Professional Documents
Culture Documents
A Case Study of A Decision Support System On Mango Fruit Maturity
A Case Study of A Decision Support System On Mango Fruit Maturity
A Case Study of A Decision Support System On Mango Fruit Maturity
Abstract
Mango fruit maturity can be difficult to determine from external attributes.
Assessment of parameters of fruit on tree (dry matter, internal flesh colour) relevant
to estimation of fruit maturity was undertaken with a handheld (near infrared
spectroscopic) system. Measurement error on dry matter was low (typical RMSEP
0.6% DM). Repeated measurements on the same individual fruit from 78 different
blocks across two farms demonstrated that each piece of fruit was on a similar, but
individual, maturation trajectory, with a time offset. The offset was presumably
related to date of pollination or environmental conditions around the fruit (e.g.,
inner or outer canopy). A non-linear indexed regression model, coupled with the use
of a ‘biological shift factor’, was used to describe the time series data. Estimated
biological shift factors were larger for dry matter than flesh colour, indicative of an
earlier change in dry matter, albeit at a lower rate. Differences between blocks
within a farm and between two farms were small, indicating the maturation
processes were independent of local conditions. This technique could be used to trace
the source of variation within a block (e.g., to location in canopy or plant water
status), towards the goal of reducing this variation, leading to crops of greater
uniformity.
INTRODUCTION
The readiness of the mango crop (Mangifera indica) for harvest is typically judged
using several complimentary approaches. Fruit shape and skin colour of fruit is useful for
only some cultivars, heat sums can be used to give an approximate window for harvest,
and fruit dry matter (DM) and flesh colour (FC) can be used in the final decision to
harvest (Diczbalis and Wicks, 1997; Johnson and Hofman, 2009). Fruit DM is an
indication of total carbohydrate content, with the DM of immature fruit increasing as
photosynthate is imported into the fruit. After harvest and during ripening, fruit DM is
effectively constant (with change mostly related to water loss by the fruit), although
starch reserves are converted to soluble sugars (TSS). DM at fruit harvest thus relates well
with TSS of fully ripened fruit (e.g., Subedi and Walsh, 2011). Mango fruit DM is thus an
index of maturity, but the level of DM that signifies harvest readiness is expected to vary
by growth conditions (canopy photosynthetic capacity). Mango FC is associated with
carotenoid content. The level of this attribute associated with maturity depends on the
cultivar (with some cultivars accumulating more of this pigment than others) and growing
conditions. Flesh colour also continues to change postharvest, during ripening. Therefore
specifications on both DM and FC attributes in relation to fruit maturity need to be
cultivar and possibly growing condition specific.
Near infrared spectroscopy (NIRS) can be used to assess the parameters of FC and
DM non-invasively, e.g., Sarawong et al. (2004), Subedi et al. (2006), allowing repeated
measures of fruit on tree, over time. Our group has been involved in the development of
both in-line and hand-held NIRS technology, routinely achieving an R>0.9 and a RMSEP
<0.8 for assessment of DM of mango fruit (independent of the calibration set). Given
acceptable measurement accuracy, the NIRS technology can be used to guide orchard
management decisions around timing of harvest, and as a tool supporting investigation
into the process of fruit ripening.
Much previous work on the rate of fruit development has been based on data
obtained using destructive techniques, using mean values over a number of individuals.
Modelling
Dry matter accumulation within fruit on tree was assumed to increase in an
196
asymmetrical sigmoidal fashion (Tijskens et al., 2006). This behaviour is often modelled
using the completely empirical Richard function (Richards, 1959; Martinez et al., 2008).
The same behaviour can, however, be described by an uncoupled autocatalytic reaction
(Tijskens et al., 2006). Similarly, change in fruit colour is frequently modelled using a
logistic function based on the usually observed sigmoidal behaviour (Tijskens et al.,
2008). A simple logistic equation was therefore used for both attributes, neglecting the
different curvatures at the start and the end of development for dry matter. When
expressed in terms of biological shift factor (for deduction, see Tijskens et al., 2005), the
behaviour can be described as shown in Equation 1.
Varmax Varmin
Var k v Varmax Varmin ( t t )
Varmin (1)
1 e
For dry matter analysis, Var is dry matter content (DM, in % w/w) and kv = kDM
(day-1), the rate constant of the process of DM accumulation. For FC analysis, Var is the
colour, and kv=kc (day-1), the rate constant of the process of colour development. For
both variables, t is time (days) and Δt the biological shift factor (days) relative to the
midpoint of the logistic behaviour. Subscript min refers to the lower and max to the upper
asymptotic value at -infinite time and +infinite time, respectively.
