CAN WE PREDICT PEELING PERFORMANCE OF PROCESSING TOMATOES?

CAN WE PREDICT PEELING PERFORMANCE OF PROCESSING TOMATOES? ELISABETH GARCIA 1, MITCHELL R. WATNIK 2 and DIANE M. BARRETT 1,3 1 Department of Food Science and Technology University of California, Davis One Shields Avenue Davis, CA 95616-8598 2 Statistical Laboratory University of California, Davis One Shields Avenue Davis, CA 95616-9936 Accepted for Publication October 28, 2005 ABSTRACT Tomato processors are increasingly interested in being able to predict whether tomatoes will peel well, and therefore, yield high-value processed tomatoes. We describe two statistical models for peeling applied to multiple years of data. One model is appropriate for perfect or defect-free tomatoes, and the second model is valid for the normal population of tomatoes obtained following mechanical harvesting. The ability to peel perfect tomatoes was significantly affected by exposure of tomatoes to temperatures greater than 100F, by fruit weight and by pericarp wall thickness. The peelability of a normal population of tomatoes was influenced by tomato weight and width as well as degree-days and exposure to temperatures greater than 90F. Thickness of the pericarp walls and red layer positively affected the peelability of normal tomatoes. The ability to predict tomato peelability using statistical models may improve the quality of processed tomatoes and may result in more efficient commercial peeling operations. INTRODUCTION Processing tomatoes are the most widely grown vegetable in California, averaging 11 12 million tons harvested annually. The value of the tomato crop was estimated at over 600 million dollars for the 2004 harvest (USDA/ERS 3 Corresponding author. TEL: (530) 752-4800; FAX: (530) 752-4759; EMAIL: dmbarrett@ ucdavis.edu 46 Journal of Food Processing and Preservation 30 (2006) 46 55. All Rights Reserved. 2006, The Author(s) Journal compilation 2006, Blackwell Publishing

PEELING PERFORMANCE OF PROCESSING TOMATOES 47 2003). The California tomato crop comprises most ( 94%) of the processing tomatoes grown in the U.S. and approximately 40% of the world s production. Before 1964, processing tomatoes were hand harvested; but by 1969 the entire California crop was mechanically harvested (Brandt and French 1983). While the introduction of mechanical harvesting brought many advantages, it also requires special demands from tomato breeding and cultivar (cv.) selection programs. Tomato cvs. were traditionally bred to contain high soluble solids, to produce high viscosity paste and to retain the red color after processing. To withstand the rigors of mechanical harvesting, the peels of tomatoes must also be resistant to the physical handling associated with mechanical harvesting. However, it is undesirable to develop tomato cvs. with peels that are difficult to remove and result in small yields recovered by the peeling process. It is a challenge for breeders to produce cvs. that can tolerate the mechanical damage inherent in mechanical harvesting, while maintaining the physical requirements for efficient peel removal and high yields of superior-quality peeled tomatoes. Since the 1980s, the whole peeled and diced tomato markets have received increased attention, with more demand for high-value ingredients for ethnic cuisine, salsa, pizza and other formulated foods. Tomato processors often retain historical observations on the peeling performance of specific tomato cvs., particular growers as well as growing locations. They base decisions for future peeling performance and applications on these observations. In addition to the differences in physical and chemical attributes of tomato cvs., maturity is another factor that affects peelability. With the advent of once-over mechanical harvesting of tomatoes, the optimal harvest date is selected based on estimates of the percentage of mature ripe tomatoes in a field. In general, harvesting is carried out when 90% of the tomatoes in a field are red. However, some tomatoes will be overmature and some may be less than the ideal degree of maturity for peeling and processing. Defects such as broken tomatoes are prone to occur when soft, overmature tomatoes are harvested, while undercolor occurs when tomatoes are harvested immature. As part of a decade-long dedication to research on the quality of processing tomatoes, our laboratory has focused on peelability issues. In an attempt to avoid subjective evaluations and to add to historical data, we determined objective analyses as indicators of the ease of peeling. Physical attributes of tomatoes such as weight, shape, tomato cv. and maturity were investigated as well as growing location and the presence of tomato defects (Barrett et al. 2006; Garcia and Barrett 2006). The accumulated data were collected over several years, although not all variables were determined every year, in part because of insignificant results and in part because of physical and practical constraints. Our ultimate goal was to develop a model applicable to acceptable

