'Jade' Fruit Trees Grown in Northern Thailand

Development of Chill Unit Calculation for Peach ‘Jade’ Fruit Trees Grown in Northern Thailand N. Kamyun Department of Mathematics School of Science, University of Phayao Muang, Phayao Thailand

A. Pichakum Department of Plant Science Faculty of Science, Mahidol University Ratchawithi, Bangkok Thailand

K. Bunwong and M. Tiensuwan Department of Mathematics Faculty of Science, Mahidol University Ratchawithi, Bangkok Thailand

E.J. Moore Faculty of Applied Science King Mongkut’s University of Technology North Bangkok Thailand

Keywords: average daily temperature, chilling requirement, dormancy, flowering date Abstract Temperate fruit trees were introduced for opium substitution and reforestation in northern Thailand more than 40 years ago. Various cultivars have shown good adaptability with an annual life cycle similar to those growing in temperate areas. For good cultural management and logistics, the time of dormancy releasing should be predicted. In North America and Europe, the time of dormancy releasing is predicted using chill unit models that have been developed for temperate zone climatic conditions. Three commonly-used chill unit models in North America and Europe are the Utah model, the North Carolina model, and the Low chilling model. In the present paper, these models are evaluated and compared for use under the highland tropical zone conditions of northern Thailand. The peach (Prunus persica) ‘Jade’ grown at Angkhang Royal Agricultural Station, Thailand was selected as the model plant. During the period 2009-2011, the opened flower percentages were continuously observed at 3-day intervals and four randomly placed data loggers were used to record air temperatures in the tree canopies at 15minute intervals. The results showed that a modified version of the Utah model gave reasonable flowering prediction. The Box-Jenkins method was used to develop a seasonal ARIMA model to analyze temperature time series data in order to forecast daily temperature and predict chill unit accumulation. The predicted temperature and chill unit data could be used to predict flowering of ‘Jade’. INTRODUCTION The Angkhang Royal Agricultural Station is located on Doi Angkhang in the Tanao Sri Mountain area of northern Thailand. The station was initiated by King Rama IX in order to improve hill-tribe living conditions, to solve the problem of heroin, and to prevent slash and burn cultivation (Angkhang Royal Agricultural Station, 2010; Royal Project Foundation, 2007). The station is located at 19°54’32”N latitude and 99°02’50”E longitude and is about 1,400 m a.s.l. The climate at the station is classified as highland tropical. As the Doi Angkhang environment seemed suitable for temperate fruit trees, they were one of the plants chosen for the opium replacement. Deciduous fruit trees normally develop their vegetative and floral buds in the spring and summer. During the winter, they enter a dormancy phase characterized by loss of leaves, reduced activity, and a tolerance to temperatures much below freezing (Nyeki and Soltesz, 1996). These fruit trees require a certain number of what are called “chill hours” or “chill units” (CUs) of cold weather in order to break floral buds successfully. When the buds have accumulated sufficient CUs, they are ready to grow in the spring (Westwood, 1978; Jackson and Looney, 1999; Fraisse and Whidden, 2010). The dormancy breaking of peaches and nectarines in northern Thailand was also investigated Proc. IXth IS on Temperate Zone Fruits in the Tropics and Subtropics Eds.: U. Poonnachit et al. Acta Hort. 1059, ISHS 2014

