Evaluating Winter Barley Cultivar Using Data Envelopment Analysis Models

Unlike feed barley, malting barley must meet a specific set of quality standards for acceptability by maltsters. Multiple quality criteria in addition to the grain yield makes ranking of genotypes challenging. The objective of this study was to apply data envelopment analysis (DEA) models to rank the efficiencies of 27 winter barley entries based on grain yield and quality indices. Four methods of DEA including Charnes, Cooper and Rhodes (CCR), Färe and Grosskopf (FG), Banker, Charnes and Cooper (BCC), and Seiford and Thrall (ST) used for the ranking. Testing trial included 14 two-rows and 13 six-rows winter barley. All entries except two, demonstrated high winter survival ratings. Overall, the six-row cultivars out-yielded the two-row cultivars by 18%. However, in terms of brewing quality, the two-row entries performed better than six-row entries and had 40% lower thinness, 12% higher plump, and lower grain protein content. The six-row entries had 32% higher germinative energy than two-row entries. The ranking by four models were not similar, however, SU-Mateo and Calypso had the highest efficiency (1.0) by all four models followed by Wintmalt and Vincenta.


Introduction
In the United Sates, unlike previous decades, the majority of barley is currently grown for malting purposes, rather than for animal feed (Mallet, 2014).The shift in the purpose of cultivating barley is at least partly due to significant increasing demand for local food and beverage consumption.Small malt houses and craft brewers are constantly looking to locally produced barley grain production to meet their demand.
Unlike feed barley, malt barley must meet a specific set of quality standards for acceptability by maltsters.Since malt barley commands a significantly higher price than feed barley, moderate high-quality yield is preferred to the high yielding but low quality barley (O'Donovan et al., 2011;USDA, 2015).
The brewing barley must be free of disease, frost damage, pre-harvest sprouted, weathering, heat damage during storage, chemical residues, smut and ergot, and broken kernels (BMBRI, 2010).Moreover, the protein content of the grains should be around 105-125 g kg -1 since higher protein levels reduce the amount of extractable malt (McFarland et al., 2014).In general, six-row barley varieties usually have higher protein content ranging from 90-120 g kg -1 , compared to two-row barley varieties, which range from 90-110 g kg -1 (Magliano et al., 2014).Higher protein levels is often associated with lower starch content.Starch is the principal contributor to malt house extract, and high starch content results in more beer produced from a given amount of malt, although some small-scale breweries are minimally concerned with malt house extract efficiency.Barley grain must have a minimum of 70% plump kernels (Wrigley, 2010).Percent plump is the percentage of all kernels retained above the 0.24 × 1.91 cm screen.Higher plump is desirable, as plumb kernels produce higher amount of extractable malt by weight (Johnson & Nganje, 2000).Percent thin is the percentage of kernels passing through the 0.20 × 1.91 cm screen.Unlike plump, high percent thin is considered undesirable due to low malt extraction by weight.Kernels need to be greater than 85% plump and of uniform size to produce more malt.Test weight, a measure of grain plumpness, is also an indicator used by malsters for evaluating malt quality.Higher test weight is favorable by maltsters.However, acceptable values of protein and plumpness might be modified depending on the quality of harvested malting barley (Smith et al., 2016).Malting barley must not have a falling number below 200 (Darby et al., 2014), since low falling number indicates degradation of the grain starches due to increased alpha-amylase activity from pre-harvest sprouting (MacArthur et al., 2009).Barley grains should have a minimum germinative energy (kernels germinated after 72 hours) of 95%, since un-germinated grains do not convert the stored starches to simpler sugars, mainly maltose, which is required for malt production.Deoxynivalenol (DON) toxin in grain, which is due to fungal infestation, is a major criterion for acceptability of barley for brewing.DON level above 1.0 ppm results in rejection by malt houses, though tolerance from the malt houses may vary depending on the quality of the grain supply that year (Smith et al., 2016).Acceptable DON level for most craft cultivar breweries is below 0.5 ppm.This is primarily due to the 'gushing', or over foaming that can occur in the finished beer when even very low levels of the mycotoxin may present (Johnson & Nganje, 2000;Mallet, 2014).Test weight is an important component of evaluating quality and value of barley grain.Falling number is not a standard quality measurement at malt houses.However, research indicates that a falling number of 220 seconds and greater is associated with high malt barley quality.
Agronomists constantly perform testing trials to evaluate and select high yielding cultivars to be used by growers in various regions.However, evaluation of malting barley genotypes is quite challenging.In testing trials of malting barley, the ranking and selection of the entries must include several quality parameters, in addition to the grain yield and possibly the tolerance to a specific stress condition such as winter hardiness.
DEA is a non-parametric technique used to determine inefficiencies in evaluating systems with multiple inputs and outputs (Banker et al., 2011).This technique has been employed since its introduction in 1978 (Charnes et al., 1978) in a wide range of industries, most commonly in banking, healthcare, human resources and education.Applications of this model to agronomic systems are more limited (Emrouznejad et al., 2008).The applicability of this model to evaluation of genotypes in agronomic research has recently been demonstrated (Etemadi et al., 2017), where the efficiency of eight faba bean cultivars calculated based on four agronomic traits.In applying this methodology to a cultivar trial, when all traditional agronomic inputs (labor, fertilizer, etc.) are equal across treatments, undesirable traits (excess protein, high DON levels, thin kernels) can be considered as inputs and desirable traits (high grain yield, test weight, plump kernels, germination energy and falling number) as outputs.The standard extension testing trials often conducted in a single location, and the results disseminated to the growers in that region.Local growers may find multiple rankings of cultivars based on various characteristics, an inefficient way to select the favorable cultivar.Providing a single, regionally specific ranking will help simplify selection of the most appropriate cultivar for the growers needs.
The main objective of this study was to apply DEA models to evaluate and rank the efficiencies of 27 barley entries based on their grain yield and quality indices including test weight, protein level, DON, percentage of plumpness and thinness.