48 E. GARCIA, M.R. WATNIK and D.M. BARRETT prediction of performance of processing tomatoes during the peeling process as well as prediction of yield as whole peeled or diced tomatoes. In this article, we describe the development of a statistical model for peeling applied to selected data subsets of processing tomatoes produced in the 1995, 1996 and 2000 2002 seasons. Two models were developed, one for ideal perfect batches of tomatoes free of defects and another for normal batches of tomatoes with observed levels and types of defects in selected seasons. MATERIALS AND METHODS In the 1995 season, seven tomato (Lycopersicon esculentum) cvs., namely Halley 3155 (Orsetti Seed Co.), H 3044, H 8892 (Heinz Tomato Products), FM 9208 (Ferry Morris Seed Co.), LaRossa (Rogers Seed Co.), Brigade (Asgrow) and Nema 512 or N 512 (Seminis Seed Co.) grown on the University of California Vegetable Crops Experiment Station in Davis were used. In the 1996 season, 10 cvs., for example, Halley 3155, BOS 8066 (Orsetti Seed Co.), H 8892, H 9280, H 3044 (Heinz Tomato Products), FM 9208, Brigade, LaRossa, HyPeel 45 and Sun 6117 also grown at the Experiment Station were used. Quality attributes, peelability and tomato yield as affected by maturity of selected cvs. studied in 1995 and 1996 are presented in Garcia and Barrett (2006). In this study, we included only the red, mature tomatoes for modeling and prediction purposes. In 2000, tomato cvs. APT 410 (formerly Asgrow Vegetable Seeds, now Seminis), AB 4077 (AB Seeds Ltd), Red Century 32 (Orsetti Seed Co.), HyPeel 303 (formerly Peto Seed, now Seminis), HyPeel 45 (formerly Peto Seed, now Seminis) and CXD 179 (Campbell Soup Co.) were also included. In the 2001 and 2002 seasons, only Halley 3155 cv. tomatoes were used. Tomatoes were hand harvested and washed as previously described (Barrett et al. 2006). Physical attributes determined in subsamples of 10 tomatoes per batch included weight, height, width, pericarp wall thickness and red layer thickness (Barrett et al. 2006). Notably, some of these determinations are destructive; therefore, tomatoes could not be further peeled. Therefore, physical characteristics in the data sets are means of representative tomatoes from each batch prior to peeling. All tomatoes were steam peeled, and percent peelability and product yield were calculated as previously described (Garcia and Barrett 2006). Peeling was carried out on batches of 20 tomatoes. Two categories of tomato batches were considered, namely perfect or nondefective, where tomatoes were sorted to remove any type of defect, and normal batches, where tomatoes were not sorted for defects and all harvested tomatoes were included.

PEELING PERFORMANCE OF PROCESSING TOMATOES 49 Selected types and numbers of defects were present in the seasons studied (Barrett et al. 2006). A total of 2840 and 7600 tomatoes were peeled in the perfect and normal categories, respectively. The cvs. and the number of tomato batches included in each category are listed in Tables 1 and 2. Statistical analyses were carried out using logistic regression models, which provided mean comparisons (SAS 2000). In the logistic regression models, both forward addition and backwards elimination model selection techniques were employed to identify variables that may be useful. While logistic regression models likely were excessive in size (in statistical terms, referred to as over-fitting), our purpose was to identify tomato characteristics associated with peelability to assay. High statistical power resulting from the use of logistic regression models is preferable to low statistical size. Because some of the explanatory characteristics were nonlinearly associated with peelability, we also employed classification and regression trees (CART) modeling techniques (Venables and Ripley 1998) to identify possible thresholds for select variables that may be associated with peelability. These artificial categorical variables were then included in the logistic model. The CART models were performed using the S-PLUS computing package (Insightful Corp., Seattle, WA) (S-PLUS 2003). We obtained the means of percentage peeled tomatoes (76.3% [1995] and 64.3% [1996]) and percentage whole peeled tomato yield (50.7% [1995] and 36.5% [1996]). Overall peelability was considered acceptable when percentages obtained were equal to or greater than the calculated averages for each respective year. RESULTS AND DISCUSSION Tomato peelability is affected by the presence of tomato defects (Barrett et al. 2006). When only perfect tomatoes (i.e., exempt of any defect or imperfection) were assayed, the percentage of tomatoes peeled as well as the yield of whole peeled tomatoes were generally greater than when normal batches of tomatoes were peeled. However, results for normal tomatoes may be more applicable to realistic commercial peeling of tomatoes. Table 1 and Table 2 represent perfect and normal tomatoes, respectively. The tables present the group means for the seasons evaluated by year, cv., physical measurements, peelability and whole peeled tomato yield. Our main interest was to identify simple indices that will better enable processors to select tomatoes that are easily peeled. These indicators will consequently allow for rapid inspection and directing of incoming loads to manufacturing of value-added tomato products rather than paste. Important outcomes considered in this study were the percentage of peeled tomatoes using a standard

