Within Plant Sampling Procedures Fruit Variation in Kiwifruit Vines

Annals of Botany 78: 289 294, 1996 Within Plant Sampling Procedures Fruit Variation in Kiwifruit Vines D. B. MILES, G. S. SMITH and S. A. MILLER The Horticulture and Food Research Institute of New Zealand, Ruakura Research Centre, Pri ate Bag 3123, Hamilton, New Zealand Received: 5 June 1995 Accepted: 4 January 1996 Costs of fruit assessment often dictate that only a sample of fruit is taken for analysis. To ensure the sample results are consistent with those expected from the whole fruit population, selected fruit must be representative of the fruit population under investigation and the sample estimate should be unbiased. Samples of fruit from kiwifruit (Actinidia deliciosa var. deliciosa) vines grown on T-bar trellis systems were harvested and the mass of each individual fruit within the sample recorded. The choice of an individual fruit was completely objective and consisted of systematically selecting fruit based on their hierarchical location within the vine. Fruit not included in the sample were also harvested and weighed. The fruit were assigned to a zone based on their position within the vine canopy. Comparison of the sample estimates with the expected values showed the sample gave a reliable estimate of the average harvest mass. Errors in sampling technique were identified by the comparison of the locations of the sampled fruits with the those from the expected sample set. These operator errors were divided into two classes; incorrect lateral choice within a cane and wrongly sampled fruits within an individual lateral. With respect to fruit weight, analysis of the two classes of error indicated that neither source had any significant effect on the reliability of the sample. 1996 Annals of Botany Company Key words: Fruit sampling, fruit variability, plant architecture, Actinidia deliciosa, kiwifruit. INTRODUCTION Studies have investigated the relative efficiencies of fruit sampling designs that are based on between and withinplant components of variance (Schultz, 1955; Schultz and Schneider, 1955; Marini and Trout, 1984). Those designs identified the optimal number of sample units at each experimental stratum necessary to achieve a specific level of confidence. Data collection costs for sampling designs vary in response to the resources required to sample between plants as compared to within plant. The cost of sample information may be optimized without the loss of accuracy (Schultz, 1955). Spatial dependencies between fruit and the effect of these on sample design and methods of estimation have been studied (Monestiez, Habib and Audergon, 1990; Audergon, Monestiez and Habib, 1991, 1993; Habib et al., 1991). The existence of systematic variability within the plant has been demonstrated further and quantified by Smith et al. (1994). Sampling in the presence of these systematic trends, specifically, harvesting fruits from selected positions that are not truly representative of the overall vine, leads to a biased sample. Without estimation procedures that acknowledge this bias, results based on estimates from these samples provide misleading information. For example, data presented by Smith et al. (1994) suggest that the standard method for assessing harvest Brix in fruit from kiwifruit [Actinidia deliciosa (A. Chev.) C. F. Liang et A. R. Ferguson var. deliciosa] vines by selecting fruits in the upper canopy (Harman, 1981) is likely to result in an overestimation of the vine Brix level. Parts of a vine that, when harvested, 0305 7364 96 090289 06 $18.00 0 provide satisfactory estimates of harvest yield variables for the whole vine have been identified (Henzell, Gravett and Allison, 1994). In combination, these studies highlight the need to examine the process of physically sampling the required number of fruit from a population spatially distributed about the vine canopy. The object of the study, therefore, has been to illustrate the process of sample design and investigate the practicability and accuracy of the field application of a suggested sampling scheme. MATERIALS AND METHODS Plant material Measurements were made during the 1993 1994 growing season. Five 8-year old kiwifruit vines (cultivar Hayward grafted on seedling rootstock) growing near Hamilton, New Zealand on a Horitiu sandy loam soil (Typic Vitrandept thermic) were used in the study. The vines were supported on T-bar trellises (Sale, 1990) and planted in rows 5 m apart. Each vine occupied an area of 25 m. Early in the season, approximately 6 weeks after anthesis, fruit were thinned according to commercial practice (Sale, 1990) by removal of small or misshapen fruit and the lateral (side) fruit from double or triple clusters. A method similar to Smith, Curtis and Edwards (1992) was used to record the vine information. Instead of a theodolite, a modified Polhemus-3-Space-Fastrak system was used to map the three-dimensional coordinates of the trunk, cordon, fruiting canes and laterals, and fruit (Smith and Curtis, 1996). 1996 Annals of Botany Company

