Sustainable cropping systems for dairy farms in the Northeastern US

Figure 1 illustrates the three crop rotations and the cropping system practices that we integrated into the manure and weed management strategy comparisons. Crop management details by year such as crop varieties, planting dates, seeding rate and depth, nutrient applications and pest management activities are described in detail in the management tables in the Annual Reports for 2010, 2011 and 2012. For an example, the 2012 management tables are included here with Figure 1.

Each main crop entry plot in the experiment measures 120 feet x 90 feet (27.4 × 36.6 m), and is split into four split-split-plots measuring 90 ft x 30 ft (27.4 × 9.1 m). Plots are surrounded by grass alleyways that separated plots from the others by 30 feet (9.1 m). Each rotation is replicated four times, crop entries within each rotation are arranged in a randomized block design. The soil at the site is predominantly a Murrill channery silt loam, with smaller areas of Buchanan channery loam and Hagerstown silt loam.

In spring 2010, 10 soil cores were collected from random locations in each split-split plot at 0-5 cm and 5-15 cm depth. In spring 2013 this soil sampling was repeated (different # cores). These samples were dried, ground analyzed for pH, soil Carbon, and agronomic soil fertility analysis using Mehlich 3 extraction. Also in spring 2010 and 2013, 5 random soil cores/split-split plot were also collected to 15.24 cm, stored in air-tight containers in a cooler, until they could moist sieved and analyzed for water stable aggregates. In fall 2010 and 2011, 8 soil cores were also collected at 15 cm depth for soil fertility analysis.

Each time liquid dairy manure was applied, a sample was analyzed for nutrient concentration to determine rate of nutrient applications and for nutrient management planning. Pre-sidedress soil nitrate test soil samples were collected when corn was about 12 inches (30.5 cm) tall or before the end of June by collecting three random soil cores/split-split plot to a depth of 12 inches (30.5 cm) when manure was broadcast. Where manure was injected, three samples were collected from random locations within the injected manure band and three from outside of the band. Samples were homogenized, dried and sent to the PSU Soil Analytical Lab for analysis.

Crop yields were collected for each crop entry point in the FORAGE and GRAIN rotation in 2010-2012; data was analyzed with a split-plot, mixed ANOVA model using PROC MIXED of SAS for all crops that received manure or weed management in the rotation. For analysis of split-split plot treatments, such as green manures and canola with or with or without oats, data were analyzed with a split-split plot model. Forage crop yields for all harvests were analyzed with a repeated measures split-plot, PROC MIXED model of SAS.

Crop yields were sampled from the middle of each split-split 90 foot long plots and the sampling width varied with crops. For instance a forage plot harvester sampled the middle 4 feet from the forage crops, while a two row plot harvester samples the middle two rows of corn silage and corn grain, a farm scale combine fitted with a weighing basket harvested the middle 13 feet of canola, soybeans, and wheat. Cover crops and green manure dry matter production was sampled from two 0.5 m2 (5.382 ft2) quadrats per split-split plot. Before cover crop termination, cover crops were sampled for biomass and separated into percent cover crop and weeds. Nutrient analyses were performed on cover crop tissue samples to estimate nitrogen contribution at the time of corn planting. Samples are dried and yields are reported on a dry weight basis.

At every crop harvest in the NESARE Dairy Cropping Systems Trial, we collect three subsamples for forage or feed quality analysis from each of our main management treatments that compare two strategies for either manure in the Forage rotation, or weed management in the Grain rotation. In many cases, processing is needed to prepare a crop sample that is representative of on-farm storage or processing.

Ensiling Forage Crops:
To simulate the fermenting that occurs in the ensiling process, subsamples of forage crops, including alfalfa, alfalfa, grass and companion crop mixtures, corn silage, and red clover, were vacuum sealed and stored at 30 degrees C for at least three weeks to simulate the fermentation that occurs in the ensiling process. The dairy scientist on our team, Virginia Ishler, noted that the forage quality analyses of our simulated silage samples were very similar to the forage quality ensiled at the Penn State Dairy Complex at University Park.