Experimental Site
This trial was conducted at the University of Massachusetts Agricultural Experiment Station Farm in S. Deerfield, MA (42 N, 73 W).Soil at this site in the Connecticut River Valley is characterized as fine Hadley loam.Soil sample from the top 15 cm of soil collected by sampling a 5 × 4 grid of 20 sub-samples across the block.The block was then amended appropriately as recommended by the University of Massachusetts soil-testing lab for barley production.On September 25, 2015, all barley entries were planted into the experimental block, which had been left fallow over the preceding season, following a 2014 winter rye cover crop.Experimental plots were 0.5 m wide and 1.5 m long and seeded at 123 kg ha -1 .

Experimental Layout
Treatments consisted of 27 winter cultivars (Table 1) supplied to the University of Massachusetts through its participation in the national Winter Malting Barley Trial, organized by the University of Minnesota.All entries replicated three times in a randomized complete block design.The variety Wintmalt was planted as buffer strips to prevent edge effect.

Harvest and Quality Analysis
All entries harvested using a 1995 ALMACO SPC20 plot combine on July 21, 2016.Grain yields standardized to 13.5% moisture.Sub samples from each plot were stored in a 37.8 °C until being processed and shipped to the University of North Dakota cereal grains testing facility for malt quality analysis.
Total protein (g kg -1 ), color and moisture, based on a dry-matter, was determined by near infrared transmittance on an Infratec 1241 grain analyzer (FOSS, 8091 Wallace Rd., Eden Prairie, MN 55344, USA).Barley plumpness (100 g) kernel assortment was determined using a Sortimat (Pfeuffer GmbH, Flugplatzstraße 70, 97318Kitzingen, Germany) (Methods of Analysis of the American Society of Brewing Chemists, 2004).Deoxynivalenol (DON) expressed in ppm and was determined by gas chromatography (Gas Chromatography with Electron Capture Detection by Tacke and Casper: Journal of AOAC International, Vol. 79, No. 2, 1996).Falling number was determined by a TecMaster RVA unit (PertenInstruments, Hägersten, Sweden).Germinative Energy was determined by ASBC method Barley 3-C (Methods of Analysis of the American Society of Brewing Chemists, 2004).Thin percentage is the percentage of kernels passing through the 14.3 × 8.6 cm screen.