50 E. GARCIA, M.R. WATNIK and D.M. BARRETT TABLE 1. PHYSICAL ATTRIBUTES, PEELABILITY, TOMATO YIELD, TEMPERATURE AND DEGREE-DAYS OF TOMATO CULTIVARS Year Cultivar No. batches Weight (g) Width Pericarp Red layer Peeled tomatoes (%) Whole peeled yield (%) Number of days temperature above Degree-days 90F 100F 1995 H 3044 2 59.52 45.58 5.77 2.18 70.6 27.4 45 9 3566 1995 FM 9208 3 87.50 50.55 7.81 2.57 91.1 59.3 47 9 3684 1995 LaRossa 3 77.14 44.53 6.31 2.54 78.3 48.3 47 9 3800 1995 H 8892 3 78.02 47.55 6.80 2.71 45.0 28.0 49 9 3868 1995 Brigade 3 80.00 50.44 7.60 2.57 75.0 56.7 48 9 3832 1995 Halley 3155 3 89.63 50.94 7.17 2.25 81.7 60.6 51 9 4021 1995 H 512 3 79.71 50.56 5.95 2.13 58.3 39.9 47 9 3684 1996 H 3044 2 62.13 47.83 6.06 2.19 67.5 34.1 50 13 3268 1996 FM 9208 3 80.76 50.38 7.70 2.49 87.5 42.8 53 15 3392 1996 LaRossa 3 82.97 44.88 7.62 2.89 40.0 23.1 54 15 3515 1996 H 8892 3 74.70 48.37 7.02 2.97 65.0 35.5 54 15 3543 1996 Brigade 3 73.18 49.96 7.34 2.41 61.7 29.9 55 15 3577 1996 Halley 3155 3 100.89 52.02 8.05 2.78 75.0 54.3 56 15 3611 1996 BOS 8066 3 92.82 53.49 6.93 2.63 53.7 28.2 51 14 3313 1996 H 9280 3 72.82 48.81 7.02 2.71 60.0 37.6 49 12 3222 1996 HyPeel 45 3 91.28 53.26 7.61 2.55 67.9 44.3 54 15 3487 1996 Sun 6117 3 87.96 53.45 6.72 2.09 66.7 37.7 52 15 3356 2001 Halley 3155 93 74.60 45.70 6.70 2.10 84.5 63.3 51 5 3187 Each batch had 20 tomatoes. Only defect-free tomatoes were peeled.

PEELING PERFORMANCE OF PROCESSING TOMATOES 51 TABLE 2. PHYSICAL ATTRIBUTES, PEELABILITY, YIELD AND TEMPERATURE EXPOSURE INFORMATION ON TOMATO CULTIVARS Year Cultivar No. batches Weight (g) Width Pericarp Red layer Peeled tomatoes (%) Whole peeled yield (%) Number of days temperature above Degree-days 90F 100F 2000 Halley 3155 21 90.1 45.5 8.17 2.80 66.9 43.9 26 3 2719 2000 Halley 3155 6 72.3 48.6 7.76 2.87 77.3 57.3 27 3 2930 2000 Halley 3155 2 74.4 49.1 8.01 3.19 70.0 50.7 46 7 3015 2000 Halley 3155 26 65.2 45.1 7.31 2.84 55.6 41.3 50 7 3284 2000 Halley 3155 18 88.6 50.6 8.01 2.84 42.2 32.7 59 7 3543 2000 APT 410 6 82.0 49.4 7.90 2.85 61.8 49.1 27 3 2930 2000 APT 410 5 56.3 46.0 6.95 2.85 78.0 54.5 46 7 3015 2000 APT 410 12 70.0 48.2 7.65 2.50 76.7 55.9 47 5 3006 2000 AB 4077 6 75.9 49.4 7.76 2.80 65.0 35.9 47 5 3006 2000 AB 4077 4 86.1 50.0 7.93 2.83 90.0 66.0 59 7 3543 2000 AB 4077 6 82.8 51.8 8.26 3.02 78.3 57.7 27 3 2930 2000 Red Century 32 18 80.2 49.1 7.77 2.79 70.6 41.2 26 0 2595 2000 HyPeel 303 4 90.8 51.2 7.63 2.80 36.3 25.5 31 0 2834 2000 HyPeel 45 10 68.3 46.5 7.43 2.41 72.5 44.9 46 7 3015 2000 CXD 179 5 64.8 44.8 8.58 3.24 77.0 54.7 46 7 3015 2001 Halley 3155 87 74.3 45.9 8.37 2.78 51.1 39.0 51 5 3187 2002 Halley 3155 144 92.9 51.9 7.61 2.60 33.0 25.4 67 5 4247 Each batch had 20 tomatoes. Peeled tomatoes were representative of normal batches of harvested tomatoes, including different types of defects.