147

by Boonprakob et al. (2006). There are three commonly used models for CUs, namely, the Utah model (UM) (Richardson et al., 1974), the Low chilling model (LCM) (Gilreath and Buchanan, 1981), and the North Carolina model (NCM) (Shaultout and Unrath, 1983). The UM was originally developed to predict dates of bud breaking for ‘Red-haven’ and ‘Elberta’ peach cultivars in Utah, the United States. The LCM was developed for ‘Sungold’ nectarines grown in Florida, the United States, and the NCM was developed for the ‘Starkrimson Delicious’ cultivar of apple grown in North Carolina, the United States. These models have been used to predict behavior of fruit trees in different countries, for example, they have been used to predict first blooming and fruit volumes for two varieties of peach cultivars in subtropical Australia (Weinberger, 1950), for predicting bud burst of crop and forest species (Cesaraccio et al., 2004), and for finding chilling requirements for grapes in Southern Tunisia (Mohamed et al., 2010). The main objective of this study was to test if CUs can be used to predict full bloom dates for ‘Jade’ peach grown at the Angkhang Royal Agricultural Station in Thailand. In order to calculate CUs for the three models, detailed temperature data for Doi Angkhang was obtained at 15-min intervals from automatic data loggers. The calculated CUs for the three models were then compared with data obtained for bud break and full bloom by counting the percentage of buds breaking at 3-day intervals for two flowering seasons. Two possible methods for predicting full bloom dates from the CUs have been developed. One method uses an ARIMA model of the temperature time series data to predict temperatures, CUs and full bloom dates. A second method uses average daily temperatures and weather forecast data from meteorology stations near Doi Angkhang to predict CUs and full bloom dates. MATERIALS AND METHODS Collection of Temperature Data Air temperatures were recorded in the peach tree canopies by automatic data loggers every 15 min from April 1, 2008 to March 31, 2011. Collection of Peach Flowering Data ‘Jade’ trees were cultivated separately from other cultivars of peach trees, and observations of flowering were made in two flowering seasons during the periods April 1, 2009 to March 31, 2010 and April 1, 2010 to March 31, 2011. At 3-day intervals during these periods, 10 trees were randomly selected and then 10 branches with flower buds were selected from each tree. Each branch was then sampled and labeled, the number of blossoms counted, and the branch then discarded to prevent double counting. The total number of blossoms for the 100 branches was then recorded. At the end of each flowering season, the total number of blossoms counted in all of the 3-day intervals was calculated and the numbers for each 3-day interval presented as a flowering percentage as a function of time. Calculation of CUs Hourly and daily CU values for UM, LCM, and NCM were calculated from the logged 15-min temperature data as follows. Hourly average temperatures were calculated from the 15-min temperature data and then weighting factors for each of the three models (Kamyun and Bunwong, 2011) were used to calculate the hourly CUs from the hourly temperature averages. The hourly CU values for each day were then added to obtain daily CU values. However, if a daily CU was calculated to be negative, then the daily CU value was recorded as zero. Consequently, the accumulated daily CUs were equal to zero until the temperatures dropped sufficiently and positive CUs began to accumulate. Time Series Analysis of Temperature Data Time series analysis comprises methods for analyzing time series data in order to 148

extract meaningful statistics and other characteristics of the data. A standard method for time series analysis is the Box-Jenkins method (Box and Jenkins, 1976), and autoregressive integrated moving-average (ARIMA) models are perhaps the most general Box-Jenkins type models (Box and Jenkins, 1976). ARIMA Models In this section, only a brief summary is given of the construction of an ARIMA model for the temperature time series data obtained from the temperature loggers. For a detailed analysis, see Kamyun and Bunwong (2011). As explained in Wei (1990), an ARIMA model is used for a time series which contains a trend or seasonality, nonconstant variances or other nonstationary phenomena. A general ARIMA (p, d, q) is defined by the order p of the autoregressive term, the order of differencing d and the order of the moving average term q. An analysis of the sample autocorrelation function (ACF) and sample partial autocorrelation function (PACF) of the temperature logger data showed that differencing was required because the ACF and PACF characteristics did not match the characteristics given in Wei (1990) for an ARMA model, i.e., that ACF decays very slowly and sample PACF cuts off after lag 1. The degree of differencing required to obtain a stationary series for the temperature data was determined by analyzing the ACF and PACF of series with different orders of differencing. The next step in the identification process was to find initial estimates for the orders of the parameters, p, q. Since the data showed a seasonal trend it was first necessary to apply a variance-stabilizing transformation and differencing (Ljung and Box, 1978; Wei, 1990). The orders of p and q were then obtained by matching the ACF and PACF patterns in the transformed series with the patterns in Wei (1990). One or more models that seemed to provide statistically adequate representations of the available data were then tentatively chosen and precise estimates of parameters of these models were obtained by a least squares fit (Box and Jenkins, 1976). The best of these tentative models was then selected using Akaike’s Information Criterion (AIC) and Schwartz’s Bayesian Criterion (SBC) (Akaike, 1973, 1974; Wei, 1990). As a final step, a diagnostic check of the original time series data against the predicted data from the best model was carried out to verify that the model assumptions of white noise and statistical insignificance of model ACF and PACF were satisfied (NIST, 2010). The packages SAS and SPSS were used to carry out the time series analysis. RESULTS AND DISCUSSION CUs, Chill Accumulation, and Flowering Percentages for October 2009 - March 2010 The calculation of daily CUs was started on October 22, 2009 and ended on March 31, 2010. The daily CUs for each of the three models were calculated and compared with the data for flowering percentages. In this paper, only a summary of the results is given. A more detailed comparison of the daily CUs for the three classical CU models and the flowering percentages can be found in Kamyun and Bunwong (2011). In Figure 1, the graphs labeled J1, J2, and J3 show the relationships between CU accumulations and flowering percentage accumulations for UM, NCM, and LCM, respectively. The solid lines in Figure 1 show a best fit to the data for a nonlinear regression function of the form: y  a  b (1  e  ( x  c ) / d )

