Predicting Harvest Labour Allocation in Bell Pepper Production

In the production of tomatoes in greenhouses, a more or less constant number of tomatoes are ready for picking at each harvest time. In the production of bell peppers, however, large fluctuations exist in number of fruits ready to pick. This major problem in bell pepper production is called ‘flushing’: weeks with high yields are alternated by weeks with low yield. This irregular harvest pattern makes it difficult for growers to meet regular weekly demands. As this cyclic production pattern is more or less the same for all growers, it results in weeks with a high market supply and low prices alternating with weeks with a low market supply and high prices. Flushing also results in strong fluctuations in labour demand in the greenhouses. Hence, avoiding an irregular pattern of sweet pepper production is of great economic importance. Flushing makes it very difficult for the growers to allocate sufficient labour force at a certain time in production. Both allocating too many and too few pickers will have financial consequences for the grower. The aim of this study is to predict the optimal harvest date based on repeated colour measurements of the growing bell peppers. The fruit ripening of bell peppers has been followed by colour imaging in the greenhouse. A commercial digital camera was used to record the development of colour for two cultivars of bell peppers over time. The images were analysed by image processing to obtain the R, G and B values. At the same time the time to harvest was recorded for the same fruits. Using these data a model was developed, on the basis of an exponential function towards a fixed asymptotic end value, to describe the colour development during fruit growth in terms of R/G ratio and to predict the optimal harvest date of the fruits. INTRODUCTION Unlike tomatoes, bell peppers (Capsicum annuum) show a highly irregular yield pattern in number of ripe fruits (Heuvelink and Körner, 2001). The presence of developing fruits inhibits subsequent set and growth of new fruits both by competition for limited assimilates as well as by dominance (ill-balanced control) due to the production of plant growth regulators (Marcelis and Baan Hofman-Eijer, 1997). This results in the (weekly) fluctuation of the number of harvestable fruits. Since these fluctuations can be quite large (see Fig. 1) it will become difficult to estimate the number of harvesters necessary at a certain point in time. By monitoring the ripening fruit, growing in a greenhouse, taking pictures with a digital camera and assessing the colour of the bell peppers on the picture, the development of ripening was followed. Based on these repeated colour measurements a model for the (colour) ripening is proposed. The model aims to predict the number of harvestable fruits for each week during a growing season. Of course there are still quite a number of difficulties, and the method is certainly not ready for commercial use right now, but the prospects of using commercial digital cameras for online monitoring of growth is promising. MATERIALS AND METHODS Preliminary experiments were conducted in 2001 on two cultivars: one red (exp. Type 444) and one yellow (Bosanova) colouring cultivars, with 4 samplings in spring and autumn season. More extended experiments were conducted in 2002 on the same two cultivars (red and yellow) for a complete season from early June until late November. 1435 Proc. 5 Int. Postharvest Symp. Eds. F. Mencarelli and P. Tonutti Acta Hort. 682, ISHS 2005 A commercial digital camera was used to picture the growing fruit. Later, the pictures were evaluated by computer imaging into the RGB values. By using a commercial digital camera under non-standardised conditions in the greenhouse, a large variation will be introduced in the data. Using a ratio of colour aspects greatly reduced this large variation. The ratio of R/G was therefore used to model the ripening fruits. The fruits were harvested when approximately red or yellow for 90% of the surface. The harvest date and date of incipient coloration were recorded. Additional information was gathered during the experiments, like e.g. average temperature and light, number of ripening fruits on the same plant, either above or below the actual fruit etc. However, all these data and information have not yet been analysed properly. RESULTS AND DISCUSSION Model Development and Data Analysis The ratio of Red to Blue (R/G) turned out to increase exponentially with growing time starting from an asymptotic value, which represent the most green colour that cultivar can obtain (data not shown). However, since the starting point for measurement in time is located somewhere during the season, the number of days to harvest is a much more elegant and appropriate way to express time. By this conversion, the time is simple reversed. The general equation for exponential behaviour can be described as: fix t k fix 0 Col e ) Col Col ( Col + ⋅ − = ⋅ − Eq. 1 with the time t represented as days to harvest, the rate constant k is positive, expressed against Julian dates, the rate constant is of course negative. Col is the colour ratio (R/G), indices 0 for initial and fix for the asymptotic value. Equation 1 describes the colouring process for each bell pepper. However, it is not very useful for describing the harvest date for flushes of bell peppers. For instance, not all bell peppers are harvested when they reach the same colour. Actually, the grower uses a target colour (Coltar). The colour of each bell pepper is assessed and harvested when exceeding Coltar. Furthermore, as bell peppers tend to be harvested in distinct batches (‘flushes’), it seems convenient to use a batch approach. This approach allows to describe and to include the biological variance in ripeness during growth. This approach has been used frequently (Tijskens, 1997, 2000; Schouten et al., 2003, 2004). Applying Coltar and the batch approach transforms Eq. 1 into Eq. 2: fix ) t t ( k fix tar Col e ) Col Col ( Col + ⋅ − = Δ + ⋅ − Eq. 2 Index tar stands for target. Coltar was arbitrarily set to .75, the value that the majority of individual fruit had reached at the moment of harvest. The time shift factor Δt is a stochastic variable, with a value for each individual in the complete batch. It represents the biological variance in reaching ripeness during growth (Tijskens et al., 2003). After intensive testing in preliminary data analysis, searching for additional effects of temperature, light, row and plant number, and the season (early, late), it was concluded that these factors do not really contribute to the variation in Δt (data not shown). So the stochastic variable Δt was only estimated for the fruit number, irrespective of previously mentioned factors. The analyses were conducted with the non-linear mixed effects procedure (nls) in the statistical package S-Plus (Mathsoft Inc, USA). The rate constant k and the asymptotic value Colfix were estimated in common for all individuals and all circumstances. In Table 1 the results of the analysis is shown for the parameters in common (Colfix and the rate constant k) and for the mean value of the stochastic variable Δt. The stochastic properties of Δt are shown in Table 2. The red cultivar clearly fits better to the model than the yellow cultivar Bosanova,