52 E. GARCIA, M.R. WATNIK and D.M. BARRETT steam peeling method (Garcia and Barrett 2006) and the yield of whole peeled tomatoes. The definition of a batch considered acceptable for peeling was based on the median percentage of both peeled tomatoes and whole peeled tomato yield; anything larger than these medians was considered acceptable. For the perfect batches, medians were 80% and 57% for percentage peeled tomatoes and percentage whole peeled tomatoes, respectively. Medians larger than the means are indicative of a few defect laden tomatoes in the data set. For the normal tomatoes, the medians were considerably smaller than the perfect tomatoes, for example, 55% peeled tomatoes and 36% whole peeled tomatoes. Perfect Tomatoes Initial research with tomato peelability included many more variables than those included in this study, such as tomato height, ratio height/width, stem scar diameter, shoulder height, number of tomato locules, and number of seeds and cracks at the stem scar. Several of these variables were later omitted because they were not associated with peelability; only variables that exhibited association with peelability in previous analyses remained (Table 1) in the study. A comparison of the peelability and yield of Halley 3155, the only cv. assessed in three seasons (1995, 1996 and 2001), demonstrated a dramatic yearly variability. The 1996 season resulted in the poorest peeling, with a tomato yield of 54.3% in contrast to 60.6% and 63.3% for 1995 and 2001, respectively. The poor 1996 season for tomato peelability and yield held true for other cvs. as well. Overall, the growing year exhibited a large impact on the selected tomato quality of cvs. Also, the influence of cv. is observed as widely variable in selected tomato peelability and whole peeled yield in 1995 and 1996 (Table 1). When the variables listed in Table 1 were included in the statistical CART modeling, variables selected as predictors of acceptable peeling were tomato weight less than 88.3 g, tomato pericarp thickness less than 6.5 mm and fewer than 7 days exposure to temperatures greater than 100F. Logistic regression model analyses with backward variable selection suggested the only significant factor was production year. When using forward variable selection, the pericarp thickness threshold provided an index of potentially good peeling tomatoes. Observations from previous research confirm that large tomatoes and/or tomatoes with overly thick walls are more difficult to peel. Normal Tomatoes Using Halley 3155, the only cv. evaluated in 2000, 2001 and 2002 for reference, the normal batch data for the 2002 season exhibited the smallest

