Forecasting Harvest Area and Production of Strawberry Using Time Series Analyses

Transcription

1 Gazosmanpaşa Ünverstes Zraat Fakültes Dergs Journal of Agrcultural Faculty of Gazosmanpasa Unversty Araştırma Makales/Research Artcle JAFAG ISSN: E-ISSN: (2017) 34 (3), do: /jafag4298 Forecastng Harvest Area and Producton of Strawberry Usng Tme Seres Analyses Melekşen AKIN 1* Sadye PERAL EYDURAN 1 1 Igdr Unversty, Agrcultural Faculty, Department of Hortculture, Igdr *e-mal: aknmeleksen@gmal.com Alındığı tarh (Receved): Kabul tarh (Accepted): Onlne Baskı tarh (Prnted Onlne): Yazılı baskı tarh (Prnted): Abstract: Ths study was conducted to model the harvest area and producton of strawberry n Turkey usng FAOSTAT data from perod of to forecast strawberry harvest area and producton for perod. Non-statonary tme seres of strawberry harvest area and producton for perod were transformed nto statonary tme seres after takng the frst dfference of the tme seres. Three Autoregressve Integrated Movng Average (ARIMA (1,1,0), ARIMA (1,1,1) and ARIMA (0,1,1)) and three Exponental Smoothng (Holt, Brown and Damped) models were used comparatvely for tme seres data sets on strawberry harvest area and producton. Holt exponental smoothng model showed the best forecastng and Brown exponental smoothng model was the most approprate forecastng model for strawberry harvest area and producton from the tested sx models. We forecasted that the strawberry harvest area s gong to be ha n 2016 and wll ncrease to ha n The strawberry producton forecasted sgnfcant ncrease for the perod, from tons to tons. Brefly, the present forecastng results mght help polcy makers to develop macro-level polces for food securty and more effectve strateges for better plannng strawberry producton n Turkey. Keywords: Strawberry, producton, harvested area, exponental smoothng, tme seres Zaman Sers Analz Yöntemlern Kullanarak Türkye dek Çlek Hasat Alanı ve Üretmnn Tahmnlenmes Öz: Bu çalışma, yılı FAOSTAT verlern kullanarak yılı çn Türkye dek çlek hasat alanı ve üretmn tahmnlemek amacıyla yapılmıştır dönemne at çlek hasat alanı ve üretm zaman serler, zaman serlernn brnc dereceden farkının alınmasıyla durağan hale getrlmştr. Çlek hasat alanı ve üretmn modellemek çn üç bütünleştrlmş otoregresv hareketl ortalama (ARIMA (0,1,1), ARIMA (1,1,0) ve ARIMA (1,1,1)) ve üç üstsel düzleştrme (Holt, Brown ve Damped) yöntem kıyaslanmıştır. Çlek hasat alanı ve üretmn tahmnlemede test edlen altı yöntemden Holt üstsel düzleştrme teknğ en y projeksyonu gerçekleştrmş olmasına rağmen, Brown model en uygun yöntem olarak öne çıkmıştır. Bu sonuçlar doğrultusunda, 2016 yılında hektar olan çlek hasat alanının 2025 yılında hektara yükseleceğ; 2016 yılında ton olan çlek üretm mktarının se 2025 yılında tona doğru artış göstereceğ öngörülmüştür. Kısacası, bu çalışmadan elde edlen sonuçların gıda güvenlğn sağlamak çn makro düzeyde poltkaların gelştrlmesne ve Türkye dek çlek üretmnn daha etkn br şeklde planlanmasına yardımcı olacağı düşünülmektedr. Anahtar Kelmeler: Çlek, üretm mktarı, hasat, üstsel düzleştrme, zaman serler 1. Introducton Strawberres are very rch source of carotenods, vtamns, phenols, and flavonods whch play an mportant role n human nutrton. Strawberres show very hgh antoxdant actvty and have postve health effects. There s a hgh demand on berres due to ther potental for reducng the rsk of chronc dseases such as 18 cancer, cardovascular dseases, and stroke (Wang and Lnn, 2000). Therefore, t s mportant to provde food sustanablty for a healthy det and healthy next generatons. Turkey strawberry harvest area was ha wth a producton of tons n the 2000 year, but t showed an dramatc ncrease wth ha harvest area and tons producton n the 2014 year