DEA Analysis: Generalized DEA/AR Model
In this study, GAMS software utilized to evaluate the efficiency of malting barley entries through four models including CCR, BCC, FG, and ST.We considered x ij (i = 1, …, m) and y rj (r = 1, …, s) as the positive inputs and outputs of DMU j (j = 1, …, n), thus the generalized DEA model can be formulated as below (Yu et al., 1996): (1) where, δ 1 , δ 2 , δ 3 are binary parameters: However, in reality additional information or assumptions may rise that should be considered when using these models.For example, some of the weights in a DEA problem might need to be restricted.For this reason, Zhou et al. ( 2012) developed model ( 1) into the following model: (2) Here, the assumption is that 0 ≤ C pq L < C pq U and 0 ≤ D pq L < D pq U .
However quite often, one of the inputs, say x 1 , can be considered as an "input numeraire" and one of the outputs, say y 1 , as an "output numeraire" (Thompson et al., 1992).In this case, the assurance regions are simplified as: (3)

Statistical Analysis
Data were analyzed using PROC GLM in SAS version 9.4, and Tukeys HSD at P ≤ 0.05 determined the significance of relationships between seeding rates to any of the measured indices.

Weather
Although the total precipitation in fall 2015 was almost similar to the norm for the experimental location, the main winter barley growing season (March-July 2016) was significantly dryer (280 mm) than the norm (452 mm) for the experimental location.
The 2016 growing season was also warmer than the norm for the location where crop collected 306 more thermal units (Data not shown).Therefore, the spring green-up occurred earlier thus the earliest maturity entries (Throughbred and MW11S4024-004), headed in mid-May, that was about 6 days earlier than the average for this location.

Genotypes Grain Yield
In this study, 14 two-row and 13 six-row genotypes evaluated (Table 1).The tested genotypes varied significantly in terms of grain yield and quality (Table 2).All tested entries except DH130004 and DH130718 demonstrated high winter survival ratings (Table 1).Consequently, these two non-hardy enough entries produced the lowest grain yield.On average, six-row cultivars out-yielded the two-row cultivars by approximately 18%.This could be partly due to the poor winter survival of DH130004 and DH130718, which were two-row cultivars.
Due to more uniform germination, the two-row type malting barley preferred by local maltsters and craft brewers.

Test Weight
Test weight indicates how well the kernels filled during the grain-filling period and commonly used as one of the grain quality indicators.The higher the test weight the higher the quality for brewing purpose.Great variations in test weight of brewing barley genotypes have been reported (Brouwer et al., 2016;Kratochvil, 2017).However, environmental factors especially conditions during grain filling period could also dramatically influence the test weight of barley grains (Stevens et al., 2015).In current study, test weight of entries ranged as low as 49.9 (05ARS561) to 62.2 kg hl -1 in Puffin (Table 2).Other entries, which ranked high in test weight, included Calypso, SU-Mateo, and Thoroughbred with 61.6, 60.8, and 60.4 kg hl -1 , respectively.Pasynkov et al. (2017) considered a test weight of 68-73 kg hl -1 as the best for brewing purpose in Czech.Obviousley none of the tested entries were in this range.