Extracting Oil from Canola:
For canola, we used an Oekotec single-dye press to extract oil from the grain, leaving a high protein meal that can be fed to dairy cows. This “cold” press extraction method can result in a canola meal that has a higher than desired fat content for the dairy ration. We found that drying the canola seed down to 5.5 to 7% moisture before pressing the seed was particularly important in order to maximally extract oil and produce a meal with the lowest amount of fat as possible with the mechanical Kern Kraft press (Fig. 3).

Roasting Soybeans in the Laboratory:
For soybeans, we simulated roasting soybeans by putting raw soybeans in a muffle furnace at 146 degrees C for 30 minutes followed by cooling them in a funnel dryer for 30 minutes. This is the temperature and time period recommended for a commercial roaster by Hsu and Satters, J Dairy Sci., 1995, which we adapted to the laboratory setting. A comparison between three samples each of roasted and raw beans in 2011 revealed that roasting the beans in the laboratory dropped the % soluble protein from 73 % to 21 % (Fig. 4). According to Dairy One Lab’s Feed Composition Library, this is what is expected for roasted soybeans (http://www.dairyone.com/Forage/FeedComp/disclaimer.asp).

Forage and Feed Quality Data Analysis
To compare the main management treatment effects on crop quality we compared a common set of forage and feed quality variables, including % crude protein (CP), % neutral detergent fiber (NDF), and net energy of lactation (NEL/Mcal/lb) and used the same statistical model that we used to analyze crop yields.

The virtual dairy operation was designed to represent a typical Pennsylvania tie-stall barn for the lactating herd and a bedded pack for young-stock and dry cows. Upright silos and Ag Bags are used to ensile forages. All corn grain, soybeans and canola meal are fed to the herd, as a total mixed ration. Rations for all the animal groups are formulated based on the 2001 NRC model and reflect very closely to what is fed at the Penn State dairy herd. Income over feed costs is monitored monthly for each scenario (BMSH or IMRH) to evaluate the impact of forage quality and quantity on profitability for the lactating cows. A cash flow plan is being developed for each scenario to evaluate the effect on purchased feed costs and how the cost of producing home raised feeds influences the breakeven income over feed costs/cow and breakeven net margin/cwt.

The two treatments were compared primarily in the GRAIN rotation. The “Standard Herbicide” (SH) treatment utilized broadcast herbicide applications to manage weeds in corn grain, soybeans, and alfalfa. The “Reduced Herbicide” (RH) treatment combined banding herbicide, rolling a cover crop with a roller crimper to form a weed-suppressive mat, using a high-residue cultivator for weed control in corn and soybeans; and planting companion crops for weed suppression when establishing an alfalfa-orchardgrass mix (Fig. 1). The SH alfalfa was seeded to only alfalfa as a forage; RH alfalfa was seeded to alfalfa, orchardgrass, triticale (x Triticosecale), and peas (Pisum sativum L.) as forage species. In addition, in the RH treatment, alfalfa was terminated with a moldboard plow before planting canola, while no-till and herbicides were used to terminate alfalfa in the SH treatment (Fig. 1).