where a, b, c, d are parameters, y is flowering percentage accumulation and x is CU accumulation. The standard error and the R2 residual values given in Figure 1 show that the exponential fit to the data is appropriate and makes reasonable predictions. Figure 1, it 149

was estimated that full bloom occurs for approximately 164 UM CUs (Table 1). CUs, Chill Accumulation, and Flowering Percentages for October 2010 - March 2011 For the second season, the calculation of daily CUs started from October 11, 2010 and ended on March 31, 2011. The daily CUs for each of the three models were then compared with the data for flowering percentages. The results for the second season were similar to those for the first season, but there were differences of detail. For the second season, the CU values remained zero before October 17, 2010 because the temperatures were still above the effective chilling temperature (55°F or 12.7°C for UM). During winter (October 17, 2010 - March 31, 2011), temperatures dropped into the chilling zone. The CU values then became positive and began to accumulate. Figure 2 shows the cumulative CU curves for the three classical CU models for the period October 2010 to March, 2011 and Figure 3 shows the corresponding flowering percentages (gray bars). It can be seen from Figure 2 that each CU curve gradually increases and tends to a specific value implying the end of winter. It can also be seen that UM gives the lowest CU values while LCM gives the highest values. As for the first season, it was found that a nonlinear regression function of the form given above again gave a good fit to the data for flowering percentage and CU accumulation. For this second season, full bloom occurred for UM CU accumulation of 104.5 (Table 1). Summary of CU Data for the Two Seasons A comparison of the CU results for the three models and the two seasons are shown in Table 1 and Figures 4 and 5. It can be seen that UM CU requirements for full bloom were in the range 104.5-164 and that full bloom occurred on December 15, 2009 and January 6, 2011. From an analysis of the results for the three models, it was concluded that UM gave the best estimates of the full bloom dates. The results also suggest that CUs can be useful for predicting full bloom dates for Jade peach trees in Thailand. However, as stated in Parker and Werner (1993), the chilling requirements for different cultivars of peach trees can vary from approximately 200 to more than 1000 chilling hours, with cultivars with the lower requirements being better suited to warmer climates and cultivars with higher requirements being better suited to colder climates. The low value obtained for ‘Jade’ suggests that this cultivar is suited to the highland tropical conditions of Thailand. However, because of the large variations in CU requirements for different cultivars, it is clear that the CU models developed for cultivars suitable for the temperate conditions of North America and Europe will require adjustment for the cultivars suitable for growing in northern Thailand. Further, as the data reported in this paper is limited to two years, a detailed comparison of the models over a longer period of time will be required before definite conclusions can be made of the comparative effectiveness of the three CU models under the tropical zone conditions of Thailand. Predicting Flowering Data One method that was tested for predicting flowering data was based on an ARIMA time series model of the temperature logging data. The best estimated model for the data was found to be the ARIMA (2,1,1) model shown in Table 2. Tests of this ARIMA model showed that it could predict temperatures ahead for a short period of up to about 15 days. It can therefore give predictions for CU accumulation and full bloom dates up to about 15 days ahead. As an alternative, tests were carried out to check if forecasts of maximum, minimum or average temperatures could be used as a simple method of forecasting CU accumulation and therefore dates of full bloom. These temperature forecasts are routinely supplied by meteorology stations for up to a week ahead. The graph in Figure 6 was obtained by computing the UM daily CUs. This figure suggests that average daily temperatures might give a simple method for estimating UM daily CUs. Moreover, a 150

