Predicting Key Phenological Stages for 17 Grapevine Cultivars (Vitis vinifera L.)

Transcription

1 Predicting Key Phenological Stages for 17 Grapevine s (Vitis vinifera L.) Diana Zapata, 1,2,3 * Melba Salazar-Gutierrez, 1 Bernardo Chaves, 1 Markus Keller, 2 and Gerrit Hoogenboom 1,4 Abstract: Weather conditions have a significant impact on crops, and temperature is one of the main factors that controls plant development. Thermal time models based on temperature have been applied to predict the development of many species. To implement these models, determination of an appropriate base temperature (T b ) is required to characterize the differences among developmental stages and cultivars. The goal of this study was to determine the unique T b and degree-days (DD) to predict budbreak, bloom, and veraison for 17 cultivars. T b s were estimated with the minimum variance method using phenological data collected over 23 years in Prosser, WA. T b increased throughout grapevine development and ranged from 6.1 to 8.4 C for budbreak, from 7.2 to 10.5 C for bloom, and from 9.4 to 12.8 C for veraison. Starting DD accumulation on 1 Jan and using the T b s estimated for each cultivar, the duration to budbreak ranged from 78 to 180 DD, from budbreak to bloom ranged from 240 to 372 DD, and from bloom to veraison ranged from 556 to 800 DD. Errors in prediction varied between 4.8 and 7.8 days to budbreak, between 1.9 and 5.5 days to bloom, and between 7.1 and 12.4 days to veraison. Based on the errors in prediction, models that used an estimated T b specific for a phenological stage performed better than models that had a fixed T b of 0 and 10 C. The estimated thermal time parameters provide a simple approach for characterizing differences among cultivars and assist growers and industry in implementing management practices through simple decision support tools based on thermal time models. Key words: decision support, degree-days, development, growth stages, phenology, phenophases Plant development is described through phenology, which refers to cellular differentiation and initiation of new plant structures or phenological stages and the ways in which these processes are influenced by environmental conditions (Chuine et al. 2013). The modified Eichhorn Lorenz (E-L) scale (Coombe 1995) describes the vegetative and reproductive stages of grapevine (Vitis vinifera L.) and identifies seven major phenological stages: budbreak, shoot development, flowering, fruit set, berries pea-sized, veraison, and harvest. Plant growth and development are primarily controlled by temperature and by the quantity of solar radiation intercepted. Temperature is the main factor that controls the rate of plant development as it affects essential biochemical processes and plays a role in determining the appearance of phenological stages and the length of the phenophases (Johnson and Thornley 1985, Kwon et al. 2008, Caffarra and Eccel 2010, Nendel 2010). 1 AgWeatherNet Program, Washington State University, Prosser, WA 99350; 2 Department of Horticulture, Irrigated Agriculture Research and Extension Center, Washington State University, Prosser, WA 99350; 3 present address: Department of Soil and Crop Science, Texas A&M University, College Station, TX 77843; and 4 present address: Institute for Sustainable Food Systems, University of Florida, Gainesville, FL *Corresponding author (cer.diana@gmail.com) Acknowledgments: This research was funded in part through support provided by the Washington Grape and Wine Research Program. The authors thank Lynn Mills for the collection of phenological data. Manuscript submitted Aug 2015, revised Apr 2016, accepted Aug 2016 Copyright 2017 by the American Society for Enology and Viticulture. All rights reserved. doi: /ajev Temperature-based phenological models, such as thermal time models, can play an important role in characterizing differences among species (Parker et al. 2013) and in predicting plant development under different environmental conditions (Caffarra and Eccel 2010). In addition, thermal time has been used to study the physiological processes that control the duration of phenological stages in grapevines (Duchêne et al. 2012). The degree-days (DD) model is a thermal time model based on the difference between daily mean air temperature and a threshold value known as the base temperature (T b ) (Arnold 1959). Due to the simplicity of its calculation and implementation, the DD model has been applied extensively for predicting the development of numerous species (Olivier and Annandale 1998, Snyder et al. 1999, Marra et al. 2002, Ojeda-Bustamante et al. 2004, Ruml et al. 2010). Physiologically, the T b is defined as the temperature below which plant development ceases. Above this threshold temperature, the accumulated DD are effective for the plant to reach a given developmental stage (Arnold 1959, Snyder et al. 1999). However, most thermal time models implemented so far have used the same T b for different phenological stages. In addition, it is known that T b can vary widely among cultivars of many crops (Baker et al. 2001, García de Cortázar-Atauri et al. 2009, Ruml et al. 2010, Ma et al. 2012, Vršič et al. 2014). The changes in temperature that occur during the growing season suggest that plants require specific conditions for each developmental stage; these conditions could be determined by estimating T b. Furthermore, accurate prediction of the critical phenological events during a crop life cycle requires determination of the most appropriate T b (Hoover 1955, Parker et al. 2013). 60

