Comparison of different methods of grapevine yield prediction in the time window between fruitset and veraison

摘要

Aim: To compare grape yield prediction methods to determine which provide the best results in terms of earliness of prediction in the growing season, accuracy and precision.Methods and results: The grape yields predicted by six models – one for use at fruitset (FS), two for use at veraison (V1 and V2), and three for use during the lag phase (LP40, LP50 and LP60) – were compared to field-measured yields. Regressions for the yield predicted by each model were constructed. The V1 and V2 models had the highest R2 (0.75) and efficiency index (EF; 0.67-0.71) and the lowest RMSE values (±16-17%, or veraison. This study provides predictive models that can be used at these different times of the growth cycle. IntroductionThe prediction of grape yields is necessary to prevent under- and over-cropping and thus help produce healthy plants and optimum amounts of fruit each season. Predicting yields accurately and early in the grapevine growth cycle is important since it allows adjustments of cluster load to be made (cluster thinning), thus promoting ripening and better grape quality. Yield predictions also allow wineries to determine their space, machinery and staff requirements during the harvest period. However, yields are affected by region, weather, soil conditions, cultivar, rootstock, vine heterogeneity, etc., and significant variations in vineyard yield may be recorded between years, and even between plants (Clingeleffer et al., 2001; Sabbatini et al., 2012); yields can, therefore, be difficult to predict.Grapevine reproductive structures present in one year begin their development in the previous growth cycle. For example, the cluster primordia of any year in question always begin their development at the end of spring/early summer inside the buds of the previous year. Their differentiation is halted during winter when the buds are dormant, but continues in the following spring over a short period just before budbreak (Howell, 2001; May, 2000 and 2004). Thus, grapevine reproductive behaviour is affected by the environmental conditions of both the present and previous year. This needs to be taken into account when vineyard management decisions are made.In recent years, a number of methods have been suggested for predicting vineyard yields. Some indirect real-time methods (Tarara et al., 2005) involve placing load cells on row support wires. Variations in the tension of the wire provide indications of the crop level at the moment of measurement. Such information on the dynamics of berry growth can be used to inform management decisions during the growth cycle. However, berry growth dynamics prior to ripening may vary greatly between years; this may require certain adjustments in any function used to predict yield (Tarara and Blom, 2009; Tarara et al., 2014).Several authors (Dobrowski et al., 2003; Dunn and Martin, 2004; Martínez-Casasnovas and Bordes, 2005; Nuske et al., 2011; Diago et al., 2012) have constructed models for making yield predictions based on digital, aerial or satellite images. All have shown good predictive capacity but require costly imaging and remote sensing operations. Vineyard's yields can, however, be predicted using more traditional methods based on yield components and information on phenological and climatological variables collected over the years. These methods require the inspection of the number of clusters per vine, the number of berries per cluster, and berry weight (Dunn, 2010; Sabbatini et al., 2012). The first two variables can usually be determined accurately by sampling at veraison, i.e., quite early in the growth cycle. The main source of variation lies in the predicted berry weight; the quality of any yield prediction is therefore strongly determined by how accurately this can be forecast.Berry weight can be anticipated in several ways. A relatively simple and commonly used method is to rely on historical data for average berry weight at harvest (Dami, 2006; Barajas et al., 2010). However, such method may not always be very accurate since it usually does not take into account all the variables that might affect a crop in any particular year. Other methods employ the idea that berry weight at a particular phenological stage is related to its final weight via a coefficient. Sabbatini et al. (2012) described a berry weight prediction method based on the idea that, during the lag phase, berry weight is approximately 50% of its final weight (Coombe and McCarthy, 2000). Thus, growers could predict yields at harvest by simply multiplying the number of plants by the average number of clusters, and multiplying this figure by double the average lag phase cluster weight (obtained by sampling). However, this requires the lag phase to be accurately identified (Sabbatini et al., 2012). Further, the 50% value suggested may differ from region to region.Barajas et al. (2010) and Nuske et al. (2011) suggested that yields can be predicted from the simple relationship between final berry weight