To quantify weed severity and treatment effects, weed density and biomass were collected throughout the growing season. Weeds were sampled both from the resident weed population and in 2011 and 2012 from subplots within the crop plots that were supplemented with three weed species (giant foxtail (Setaria faberi Herrm.), smooth pigweed (Amaranthus hybridus L.), and common ragweed (Ambrosia artemiisifolia L.), each at a rate of 1500 seeds per m2, in fall of 2010 and 2011. Weeds in corn grain, soybean, and canola were sampled for density at 4 and 8 weeks after planting, and then for biomass at 12 weeks after planting. Weed density was measured by identifying and counting weeds in two randomly placed 0.7 m2 quadrats from the resident weed population and/or from supplemented weed subplots. Weed biomass was sampled by collecting above-ground weed biomass from two randomly placed 2.48 m2 areas between crop rows from resident weeds (0.7 m2 quadrats in weed subplots), drying the samples at 60°C for at least 48 hours and weighing. For alfalfa, sampling occurred at the time of alfalfa harvest. Weeds in forage crop entry points were quantified by separating aboveground biomass into forage and weeds, weighing dry biomass, and calculating percent composition of each within the forage crop. Weed species were separated by species, and also into four categories by life cycle: annual broadleaf, annual grass, perennial broadleaf, and perennial grass; data was analyzed with a split-plot, mixed ANOVA model using PROC MIXED of SAS.

Within the inject and broadcast manure management treatments, we compared two green manure crops for nitrogen production prior to planting corn silage (Corn silage-Ca). While designed as a comparison between green manure crops, the red clover (RC)-hairy vetch (HV) comparison between winter wheat and corn for silage (Fig. 2), also represents two different weed management systems. Red clover was drilled in one half of the wheat plots in early spring of each year; hairy vetch was planted in the other half 4 to 6 weeks following wheat harvest. The mulch provided by cover crops can suppress weeds during the fall and winter and in the subsequent corn; to lengthen the persistence and improve the weed suppression of the mulch in the HV treatment, triticale was added at 34 kg*ha-1 in mixture beginning in 2011.

The wheat that would be rotated to HV received an early April herbicide application of 0.04 kg*ha-1 Harmony Xtra (thifensulfuron-methyl + tribenuron-methyl). Red clover has the capacity to compete with weeds in wheat and following wheat harvest and herbicide use in wheat could damage RC, so no herbicide was used in the RC wheat. Prior to seeding the HV, a burndown application of 1.26 kg*ha-1 glyphosate plus 0.56 kg*ha-1 2,4-D LVE was applied about 14 days prior to sowing hairy vetch and triticale. Thus, two extra herbicide applications (spring wheat and late summer burndown) were applied to the HV plots that were not needed with RC.

In the forage rotation, weed biomass in wheat was sampled in June before wheat harvest, and weed density and biomass after wheat harvest, but before hairy vetch seeding. Weeds in wheat and in corn silage were sampled both from the resident weed population and from subplots within the crop plots that were supplemented with three weed species (see Grain Rotation Methods) the preceding fall.

Weeds and the cover crops in corn were managed with the preemergence herbicides (glyphosate plus 2,4-DLVE plus dicamba) and with glufosinate postemergence. In 2011, halosulfuron plus nicosulfuron was added to glufosinate for improved yellow nutsedge (Cyperus esculentus L.) and orchardgrass control.

Plot-scale lysimeters were constructed to allow calculation of nutrient balances associated with injected and broadcast (unincorporated) manure applications. Construction of twelve hydrologically-isolated plots (each ~400m2) was completed in the late spring of 2012 on a site overlying shallow bedrock. Drainage tiles were installed in each plot at the level of the bedrock to intercept groundwater flow. Earthen bermes were constructed around each plot to direct surface runoff water to tile outlets. Tiles collecting both surface and ground water from each plot were directed to collection houses. Full instrumentation of the site was finalized in the spring of 2012, including the installation of flumes and pressure transducers in the houses to automate water flow measurement. An automated weather station has been installed, and instrumentation was also added to track soil moisture at 8, 16 and 30 inches in four of the 12 lysimeters. A landscaping project to improve appearances and access for demonstration events was completed in 2012. After observing evidence of bypass flow following the spring manure application, a smoke tracer study was conducted to identify rodent burrows and other macropores that were potentially biasing subsurface runoff monitoring. These macropores were then sealed with bentonite.
In addition to infrastructure for water collection and associated nutrient loss measurements, nitrogen gas flux measurements were made using portable chambers. For nitrous oxide (N2O) measurements, two chambers bases were installed in each plot shortly after manure application. We then sampled twice per week for four to six weeks after manure application, but switched to less frequent sampling as soil dried and N2O emissions diminished. Each time samples were taken, chamber tops were deployed and a series of three samples were taken from each chamber over the course of about 30 min. Samples of ambient air were also taken to establish time 0 baseline. Nitrous oxide concentrations of the samples are determined by gas chromatograph. Emission rates are determined from the slope of the regression of nitrous oxide concentration verse chamber deployment time.