2 (FAOSTAT, 2017). Increasng strawberry producton area could also help develop rural areas by ncreasng farmer revenues. It s mportant to establsh polces for ncreasng strawberry producton for future sustanablty, export ncomes, and food safety. Therefore, projecton studes are very useful tool for forecastng strawberry harvest area and producton, as well as to determne approprate polces for the future. There are only a few studes on forecastng crop producton for the next years (Masuda and Goldsmth, 2009; Semerc and Ozer, 2011; Suresh et al., 2012; Celk et al., 2013; Borkar, 2016; Celk et al., 2017; Karadas et al., 2017a, b) usng ARIMA and exponental smoothng methods tme seres analyss. To our knowledge, strawberry harvest area and producton of Turkey are not forecasted. Therefore, the man am of ths study was to model strawberry harvest area and producton of Turkey usng tme seres models for the perod of wth the am to forecast strawberry harvest area and producton for the next perod of Ths study could help establsh macro-level polces for food securty and sustanablty for the future. 2. Materal and Methods Tme seres data sets on annual harvest area (ha) and producton for the perod were downloaded by FAOSTAT database to obtan annual forecastng values for the next years. Three Autoregressve Integrated Movng Average (ARIMA (1,1,0), ARIMA (1,1,1) and ARIMA (0,1,1)) and three Exponental Smoothng (Holt, Brown and Damped) models were compared (Celk et al., 2017; Karadas et al., 2017a,b`). To fnd the best one among the tested sx canddate models, we used the followng model ft statstcs: 3. Results and Dscusson 3.1. Strawberry harvest area In the present study, tme seres analyss was used to analyses strawberry harvest area from the perod n order to forecast harvest area for the next years. Graph of the Root Mean Square Error, RMSE n 1 Y Yˆ Mean Absolute Percentage Error, MAPE = n n Y - Yˆ Y n =1 Maxmum Absolute Percentage Error, ) y - y MaxAPE = max *100, =1,2,,N y Mean Absolute Error (MAE), 1 MAE = Where M y - y ) n =1 s the error varance. Mean percentage error (MAPE) s defned as a measure of how much a dependent seres changes from ts model-predctve performance level. The root means square error (RMSE) has been employed as a model ft crteron n order to measure model performance. Maxmum absolute percentage error (MaxAPE) represents the largest forecasted error, expressed as a percentage. Statstcal analyss of annual harvest area (ha) and producton was performed by IBM SPSS program (verson 23). strawberry harvest area for the perod s gven n Fgure 1. The graph showed an ncreasng trend. Tme seres graphs of autocorrelaton (ACF) and partal autocorrelaton functons (PACF) are presented n Fgure 2 to understand ths trend better. 2 19

3 Several terms avalable n ACF graph exceeded confdence nterval, whch mples that there was a tme seres data trend (Fg. 2). To remove the tme seres from the trend and to make the data statonary, the frst-degree dfference of the tme seres was taken. ACF and PACF graphs for the frst dfference tme seres are gven n Fgure 3. The dfference seres were found statonary (Fg. 3). Some ARIMA and exponental smoothng models such as statonary R 2, R 2, RMSE, MAPE, MAE, and BIC were evaluated as a goodness of ft crtera. Fgure 1. Graph of strawberry harvest area (ha) for the perod Şekl yılları arasındak döneme at çlek hasat alanı (ha) grafğ Fgure 2. ACF and PACF graphs of strawberry harvest area (ha) for the perod Şekl yılları arasındak döneme at çlek hasat alanı (ha) ACF ve PACF grafkler The results are summarzed n Table 1. Holt (Table 1). In addton, Pektas (2013) advsed was the best model that showed the greatest R 2 usng BIC model as a ft crteron. Holt lnear and the smallest RMSE, MAPE, MAE, and BIC 20