2 Predicting Key Phenological Stages for 17 Grapevine s 61 Experimental determination of T b is laborious as it requires evaluation for a wide range of temperatures to obtain accurate values. Moncur (1989) estimated T b for budbreak and leaf appearance of grapevine cultivars under controlled conditions and found differences between phenological stages; however, the T b s were obtained by extrapolation outside the estimated range, which increased the uncertainty. Several methods for estimating T b use phenology and temperature data collected under field conditions, and selection of T b is based on finding the temperature that provides the minimum variability in DD (Yang et al. 1995). For example, Oliveira (1998) estimated T b for budbreak and flowering for V. vinifera cv. Touriga Francesa using different statistical approaches including the least standard deviation in DD, the regression coefficient, and the variation in days. Oliveira (1998) concluded that the least standard deviation method was most appropriate for estimating T b. Estimation of T b by the least standard deviation in DD also showed the lowest error in predicting veraison, compared with the regression coefficient and the development rate methods for four grape cultivars (Zapata et al. 2015). Overall, a T b of 10 C has been used widely and somewhat arbitrarily to describe grapevine growth and development (Buttrose and Hale 1973, van Leeuwen et al. 2008, García de Cortázar-Atauri et al. 2009, Vršič et al. 2014). However, the T b of 10 C was fixed as a standard threshold with the purpose of grouping climatic regions in California rather than being used to describe plant development (Amerine and Winkler 1944, Jones et al. 2010). Some studies found that a T b of 0 C for grapevine provided the best performance in terms of prediction errors in days and model efficiency (Nendel 2010, Parker et al. 2011). Lopes et al. (2008) arbitrarily assumed a T b of 3.5 C to determine thermal time for budbreak and assumed 10 C for later stages without distinguishing among cultivars. Jones and Davis (2000) identified budbreak, flowering, veraison, and harvest as the most important stages in grapevine development and indicated that the timing of these phenological stages varies with cultivar, climate, and geographic location. Therefore, accurate prediction of the occurrence of these stages can help with executing crop management practices and with scheduling labor requirements and equipment. For example, knowing when budbreak will occur would be useful for planning pruning that is needed to adjust crop load; after fruit set, leaf removal and cluster thinning are conducted to improve vine balance and berry size; before and after veraison, irrigation is applied to control crop growth, enhance desired fruit traits and ensure vine health; and throughout the season, pesticides are applied to minimize pest and disease pressure. After berry ripening, when sugar, titratable acidity, ph, and berry flavor reach desirable levels, machinery and equipment scheduling is required for both the vineyard and winery. In addition to their use as a tool in decision-making for grapevine management, phenological models can be used to select suitable sites for grape production, to optimize vineyard production systems for resource use, to estimate costs and environmental impact, to improve the overall long-term sustainability of grape production, and to maintain a balance between grape quality and quantity by executing management practices at the proper time (Campos et al. 2010, Fila et al. 2012). The goal of this study was to determine the T b and duration in thermal time required to reach budbreak, bloom, and veraison for different red and white grapevine cultivars and to use these parameters to predict the successive appearance of the key grapevine stages as a function of cultivar and temperature. Materials and Methods Weather and phenological data. Phenological data collected for 17 winegrape cultivars from 1990 to 2013 by the Viticulture Program at Washington State University, Prosser, WA, were used in this study. The selected red cultivars included Cabernet franc, Cabernet Sauvignon, Lemberger, Malbec, Merlot, Meunier, Pinot noir, Syrah, and Zinfandel; the white cultivars included Chardonnay, Chenin blanc, Gewürztraminer, Muscat blanc, Pinot gris, Sauvignon blanc, Semillon, and Riesling. The phenological data that were collected corresponded to the day of year (DOY) when a particular phenological stage, ranging from budbreak to leaf fall, was observed in the experimental vineyard (lat N; long W; 265 to 364 m asl). Three vineyards were planted in 1983, 1999, and 2010 in north south-oriented rows, with a vine spacing of 2.7 m between rows and 1.8 m within rows. Vines were spur-pruned and trained to a bilateral cordon. The vineyards were drip-irrigated using regulated deficit irrigation. The soil type was characterized as a fine sandy loam. Three stages, budbreak, bloom, and veraison, were chosen because they correspond to the periods when major changes in phenology of the grapevine occur. These stages are identified as 04, 23, and 35, respectively, on the modified E-L scale (Coombe 1995). Budbreak was defined as the stage at which green leaf tissue was visible on 50% of the previously dormant buds; bloom was considered to start when 50% of the flower caps had dropped, and veraison corresponded to the beginning of ripening (softening or color change on 50% of the berries), when berries soften, enlarge, start accumulating sugar, and change color to translucent (white cultivars) or to red-purple (red cultivars). The appearance dates of the phenological stages for the cultivars that were studied were analyzed using descriptive statistics, analysis of variance, and Fisher s least significant difference (p < 0.05) (ver. 9.4; PROC GLM; SAS Institute, Inc.). Weather data were recorded by two automated weather stations located ~750 to 868 m from the vineyard (Station 1: lat N; long W; 260 m asl; Station 2: lat N; long W; 365 m asl). The weather stations were located in areas with similar environmental and land use conditions. Daily minimum (T min ) and maximum (T max ) air temperature data were downloaded from the AgWeatherNet Portal (www. weather.wsu.edu). The phenological database was divided into two groups of data sets that were used for model calibration and evaluation. Cluster analysis was used to group similar years based on the annual T min and T max using Ward s minimum-variance method (SAS proc. Cluster, Tree). Normal,