290 Miles et al. Fruit Variation in Kiwifruit Vines S 1 2 N most occasions, however, cost of assessment often requires that only a sample of fruits are harvested from each vine. 2 FIG. 1. Schematic representation of a kiwifruit vine trained on a T-bar trellis showing the division of the canopy into various combinations of height (Bottom Middle Top), side (West East) and axial (Central Outer) zones. The canopy was divided into zones (Smith et al., 1994). The horizontal plant axis was divided into a central zone and an outer zone. The outer zone combined fruit from the north and south ends of the canopy. Three zones were also defined along the vertical plant axis. East and west fruit were separated (Fig. 1). The date of harvest was determined using the standard sampling procedure of Harman (1981). Fruit were picked once the average Brix reading for a sample of fruit from adjacent vines had exceeded the industry standard of 6 2% soluble solids. A sample of fruit from each experimental vine was then harvested. Each fruit was weighed and packed into single-layer wooden boxes, as used commercially; these were then placed in a coolstore at 0 C for 12 weeks. The remaining fruit were strip-picked and individually weighed. All fruit were harvested on the same day. The fruit were stored at 0 C for 12 weeks. Sample definitions The design identifies the experimental material required for the sample (numbers of plants and fruit) along with a method of estimating the characteristic of interest. Sampling scheme refers to the procedure or instructions that define the selection of individual fruit from a plant canopy. The sample bias or accuracy is a measure of the departure of sample estimates averaged over a large number of independent repetitions from the expected population parameter (Cochran, 1977). The precision gives an indication of the repeatability of the sample design and measures the size of deviations from the mean of repeated applications of the sampling procedure (Cochran, 1977). Sample design For a given precision, the number of sampled units depends on the intrinsic variability of the sample material. If the whole vine is harvested only one variance is relevant for comparing differences in the harvest variates, total number of fruit, total weight, average fruit weight etc. There is complete information for each vine so the within vine variation does not contribute to the experimental error. On 4E 3E 5E 4E Sample size One objective of the trial was to examine differences between canopy zones. The appropriate error from the analysis of variance to examine these comparisons involves two components of variance, the vine zone interaction, σ vz, and within zone or the sampling error, σ e. The between vine component, σ v, is not required since the comparison does not involve vine differences. The true difference δ between any zones may be expressed as δ d t α / σ d where d is the experimental difference, t α the upper tail / probability level from the Student s t distribution and σ d an estimate of the standard error of the difference between zones. The detectable difference in average quality between canopy zones is then not less than t α / 2 r σ vz σ e γ where r is the number of vines and γ the number of fruit sampled from each canopy zone. An increase in r or γ will lead to an improvement in the precision of the estimate obtained from the sample. Sampling scheme When predicting the mean response, standard estimation methods such as the sample average require the selection of an unbiased sample. An unbiased sample can be partially assured by a random selection of fruit. However, the costs involved in taking a true random sample of fruit from the vine may be excessive as such an operation would involve numbering all the fruit and then drawing a random sample. A subjective random sample will be inadequate as the sample will generally contain too many central and too few extreme values. It is important, therefore, that a sampling scheme contain a random element; be completely objective in the selection of fruit and be simple, flexible, easy to understand and apply. Smith et al. (1994) investigated fruit quality as it relates to canopy position and found strong systematic trends within the vine structure. The most significant trends were found to be; distance E or W of the cordon; displacement of the cane from the trunk in a N or S direction; height of fruit above ground; inner s. outer fruit and difference between E and W sides. Non spatial factors were thickness and length of canes, especially differences between fruit from canes and spurs. The structure of the vine can be used to draw a quasirandom sample that approximately covers these trends. If the vine is represented as a hierarchical system of canes, laterals within canes and fruit within laterals, then a systematic sample of fruit can easily be obtained from