4 model can be used for Ljung-Box Q= and p=0.063>0.05. Model parameters from exponental smoothng coeffcents of the Holt model on harvest area are provded n Table 2. Parameter coeffcents of Holt lnear model were estmated as α = and, and α coeffcent was sgnfcant (P<0.01). The relatonshp degrees of ACF and PACF, accordng to the Holt model, were wthn the confdence lmts (Fg. 4). Although 13th lag was found slghtly hgher than the confdence lmt, t was not consdered as a problem. Resdual terms were whte nose (Fg. 4). It was determned that the two seres were n agreement wth each other (Fgure 5). Forecasts for strawberry harvest area for perod are provded n Table 3. There was an ncreasng trend for the harvest area of perod, from ha to ha. Fgure 3. Strawberry harvest area ACF and PACF graphs of the frst dfference seres Şekl 3. Çlek hasat alanı brnc derece fark serlernn ACF ve PACF grafkler Table 1. Strawberry harvest area model goodness of ft statstcs Çzelge 1. Çlek hasat alanı uyum test statstkler Ft Statstc ARIMA(1,1,0) ARIMA(1,1,1) ARIMA(0,1,1) Holt* Brown Damped Statonary R R RMSE MAPE MAE BIC *Ljung-Box Q= and p=0.063>0.05 Table 2. Strawberry harvest area exponental smoothng model parameters Çzelge 2. Çlek hasat alanı üstsel düzleştrme model parametreler Estmate SE t Sg. Alpha (Level) Gamma (Trend)

5 3.2. Strawberry producton The tme seres analyss was performed to forecast strawberry producton for the perod. Annual producton graph for the perod showed a trend (Fg. 6). Graphs of autocorrelaton (ACF) and partal autocorrelaton functons (PACF) were provded to understand ths trend better (Fg. 7). Most of the terms avalable on the ACF graph exceeded the confdence nterval, whch mples that there was a tme seres data trend (Fg. 7). To elmnate the trend and make the data statonary, the frstdegree dfference of the tme seres was taken. ACF and PACF graphs for the frst dfference tme seres are shown n Fgure 8. Fgure 4. Strawberry harvest area ACF and PACF graphs of resduals Şekl 4. Çlek hasat alanı ACF ve PACF hata termler grafkler Fgure 5. Strawberry harvest area graph of the observed and forecasted (ft) seres Şekl 5. Çlek hasat alanı gözlenen ve tahmnlenen serlern grafkler 22

6 Some ARIMA and exponental smoothng models such as statonary R 2, R 2, RMSE, MAPE, MAE, and BIC were evaluated as a goodness of ft crtera. The results are summarzed n Table 4. Brown lnear exponental smoothng method was determned as the best method that had the greatest R 2 and the smallest BIC (Table 4). Brown lnear model can be used for Ljung-Box Q= and p=0.791>0.05. Fgure 6. Graph of strawberry producton (tons) for the perod Şekl yılları arasındak döneme at çlek üretm (ton) grafğ Fgure 7. ACF and PACF graphs of strawberry producton (tons) for the perod Şekl yılları arasındak döneme at çlek üretm (ton) ACF ve PACF grafkler Table 3. Forecastng strawberry harvested area (ha) values for the perod Tablo yılları çlek hasat alanı (ha) tahmn Year Area (ha)

7 Model parameters ncludng smoothng coeffcents belongng to the Brown model are presented n Table 5. Parameter coeffcent of the Brown lnear model was estmated as α = (P<0.01) (Table 5). Fgure 8. Strawberry producton frst dfference seres ACF and PACF graphs Şekl 8. Çlek üretm brnc derece fark serlernn ACF ve PACF grafkler The relatonshp degrees of lags n ACF and PACF graphs of resduals accordng to the Brown model were wthn the confdence lmts; therefore, resduals were whte noses (Fg. 9). The observed seres was n agreement wth the seres contanng ftted values (Fg. 10). Fgure 9. Strawberry producton ACF and PACF graphs of resduals Şekl 9. Çlek üretm ACF ve PACF hata termler grafkler 24