linear regression model of the form y  57.11  0.956 x , was found to give a good fit (R2 = 0.793, standard error 1.21) for the relationship between average daily temperatures (x) and daily CUs (y). Therefore, the average daily temperature forecasts from meteorology stations near Doi Angkhang could be useful for predicting UM CU accumulation and dates of full bloom for ‘Jade’. CONCLUSIONS The research in this study was aimed at finding if CUs can be used to estimate full bloom dates for temperate peach trees at the Angkhang Royal Agricultural Station in Thailand. Available temperature data was used to compute CU accumulations for each of the UM, the NCM and the LCM. Then the CU accumulations were compared with the observed full bloom date and the percentage of total flowering for ‘Jade’ peach trees. It was found that the UM gave the best prediction of full bloom date and that the ‘Jade’ trees required a minimum of 104.5 UM CUs before full bloom. However, because the models have only been tested over two years on one species (‘Jade’), it is clear that further research is required on CU requirements for all peach cultivars being grown in northern Thailand. Two temperature prediction methods for CU accumulation were tested: a BoxJenkins ARIMA time-series model and a prediction based on average daily temperatures. It was found that the ARIMA model could give reasonable temperature predictions for up to 15 days ahead and therefore it could be used to estimate CU accumulation and full bloom date for up to 15 days ahead. However, the ARIMA model is reasonably complicated. For the alternative model, it was found that average daily temperatures gave reasonable estimates for UM daily CUs. It was also found (results not shown) that maximum daily temperatures can give reasonable estimates for NCM daily CUs. Since average and maximum daily temperatures are routinely supplied by meteorology stations near Doi Angkhang for up to a week ahead, these forecasts can provide a simple method of predicting CUs and full bloom dates for ‘Jade’ peach trees. ACKNOWLEDGMENTS Thanks to Mahidol University and the Royal Project Foundation for partial financial support. Appreciation is also extended towards the Department of Mathematics and the Department of Plant Science, Faculty of Science, Mahidol University and to the Angkhang Royal Agricultural Station, Thailand for their support and facilities. Literature Cited Akaike, H. 1973. Information theory and an extension of the minimum likelihood principle. Proc. 2nd International Symposium on Information Theory. Akademiai Kiado, Budapest. 2-8 September 1971. p.267-281. Akaike, H. 1974. A new look at the statistic model identification. IEEE T. Automat. Contr. 19:716-723. Angkhang Royal Agricultural Station 2010. www.angkhangstation.com/index/index.php. Boonprakob, U., Promchot, S., Thaisamak, S. and Mabangkrut, N. 2006. Performance of low-chill stone fruit from different germplasms in highlands of sub-tropical Asia. Proc. Netpro Update. NSW, Australia 16-17 May. Box, G.E.P. and Jenkins, G.M. 1976. Time Series Analysis Forecasting and Control. Holden-Day, San Francisco. Cesaraccio, C., Spano, D., Snyder, R.L. and Duce, P. 2004. Chill and forcing model to predict bud-burst of crop and forest species. Agri. Forest. Meteorol. 126:1-13. Fraisse, C.W. and Whidden, A. 2010. http://edis.ifas.ufl.edu/ae452. Gilreath, P. and Buchanan, D. 1981. Rest prediction model for low chilling Sungold nectarine. J. Amer. Soc. Hort. Sci. 106:426-429. Jackson, D.I. and Looney, N.E. 1999. Temperate and Subtropical Fruit Production. CABI, London. Kamyun, N. and Bunwong, K. 2011. Flowering estimation for Jade and Tropic Beauty 151

