The drier conditions observed over this region are in line with the climate change patterns observed since 1900 across the Midwest (Yu et al., 2018). Although precipitation levels in the eastern Midwest were below average, they were still higher than the western Midwest region in 2022 as is typical (Marvel et al., 2023). Actual evapotranspiration (Eta) also varied across locations ( Table 1 , Figure 3 ). The relationship between Eta and precipitation determines the level of agricultural drought (Wilhite and Glantz, 1985). KS-TLI and NE are often at risk of drought because they have higher probability of seasonal crop moisture deficit (Wilhelmi and Wilhite, 2002; Logan et al., 2010), more so than the other locations in our study (Yu et al., 2018) ( Supplementary Table S5 , Supplementary Figure S3 ).

Of the climate variables identified after model selection, cumulative precipitation 60 days before anthesis (Prpc60Anthesis) was the strongest predictor of grain and summer forage yields ( Table 6 ). According to our multiple linear regression models, for every 1 mm increase in Prcp60Anthesis, we would expect 4.7 ± 0.5 and 50.9 ± 2.6 kg ha^-1 grain and forage increase respectively when other environmental conditions are held constant (Equations 2, 3).

In previous studies, temperature was more strongly related to grain and summer forage yields than precipitation (Culman et al., 2023), with precipitation explaining 16% and 6% of variability in grain and summer forage yields respectively. In this study, narrowing the window of precipitation to relate to yield may have strengthened this relationship as the 60 days before anthesis explained 21% to 45% of the variability of grain and summer forage yields ( Table 7 ). The importance of this period of precipitation before anthesis is demonstrated by some of the observed trends. For example, while the MN-ROS location had a similar amount of annual cumulative precipitation as several other locations in 2021 and 2022, it had the lowest yields in the study perhaps because it also had 47% lower precipitation in this window compared to the average across locations. This may also explain why the WI-MFAI had significantly lower yields than the WI-UW location despite similar historical climate; the WI-MFAI precipitation before anthesis was 35% below average across locations while WI-UW was 12% above. Locations with 189–195 mm precipitation in the 60 days before anthesis had typical or higher than expected yields for the region, whereas locations with lower (MN-ROS, WI-MFAI) and higher (OH) precipitation in this period had significantly lower yields than expected for the region ( Supplementary Table S1 ). We also observed anomalies in these trends. For example, despite having the lowest annual precipitation in 2021 and 2022 and high Eta ( Table 1 ), grain yields from the NE location were among the highest in this study.

Narrowing the window of precipitation analyzed with grain yield resulted in greater correlation with grain yields and may be a meaningful starting point for identifying a ‘critical period’ for IWG. Common annual crops have established critical periods around anthesis when any water, temperature, or light stress influences the number and size of grain (Slafer and Andrade, 1993; Andrade et al., 1999). For example, the critical period for wheat is 20 days before and 10 days after anthesis (Fischer, 1985), while for corn, it is 15 days before and after anthesis (Andrade et al., 1999). Our results point to the potential need to investigate a much longer precipitation window (189–195 mm in the 60 days before anthesis) than for annual crops to infer potential weather impacts to grain and summer forage yields.

The analyses in this study on climate impacts are limited by having a single data point for precipitation and temperature within each location despite wide variation in yield both within and among locations. This could have resulted in the nonsignificant or negative relationships between some climate factors and grain and fall forage yield that seem contradictory to our ecological understanding of plant growth during moisture limitation ( Table 7 ). Thus, these results are not conclusive about the relationship between precipitation and IWG yield or the critical period for IWG. They do however provide preliminary data from which future determination of the critical period for IWG could be determined. Our results also demonstrate that many factors together influence yield across and within a given climate and that multi-site modeling efforts, with multidisciplinary datasets, are needed to synthesize and elucidate these patterns.