8 Table 6 presents forecastng results of strawberry producton. It s estmated that the strawberry producton from 2016 to 2025 would ncrease from tons to tons wth the proporton of %. To our knowledge, there s no research avalable on forecastng strawberry harvest area and producton n Turkey for plannng the securty and sustanablty of ths crop for the future. Besdes, there s a lmted number of studes on predctng producton of other crops (Semerc and Ozer, 2011; Masuda and Goldsmth, 2009; Celk, 2013; Amn et al., 2014; Borkar, 2016). In the current study, Brown exponental smoothng model revealed that strawberry producton ncreased sgnfcantly from tons to tons for the perod. Sunflower producton ncreasng trend was also reported for the perod (Semerc and Ozer, 2011). The worldwde ncreasng soybean producton trend was reported for the perod usng Damped exponental smoothng method (Masuda and Goldsmth, 2009). Producton amounts of pstachos, walnuts, hazelnuts, almond and chestnuts n Turkey were forecasted for the perod and ncreasng producton trends were noted usng dfferent ARIMA models (Celk, 2013). Celk et al. (2017) used ARIMA (0,1,1) model to forecast annual groundnut producton for the perod and reported ncreasng trend for the groundnut producton. Holt exponental smoothng method showed ncreasng trend for sunflower and sesame producton n Turkey for the perod. Soybean ranged from tons to tons, and sunflower ranged from tons to tons (Karadas et al. 2017a npress). Karadas et al. (2017b n press) forecasted tons ncrease n cotton lnt producton for the perod usng Holt exponental smoothng method. Although there are a few studes on forecastng some crop productons for the next years, predcton studes on economcally mportant crops should be ncreased for developng better polces for food securty and sustanablty, as well as better producton plannng. Table 4. Strawberry producton amount model ft statstcs Tablo 4. Çlek üretm mktarı uyum test statstkler Ft Statstc ARIMA(1,1,0) ARIMA(1,1,1) ARIMA(0,1,1) Holt Brown* Damped Statonary R R RMSE MAPE MAE BIC *Ljung-Box Q= and p=0.791>0.05 Table 5. Strawberry producton exponental smoothng model parameters Şekl 5. Çlek üretm üstel düzeltme model parametreler Estmate SE t Sg. Alpha (Level and Trend) Table 6. Forecastng strawberry producton for the perod Tablo yılları çlek üretm (ton) tahmn Year Forecast (tons) 25

9 Fgure 10. Strawberry producton graph of the observed and forecasted (ft) seres Şekl 10. Gözlenmş ve tahmn edlmş serlern çlek üretm grafkler 4. Concluson Non-statonary tme seres of strawberry harvest area and producton for the perod were transformed nto statonary tme seres after takng the frst dfference of the tme seres. Holt exponental smoothng model was the best forecastng model, among the tested sx models, for predctng strawberry harvest area for the next perod. However, Brown exponental smoothng model was descrbed as the most sutable for forecastng strawberry producton for the perod. Harvest area predcton ranged from ha to ha for the perod. Forecastng results of strawberry producton for the perod showed a sgnfcant ncrease, from tons to tons. As a result, the present predcton results mght help polcymakers to develop macro-level polces for food securty and more effectve strateges for better plannng strawberry producton n Turkey. References Amn M, Amanullah M, Akbar A (2014). Tme seres modelng for forecastng wheat producton of Pakstan, The Journal of Anmal Plant Scence, 24(5): Borkar P (2016). Modelng of groundnut producton n Inda usng ARIMA model. Internatonal Journal of Research n IT & Management. 6(3): Celk S (2013). Modellng of producton amount of nuts frut by usng Box-Jenkns technque. Yuzuncu Yl J. Agr. Sc. 23(1): Celk S, Karadas K, Eyduran E (2017). Forecastng groundnut producton of Turkey va ARIMA models. The Journal of Anmal and Plant Scence. FAOSTAT (2017). Statstcal database of the food and agrculture organzaton of the Unted Natons. Karadas K, Celk S, Eyduran E, Hopoglu S (2017a). Forecastng producton of some ol seed crops n Turkey usng exponental smoothng methods. The Journal of Anmal and Plant Scence (n press). Karadas K, Celk S, Hopoglu S, Eyduran E, Iqbal F (2017b). A survey of the relatonshp between producton amount, cultvaton area and yeld of cotton lnt n Turkey usng tme seres analyss. The Journal of Anmal and Plant Scence (n press). Masuda T, Goldsmth PD (2009). World soybean producton: area harvested, yeld, and long-term projectons. Internatonal Food and Agrbusness Management Assocaton. 12(4): Pektas A (2013). SPSS le ver madenclg. Dkeyeksen Yayın Dagtm, Yazlm ve Egtm Hzmetler San. ve Tc. Ltd. St.; Istanbul. Semerc A, Ozer S (2011). Turkye de aycceg ekm alanı, üretm mktarı ve verm değernde olası değşmler. Journal of Tekrdag Agrcultural Faculty, 8(3): Suresh K, Kran R, Grdhar K, Sampath K (2012). Modellng and forecastng lvestock feed resources n Inda usng clmate varables. Asan-Australasan Journal of Anmal Scence, 25(4): Wang SY, Ln HS (2000). Antoxdant actvty n fruts and leaves of blackberry, raspberry and strawberry vares wth cultvar and developmental stage. J. Agr. Food. 26