To measure ammonia emissions, were used a 30 x30 x 4 inch chamber (with a fan for air recirculation) in conjunction with a photoacoustic gas monitor. Two or three chamber bases were installed in two replicate plots each of injected and broadcast manure. At each sampling point, we measure immediately after manure application then about 4, 12, and 36 hours after manure application (ammonia emission subside quickly as manure dries).

Both of the FORAGE and GRAIN rotations relied on scouting to manage early season pests and were planted with non-Bt maize varieties. The simple corn-soybean rotation that served as a low-diversity, high-external-input ‘Control’ rotation, was planted with a Bt corn variety, received a pre-emptive application of pyrethroid insecticide in the spring, and was not preceded by a cover crop.

Stand establishment and early season herbivory
To assess maize establishment and damage from early season pests, a three meter stretch of row from the east and west halves of each split-split-plot was randomly selected when maize reached both two and five leaf stages. All samples were taken at least 1.5 m from plot edges. In each sample, the number of maize plants was recorded and each plant was inspected for slug damage, rating them on the following scale: 0: no damage, 1: < 25% leaf area removed, 2: 25-50% leaf area removed, 3: 50-75% leaf area removed, and 4: > 75% leaf area removed (similar to Byers and Calvin 1994). In 2010, a test of this rating system revealed that damage scores of seven independent observers were highly correlated with one another (mean r = 0.86, SD = 0.05). Because slugs feed using a scraping motion, slug damage is distinct from other pest damage, and is also often accompanied by a slime trail that aids in identification. In addition to slug damage, the number of plants severed by cutworms, along with the number of plants damaged by other caterpillars or billbugs.

Slug activity-density
To measure slug activity-density, square foot pieces of roofing material (made from Owens Corning Rolled Roofing Material, color: Shasta White) as artificial slug shelters were used as shelter traps. While shelter traps provide only a relative measure of slug activity (Byers, Barratt, and Calvin 1989), they are the most feasible sampling method for a large experiment such as this. Starting in mid-April, shelters were randomly placed in the east and west halves of each split-split-plot, at least 1.5 m from plot edges and each other. Shelters were removed to allow for field operations (e.g. manure spreading) but otherwise were left them in the plots continuously. Vegetation or residue was pushed aside so that the shelters laid flat on the soil surface. In 2010, shelters were removed from late June to mid-August during the hottest part of the summer when slug activity on the surface is generally low (Port and Port 1986, Eskelson et al. 2011). In 2011, shelters remained in plots from mid-April to November. Because slugs often leave artificial shelters as they heat up during the day (Hommay et al. 2003), shelters were checked in the morning, with sampling typically completed by noon, and with progressively earlier sampling as temperatures increased through the season. The last sample occurred in each treatment shortly before harvest, in early September for maize grown for silage and in late October for maize grown for grain. Slugs were recorded by species using two references (Chichester and Getz 1973, McDonnell et al. 2009), and voucher specimens of the three species were deposited with the Frost Entomological Museum at the Pennsylvania State University.

European corn borer activity
To assess damage caused by the European corn borer, maize plants were sampled in late August, shortly before silage harvest. We chose this time to capture cumulative damage from first and second generation larvae (Penn State Agronomy Guide 2011). Other demands on the experiment precluded destructive sampling, so we measured evidence of tunneling that was visible on the outside of the stalks and ears (similar to Witmer et al. 2003). In the fourth and eighth row of each split-split-plot, we sampled three sets of two plants spaced roughly 9 m apart for a total of twelve plants sampled per split-split-plot. Each plant was inspected from top to bottom for entrance holes, and we recorded the number of holes per plant.