3 62 Zapata et al. cold, and warm years were identified and used to randomly arrange the two data sets to include years with the three climate responses that represented the range of weather conditions for the Pacific Northwest. Estimating base temperature. To assess the role of temperature in grapevine development, Pearson s correlation test (p < 0.05) (SAS proc. Corr) was performed between daily mean air temperature (T i ) for successive phenological stages and the date on which a stage was observed for each cultivar. T i was calculated as T i = (T max + T min )/2. DD was calculated as the sum over time of the difference between T i and T b (Equation 1) (Arnold 1959). Base temperatures were estimated individually for the three phenological stages and for each cultivar based on the duration of each stage (phenophase). Thus, T b for budbreak was estimated from the date on which first swell (a phenological stage that occurs prior to budbreak) was observed; T b for bloom was based on budbreak occurrence, and T b for veraison was based on bloom. Solving Equation 1, DD can also be expressed as the relationship between the accumulated temperature for a phenophase, the number of days to reach the phenophase (n), and a T b value, as shown in Equation 2. Eq. 1 Eq. 2 If T i < T b, then T i = T b and no DD accumulate. A ceiling temperature of 32 C for T i was used as an upper threshold because temperatures above that threshold can induce heat stress in grapevines (Jackson 2000). This relationship was used to estimate the temperature that minimized the standard deviation of DD through an iterative process of the T b. Nonlinear optimization using the generalized reduced gradient (GRG) algorithm in SOLVER (Microsoft Office Excel 2007) was applied to select the temperature with the least variation (Yang et al. 1995). Model development. Five combinations of parameters that included the starting date (t 0 ), T b, and the estimated DD for each phenological stage were compared. The first combination (Set 1) began the accumulation on 1 April, and T b was fixed at 10 C for each growth stage. Sets 2, 3, and 4 started on 1 Jan to estimate T b for budbreak, bloom, and veraison, and to calculate DD for each stage. Set 5 started on 1 Jan and used the specific T b and DD values estimated for the individual stages to predict the consecutive appearance of the three stages. Therefore, for budbreak predictions, the beginning of DD accumulation was fixed at 1 Jan; subsequent bloom prediction was based on the initial budbreak prediction, and veraison was predicted based on the bloom prediction. Model performance. For model calibration and evaluation, the predicted date for a specific phenological stage and the length of the phenophase (in days) between successive stages, obtained with each combination of parameters, were statistically compared with the observed dates. The statistical measurements of the root mean square error (RMSE) (Equation 3), mean bias error (MBE) (Equation 4), and agreement index (d) (Equation 5) (Willmott et al. 2012) were used to evaluate each model. Eq. 3 Eq. 4 Eq. 5 Where P i and O i are the predicted and observed dates for year n, and is the average observed date. Student s t-test (Equation 6) was used to assess whether the slope of the regression of observed versus predicted date was significantly different from 1, which corresponds to the slope of the standard, 1:1 (SAS proc. Reg). Eq. 6 Where t is the t distribution with n 2 degrees of freedom, is the estimated slope, β 0 is the slope being tested (β 0 = 1), and is the standard error of the slope coefficient. Results Weather and phenology. Daily weather data for the period of this study from 1990 to 2013 showed that the monthly mean T min ranged from 3.2 to 12.8 C, and the monthly mean T max ranged from 3.6 to 31.1 C, in December and July, respectively (Table 1). Large temperature fluctuations (averaging 18.4 C) occurred in July (summer), and the lowest variation (6.8 C) occurred in December (winter). Cluster analysis using average annual temperature was conducted to group warm, normal, and cold years. Three homogenous groups that accounted for 77% of the variation in temperature were identified and the results were displayed in a dendrogram (Figure 1). Cluster one, representing cold years, had an average annual T min and T max of 4.0 and 16.7 C. Cluster two, representing normal years, had an average annual T min and T max of 4.7 and 17.5 C. Cluster three (warm years) had an average annual T min and T max of 5.2 and 18.3 C. Based on these clusters, two sets of data were created for model calibration and evaluation, and each set included years with the three climatic responses. The period of record and the number of years of observed data that were used for each cultivar are presented in Table 2. The date of appearance of each phenological stage varied among cultivars (Table 3). The mean dates for budbreak were significantly different for Cabernet Sauvignon, Malbec, Merlot, Syrah, and Zinfandel, and ranged from 4 April to 5 May. The mean bloom date ranged from 27 May to 28 June, and only Syrah showed to be different from the other cultivars

