Spatial variation in ‘Hayward’ kiwifruit dry matter content within a growing region across seasons

Abstract

Between-orchard variation in fruit dry matter (DM) content at harvest of 341 commercial ‘Hayward’ kiwifruit orchards was spatially modelled across a production region over consecutive growing seasons (2003–08). Within seasons there were spatial patterns to the distribution of fruit DM between orchards. The temporal consistency of the spatial variation enabled segregation of the production region into two geographic zones which contained orchards that consistently produced fruit of distinct DM both within and across seasons. The differences in fruit DM between geographic zones were statistically significant, but were not of sufficient magnitude to be of commercial interest. Although the location of an orchard within the production region had an effect on the DM of the fruit produced, orchard location was not predictive of fruit DM. Therefore, it is concluded that in this situation zonation between orchards should not be where the effort in managing variation in fruit quality is concentrated.

Keywords:

• kiwifruit
• spatial variation
• fruit dry matter content
• zonal management

Introduction

Variation in fruit quality is a natural phenomenon that is influenced by a range of orchard and post-harvest factors. Improved understanding of how fruit quality varies within and between fruit populations is required to enable management of such variation. Knowledge of the nature of variation in fruit quality is potentially valuable as a research, production and ultimately marketing tool.

There is a direct link between ‘Hayward’ kiwifruit dry matter content (DM: the ratio of fruit dry weight to fresh weight expressed as a percentage) at harvest and consumer preference for fruit when ripe, with consumers preferring higher DM fruit (Lancaster 2002; Burdon et al. 2004). Supplying markets with fruit of consistently high DM is a major industry goal. Consequently, fruit DM at harvest is used as a measure of commercial acceptability within the New Zealand kiwifruit industry with growers paid more for the production of high DM kiwifruit.

At harvest, Hayward kiwifruit typically have DM in the range of 12–20%, with most fruit having a DM content in the range of 14–17% (Burdon et al. 2004). Variation in any kiwifruit quality characteristic is a natural phenomenon influenced by a range of pre- and post-harvest factors. A component of this variation arises from within the individual plant (Smith et al. 1994). Additional variability is introduced between individual plants, between areas within an orchard, and between geographically separated orchards by differences in management, site, plant material, environment and climate (Walton & De Jong 1990; Praat et al. 2003; Woodward et al. 2007).

An understanding of the magnitude and spatial component of within-region variation is of interest as such knowledge will help to focus management practices on minimizing the most significant sources of variation, and have implications for the potential application of approaches associated with precision agriculture. The applications of precision agriculture—the most common being zonal management—are dependent on the existence of consistent spatial variability in crop characteristics.

The objective of the present study was to quantify the magnitude and spatial component of between-orchard variation in fruit DM content within the Te Puke growing region across consecutive seasons to assess the potential for zonal management strategies. Two hypothesises were tested. First, orchards that consistently produce high or low fruit DM outcomes can be identified and that such orchards are spatially aggregated within the Te Puke growing region (consistent spatial variation). Second, the spatial aggregation of such orchards was temporally consistent with orchard location being predictive of orchard DM outcomes at commercial harvest across years (consistent spatio-temporal variation). The identification of consistent spatio-temporal variation in DM across the growing region could enable crop management at the geographic level. At the orchard level, the management programmes of orchards located within lower performing areas could focus on practices known to favour the production of higher DM fruit such as summer trunk girdling and low vigour canopy management (Thorp et al. 2003; Patterson & Currie in press). At an industry level, harvest planners could target orchards within geographic zones to source fruit of distinct characteristics for specific markets.

Materials and methods

Study area

The study utilized harvest information collected from March 2003 through to June 2008 with Actinidia deliciosa (A. Chev.) C. F. Liang et A. R. Ferguson var. deliciosa ‘Hayward’ kiwifruit harvested from 341 conventionally managed commercial orchards in the Te Puke region (37°49′S, 176°19′E), New Zealand (Fig. 1A).

Figure 1  A, Geographic location of the Te Puke kiwifruit growing region within New Zealand; B, location of model orchards (○) and test orchards (▴) within the Te Puke growing region.

In the New Zealand kiwifruit industry, the majority of kiwifruit is grown in the Bay of Plenty province with the bulk of production centred within the Te Puke region, the climatic and soil characteristics of which have been previously described (Snelgar et al. 1993; Manson & Snelgar 1995; Salinger & Kenny 1995; Snelgar et al. 2007; www.gisportal.landcareresearch.co.nz ).