Predatory arthropod activity-density
We used pitfall traps to measure the influence of treatments on the activity-density of ground-dwelling, predatory arthropods. We placed two traps in the seventh row of each split-split-plot, spaced equally along the row (~ 9 m apart), for a total of eight traps/plot. Each trap comprised a 16-oz plastic deli container (11.5-cm wide, 8-cm tall Reynolds Del Pak ®) that was sunk into the ground so that the edge was level with the soil surface. The lip was removed from an identical container, which was placed inside the first so that it could be easily removed to empty the trap without disturbing the surrounding soil. A white plastic plate (18-cm diameter) supported by nails (8.5-cm long) served as a trap cover and the killing agent was 50% propylene glycol. Traps were opened for 48 h at a time, with the first sample occurring roughly two weeks after corn planting and additional samples about every two and a half weeks until early September (maize silage harvest). The two samples from each split-split-plot were combined into one sample, strained through a fine mesh sieve (1-mm openings) and then transferred into 80% ethanol for later identification. Predatory arthropods were identified to the lowest level possible, mostly to family except for Carabidae which were identified to species or genus. We identified carabid species using a regional publication (Bousquet 2010) and a reference collection at the Pennsylvania State University from previous studies of carabids in central Pennsylvania (Leslie et al. 2007, 2009). Other arthropods were identified with trusted references (Kaston et al. 1972, Triplehorn and Johnson 2005, and Ubick et al. 2009). Voucher specimens of carabids and other arthropods identified in this study have been deposited with the Frost Entomological Museum at the Pennsylvania State University.

Predation on sentinel caterpillars
In 2011, we used sentinel waxworm caterpillars (Galleria mellonella) to assess predation (after Lundgren et al. 2006). Once in June (6/14/11) and again in July (7/21/11), I deployed eight caterpillars along the eighth row of each split-split-plot. We pinned caterpillars (mean mass: 0.20g, SD: 0.04g) through their final abdominal segment to a small piece of modeling clay that was buried in the field so that the caterpillar rested on the soil surface. In initial laboratory and field experiments, caterpillars survived in this condition for over 24 hours. Every other caterpillar was enclosed in a cylindrical, hardware cloth cage (9.5-cm tall, 11.5-cm diameter, mesh size: 1.3 cm) topped with a plastic lid to exclude vertebrates. On each sample date, I made two assessments of predator activity: one started at 8:30, and the other at 20:30. Sentinel caterpillars were checked 3 and 12 h after being placed in the fields. After 12 h, any caterpillars that had been attacked or were missing or compromised were replaced with new caterpillars. During night-time sampling, observers used red headlamps or lights covered in red cellophane to minimize disturbance to predators. We recorded caterpillars as whole and alive, partially eaten, or missing at each time point. Caterpillars that died but showed no signs of predation were recorded and excluded from the analysis. In addition, observers recorded the presence of predators in proximity to or eating the caterpillars.

We implemented scouting protocols to guide IPM decision-making for key pest species including potato leafhopper in alfalfa and early-season pest species attacking corn and the later-occurring European corn borer. For instance, leaf hopper populations were sampled weekly in alfalfa and alfalfa mixture stands using a standard sweep net approach at the split-plot level. Management decisions to cut or spray stands were based on whether populations had surpassed economic thresholds or not (based on recommendations in Penn State’s Agronomy Guide). Stand counts and damage assessments were implemented in corn at V2 and V5 stages to monitor early season pests, while evidence for corn borer damage was assessed prior to corn harvest each fall.