4 Predicting Key Phenological Stages for 17 Grapevine s 63 Table 1 Monthly mean minimum (T min ) and maximum (T max ) air temperature ( C) and standard error (SE) near the experimental vineyard in Prosser, WA from 1990 to Air temp ( C) Jan Feb March April May June July Aug Sept Oct Nov Dec T min Mean SE T max Mean SE Figure 1 (A) Dendrogram and (B) scatter plot of cluster analysis that identified years according to their climate response based on Ward s minimumvariance method using the average annual minimum and maximum temperature (1990 to 2013) for Prosser, WA. Table 2 Description of the calibration and evaluation data sets by cultivar, indicating the years for which observed data were available for budbreak, bloom, and veraison. Calibration Evaluation Years n Years n Cabernet franc 1990, 1992, 1993, 1995, , 2003, , 1994, , 2002, 2004, Cabernet 1990, 1992, 1993, 1995, , 2003, 2005, , 1994, 1997, 1998, 2002, 2004, 2006, Sauvignon Lemberger 1990, 1992, 1993, 1995, , , 1994, 1997, 1998, Malbec 1990, 1992, 1993, 1995, , , 1994, 1997, 1998, Merlot 1990, 1992, 1993, 1995, , 2003, 2008, 2010, , 1994, 1997, 1998, 2002, 2007, 2009, Meunier 1991, 1993, 1995, 1998, , 1997, 2000, 2001, Pinot noir 1992, 1993, 1995, , 2003, , 1994, , 2002, 2004, Syrah a 2005, 2006, 2007, 2008, 2009, Zinfandel 1992, 1993, 1995, , 2003, , 1997, 1998, 2002, 2004, Chardonnay 1990, 1992, 1993, 1995, , 2003, 2008, 2010, , 1994, , 2002, 2004, 2007, 2009, 2013 Chenin blanc 1990, 1992, 1993, 1995, , 2003, , 1994, 1997, 1998, 2002, 2004, Gewürztraminer 1990, 1992, 1993, 1995, , 2003, , 1994, , 2002, 2004, Muscat blanc 1990, 1992, 1993, 1995, , 2003, , 1994, 1997, 1998, 2002, 2004, Pinot gris 1992, 1993, 1995, , 2003, , 1997, 1998, 2002, 2004, Sauvignon blanc 1990, 1992, 1993, 1995, , 2003, , 1994, , 2002, 2004, Semillon 1990, 1992, 1993, 1995, , 2003, , 1994, 1997, 1998, 2002, 2004, Riesling 1990, 1992, 1993, 1995, , 2003, 2005, , 1994, , 2002, 2004, 2006, a For Syrah, the available phenological records included fewer years and were used only for calibration. 10

5 64 Zapata et al. due to its late bloom date. Mean veraison date ranged from 18 July to 27 Aug and showed significant differences among cultivars. Overall, we found that early or late development was not consistent among growth stages, and although some cultivars showed an early budbreak, this does not necessarily mean that they have early bloom and veraison (Table 3). There were significant differences in the duration of phenophases among cultivars (p < 0.05) (Table 4). The average duration from budbreak to bloom ranged from 44 (Pinot gris) to 57 days (Syrah), the average duration from bloom to veraison Table 3 Descriptive statistics for appearance dates (day of year [DOY]) for the three phenological stages. Budbreak Bloom Veraison Mean SE Min Max Mean SE Min Max Mean SE Min Max Cabernet franc Cabernet Sauvignon Lemberger Malbec Merlot Meunier Pinot noir Syrah Zinfandel Average red Chardonnay Chenin blanc Gewürztraminer Muscat blanc Pinot gris Sauvignon blanc Semillon Riesling Average white Total average Table 4 Descriptive statistics for the phenophases for different cultivars. Units are duration in days. Budbreak Bloom Bloom Veraison Budbreak Veraison Mean SE Min Max Mean SE Min Max Mean SE Min Max Cabernet franc Cabernet Sauvignon Lemberger Malbec Merlot Meunier Pinot noir Syrah Zinfandel Average red Chardonnay Chenin blanc Gewürztraminer Muscat blanc Pinot gris Sauvignon blanc Semillon Riesling Average white Total average

6 Predicting Key Phenological Stages for 17 Grapevine s 65 ranged from 59 (Muscat blanc) to 68 days (Chenin blanc), and the average duration from budbreak to veraison ranged from 106 (Malbec) to 120 days (Chenin blanc). Phenological observations were related to the weather responses for each year based on the cluster analysis. There were also differences in the appearance dates of the phenological stages of the warm, normal, and cold years as indicated by the mean observed DOY (Figure 2). Warm years showed a trend toward acceleration of the occurrence of the phenological stages compared to cold years, which in some cases did not have a strong effect on phenology and showed no differences in appearance dates. The effects of temperature on acceleration or delay of grapevine development varied depending on the cultivars, stages, and their interaction. For example, while there were no differences in budbreak dates among years for Cabernet Sauvignon, veraison dates were significantly different (Table 3). Bloom varied slightly among the temperature clusters, which could indicate that there was no sensitivity to environmental signals and that this reproductive stage in grapevines is genetically controlled. The correlation analysis showed the same results, in which the appearance date for the various stages was negatively correlated with mean air temperature of the phenophase across the 17 cultivars (Figure 3). However, this correlation was only significant for budbreak and bloom (p < 0.05), and for veraison only in Zinfandel and Pinot gris. The strength of the association, as indicated by Pearson s correlation, varied among cultivars. Base temperature and degree-days. The T b values obtained using the minimum variance method varied among Figure 2 Mean and standard error for the appearance of budbreak, bloom, and veraison for red and white cultivars for the three temperature clusters. DOY, day of year. Figure 3 Coefficients of Pearson s correlation between mean temperature for the three phenophases (1990 to 2013) and day of appearance of budbreak, bloom, and veraison for red and white cultivars. *95% significance.