Due to the limited range in the data of both variables assessed (from stone-
hardening stage to commercial harvest), neither of the asymptotic values could be
estimated directly. In preliminary analyses, the minimum value for FC tended consistently
to a value of almost zero, and was therefore fixed to zero. For DM, this minimum value
was fixed to the value of 4. For both variables, the upper asymptote seemed to vary over
the individual fruit in a batch. This stochastic parameter was therefore considered in two
parts: one part that varies per fruit and another that is common to all fruit. The maximum
value observed per fruit (max(Var)IFN), assessed just prior to fruit harvest, was assumed
to be an indication for the first part. The second part, which is to be estimated, was then
added to this value per fruit to assess a first estimate of the maximum value attainable by
each piece of fruit (Varmax) (Eq. 2). The fixed component of DMmax (d.Varmax) was
empirically determined at a value of 10 to achieve the best results across all blocks and
both farms. When a fruit dropped off the tree before commercial harvest, the mean value
of DMmax was used for that particular fruit. Thus a slightly different value of DMmax was
used for each fruit.
Re-arranging Equation 2, the stochastic parameters (∆t and Varmax) are separated
(Eq. 3). The left side of Equation 3 contains all variation in end value (Varmax), while the
right hand side contains all the variation in biological time (t+Δt). This arrangement
ensures that in graphical representations all the biological variation accounted for by the
model is properly included, while the remaining (mainly measuring error) variation
remains visible as the width of the cloud of points around the simulated lines.
Var Varmin 1
Vars tan (3)
Varmax Varmin 1 e vk Varmax Varmin ( t t )
Statistical Analysis
All statistical analyses were conducted using the non-linear regression procedure
(‘nls’) of the package R. Rate constants (kDM and kc) were estimated in common for all
individuals within a block (fixed effects), while the biological shift factor Δt was
estimated for individual fruit (random effects). The normality of distributions was tested
using the Shapiro-Wilk test in the package R. All model development was conducted
197
using Maple (Release 12 Maplesoft, Waterloo Maple Inc., Waterloo, Ontario, Canada), a
software package for symbolic manipulation.
Application Development
The vis-NIR instrument performed well under field conditions (high ambient
temperature and light), courtesy of an internal referencing step associated with every
sample and the interactance optical geometry employed. Applications (data not shown)
for non-invasive assessment of mango fruit DM and FC include consideration of fruit
maturity in context of (i) within tree variation (guiding decisions on whether to ‘strip’ or
selectively’ harvest fruit), (ii) selective picking effectiveness (comparison of harvested to
remaining fruit); (iii) within orchard variation (towards a decision on timing of harvest);
(iv) repeated measures over time to estimate rate of change (towards a decision on timing
of harvest); (v) analysis of incoming lots at the pack house as quality control on harvest
timing; and (vi) development of postharvest knowledge (e.g., relationships between DM
at harvest and fruit shelf life, and final ripe eating quality).
These practical applications are underpinned by an understanding of the sources of
variation in fruit maturation, and of the rate of fruit maturation. Therefore our attention
focussed on the modelling of maturation, as indexed by DM and FC.
198
likely that these differences are a coincidence of sampling and of measurement variation
in the data.
The effect of data standardisation, as described in Equation 3, was minimal
(Figs. 1 and 2). This comparison indicates that the effect of the relatively small variation
in asymptotic end value is not very important from a statistical point of view. The
remaining variation is ascribed to measurement errors, including the uncertainty of the
vis-NIR measurement.
The mean values of the biological shift factor (Δt in Table 1) differed between
blocks by up to about 10 d for both quality aspects. This result is consistent with an effect
by location in the orchard on the average maturity stage of the fruit. Standard deviations
of the factor for DM were remarkably similar for all blocks (Table 1), while that of colour
was prone to larger differences. The latter result may be a consequence of the lesser
reliability of the colour data (more technical error). The mean biological shift factor can
therefore be used as an indicator to map the average maturity per location over seasons,
while the standard deviation can be used to decide on the need for repeated harvests.
All distributions of the biological shift factor appeared to be normally distributed
as indicated by the p-value (Shapiro-Wilk test, p[Δt]>0.05) (Table 1). The biological shift
factor (Δt) for DM ranged from -17 to -35, while the standard deviation over all values
(σ[Δt]) was quite large and very variable (varying from 2.4 to 8.7 days across blocks).
The biological shift factor for FC ranged from about -66 to -39 days across all blocks,
while its standard deviation ranged from 3.3 to 10.5 days. Thus quite a large difference in
maturity existed between blocks.
CONCLUSION
Mango fruit DM was reliably assessed with the vis-NIR system, while FC was
more prone to measurement error. Repeated, non-destructive measurement of fruit on-tree
is useful in assessing the maturity of individual fruit, harvest batches and even whole
orchards. While the approach of Jordan and Loeffen (2013) can be used to judge harvest
readiness using such data, indexed non-linear regression analysis, as reported in this
study, can be used to trace the source of variation in a batch of fruit or an orchard, given
appropriate sampling strategies. Such information could be used to reduce this variation,
leading to fruit crops of greater uniformity of maturity.