Data analysis

Industry databases pertaining to the 2003–08 harvest seasons were supplied by ZESPRI International Ltd (Mount Maunganui, New Zealand) and collated to obtain orchard geographic location, altitude and fruit DM content at harvest maturity clearance for conventionally managed ‘Hayward’ orchards located within the Te Puke growing region (Mowat & Kay 2007). Fruit DM data were based on 90 fruit samples taken randomly from orchard maturity areas and assessed for DM content using standard laboratory methodology (Burdon et al. 2004). An orchard maturity area represents a management unit within an orchard and may comprise a single orchard block or group of blocks that is harvested as a discreet unit (Mowat & Kay 2007). As the composition of an orchard's maturity areas may vary between seasons, individual fruit measurements across all maturity areas within an individual orchard were combined to produce an orchard average fruit DM value per season. Commercial harvest operations typically run from late March to early June for kiwifruit orchards within the Te Puke growing region, therefore the DM measurements were taken across a 10-week period.

The resulting data set consisted of two orchard groups within the Te Puke growing region: model orchards and test orchards (Fig. 1B). The model orchard dataset comprised of 288 individual orchards with 6 consecutive years of harvest DM information (2003–08) that were used to model the spatial distribution of between-orchard variation in fruit DM. The test orchard dataset consisted of 53 orchards lacking consecutive seasonal information which were used to assess the predictive power of the spatial models developed using the model orchards.

Geostatistics assumes normality in the variables modelled; Kolmogorov-Smirnov tests were used to confirm that orchard average fruit DM were normally distributed between orchards both within seasons and across seasons (data not shown).

For the spatial analysis, each season's DM measurements were normalized within the overall distribution for that season (µ = 0, σ = 1) to facilitate between-season comparisons independent of any seasonal effect. The spatial component of between-orchard variation in fruit DM was modelled by calculating variograms from normalized average orchard values for each season and then interpolated using block kriging with a global variogram, on to a 100 m grid with VESPER software (Minasny et al. 2005).

The temporal persistence of the spatial patterns of between-orchard variation in fruit DM across the Te Puke growing region was investigated using k-means clustering (Bramley & Hamilton 2004; Bramley 2005; Woodward et al. 2007). Interpolated normalized DM values for each geographic point in the interpolation grid were grouped into zones consistently producing fruit of distinct DM by k-means clustering, using SPSS v13 software (SPSS Inc, Chicago, IL, US), in which each season was used as a variate in the clustering.

Average fruit DM was compared between orchards located within each geographic cluster both within seasons and between seasons using one-way ANOVA with DM having two levels (one for each cluster group) in SPSS v13 software.

The validity of spatial models was assessed by a nearest neighbour analysis where interpolated predictions (derived from the model orchard dataset) were correlated with test orchard DM outcomes at commercial harvest. Pearson product moment correlation coefficients were calculated as orchard average DM at commercial harvest was normally distributed between orchards and across seasons.

Results

Summary statistics of between-orchard variation in fruit DM at harvest are presented in Table 1. Season 2007 produced the highest average fruit DM at harvest but DM values between orchards were the most variable. Conversely, season 2005 produced the lowest average fruit DM across orchards, but variability between orchard DM values were the least. By comparing the variation between seasons (temporal) with that within a season (spatial), the relative importance of both components of variation was estimated. Comparison of the measures of variance (Table 1) would suggest that variation between seasons (~temporal variation) was similar to the level of variation between orchards within seasons (~spatial variation).

Table 1  Summary statistics of Hayward kiwifruit orchard fruit dry matter content within the Te Puke growing region across consecutive seasons (2003–08). Measures of variation are between-orchard for each individual season and between-season for the 2003–08 figure.

Season	Average	Standard deviation	Coefficient of variation
2003	16.26	0.47	2.89%
2004	16.25	0.48	2.95%
2005	15.99	0.45	2.84%
2006	16.63	0.52	3.15%
2007	17.13	0.53	3.07%
2008	17.08	0.52	3.02%
2003–08	16.56	0.47	3.93%

Spatial techniques were used to provide a visual representation of fruit DM distributions between orchards across the growing region and to identify geographic zones consistently producing fruit of distinctly different DM at commercial harvest. Spatial structure in between-orchard variation in DM was evident in the six seasons investigated, along with some temporal consistency to these spatial distributions across seasons (Fig. 2). Within each individual season, lower DM values were typically associated with orchards located in the south of the Te Puke growing region and higher DM values with orchards located within the centre of the Te Puke growing region.

Figure 2  Modelled spatial variation between ‘Hayward’ kiwifruit orchards average fruit DM content at commercial harvest within the Te Puke growing region over consecutive growing seasons (2003–08).