7 66 Zapata et al. cultivars and phenological stages (Table 5). The lowest T b value was found for budbreak, with an increase for both bloom and veraison. The mean T b was 7.2 C for budbreak, 8.7 C for bloom, and 11.0 C for veraison. The standard error of the estimated T b s for each stage also increased from budbreak to veraison. Among red cultivars, Cabernet Sauvignon had the highest T b, with values of 8.3 C for budbreak, 10.4 C for bloom, and 12.5 C for veraison. In general, white cultivars had lower temperature values for budbreak and bloom, but not for veraison. Among white cultivars, Riesling achieved the highest thresholds with a T b of 7.6 C for budbreak, 9.6 C for bloom, and 11.6 C for veraison. The DD were significantly different among cultivars independent of the T b used in the calculations (p < 0.05). Set 1, with the parameters t 0 = 1 April and T b = 10 C for veraison, and Set 4 (t 0 = 1 Jan and T b estimated) underestimated thermal time and had the lowest DD because either the accumulation started late, on 1 April (Set 1), or the T b used corresponded to an estimated high value for veraison (Set 4) (Figure 4). For all cultivars, the average thermal time using the standard of 1 April and 10 C (as in the Winkler index) was 27 DD for budbreak, 254 DD for bloom, and 711 DD for veraison. DD calculated with Set 5, which used the parameters t 0 = 1 Jan and T b values specific to budbreak, bloom, and veraison, were 113 DD for budbreak, 316 DD for bloom, and 647 DD for veraison (Table 5). Sets 1 and 4 had the same DD requirements for budbreak, and slightly different requirements for bloom, as mean temperatures above 10 C generally do not occur until April, and the mean T b estimated for veraison in Set 4 and used to predict all phenological stages was 11 C. Model assessment. The five combinations of T b and DD parameters were used to predict successive phenological stages and to compare the predicted dates, using the RMSE, MBE, r 2, and d-index. Overall, the prediction errors in days, based on the calibration or evaluation data sets, were similar as indicated by the RMSE (Tables 6 and 7). The d-index values ranged from 0.5 to 0.9 for the calibration data set and from 0.2 to 1.0 for the evaluation data set, and were similar among the tested sets of parameters (Table 8). The MBE indicated that the tested sets of T b and DD under- or overestimated budbreak, depending on the cultivar (Figure 4). The plots for the predicted and observed dates and their comparison with the 1:1 line are presented in Figures 5 and 6. The coefficient of determination and significance t-test of the slope varied widely between data sets. Overall, bloom had the best performance based on the r 2, while veraison had the lowest r 2 values. The budbreak and veraison stages showed large RMSE and MBE compared to bloom. The predicted errors in dates to budbreak ranged from 3 to 14 days for the calibration data set and from 3 to 11 days for the evaluation data set (Tables 6 and 7). Errors in days were larger when using the estimated T b for veraison (Set 4), with values up to 14 days. The best performance (error ranging from 4 to 9 days) was observed when the estimated T b for budbreak (Sets 2 and 5) was used for the prediction. The lowest RMSE in dates was obtained for Pinot noir (4 days), whereas Cabernet Sauvignon showed the largest error (10 days). For the evaluation data set, the d-index was similar among the tested sets of T b and DD (Table 8). Table 5 Base temperature (T b ; C) estimated with the minimum variance method and growing degree-days using the estimated T b from 1 Jan (GDD est ) and 10 C from 1 April (GDD 10 ) for each phenological stage and cultivar. Budbreak Budbreak Bloom Bloom Veraison Tb GDDest GDD10 Tb GDDest GDD10 Tb GDDest GDD10 Cabernet franc Cabernet Sauvignon Lemberger Malbec Merlot Meunier Pinot noir Syrah Zinfandel Average red Chardonnay Chenin blanc Gewürztraminer Muscat blanc Pinot gris Sauvignon blanc Semillon Riesling Average white Total average

8 Predicting Key Phenological Stages for 17 Grapevine s 67 Predictions of bloom dates showed the best performance among the phenological stages, independent of the set of T b and DD used. Errors in days to bloom was the same for the calibration and evaluation data sets and ranged from 1 to 6 days (Tables 6 and 7). Lemberger had the lowest RMSE for days to bloom (2 days), whereas Cabernet Sauvignon and Semillon had the largest RMSE (6 days). The MBE for days to bloom indicated that, on average, the DD models underestimated the actual observations, with a range from 2 to 1 day. The d-index for days to bloom performed poorly for the Figure 4 Mean bias error (MBE) for prediction of budbreak, bloom, and veraison using five sets of parameters. Set 1: t 0 = 1 April, T b = 10 C; Set 2: t 0 = 1 Jan, T b = estimated value for budbreak; Set 3: t 0 = 1 Jan, T b = estimated value for bloom; Set 4: t 0 = 1 Jan, T b = estimated value for veraison; Set 5: t 0 = 1 Jan, T b = estimated value for budbreak, bloom, or veraison. Table 6 Root mean square error (units of days) in predicting the appearance date for budbreak, bloom, and veraison for five sets of parameters for the calibration data set. Budbreak / Set a Bloom / Set a Veraison / Set a Cabernet franc Cabernet Sauvignon Lemberger Malbec Merlot Meunier Pinot noir Zinfandel Average red Chardonnay Chenin blanc Gewürztraminer Muscat blanc Pinot gris Sauvignon blanc Semillon Riesling Average white Total average a Set 1: Starting date (t 0 ) = 1 April, base temperature (T b ) = 10 C; Set 2: t 0 = 1 Jan, T b = estimated value for budbreak; Set 3: t 0 = 1 Jan, T b = estimated value for bloom; Set 4: t 0 = 1 Jan, T b = estimated value for veraison; Set 5: t 0 = 1 Jan, T b = estimated value for budbreak, bloom, or veraison.