Variation between fruit across the orchard can be attributed to differences in
development stage, most probably modulated by date of flowering and the amount of
light and temperature received. The estimated biological shift factors for DM were much
larger (less negative) that for colour (Tables 1 and 2). Thus DM changes earlier than
colour, although at a lower rate. No appreciable differences were found between two
different farms, a result that indicates either the maturation processes are generic, and
199
mostly independent of local circumstances, or that the management conditions of the two
farms were very similar.
ACKNOWLEDGEMENTS
The support of OneHarvest P/L, Hortical P/L and Horticulture Australia Ltd. is
greatly appreciated, and discussions with O. van Kooten and H. Schepers gratefully
acknowledged. Technology development partner, Integrated Spectronics P/L of Sydney,
Australia, has ceased trading, with related technology now pursued through CID Inc.,
USA.
Literature Cited
Diczbalis, Y. and Wicks, C. 1997. Mango - Harvesting time - Northern Territory. Gordon,
N.S.W.: Horticultural Research & Development Corporation, viii, 28 p.30 CM.
Bibliography: 20-22.
Jordan, R.B. and Loeffen, M.P.F. 2013. A new method for modelling biological variation
using quantile functions. Postharvest Biol. Tech. 86:387-401.
Johnson, G.I. and Hofman, P.J. 2009. Postharvest technology and quarantine treatments.
In: R.E. Litz (ed.), The Mango: Botany, Production and Uses. 2nd edition. CABI.
Martinez, A.S., González, R.S. and Terçariol, C.A.S. 2008. Continuous growth models in
terms of generalized logarithm and exponential functions. Physica A: Statistical
Mechanics and its Applications 387:5679-5687.
R Development Core Team. 2005. A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna. (http://www.R-
project.org).
Richards, F.J. 1959. A flexible growth functions for empirical use. J. Exp. Bot. 10:290-
300.
Saranwong, S., Sornsrivichai, J. and Kawano, S. 2004. Prediction of ripe-stage eating
quality of mango fruit from its harvest quality measured nondestructively by near
infrared spectroscopy. Postharvest Biol. Technol. 31:137-145.
Schouten, R.E., Jongbloed, G., Tijskens, L.M.M. and van Kooten, O. 2004. Batch
variability and cultivar keeping quality of cucumber. Postharvest Biol. Technol.
32:299-310.
Subedi, P.P., Walsh, K.B. and Owens, G. 2006. Prediction of mango eating quality at
harvest using short-wave near infrared spectroscopy. Postharvest Biol. Technol.
43:326-334.
Subedi, P.P. and Walsh, K.B. 2011. Assessment of sugar and starch in intact banana and
mango fruit by SWNIR spectroscopy. Postharvest Biol. Technol. 62:238-245.
Tijskens, L.M.M., Heuvelink, E., Schouten R.E., Lana, M.M. and van Kooten, O. 2005.
The biological shift factor. Biological age as a tool for modelling in pre- and
postharvest horticulture. Acta Hort. 687:39-46.
Tijskens, L.M.M. and van Kooten, O. 2006. Theoretical considerations on generic
modelling of harvest maturity, enzymes status and quality behaviour. International
Journal of Postharvest Technology and Innovation 1:106-120.
Tijskens, L.M.M., Konopacki, P.J., Schouten, R.E., Hribar, J. and Simčič, M. 2008.
Biological variance in the colour of Granny Smith apples. Modelling the effect of
senescence and chilling injury. Postharvest Biol. Technol. 50:153-163.
200
Tables
Table 1. Results of the indexed non-linear regression analysis of mango fruit dry matter (DM) accumulation per block, from data of
individual mango fruit on tree. Reported parameters relate to Eqn. 1. Parameters DMmax, σ[DMmax], kDM, and Δt were estimated for all
56 and 22 blocks of the two farms, with results for 17 and 8 blocks from farms 1 and 2, respectively, presented here.
Table 2. Results of the analyses on DM and colour using Equation 1 for data combined over all blocks for each farm.
201
201
Figures
Fig. 1. Dry matter plotted against calendar time (top left panel) and biological time (top
right panel), and standardised dry matter (Eq. 2 plotted against biological time
(t+Δt) (bottom left panel). Data are presented for one block of one farm (Block 14,
Farm 2). Each line and symbol represents an individual fruit, repeatedly assessed,
with 20 fruit assessed (four fruit from each of five trees from the block).
Measurements began at the stone-hardening stage of fruit development. Frequency
distribution of the biological shift factor (Δt) (bottom right panel) represents for all
fruit (four fruit per tree, five trees per block, 78 blocks across two farms).
202
Fig. 2. Change in fruit skin colour, with graph legends as for Figure 1.
203
204