The interpolated spatial predictions were clustered to identify geographic zones that contained orchards producing consistently different DM outcomes at commercial harvest across years.

Clustering by DM produced a two-cluster solution (Fig. 3). Model orchards located within the cluster 1 geographic area (located in the north east and south west of the Te Puke growing region) consistently produced fruit with a lower average DM at commercial harvest than model orchards located in cluster 2 both within seasons and across seasons (Table 2). The average altitude of orchards located within the north-east portion of cluster 1 was 15 masl (metres above sea level), compared with the 193 masl of orchards located within the south-west portion of cluster 1 (Table 3). The average altitude of orchards located within cluster 2 was 103 masl.

Figure 3  Geographic location of ‘Hayward’ kiwifruit orchard DM groupings across the Te Puke growing region as identified by k-means clustering. Clusters 1 and 2 contained 57 and 231 of the 288 model orchards, respectively. The fruit DM and altitudes of the orchards located within each geographic cluster are summarized in Table 2 and 3, respectively.

Table 2  Comparison of average DM between orchards located within geographic clusters. Clusters 1 and 2 contained 57 and 231 of the 288 model orchards, respectively.

Season	Cluster 1	Cluster 2	Statistically significant?
2003	16.01	16.33	P<0.01
2004	15.93	16.33	P<0.01
2005	15.69	16.07	P<0.01
2006	16.41	16.68	P<0.01
2007	16.92	17.18	P<0.01
2008	16.80	17.15	P<0.01
2003–08	16.30	16.62	P<0.01

Table 3  Average altitude of model orchards located within geographic clusters.

Geographic cluster		Altitude (masl)
Cluster 1:	North-east zone	15
	South-west zone	193
Cluster 2		103

How predictive is orchard location of orchard DM at commercial harvest? The spatial models derived from the model orchard dataset were correlated with the actual harvest outcomes of the test orchards (Table 4). The predictive power of the spatial models varied with season; there was no significant correlation between predicted and actual DM values in season 2005, for all other seasons there was a statistically significant although weak correlation.

Table 4  Correlations between modelled spatial predictions and actual test orchard average fruit DM at commercial harvest within and between seasons (2003–08).

Season	Pearson correlation	R ^2
2003	0.403(*)	0.16
2004	0.458(**)	0.21
2005	0.133	0.02
2006	0.404(**)	0.16
2007	0.384(**)	0.15
2008	0.327(*)	0.11
2003–08	0.347(**)	0.12
*Correlation is significant at the 0.05 level (2-tailed)
**Correlation is significant at the 0.01 level (2-tailed).

Discussion

This study utilized a unique industrial dataset of fruit DM measurements from 341 individual orchards across 6 consecutive years; such a dataset enabled investigation of regional scale spatial variation in a perennial fruit crop rather than the within-production unit scale variation which is typically the subject of investigations into the potential applications of precision agriculture (Bramley & Hamilton 2004; Bramley 2005; Woodward et al. 2007; Aggelopoulou et al. 2010). Knowledge of spatial variation in orchard production characteristics and an interpretation of the causes of such variation is preliminary to the implementation of any zonal management strategies (Woodward et al. 2007). Furthermore, as production characteristics also respond to temporal variation between growing seasons, it is essential to assess the consistency of the spatial partitioning of the production region over time (Bramley & Hamilton 2004; Bramley 2005; Woodward et al. 2007; Aggelopoulou et al. 2010).

Within seasons there were spatial patterns to the distribution of fruit DM between orchards (Fig. 2). It must be noted that the magnitude of the between-orchard variation in DM is low in comparison with typical precision agriculture datasets. The magnitude of variation in fruit quality has previously been reported to be less than variation in fruit yield (Bramley & Hamilton 2004; Bramley 2005; Aggelopoulou et al. 2010). Variation in the current data set is low, as could be expected from a fruit-quality characteristic. In addition, the standard deviations in DM between orchards across the study period ranged from 0.45–0.53 (Table 1), considerably less than the typical within-orchard standard deviations in kiwifruit DM of around 1.0. This suggests that the magnitude of within-orchard variation in DM is greater than DM variation between orchards. Furthermore, the DM measurements used in this study came from commercial harvest clearance data and the harvest period typically extends over a 10-week period. During the later stages of the fruit growth period, DM is increasing ~0.05–0.1% per week (Snelgar et al. 2005). Thus the timing of harvest will have influenced an orchard's DM result, and management decisions such as the choice of harvest date are likely to have reduced variation between orchards. However, the dataset we used does represent the actual DM results for each orchard at harvest, and these in turn represent the actual DM profile and potential taste quality of the crop produced within the region.