9 68 Zapata et al. evaluation data set (from 0.2 to 1) compared to the calibration data set (from 0.6 to 1), and did not vary among the sets that were evaluated (Table 8). The errors in dates to predict veraison ranged from 3 to 12 days for model calibration and from 3 to 11 days for model evaluation (Tables 6 and 7). The poorest performance, according to the RMSE, was found for the prediction of veraison for Chenin blanc (10 days), Gewürztraminer (10 days), and Semillon (12 days). Pinot gris had the smallest error (3 days). Among the sets that were tested, the errors were quite similar; Table 7 Root mean square error (units of days) in predicting the appearance date for budbreak, bloom, and veraison for five sets of parameters for the evaluation data set. Budbreak / Set a Bloom / Set a Veraison / Set a Cabernet franc Cabernet Sauvignon Lemberger Malbec Merlot Meunier Pinot noir Zinfandel Average red Chardonnay Chenin blanc Gewürztraminer Muscat blanc Pinot gris Sauvignon blanc Semillon Riesling Average white Total average a Set 1: Starting date (t 0 ) = 1 April, base temperature (T b ) = 10 C; Set 2: t 0 = 1 Jan, T b = estimated value for budbreak; Set 3: t 0 = 1 Jan, T b = estimated value for bloom; Set 4: t 0 = 1 Jan, T b = estimated value for veraison; Set 5: t 0 = 1 Jan, T b = estimated value for budbreak, bloom, or veraison. Table 8 Index of agreement (d) and coefficient of determination (r 2 ) for the comparison between predicted and observed dates for the calibration and evaluation data sets, starting on 1 Jan and using the specific base temperature (T b ) and the required degree days to predict the appearance of budbreak, bloom, and veraison. Calibration Evaluation Budbreak Bloom Veraison Budbreak Bloom Veraison d r 2 d r 2 d r 2 d r 2 d r 2 d r 2 Cabernet franc Cabernet Sauvignon Lemberger Malbec Merlot Meunier Pinot noir Zinfandel Chardonnay Chenin blanc Gewürztraminer Muscat blanc Pinot gris Sauvignon blanc Semillon Riesling

10 Predicting Key Phenological Stages for 17 Grapevine s 69 the MBE approached zero, but the negative bias indicates that the sets of parameters underestimated the actual observations for most cultivars (Figure 4). Discussion A negative correlation was obtained between temperature and the start date of each phenological stage. Overall, an increase in temperature advanced the timing of the events, but this effect was not quite clear for normal and cold years for which some cultivars did not show significant differences in the dates for budbreak, bloom, and veraison. These differences can be explained in part by the criteria that were used to separate the data sets based on minimum and maximum temperature; different results could be expected if more variables, such as precipitation, were included. In addition, a cultivar stage response to temperature was observed. For example, the dates of budbreak appearance for Zinfandel were not different among years within the three clusters with a contrasting climate response, but it affected the dates for bloom and veraison. The specific response of the phenological stages to temperature found in this study suggests that a more precise characterization of grapevine cultivars must be based on the individual occurrence of the phenological stages and phenophases (Lopes et al. 2008, van Leeuwen et al. 2008, Parker et al. 2013). Duchêne et al. (2010) found that predictions of flowering and veraison improved when daily maximum temperature was considered, whereas budbreak was better predicted using daily mean temperature. The growing season temperature should be used for future classifications of years, and a comparison of phenological responses to temperature is suggested for future research. The effects of acclimation and deacclimation processes that occur during dormancy and impact grape phenology also should be considered when analyzing early and late onset of phenological stages. For example, some studies have indicated that cold-hardy cultivars tend to reactivate growth earlier in Figure 5 Comparison of predicted and observed dates for (A) budbreak, (B) bloom, and (C) veraison for red cultivars for model calibration (left) and evaluation (right). Solid line is 1:1 line. DOY, day of year. Figure 6 Comparison of predicted and observed dates for (A) budbreak, (B) bloom, and (C) veraison for white cultivars for model calibration (left) and evaluation (right) data sets. Solid line is 1:1 line. DOY, day of year.