Soil characteristics explained as much or more variance in IWG grain and forage yield data than climate variables in multiple regression models ( Tables 5 , 6 ). Soil texture was the most influential variable in explaining grain yields, where each increase in percent sand resulted in a decline of 16.7 ± 2.5 kg grain ha^-1. Soil texture strongly influences soil water holding capacity, water retention, and provision of nutrients important to plant growth, and the ability of the soil to retain or form soil organic matter (Crews and Rumsey, 2018; Crews et al., 2018; Sprunger et al., 2018). Aggregate stability had a surprisingly negative but consistent relationship with forage yield. NE had lowest aggregate stability but nearly the highest yields, where MN-ROS was the opposite with the highest aggregate stability and lower yields. This could be an inadvertent correlation with baseline soil characteristics. We expect soil aggregate stability to increase with increasing IWG stand age (Rakkar et al., 2023; Chamberlain et al., 2022; Wuest et al., 2006), just as we expect improvements in SOM and infiltration (Culman et al., 2013). Potassium (K), sodium (Na) and magnesium (Mg) also consistently explained variance in our yield data ( Tables 5 , 6 ). Very few studies have investigated the IWG requirements of macro or micro-nutrients other than nitrogen but there is evidence that IWG total biomass, grain yield, or yield components can be explained by soil elemental composition (up to 35% of variance for grain yield, 39% for total biomass, 79% for the number of grain producing stems; Muhandiram, 2023). This suggests that although we did not find that other macro or micronutrients explained variance in grain or biomass yields in this study, they could be important for IWG fertility at certain locations.

Among all fertilization treatments, the only significant responses we observed were with summer forage when averaged across locations, and fall forage yields when averaged across locations or at NE, with no response from grain yields ( Table 4 ). The lack of grain yield response to fertilizer is common in first year IWG production (Jungers et al., 2018; Reilly et al., 2022b; Pinto et al., 2022), but because these studies all used older varieties of IWG with lower yield potential than MN Clearwater, we had anticipated that we might see stronger year 1 responses. At the same time, a meta-analysis on IWG grain and biomass yields in response to N fertilizer rates showed that grain yields were not affected by increasing N rates in the first year, but a positive effect of N rates on grain yields increased as stands aged (Bianchin Rebesquini, 2023). IWG’s large and deep root system that can access nutrients deep in the soil profile (Ryan et al., 2018) could be related to the lack of fertilizer response. Endogenous N mineralization from SOM induced by tillage before planting may help explain the lack of N limitation in year one. Between 0.9 and ~10 mg N kg⁻¹ soil has been adequate for fulfilling IWG (Dobbratz et al., 2023) and, more generally, perennial grass demand (Vogel et al., 2002). Locations in this study (except KS-TLI) fell within the low end of that range ( Table 3 ). Thus, it appears that for MN Clearwater, the range of sufficient soil N to meet IWG grain needs in the first year of growth remains the same. Also, since the number of grain producing spikes is highly correlated to grain yields, it could be that domestication traits and environmental factors that control tiller production are as important as fertility management for determining grain yields (Fagnant et al., 2024). With annual removal of N in grain and forage biomass, we expect there will be a grain yield response to treatment in future years as has been observed in other studies (Bianchin Rebesquini, 2023; Pugliese et al., 2019; Pinto et al., 2022; Bowden, 2023).

Although endogenous soil N may be adequate for IWG grain production, we found evidence that IWG may have a higher first year N demand for both summer and fall forage production. We observed summer forage yield responses to N fertilizer when averaged across locations, and fall forage yield responses when averaged across locations and at the NE location ( Table S11 ). Specifically, we found that 135 kg ha^-1 of spring applied N produced ~ 19% more summer biomass than the unfertilized control. We also found that both 90 and 180 kg ha^-1 of spring applied N produced ~55% more fall forage biomass than the unfertilized control at the NE location, and ~60% more fall forage biomass when averaged across locations ( Table S11 ). Although N absorption early in the season until the point of spike and grain development may be critical for grain yields, N can continue to be taken up or reallocated in fall for vegetative forage biomass production. There have been inconsistent results in terms of summer forage yield response to fertilizer (Bowden, 2023; Pinto et al., 2022; Reilly et al., 2022b). In the case of fall forage biomass, few studies have reported yield responses to variable N rates (Bowden, 2023; Pinto et al., 2024). In summary, the fertilization strategy of IWG could depend on specific management goals of grain, forage, or dual use Kernza production given that grain versus biomass production may require different N rates or timings in the first year.