PEELING PERFORMANCE OF PROCESSING TOMATOES 53 percentages of peeled tomatoes (33.0%) and whole peeled tomatoes (25.4%). In 2000, peeling varied widely from 42 to 77% for peeled tomatoes and 33 57% for product yield, depending on the grower. The range of percentages of peeling for the tomatoes harvested in the 2000 season indicates just how poor the 2002 tomato crop was in terms of peelability. Analyses of the normal batches across these 3 years using forward stepwise selection identified only degree-days. The backward variable selection identified tomato weight, width and number of days at temperatures greater than 90F as statistical factors that significantly affect peeling. Using the CART modeling technique, the first cut identified less than 63 days at temperatures greater than 90F was a predictor of good peeling tomatoes. A second statistical cut identified less than 48 days at temperatures greater than 90F. These time/ temperature thresholds should be viewed with caution because this study did not include a wide range of climatic conditions. As indicated in Table 2, tomatoes harvested in the 2001 and 2002 seasons were exposed to the equivalent number of days at temperatures above 90F and 100F and, consequently, equivalent accumulated degree-days. The CART model effectively separated the 2001 and 2002 seasons from 2000 in terms of peelability. The effect of temperature was specific to each growing season and, therefore, was confounded with the year of production in this data set. The backwards variable selection approach for normal batches identified three CART-identified thresholds as significant explanatory variables. Thick red layer ( 2.8 mm) is an index of good peeling. Relatively thick pericarp ( 7.6 mm) and a large number of days ( 48) at temperatures greater than 90F are identified detrimental to peeling. The selection of thick red layer as a positive physiological attribute may be explained by the histological differences observed in the tomato. The cell layers located immediately beneath the peel that are more richly colored identified as the red layer can be easily removed during peeling (Juven et al. 1969). Thicker red layers may contribute to the ease of peel detachment during peeling operations. CONCLUSIONS The original objective of developing a model for application to loads of processing tomatoes to predict peelability was complex. First, commercially produced tomatoes are normal and afflicted with many defects. The perfect tomatoes exhibited stronger relationships between peelability and growth temperature and time than the normal tomatoes. Temperatures experienced during the growing season may be associated with the occurrence of defects or other undesirable physiological characteristics in tomatoes and are reflected in the statistical model, whose greater variability is exhibited in the statistical model of normal tomatoes.

54 E. GARCIA, M.R. WATNIK and D.M. BARRETT Barrett et al. (2006) concluded that select tomato defects occur at divergent levels in selected production years. Some defects exhibit a negative impact on peelability and some defects exhibit a positive impact on peelability. In addition, peeling methods (steam versus lye) and conditions (steam exposure, vacuum levels, lye residence time) are variables influencing the peeling outcome. Previous research comparing commercial peeling and the small scale nonautomated steam peeling used in our pilot plant demonstrated considerable differences in peelability and product yield. Therefore, results obtained at the pilot scale should be validated in a commercial setting before implementation. A statistical model that may be applied to processing tomato plants must include peeling results obtained with industrial equipment used under commercial peeling conditions instead of based on small pilot scale peeling equipment used under laboratory conditions. An evaluation of the usefulness of this statistical model requires the cooperation of commercial processors who are willing to share equipment time for collection of critical data with the final objective of achieving a useful model for prediction of peelability performance. Development of peeling models is possible provided that large quantities of tomatoes are available, including tomatoes of selected cvs. and processing consistency with respect to peeling conditions and assessment of physical parameters is maintained throughout the study. However, a model developed for one tomato cv. and one peeling condition provides little potential application to other plants because of variations in tomato cvs. and peeling methods and conditions. ACKNOWLEDGMENTS We would like to express our sincere gratitude to the California League of Food Processors and the members of the Tomato Research Committee for their generous support of years of research on the peelability of processing tomatoes. REFERENCES BARRETT, D.M., GARCIA, E. and MIYAO, G. 2006. Effects of fruit defects on peelability of processing tomatoes. J. Food Process. Pres. 30, 37 45. BRANDT, J.A. and FRENCH, B.C. 1983. Mechanical harvesting and the California tomato industry: A simulation analysis. Am. J. Agr. Econ. 65, 265 272. GARCIA, E. and BARRETT, D.M. 2006. Evaluation of processing tomatoes from two consecutive growing seasons: Quality attributes, peelability and yield. J. Food Process. Pres. 30(1), 20 36.

PEELING PERFORMANCE OF PROCESSING TOMATOES 55 JUVEN, B., SAMISH, Z. and LUDIN, A. 1969. Investigation into the peeling of tomatoes for canning. Israel J. Technol. 7, 247 250. S-PLUS. 2003. S-PLUS 6.2 for Windows. Insightful Corp., Seattle, WA. SAS. 2000. SAS Software, Version 8.01. SAS Institute, Cary, NC. USDA/ERS. 2003. U.S. processing tomatoes at a glance. URL http://www.ers. usda.gov/briefing/tomatoes (accessed November, 2003). VENABLES, W.N. and RIPLEY, B.D. 1998. Modern Applied Statistics with S-PLUS, 2nd Ed., Springer, New York, NY.