11 70 Zapata et al. spring compared to less cold-hardy cultivars grown in the same environment (Ferguson et al. 2014). Most phenological models are based on air temperature obtained from nearby weather stations, but environmental conditions can vary broadly within a vineyard and grapevine canopy, because of the canopy structure, row orientation, and incident solar radiation. Therefore, detailed measurements of temperature in buds, clusters, and berries should be conducted to provide more accurate information about T b (Keller and Tarara 2010). The large variation in T b for veraison among cultivars observed in our study can be explained by the precocity of fruit ripening, a genetically determined characteristic that is highly variable (Jones and Davis 2000). Previous studies have also shown that the ability of grapevine species to adapt to cool or warm conditions is likely related to lower or higher heat thresholds (Parker et al. 2011, Molitor et al. 2014). Our estimation of T b differs from previous results in which a temperature of 0 C resulted in the lowest error in prediction for flowering and veraison, which provides insights into the role of low temperatures in plant phenology in cool environments (Parker et al. 2011). Here, instead of implementing a single T b across all stages, we reported T b s for the sequential appearance of phenological stages. In grapevine, most studies assume 10 C as T b. However, this is a threshold for calculating the Winkler index, a bioclimatic index used to classify and identify suitable areas for grape cultivation rather than to make predictions about grape development (Amerine and Winkler 1944). Molitor et al. (2014) estimated the cardinal temperatures as 5, 20, and 22 C representing the lower (base), optimum, and upper temperature for the complete grapevine growth cycle. The lower threshold reported by Molitor et al. (2014) is fairly close to the T b estimated for the occurrence of budbreak in our study. An upper threshold of 22 C would likely be too low to inhibit growth under our study conditions, because historical average maximum temperatures exceeded 22 C from May through September. Once dormancy is completed, grapevine development depends mainly on increases in air temperature, which could be related to the increase in T b observed from spring to summer through an increase in the minimum threshold for the appearance of the stages. This fact can also be explained because the biochemical reactions that occur in plants require an increase in temperature to induce changes from the vegetative to reproductive phases, as observed in the transition from budbreak to veraison (Johnson and Thornley 1985). Similar to our results in which T b increased during crop development, Duchêne et al. (2010) estimated T b s of 2 C for budbreak and 10 C for flowering, but a decreased T b was observed for veraison (6 C). However, their estimates were obtained by combining phenological information for the cultivars Gewürztraminer and Riesling, without considering differences in development between the cultivars, and by extrapolation of the regression line between the inverse of the duration of the stage and the average maximum temperature. In this study, we observed that DD requirements were greater for veraison than for bloom for all cultivars. This can be explained in part by differences in crop management and water status during veraison compared to the other stages, and by clonal variability (Parker et al. 2013). Thermal time, beginning on 1 Jan, ranged from 516 to 800 DD using the estimated T b and from 653 to 768 DD using a T b of 10 C. These calculated DD were larger than those found by Sadras and Petrie (2011) in Australia, where the DD requirements starting on 1 Dec ranged from 312 to 528 DD. This lower thermal time could be explained by differences in temperature between the study regions and the use of a general T b of 10 C. Phenological studies conducted so far have assumed a common T b for all cultivars. Parker et al. (2013) classified and predicted the timing for flowering and veraison for 50 grapevine cultivars using a T b of 0 C and started accumulation on 1 March (DOY 60). Their study highlighted the importance of the use of thermal time to describe specific phenological stages, and they showed variability among cultivars which is closely related to the specific T b and DD values that we estimated that characterize early and late cultivars. Performance evaluation of stage- and cultivar-specific T b s in predicting individual phenological stages has not been widely explored in grapevine. There was similar error in the calibration and evaluation data sets used to predict phenology in this study. The RMSE s were consistent in both data sets, indicating that the method that was used to separate the data sets to include years with the three climatic responses was appropriate. Overall, the average RMSE calculated in this study for predicting budbreak (6 days) was low compared to the error of the models that used a T b of 10 C, which had RMSE greater than 10 days for Cabernet Sauvignon, Chardonnay, Merlot, Pinot noir, Riesling, Sauvignon, and Syrah (García de Cortázar-Atauri et al. 2009). The overall low RMSE for predicting bloom can be explained by the strong positive correlation between flowering and temperature (Buttrose and Hale 1973). The negative bias observed here could be due to systematic error in collecting phenological data or to variability that was not accounted for by the model. In addition to temperature, grapevine phenology is affected by factors such as winter chilling accumulation and soil moisture, which could accelerate or delay the occurrence of budbreak. Pruning time also has an impact on degree-day accumulation and could partially be responsible for differences in errors of prediction (Scarpare et al. 2012). The arbitrary fixing of the starting date could have increased the errors in prediction; however, contrasting results have been reported depending upon the crop. Ruml et al. (2011) found that 1 Jan was more correlated with phenological observations and used this as a starting date to predict bloom in apricots. Nendel et al. (2010) found that the starting date for heat accumulation should be specifically determined to accurately predict budbreak appearance. Several criteria have been considered to determine more realistic starting dates for budbreak, including the effect of chilling in predicting budbreak and the dependence of the beginning of dormancy on environmental conditions (Harrington and Gould 2015). Further research that involves chilling requirements in grapevines could help in substituting fixed starting dates. Londo and Johnson (2014) categorized