Falling Number
The entries demonstrated dramatic variations in their falling number, which ranged from 60 to 350 seconds (Table 2).A falling number of 220 s or lower indicates that some of the grains sprouted which negatively influence the malting quality (Brouwer et al., 2016).Overall, seven of the 27 entries had falling numbers above 220 seconds (Table 2).The highest falling number measured in Calypso (350), Thoroughbred (311), and SU-Mateo (310) seconds.Falling number is controlled by both genotype and environment conditions (MacArthur et al., 2009).Factors contributing to pre-harvest sprouting include temperature, and also timing and duration of precipitation following anthesis (Brouwer et al., 2016).Overall, the six-row entries performed better than two-rows in regard to the falling number (Table 2).

Protein
The protein level of entries varied between 106 and 149 g kg -1 (Table 2).MW11S4029-002 had the highest protein content (149 g kg -1 ), followed by McGregor and DH130004 which contained 148 and 146 g kg -1 , respectively (Table 2).For malting purpose, high quality barley typically contains low to moderate protein levels ranging from 90-110 g kg -1 .The AMBA guidelines ( 2018) for all malt two-row specify grain protein level below 120 g kg -1 (12%).Lower crude protein is desirable from a malting perspective since high protein levels can make beer hazy and lower starch content, which is the principal contributor to the brewed extract.In current study, majority of entries did not meet the acceptable protein level for malting.Although Calypso produced the highest grain weight among all entries, its protein content (133 g kg -1 ) can be a limiting factor for brewing acceptability in Northeast.On average, six-row barley contained 5.6 g kg -1 more protein than the two-row entries (Table 3).

Germination Energy
Malting is a germination process and barley varieties with GE values higher than 95% (72 hours) considered as ideal commercial malt criteria (AMBA, 2018).Out of 27 participated entries, fifteen entries met the current industry malting standards for seed germination (Table 2).Although four two-row entries had GE of above the acceptable standard, on average, six-row entries performed better than the two-row entries.Khokonova et al. (2015) stated that barley grains obtain the energy necessary for germination, which is crucial for mating purpose, during their post-harvest duration.

Plump Kernels
Plumpness is an important quality index for brewing barley.According to AMBA (2018), grains with 90% or higher plump produce desired extract yield, which is an important economic factor to the brewers.However, barley with plump higher than 75% may produce relatively high-quality malt (Brouwer et al., 2016).In current study, plump kernels ranged from 54.6% to 97.4% (Table 2).In this study, the two-row entries performed better, and their plum was roughly 12% higher than six-row genotypes.

Percent Thin
According to the AMBA guideline (2018), the percent thin for 2-row and 6-row barley cultivars should be below 3%.All two-row entries had acceptable percent thin whereas some of six-row entries including 6W11-0003 (6.7%), Strider (5.9%), and 05ARS561 (4%), were above the standard acceptable value for percent thin.The lowest thin percent belonged to SU-Mateo with 0.3 %, followed by Calypso, Vincenta, and Wintmalt all with 0.6%.Averaged over the entries, the two-row barley performed much better in regard to their thin percent which was roughly 40% lower than six-row entries.Barley cultivars with lower thin percent absorb water fasters, and a high percentage of thins may cause problems with malt homogeneity.

DON
DON level of barley grains indicates the degree of fungal infestation and DON values above 1.0 ppm results in rejection by malt houses.All winter barley entries in this trial exhibited DON values below the FDA's 1 ppm limit (Table 2).However, the acceptable DON value by many craft malt houses is even lower (0.5 ppm) than the standard acceptability set by AMBA (2018).This is primarily due to the over foaming of beers when barley grains are infected with even very low levels of the mycotoxin (Johnson and Nganje, 2000;Mallet, 2014).

Efficiency of Entries
In this study seven major quality traits including test weight, falling number, protein %, germination energy, plump kernel, thin %, and DON in addition to the grain yield for ranking of the entries using four DEA/AR models were considered.To analyze the data by the models, the inputs and outputs of the DEA model was predefined.

Table 1 .
List of winter malting barley entries and days to 50% heading
The larger values for grain yield, falling number, germination energy, plump kernel, and test weight was considered favorable, whereas lower values for protein percent, thinness, and DON considered desirable.As expected, none of the