The coefficient of variation in DM among 288 kiwifruit orchards ranged from 2.88–3.15% across seasons (Table 1), so although the maps in Fig. 2 illustrate a marked spatial structure they only represent a very low level of variation between orchards and overstate the magnitude of the actual spatial variation.

The temporal consistency of the observed spatial variation enabled geographic zones to be identified which contained orchards that consistently produced fruit of distinct DM both within and across seasons (Fig. 3). However, given that average DM values ranged between orchards within seasons by 2.69–3.72% DM, then the observed difference in DM between geographic clusters of 0.32% DM (Table 3), despite being statistically significant, is not of sufficient magnitude to either be of commercial interest or warrant the implementation of zonal management strategies at the geographic level.

It is important to note that the spatial structure of the geographic zones identified by k-means clustering (Fig. 3) was not an artefact of the clustering procedure. The clustering method partitioned geographic points based on interpolated fruit DM predictions without using any spatial information. Therefore, the resulting spatial structure of the clusters suggests a correlation with underlying variables that affected fruit DM. Potential variables include the influence of environment, management or a combination of both. In a comparison between kiwifruit varieties on the impact of environment versus management, it was concluded that environment has a greater influence on ‘Hayward’ DM compared with ‘Hort16A’, with DM in ‘Hort16A’ being more responsive to management compared with ‘Hayward’ fruit (Mowat & Kay 2007).

Temperature is a major driver of kiwifruit development (Salinger & Kenny 1995; Snelgar et al. 2005). Warmer spring and cooler summer growth temperatures are known to favour the production of higher DM kiwifruit at harvest (Snelgar et al. 2007). Across the Te Puke growing region there is an altitude gradient increasing from the north east through to the south west; the altitudes of the orchards included in this study varied by 280 m. From the literature we could expect fruit produced at higher altitudes to have experienced lower growth temperatures, resulting in fruit of lower DM content (Hopkirk et al. 1989; Snelgar et al. 2005).

The hypothesis that a difference in orchard altitude, and therefore growth temperatures, is driving the observed differences in DM between clusters is consistent with the majority of cluster 1 (lower DM) being located at higher altitudes in the south and west of the Te Puke growing region. However, this is at odds with the orchards forming the north-eastern portion of cluster 1 being at low altitude (Table 3). Therefore, orchard altitude may contribute to, but is not entirely responsible for, the differences in fruit DM produced between clusters.

Bramley (2001) has previously reported that patterns of grape quality variation in a vineyard closely matched variation in soil properties. As such, could differences in soil types across the Te Puke growing region be influencing between-orchard variation in fruit DM? This does not seem likely as soil maps of the Te Puke region indicate that soil type does not vary across the area studied (www.gisportal.landcareresearch.co.nz ). However, the north-eastern portion of cluster 1 borders coastal plains where soil drainage is known to be very different.

Conclusion

Can orchardists offset the impacts of environment and consistently produce high DM fruit year to year? What is the relative influence of environment versus management in determining kiwifruit DM? The answers to these questions will affect both the individual orchardist in how they approach the management of their individual orchards and the wider industry through the potential for applying zonal management strategies.

Temporally consistent spatial relationships were identified between orchards for fruit DM enabling effective zonation of the Te Puke growing region. Although the differences in fruit DM between such zones were statistically significant, they were not of sufficient magnitude to be commercially significant and warrant a change from uniform to zonal management.

We can surmise that the weak correlations between spatial predictions and actual orchard outcomes is a function of orchard management responding to vines and climate, layered on top of natural climatic variation and location effects. As a result, despite the location of an orchard within the Te Puke region having a limited effect on the DM of the fruit produced, the spatial component of between-orchard variation was exceeded by that of non-spatial site-to-site variation. This we attribute to differences in individual management practices between orchards. Patterson and Currie (in press) reported orchard management practices to have a greater influence on kiwifruit DM than seasonal influences. Overall, orchard location was not predictive of orchard DM and therefore geographic zonation between orchards should not be where the effort in managing variation in ‘Hayward’ kiwifruit DM is concentrated.

Acknowledgements

Tim Woodward was the recipient of a NZ Tertiary Education Commission Bright Futures Enterprise Scholarship and a ZESPRI Doctoral Fellowship. Industry data were made available by ZESPRI International Ltd. Project assistance was provided by The New Zealand Institute for Plant & Food Research Ltd, funded by the NZ Foundation for Research, Science and Technology (contract no C06X0202).