Cane 1 Cane 3 Cane 5 (1,1) (2,2) Miles et al. Fruit Variation in Kiwifruit Vines 291 (1,1) Cordon means of accommodating any systematic trends in fruit quality along an individual lateral that may bias the sample. Non-fruiting canes are excluded from the sample population. NF FIG. 2. The sample set corresponding to the initial sampling position of cane 1, lateral 1 and fruit 1 [i.e. (1,1,1)]., Identifies the initial fruit sampling position for the whole vine. Labelled fruit, e.g. (1,1) or (2,2) give the initial fruit sampling position for each cane., Highlight those fruit selected for the sample. NF indicates non-fruiting laterals and is an example of an identified position in the sampling sequence where no fruit was present. Only sampled canes are presented. within this structure. For example, taking every second cane, every second lateral on each selected cane and sampling every second fruit on each chosen lateral. A sample is uniquely determined by an initial fruit sampling position. The current example defines eight starting positions identified by (cane no., lateral no., fruit no.) of (1,1,1), (1,1,2),, (2,2,2). These represent mutually exclusive samples, one of which is selected at random. The sampling scheme defined by the initial fruit sampling position of (1,1,1) is illustrated in Fig. 2. The highlighted combination of canes, laterals and fruit results in a sampling fraction of 1 8. Depending on the variable of interest and number of fruit required, the sampling fraction can be adapted; e.g. every third or fifth cane, or take all fruit from a selected lateral. The choice of the appropriate combination of canes, laterals and fruit, is determined by the variation between canes along the cordon, between laterals on individual canes and between individual fruit on laterals (Smith et al., 1994). For example, an attribute with a small between lateral component of variance may be sampled by harvesting all the fruit from relatively few laterals. The position of the sampled fruits within an individual cane is uniquely determined by the initial fruit sampling position. These are identified by a lateral and fruit number. For example, in Fig. 2, on the first selected cane, cane one, sampling begins with the first fruit on the first lateral (1,1), on the second selected cane, cane three, the second fruit on the second lateral (2,2) etc. Counting from the initial sample lateral, every second fruiting lateral within the current cane is selected. From these laterals every second fruit is selected. So in Fig. 2, on cane one, the first, third and fifth fruits (if they exist) are sampled from each of the first, third and fifth laterals. On cane three, the second, fourth, etc. fruits are sampled from the second, fourth, etc. numbered laterals. The initial fruit sampling position alternates on adjacent canes. For example, if the initial sampling position on the first cane is identified by the number pair (1,1), then sampling begins on the next selected cane at position (2,2). Other combinations, e.g. (1,2) or (2,1) are similarly transformed to (2,1) and (1,2), respectively. This is a simple NF RESULTS AND DISCUSSION Sample size The number of vine replicates was fixed at five and the estimates of the components of variance for fruit harvest mass of, σ v 8 45 g, σ vz 7 5 g and σ e 28 8 g were obtained from previous field trials. Generally, the variation between parts of the vine has been found to be 2% lower than the variability between whole vines. The sample detection limit, with respect to the fruit harvest mass, indicated no substantial improvement in precision by selecting more than 15 fruit per zone (Fig. 3). To illustrate the merits of selecting more vines, fixing the number of sampled fruit at 15 per zone and increasing the number of vines from five to ten would result in a relative increase in precision of approximately 30%. Whilst for fruit harvest mass the variability is predominantly located within zone, other quality variables exhibit a more uniform variance decomposition (Table 1). A decrease in within zone variability leads to a reduction in the required number of sampled fruit per zone. For example, in comparison to harvest mass with a detection limit of 20% Precision (%) 50 30 20 10 0 10 20 30 Sample size FIG. 3. The detectable difference between canopy zones as a function of the number of sampled fruit per zone expressed as a percentage of the mean harvest mass. Lines represent significance levels of ( ) 10%, ( ) 5% and (------) 1%, respectively. TABLE 1. Between and within zone coefficients of ariation for arious quality attributes Variable (at harvest) Between vine Coefficient of variation (%) Between zone (within vine) Within zone Mass (g) 9 5 8 4 32 3 Brix 12 5 4 8 8 2 Flesh firmness 5 9 5 9 11 4

292 Miles et al. Fruit Variation in Kiwifruit Vines TABLE 2. Population and sample characteristics for har est mass (g) in each canopy zone Height zones Axial zones Side zones Mean no. of fruit Sampling fraction Population mean Sample mean S.E. of mean Top Inner 51 2 26 90 89 96 19 2 49 Outer 86 6 25 76 102 26 1 75 Middle Inner West 63 2 21 96 85 95 15 3 50 East 70 8 21 97 93 97 11 1 72 Outer West 91 4 22 101 51 103 93 2 19 East 113 0 27 105 52 106 20 1 47 Bottom West 59 2 21 102 50 108 30 2 49 East 48 6 27 99 50 15 2 24 Whole vine 592 6 23 6 99 58 101 23 0 75 Laterals 300 250 200 150 50 A was required. To obtain a sample of approximately one quarter of the fruit, the scheme consisted of harvesting every second fruit from every second lateral from all fruiting canes. A different fraction could be similarly defined for the remaining zones. A constant sampling fraction, however, has several advantages, it keeps the field application of the procedure simple and with respect to the individual zone profiles provides a self weighting estimate of the overall vine mean. For example, with