peach trees on Doi Angkhang. Proc. 16th AMM. Khon Kaen, Thailand. p.131-136. Ljung, G.M. and Box, G.E.P. 1978. On a measure of lack of fit in the time series models. Biometrika 65:297-303. Mohamed, H.B., Vadel, A.M. and Khemira, H. 2010. Estimation of chilling requirement and effect of hydrogen cyanamide on budbreak and fruit characteristics of ‘superior seedless’ table grape cultivated in a mild winter climate. Pak. J. Bot. 42:1761-1770. NIST 2010. www.itl.nist.gov/div898/handbook/pmc/section4/pmc4.htm. Nyeki, J. and Soltesz, M. 1996. Floral Biology of Temperate Zone Fruit Trees and Small Fruits. Akademiai Kiado. Parker, M.L. and Werner, D.J. 1993. Chilling requirement of selected peach varieties. North Carolina State University Leaflet No. 327. Richardson, E.A., Seeley, S.D. and Walker, D.R. 1974. A model for estimating the completion of rest for ‘Red-haven’ and ‘Elberta’ peach trees. HortScience 9:331-332. Royal Project Foundation. 2007. The Peach and the Poppy, the Story of Thailand’s Royal Project, Chiang Mai. Highland Research and Development Institute. Shaultout, A.D. and Unrath, C.R. 1983. Rest completion prediction model for Starkrimson delicious apples. J. Amer. Soc. Hort. Sci. 108:957-961. Wei, W.W.S. 1990. Time Series Analysis. Addison-Wesley, Pennsylvania. Weinberger, J.H. 1950. Chilling requirements of peach varieties. Proc. Amer. Soc. Hort. Sci. 94:304-307. Westwood, M.N. 1978. Temperate Zone Pomology. W.H. Freeman and Company, San Francisco.

Tables

Table 1. CU accumulation for UM, NCM, and LCM for full bloom of ‘Jade’ peach. Duration April 1, 2009 - March 31, 2010 April 1, 2010 - March 31, 2011

UM 164 104.5

NCM 270 387

LCM 372 585

Table 2. Estimated parameters of the ARIMA model for daily average temperature series. Model

Estimated parameters

ˆa2

ARIMA (2, 1, 1)

(11B  2 B2 )(1 B)Zt  (11B)at

2.9446

 1  0 .6 8 0 2  0 .0 5 1 3,  2   0 .0 7 0 9  0 .0 3 5 1,  1  0 .8 3 0 6  0 .0 4 2 3

152

Figures

Fig. 1. Relationship between CU accumulation for UM (J1), NCM (J2), and LCM (J3) and flowering percentage accumulation of ‘Jade’ for October 2009 - March 2010.

Fig. 2. CU accumulations for UM, NCM, and LCM for October 11, 2010 - March 31, 2011.

Fig. 3. Flowering percentage of ‘Jade’ peach for December 4, 2010 - January 21, 2011.

153

400 350 300 250 200 150 100 50 0

UM2009 UM2010

▲ Full Bloom Dates 23-Mar

9-Mar

23-Feb

9-Feb

26-Jan

12-Jan

▲ 29-Dec

15-Dec

1-Dec

17-Nov

3-Nov

▲ 20-Oct

Chill unit accumulation

Fig. 4. Full bloom period of ‘Jade’ peach for April 1, 2009 - March 31, 2010 and April 1, 2010 - March 31, 2011.

Fig. 5. UM CU accumulation for full bloom of ‘Jade’ peach; April 1, 2009 - March 31, 2010 and April 1, 2010 - March 31, 2011. Full bloom dates: December 15, 2009 (164 CU), January 6, 2011 (104.6 CU).

Fig. 6. Average daily temperature, maximum daily temperature, and minimum daily temperature for April 1, 2009 - March 31, 2010 and UM daily CUs.

154