To understand the influence of treatments on pests and predatory invertebrates, we compared response variables using linear mixed models (PROC MIXED; SAS 9.2, SAS Institute Inc. 2009) for all analyses unless otherwise stated. To improve their relevance or satisfy assumptions of normality and homoscedasticity, we transformed some of the data prior to analysis. For stand establishment, we converted the number of maize seedlings per sample at V5 into plants per acre, and then divided by the seeding rate for each treatment to calculate establishment as a percentage of the number of seeds planted. Pitfall trap catches were totaled over the season to compare cumulative predator activity-density in the different treatments. For Carabidae, we separately analyzed species that made up more than one percent of the total carabid catch in each year. Slug counts, European corn borer damage, and pitfall trap catches were square-root transformed.

We used three models to examine the influence of experimental treatments on response variables. For variables with a single measurement (maize establishment, European corn borer damage, seasonal pitfall trap catches), we used a model with block as a random factor, rotation as a fixed factor, and weed or manure split-plot treatment as a fixed factor nested within rotation. Because only the Forage rotation included split-split-plot treatments, we performed a separate analysis for this rotation, with block as a random factor, manure split-plot treatment as a fixed factor, and cover crop split-split-plot treatment as a fixed factor. Where the analysis indicated a significant treatment effect (P ? 0.05), I used Tukey’s post-hoc tests to separate means.
For response variables that were measured at multiple times (slug damage at V2 and V5, slug activity-density over the season), we used repeated measures analyses. The model was similar to that described above, but also included date and its interactions with block, rotation, and split-plot as within-subject factors. We evaluated candidate covariance structures for each response variable and selected an appropriate covariance structure by minimizing AICC scores (Wang and Goonewardene 2004, Littell et al. 2006). This resulted in using the variance components covariance structure for both slug damage and slug activity-density. Where the analysis indicated a significant date by treatment effect, I used the ESTIMATE procedure to separate means within dates (Littell et al. 2006).
The analysis of predation on sentinel caterpillars included several additional variables that necessitated a unique repeated measures analysis. The model included block as a random factor, rotation as a fixed factor, and weed or manure split-plot treatments as a fixed factor nested within rotation. In addition, cage treatment (caged or open) was included as a fixed factor, along with its interactions with the other factors. The repeated factor was time of day (day vs. night).

To examine whether slug activity-density measured via shelter traps was related to slug damage to corn seedlings, we used correlation. We averaged values of slug activity-density collected in spring (April to June) for each plot, and then related activity-density to mean slug damage scores using Pearson’s correlation (Minitab v. 16, Minitab, Inc., State College, PA). Because of differences in the timing of field operations each year, the spring average represented three sample dates in 2010 and nine sample dates in 2011.
We used a similar approach to examine whether predator activity-density measured via pitfall traps was correlated with predation on sentinel caterpillars under vertebrate exclusion cages in 2011. Because of low predation in June, this analysis was restricted to July (sentinel prey sample: 7/20/11; pitfall sample: 7/21/11-7/23/11). We used field observations of predation on tethered caterpillars to choose taxa that would be expected to correlate with predation. The taxonomic groups tested in the analysis included ants (Formicidae), the dominant ground beetle Pterostichus melanarius, all large Carabidae (? 9mm), and all predatory arthropods combined. We used Pearson’s correlation (Minitab v. 16) to test for a significant relationship between catches of these taxa and caterpillar predation as observed three and twelve hours after the start of the experiment. Both variables were averaged at the plot level.

In order to compare rotation effects on arbuscual mycorrhizal fungi, fungal colonization assessments were made on corn seedlings following different cover crops in the NESARE diary cropping systems, 10-14 days after emergence (or late V1) by clearing and staining corn roots and looking for mycorrhizal fungal structures. Additionally, data on corn seedling aboveground biomass and %N and %P in the aboveground tissues were obtained.

Similarly, in alfalfa and canola split plots in the GRAIN Rotation, corn seedlings were planted into the crops and used as trap seedlings to assess mycorrhizal colonization, aboveground biomass, and %N and %P in the aboveground tissues as described above.