significant trends in fruit attributes within the vine and an uneven distribution of fruit about the vine canopy, an arithmetic average of the zone means is not an unbiased estimate of the true vine level. Fruit 800 700 600 500 0 300 200 0 1 2 3 4 Vine B 0 1 2 3 4 Vine FIG. 4. A, The accuracy of the sampling scheme with respect to lateral choice. ( ) Represents the expected sampled laterals defined by the vine mapping, ( ) the remaining laterals and, ( ) the physically sampled laterals (see text). B, The accuracy of the sampling scheme with respect to fruit choice. ( ) Represents the expected sampled fruits given a lateral choice, ( ) the remaining fruits and, ( ) the physically sampled fruits. for a sample of 15 fruit per zone, a difference of 11% in average Brix can be detected at the 5% level of significance by sampling only five fruit per zone (Fig. 3). Sampling scheme The smallest number of fruit were located in the two bottom and inner top zones (Table 2) and suggested that to select 15 fruit from these zones a sampling fraction of 0 25 5 5 Sample accuracy A proportional within zone sample is necessary to estimate the vine mean. The height of fruit above ground is an important factor affecting fruit weight (Smith et al., 1994), therefore, a non-uniform sampling fraction particularly between height zones represented a possible source of sample bias. The deviations of the sample fraction from the expected vine fraction were modelled as coming from a χ distribution with seven degrees of freedom. The hypothesis of a uniform sampling proportion between different zones is rejected if the test statistic is greater than 14 07. The value of 1 86 obtained is clearly non-significant and indicated there was no evidence to confirm the existence of a variable zone sampling fraction and therefore confirmed that the variation in fruit harvest mass along canes was adequately covered by the sampling procedure. For fruit harvest mass, there were no significant differences on both the zone and whole vine level between the sample mean and the corresponding average based on the population of fruit within the area of interest (Table 2). Operator error Since the relative position of individual fruit within the vine structure was known the accuracy of the sampling scheme with respect to the field application could be assessed. From an average population of approximately 54 fruiting canes per vine the number of missed canes ranged from zero to seven. With respect to the choice of sample laterals within a cane, on vine three for example, of a total of 218 fruiting laterals, 106 laterals were identified as forming a correctly obtained sample (Fig. 4A). The sample

TABLE 3. Mean alues of fruit har est mass (g) from laterals that were incorrectly chosen Vine Included Omitted Difference 1 91 83 92 30 0 47 2 43 97 05 3 38 3 116 69 111 33 5 36 4 109 19 106 41 2 78 5 87 89 88 71 0 82 included 105 laterals, therefore, the proportion of laterals sampled was approximately correct (99%). However, of these sampled laterals, only 92 (88%) were laterals indicated by the sampling scheme. These departures from the correct application of the sampling scheme resulted in errors of two types. Specifically, fruit intended for the sample were omitted, whilst other fruit were incorrectly included. A comparison of the mean weight of fruit from laterals that were incorrectly omitted with that of fruit from laterals incorrectly included in the sample (Table 3) indicated no significant differences when all vines were considered. Therefore incorrect lateral choice did not bias the sample. The selection of fruit was similarly assessed (Fig. 4B), and again, incorrect choice did not bias the sample. Miles et al. Fruit Variation in Kiwifruit Vines 293 Expected frequency Expected frequency 120 80 60 20 120 80 60 20 0 90 A B 92 94 96 98 102 104 106 Harvest mass (g) 108 Sources of bias Presence of a systematic source of sampling error will lead to biased estimates. Analysis of the two identified sources of error showed that operator technique had not biased the sample. This result suggests that any errors arising from an attempt to correctly apply the sampling scheme will be random or haphazard and therefore do not affect the sample properties. In the current context, sources of systematic error that may affect the sample include harvesting only the first or last fruits from selected laterals or by only selecting laterals located in the upper vine canopy (Fig. 5A). As compared to a completely random selection of fruit from the whole vine, a sample from only the upper vine canopy would be expected to be biased by approximately 2 4 g in mean harvest mass. A similar difference would be observed between sampling from only the first or last fruits on a fruiting lateral. These results illustrate the effects of sampling either from an area that is not representative of the population of interest or application of a scheme that does not cover existing trends in fruit quality within the vine canopy. Commonly more than one fruit attribute is measured. If the standard method of assessing average soluble solids concentration for fruit from kiwifruit vines (Harman, 1981) is considered, after storage, a difference of 0 16 Brix exists between the randomly sampled fruit and those sampled from the upper canopy (Fig. 5B). This indicates that the mean soluble solids concentration will be overestimated on 94 6% of all occasions by sampling fruit from only the upper canopy. The difference between these results and those presented for harvest mass (Fig. 5A) highlight sampling difficulties when more than one fruit attribute is 0 11.2 11.4 11.6 11.8 12.0 12.2 Soluble solids concentration ( Brix) 12.4 FIG. 5. A, Expected frequency distribution of sample means (20 fruit) for harvest mass from various vine parts. Values represent the expected count of the mean harvest mass from 0 repeated random samples falling into intervals of 0 5 g. ( ) Represents the whole vine, ( ) the upper canopy (zones 1 and 2), ( ) the first lateral fruit and, ( ) the last lateral fruit. B, Expected frequency distribution of sample means (20 fruit) for Brix from various vine parts. Values represent the expected count of the mean Brix from 0 repeated random samples falling into intervals of 0 025 Brix. ( ) Represents the whole vine and, ( ) the upper canopy (zones 1 and 2). measured sampling from an area that is representative for one attribute may yield a significantly biased result for another. CONCLUSION Sampling from a population of fruit requires an objective sampling scheme with a fruit selection policy designed to cover trends in fruit quality within the vine canopy. The scheme presented in this paper gave a reliable estimate of fruit harvest mass. Analysis of operator errors in fruit selection also confirmed the reliability of the sampling scheme. Trends within the canopy differ between fruit attributes, therefore when more than one fruit characteristic is measured the choice of the sampling scheme must be carefully considered. Unlike a scheme based on the selection of fruit from a particular plant zone, the sampling scheme presented in this paper covered the variation within the whole vine, therefore, estimates of the mean level for other fruit attributes should have similar properties to those presented for fruit harvest mass. The kiwifruit vine was chosen as the experimental plant

294 Miles et al. Fruit Variation in Kiwifruit Vines because of the wide variation in fruit growth and quality that is known to occur within the canopy. The sampling methodology can, however, be equally applied to other crops that exhibit similar trends in fruit quality within the plant structure (Ah Chee and Mowat, 1993; Broom, 1996). ACKNOWLEDGEMENTS We thank John Curtis for his assistance with the mapping and data manipulations, Isabelle Gravett for her invaluable advice and the staff of the Plant Physiology Group, Ruakura Research Centre, for help with sampling the vines and processing the fruit. LITERATURE CITED Ah Chee A, Mowat AD. 1993. Variability in fruit quality of persimmon (Diospyros kaki L. cv. Fuyu). Australasian Post Har est Conference Proceedings: 41 44. Audergon J-M, Monestiez P, Habib R. 1991. Sampling in a fruit tree: a new concept applied to an apricot tree. Acta Horticulturae 293: 685 690. Audergon J-M, Monestiez P, Habib R. 1993. Spatial dependences and sampling in a fruit tree: a new concept for spatial prediction in fruit studies. Journal of Horticultural Science 68: 99 112. Broom FD. 1996. Spatial ariation, bitter pit and the quality of indi idual Braeburn fruits. DPhil thesis. University of Waikato. Cochran WG. 1977. Sampling techniques, 3rd edn. New York: John Wiley & Sons. Habib R, Tisne-Agostini D, Vanniere M, Monestiez P. 1991. Geostatistical method for independent sampling in kiwifruit vine to estimate yield components. New Zealand Journal of Crop and Horticultural Science 19: 329 335. Harman JE. 1981. Kiwifruit maturity. The Orchardist of New Zealand 54: 126 127, 130. Henzell RF, Gravett IM, Allison PA. 1994. Evaluation of sampling methods harvesting non-treated and hydrogen cyanamide treated kiwifruit vines. Scientia Horticulturae 58: 17 30. Marini RP, Trout JR. 1984. Sampling procedures for minimising variation in peach fruit quality. Journal of the American Horticultural Society 109: 361 364. Monestiez P, Habib R, Audergon J-M. 1990. Geostatistics, spatial dependencies in a tree: a new approach in fruit tree studies. Acta Horticulturae 276: 257 263. Sale PR. 1990. Kiwifruit growing. Wellington, New Zealand: GP Books. Schultz EF. 1955. Optimum allocation of experimental material with an illustrative example in estimating fruit quality. Proceedings of the American Society for Horticultural Science 66: 421 433. Schultz EF, Schneider GW. 1955. Sample size necessary to estimate size and quality of fruit growth of trees, and per cent fruit set of apples and peaches. Proceedings of the American Society for Horticultural Science 66: 36 44. Smith GS, Curtis JP. 1996. A fast and effective method of measuring tree structure in 3 dimensions. Acta Horticulturae (in press). Smith GS, Curtis JP, Edwards CM. 1992. A method for analysing plant architecture as it relates to fruit quality using three-dimensional computer graphics. Annals of Botany 70: 265 269. Smith GS, Gravett IM, Edwards CM, Curtis JP, Buwalda JG. 1994. Spatial analysis of the canopy of kiwifruit vines as it relates to the physical, chemical and postharvest attributes of the fruit. Annals of Botany 73: 99 111.