12 Predicting Key Phenological Stages for 17 Grapevine s 71 grapevine species based on low and high chill requirements, and their results were correlated with the time of budbreak. Our errors in days to bloom using estimated T b (2 to 6 days) were similar to errors reported for flowering using a T b of 0 C (3 to 7 days) by Parker et al. (2011). However, our errors in predicting veraison were higher (3 to 12 days, compared to 6 to 8 days found by Parker et al. [2011]). It should also be noted that for most phenological studies, data and observations are not collected daily, which affects the accuracy of the models. Conclusion This study found that the parameters T b and DD varied among phenological stages and cultivars, and increased progressively from budbreak to veraison. Slight differences in T b were observed between red and white cultivars. An approach using individual thermal time parameters for each stage and cultivar was able to predict the consecutive appearance of phenological stages with errors in predictions that were similar to the uncertainty reported for other models. These results contribute to an improved understanding of grapevine phenology. Such understanding is critical in planning vineyard management practices in the Pacific Northwest. This study included a large data set of years with contrasting climate conditions; further research that evaluates T b and estimated DD in different locations and environments could extend the applicability of the model and potentially reduce prediction errors. Literature Cited Amerine MA and Winkler AJ Composition and quality of musts and wines of California grapes. Hilgardia 15: Arnold C The determination and significance of the base temperature in a linear heat unit system. Proc Am Soc Hortic Sci 74: Baker JT, Leskovar DI, dy VR and Dainello FJ A simple phenological model of muskmelon development. Ann Bot-London 87: Buttrose MS and Hale CR Effect of temperature on development of the grapevine inflorescence after bud burst. Am J Enol Vitic 24: Caffarra A and Eccel E Increasing the robustness of phenological models for Vitis vinifera cv. Chardonnay. Int J Biometeorol 54: Campos I, Neale CMU, Calera A, Balbontín C and González-Piqueras J Assessing satellite-based basal crop coefficients for irrigated grapes (Vitis vinifera L.). Agric Water Manag 98: Chuine I, Garcia de Cortazar-Atauri I, Kramer K and Hänninen H Plant development models. In Phenology: An Integrative Environmental Science. MD Schwartz (ed.), pp Springer Dordrecht Heidelberg, New York, London. Coombe BG Adoption of a system for identifying grapevine growth stages. Aust J Grape Wine Res 1: Duchêne E, Huard F, Dumas V, Schneider C and Merdinoglu D The challenge of adapting grapevine varieties to climate change. Clim Res 41: Duchêne E, Butterlin G, Dumas V and Merdinoglu D Towards the adaptation of grapevine varieties to climate change: QTLs and candidate genes for developmental stages. Theor Appl Genet 124: Ferguson JC, Moyer MM, Mills LJ, Hoogenboom G and Keller M Modeling dormant bud cold hardiness and budbreak in twentythree Vitis genotypes reveals variation by region of origin. Am J Enol Vitic 65: Fila G, Di Lena B, Gardiman M, Storchi P, Tomasi D, Silvestroni O and Pitacco A Calibration and validation of grapevine budburst models using growth-room experiments as data source. Agr Forest Meteorol 160: García de Cortázar-Atauri I, Brisson N and Gaudillere JP Performance of several models for predicting budburst date of grapevine (Vitis vinifera L.). Int J Biometeorol 53: Harrington CA and Gould PJ Tradeoffs between chilling and forcing in satisfying dormancy requirements for Pacific Northwest tree species. Front Plant Sci 6:120. Hoover MW Some effects of temperature on the growth of southern peas. Proc Am Soc Hortic Sci 66: Jackson RS Wine Science: Principles, Practice, Perception. Academic Press, San Diego. Johnson IR and Thornley JHM Temperature dependence of plant and crop process. Ann Bot-London 55:1-24. Jones GV and Davis RE Climate influences on grapevine phenology, grape composition, and wine production and quality for Bordeaux, France. Am J Enol Vitic 51: Jones GV, Duff AA, Hall A and Myers JW Spatial analysis of climate in winegrape growing regions in the western United States. Am J Enol Vitic 61: Keller M and Tarara JM Warm spring temperatures induce persistent season-long changes in shoot development in grapevines. Ann Bot-London 106: Kwon EY, Jung JE, Chung U and Yun JI Using thermal time to simulate dormancy depth and bud-burst of vineyards in korea for the twentieth century. J Appl Meteorol Clim 47: Londo JP and Johnson LM Variation in the chilling requirement and budburst rate of wild Vitis species. Environ Exp Bot 106: Lopes J, Eiras-Dias JE, Abreu F, Climaco P, Cunha JP and Silvestre J Thermal requirements, duration and precocity of phenological stages of grapevine cultivars of the Portuguese collection. Ciência Téc Vitiv 23: Ma S, Churkina G and Trusilova K Investigating the impact of climate change on crop phenological events in Europe with a phenology model. Int J Biometeorol 56: Marra FP, Inglese P, DeJong TM and Johnson RS Thermal time requirement and harvest time forecast for peach cultivars with different fruit development periods. Acta Hortic 592: Molitor D, Junk J, Evers D, Hoffmann L and Beyer M A high-resolution cumulative degree day-based model to simulate phenological development of grapevine. Am J Enol Vitic 65: Moncur MW, Rattigan K, Mackenzie DH, Mc Intyre GN Base temperatures for budbreak and leaf appearance of grapevines. Am J Enol Vitic 40: Nendel C Grapevine bud break prediction for cool winter climates. Int J Biometeorol 54: Ojeda-Bustamante W, Sifuentes-Ibarra E, Slack DC and Carrillo M Generalization of irrigation scheduling parameters using the growing degree days concept: Application to a potato crop. Irrig Drain 53: Oliveira M Calculation of budbreak and flowering base temperatures for Vitis vinifera cv. Touriga Francesa in the Douro Region of Portugal. Am J Enol Vitic 49: Olivier FC and Annandale JG Thermal time requirements for the development of green pea (Pisum sativum L.). Field Crop Res 56: