SILPAKORN UNIVERSITY Science and Technology Journal SUSTJ
Copyright All rights reserved. Apart from citations for the purposes of research, private study, or criticism and review, no part of this publication may be reproduced, stored or transmitted in any other form without prior written permission by the publisher. Published by Silpakorn University Printing House. Silpakorn University, Sanamchandra Palace Campus, Nakhon Pathom 73000 Š Silpakorn University ISSN 1905-9159
SILPAKORN UNIVERSITY Science and Technology Journal Editorial Office Silpakorn University Research and Development Institute (SURDI), Silpakorn University, Sanamchandra Palace Campus, Nakhon Pathom, Thailand
Editorial Policy All articles submitted for publication will be evaluated by a group of distinguished reviewers. The editorial board claims no responsibility for the contents or opinion expressed by the authors of individual article.
Editorial Advisory Board Assist. Prof. Alice Thienprasert, Ph.D Director, Silpakorn University Research and Development Institute, Thailand Prof. Amaret Bhumiratana, Ph.D Department of Biotechnology, Mahidol University, Thailand Prof. Geoffrey A. Cordell, Ph.D Professor Emeritus, University of Illinois at Chicago, USA Prof. Kanaya Shiginori, Ph.D Department of Material and Life Sciences, Osaka University, Japan Prof. Keiji Yamamoto, Ph.D Graduate School of Pharmaceutical Sciences, Chiba University, Japan Assoc. Prof. Kriengsak Poonsuk, Ph.D Faculty of Animal Sciences and Agricultural Technology, Silpakorn University, Thailand Assist. Prof. Lerkiat Vongsarnpigoon, Ph.D National Metal and Materials Technology Center (MTEC), Thailand Assoc. Prof. Nijsiri Ruangrungsri, Ph.D Department of Pharmacognosy, Chulalongkorn University, Thailand Assoc. Prof. Petcharat Pongcharoensuk, Ph.D Department of Pharmacy, Mahidol University, Thailand Prof. Piyasan Praserthdam, Ph.D Department of Chemical Engineering, Chulalongkorn University, Thailand Assoc. Prof. Surachai Nimjirawath, Ph.D Department of Chemistry, Silpakorn University, Thailand Prof. Tharmmasak Sommartya, Ph.D Faculty of Agriculture, Bangkhen Campus, Kasetsart University, Thailand Prof. Virulh Sa-Yakanit, Ph.D Department of Physics, Silpakorn University, Thailand
Editor Assoc. Prof. Onoomar (Poobrasert) Toyama, Ph.D Faculty of Pharmacy, Silpakorn University
SILPAKORN UNIVERSITY Science and Technology Journal Editorial Board Assist. Prof. Bussarin Ksapabutr, Ph.D Faculty of Engineering and Industrial Technology, Silpakorn University Prof. Chawewan Ratanaprasert, Ph.D Faculty of Science, Silpakorn University Assist. Prof. Chockpisit Thepsithar, Ph.D Faculty of Science, Silpakorn University Assist. Prof. Jittat Fakcharoenphol, Ph.D Faculty of Engineering, Kasetsart University Assoc. Prof. Mana Kanjanamaneesathian, M.Appl.Sc. Faculty of Animal Sciences and Agricultural Technology, Silpakorn University Assist. Prof. Pramote Khuwijitjaru, Ph.D Faculty of Engineering and Industrial Technology, Silpakorn University Smith Tungkasmit, Ph.D College of Social Innovation, Rangsit University Assoc Prof. Soraya Ruamrungsri, Ph.D Faculty of Agriculture, Chiang Mai University Agnes Rimando, Ph.D U.S. Department of Agriculture, Agricultural Research Service, USA Prof. Juan Boo Liang, Ph.D Institute of Bioscience, Universiti Putra Malaysia, Malaysia Prof. Shuji Adachi, Ph.D Graduate School of Agriculture, Kyoto University, Japan Vincenzo Esposito, Ph.D Department of Energy Conversion and Storage, Technical University of Denmark, Denmark
Managing Editor Pranee Vichansvakul
Periodicity Twice yearly
All correspondence should be addressed to: Managing Editor, SUSTJ, 44/114 Soi Phaholyothin 52 Phaholyothin Road, Klongthanon, Saimai, Bangkok 10220, Thailand Telephone: 080-5996680 Fax: 66-2973-8366 E-mail address: pranee_aon1@hotmail.com Web site: http://www.journal.su.ac.th
Information about the Journal An electronic journal is provided on the web site (http://www.journal.su.ac.th). The journal is available at Silpakorn University Book Center. Telephone: 66-2223-7345, 66-2434-1792, 66-3424-4054.
Instructions to Authors Silpakorn University Science and Technology Journal (SUSTJ) is a peer review journal published twice a year in January and July by the Research and Development Institute of Silpakorn University, Thailand. SUSTJ puts together articles in Science and Technology and aims to promote and distribute peer reviewed articles in the areas of science, health science, animal science, agriculture, engineering, technology and related fields. Articles from local and foreign researchers, invited articles and review from experts are welcome.
Publication Ethics and Publication Malpractice Statement The Editor must ensure that each manuscript is initially evaluated by the editor/co-editor for the originality of the contents before forwarding to two referees for blind peer review. The review period will be from two weeks to one month (Maximum two months in extra ordinary circumstances). The Editorial Board evaluates manuscripts on their intellectual content without regard to authors’ race, gender, sexual orientation, or citizenship and must not disclose any information about a submitted manuscript to anyone other than the corresponding author, reviewers, potential reviewers, and the publisher.
Reviews should be conducted objectively without personal criticism.
Reviewers should express their views clearly with supporting arguments, identify relevant published work that has not been cited by the authors, and report any similarity or overlap between the manuscript under consideration and other published papers to the editor’s attention. Reviewers should not review manuscripts in which they have conflicts of interest resulting from competitive, collaborative, or other relationships or connections with any of the authors, companies, or institutions connected to the papers. The authors should ensure that they have written entirely original works, and has been appropriately cited or quoted the work of others. Manuscripts submitted must not have been published as copyrighted material elsewhere. Manuscripts under review must also not be submitted for consideration by other publication as copyrighted material. This should be declared in the Submission Form. By submitting a manuscript and a Submission Form, the author(s) agree that the copyright will be transferred to SUSTJ if the manuscript and associated multimedia are accepted for publication. The author(s) retain the rights to the fair use (e.g., teaching and non-profit uses) of the published material. Types of contributions Short communications, Research articles, Review articles
Preparation of manuscripts 1. The text should be double-spaced with line number on A4 and a font Times New Roman size 11 should be used. When using MS Word, insert all symbols by selecting “Insert-Symbol” from the menu and use the “Symbol” font. 2. Manuscripts should be organized in the following order: Cover page with title and authors’ names and affiliations Abstract (in English and Thai) Key Words Introduction Materials and Methods, Area Descriptions, Techniques Results Discussion Conclusion Acknowledgements References Tables and Figures Authors’ names and affiliations Full names and affiliations (marked with superscript number) should be provided for all authors on the cover page, separately from the content. The corresponding author (marked with superscript asterisk) should also provide a full postal address, telephone and fax number and an e-mail address as a footnote on the title page. Abstract First page of the content starts with Abstract, including title of the article on top of page. Provide a short abstract not more than 200 words, summarizing the question being addressed and the findings. Key Words Provide 3-5 key words or short phrases in alphabetical order, suitable for indexing. References In text references: Refer to the author’s name (without initials) and year of publication, e.g., Feldmann, 2004 (for 1 author), Feldmann and Langer, 2004 (for 2 authors), or Feldmann et al., 2004 (for more than 2 authors). Article references: References should be listed in alphabetical order of author(s). For journal, list all names of authors. Book Feldmann, H. (2004) Forty Years of FEBS, 2nd ed., Blackwell Publishing Ltd., Oxford, pp.1121-1129. Chapter in a book Langer, T. and Neupert, W. (1994) Chaperoning mitochondrial biogenesis. In The Biology of Heat Shock Proteins and Molecular Chaperones (Morimoto, R. I., Tissieres, A. and Georgopoulos, C., eds.), 3rd ed., pp. 53-83. Cold Spring Harbor Laboratory Press, Plainview, New York.
Article in a journal Hammerschlag, F. A., Bauchan, G., and Scorza, R. (1985) Regeneration of peach plants from callus derived from immature embryos. Journal of Natural Products 70(3): 248-251. Hammerschlag, F. A., Bauchan, G., and Scorza, R. Regeneration of peach plants from callus derived from immature embryos. Journal of Natural Products (in press). Article on the web Lee, K. (1999) Appraising adaptaive management. Conservation Ecology 3(2). [Online URL:www. consecolo.org/Journal/vol3/iss2/index.html] accessed on April 13, 2001. Proceedings MacKinnon, R. (2003) Modelling water uptake and soluble solids losses by puffed breakfast cereal immersed in water or milk. In Proceedings of the Seventh International Congress on Engineering and Food, Brighton, UK. Patent Yoshikawa, T. and Kawai, M. (2006) Security robot. U.S. Patent No. 2006079998. Tables and Figures Each Table and Figure must be on a separate page of the manuscript. Tables: Number the tables according to their sequence in the text. The text should include references to all tables. Vertical lines should not be used to separate columns. Leave some extra space instead. Figures: Figures should be of high quality (not less than 300 dpi JPEG or TIFF format), in black and white only, with the same size as the author would like them to appear in press. Choose the size of symbols and lettering so that the figures can be reduced to fit on a page or in a column. Submission of Manuscripts All information contained in a manuscript is a full responsibility of the authors, including the accuracy of the data and resulting conclusion. Authors are requested to send the manuscript on a CD labeled with the authors’ names and file names. The files should be prepared using MS Word only. Three copies of manuscript must be supplied. The editorial office will acknowledge receipt of the manuscript within 2 weeks of submission. The ‘accepted date’ that appears in the published article will be the date when the managing editor receives the fully revised version of the manuscript. The manuscript may be returned to authors for revision. Authors will be given 2 weeks after receipt of the reviewers’ comments to revise the manuscript. Please submit the manuscript with a CD and a submission form to the following address: Pranee Vichansvakul 44/114 Soi Phaholyothin 52 Phaholyothin Road, Klongthanon, Saimai, Bangkok 10220 Proofs Proofs will be sent to the corresponding author by e-mail (as PDF file) or regular mail. Author is requested to check the proofs and return any corrections within 2 weeks.
Silpakorn University Science and Technology Journal
Contents
Volume 7 Number 1 (January - June) 2013
Research Articles
Mortality Rate Model due to Transportation Accidents in Thailand...................................……....
Wattanavadee Sriwattanapongse, Sukon Prasitwattanaseree,
Surin Khanabsakdi and Supreeya Wongtra-ngan
Evaluation of the Shade Tolerance of Moth Bean (Vigna aconitifolia)
and Two Tropical Legume Species..........................................................................................……....
Odds Prediction of Drought Category Using Loglinear Models Based on
SPI in the Northeast of Thailand............................................................................................……....
Principal Component Analysis Coupled with Artificial Neural Networks for
Therapeutic Indication Prediction of Thai Herbal Formulae..........................................................
41
Lawan Sratthaphut, Samart Jamrus, Suthikarn Woothianusorn and Onoomar Toyama
Usage of Chitosan in Thai Pharmaceutical and Cosmetic Industries................................................
32
Wisoot Salee and Veeranun Pongsapukdee
19
Pantipa Na Chiangmai, Yupa Pootaeng-on and Thanakrit Khewaram
9
Rapeepun Chalongsuk and Namfon Sribundit
49
Silpakorn University Science and Technology Journal (SUSTJ) is now available on the following databases: Chemical Abstract Service (CAS) SciFinder Scholar (CAPLUS) International Information System for the Agricultural Sciences and Technology (AGRIS) (FAO) AGRICultural Online Access (AGRICOLA) Food Science and Technology Abstracts (FSTA) The National Science Digital Library (NSDL) CAB Abstract Directory of Open Access Journals (DOAJ) Google Scholar Thai Journal Citation Index Centre (TCI Centre)
Research Article Mortality Rate Model due to Transportation Accidents in Thailand Wattanavadee Sriwattanapongse 1*, Sukon Prasitwattanaseree 1 Surin Khanabsakdi 1 and Supreeya Wongtra-ngan 2 Biostatistics and Applied Statistics Research Unit, Department of Statistics, Faculty of Science, 2 Department of Obstetrics and Gynecology, Faculty of Medicine, Chiang Mai University, Thailand * Corresponding author: E-mail: wattanavadee.s@cmu.ac.th
1
Received July 19, 2012; Accepted September 19, 2012 Abstract The mortality from transportation accidents is a major problem that leads to loss of human lives and property. The deaths as a result of transportation accidents are now accepted to be a global phenomenon in virtually all countries concerned about the growth in the number of people killed. The objective was to model and forecast the transportation accident mortality rate in Thailand using death certificate reports. A retrospective analysis of the transportation accident mortality rate was conducted in this study. This study is based on the records of the national vital registration database for the 10-year period from 2000 to 2009, provided by the Ministry of the Interior and coded as cause-of-death using ICD-10 by the Ministry of Public Health. Multivariate linear regression was used for modeling and forecasting age-specific transportation accident mortality rates in Thailand. The transportation accident mortality increased higher in males than females. The highest was in males aged 20-29 years. The trend slightly decreases in all other ages. Having a model that provides such forecasts of transportation accident fatalities, even if based purely on statistical data analysis, can provide a useful basis for allocation of resources for transportation accident fatality prevention. Key Words: Mortality rate; Multivariate linear regression; Transportation accidents Introduction The problem of death as a result of road accidents is now acknowledged to be a global phenomenon with authorities in virtually all countries concerned about the growth in the number of people killed and seriously injured on their roads. The World Health Organization reported that about 1.2 million people were killed from road traffic accidents every year and it was estimated that approximately 3,000 people died by road traffic
Silpakorn U Science & Tech J 7 (1) : 9-18, 2013
accidents around the world on any given day. World Health Organization revealed a projection of global mortality leading causes of death from 2008 to 2030 that road traffic accidents would rise from the ninth to the fifth of the world’s leading causes of death, up from 2.2 % in 2004 to 3.6 % of global deaths (World Health Organization, 2008). Deaths resulting from road accidents have become a big problem in the developing countries such as Thailand. It reflects not only road safety
Silpakorn U Science & Tech J Vol.7(1), 2013
Mortality Rate Model due to Transportation Accidents in Thailand
in Thailand, but also in other Asian countries. In Thailand, road accidents are considered one of the top three public health problems. Despite the Government’s best efforts, there are sadly over 13, 000 deaths and more than one million injuries each year as the result of road accidents, with several hundred thousand people disabled. An overwhelming majority of the deaths and injuries involve motorcyclists, cyclists and pedestrians. Most occur between the ages of 20-24 years and 80% of them are from motorcycles. The estimated economic losses due to road accidents are over 100,000 million baht (approximately US$2,500 million) per year (Tanaboriboon 2004). The Royal Thai Government regards this problem to be of great urgency and has accorded its high priority in the national policy. The problem of road traffic injuries is indeed a highly serious one, but can be dealt with and prevented through concerted action among all the parties concerned. Thailand is a country located at the centre of the Indochina peninsula in Southeast Asia. It is divided into 77 provinces, which are gathered into four groups of regions, Central, North, North-East and South (Wikipedia 2011). It is bordered to the North by Myanmar and Laos, to the East by Laos and Cambodia, to the South by the Gulf of Thailand and Malaysia, and to the West by the Andaman Sea and the Southern extremity of Myanmar. The Thai population is estimated by the Department of Provincial Administration to be 65,479,453 (National Statistics Office, 2010). There were various studies about models to forecast transportation accident mortality rates. Sriwattanapongse and Khanabsakdi (2011) modeled the patterns of mortality rate due to traffic accident by gender, age and year. The method used linear regression and Poisson regression models to forecast mortality rates due to traffic accident likely to occur in the near future, in order to prevent the mortality rate by using suitable models. Among the models deemed suitable, the best was chosen based on the
analysis of deviance. The results of this study showed that additive effects associated with the gender, age group, and year could be used to provide a useful short-term forecast. Kardara and Kondakis (1997) identified trends of road traffic accident deaths and injury rates in Greece from 1981-1991 by using linear regression with logarithmic transformation. Road traffic accidents cause mortality and disability that lead to public health and social problems such as economic loss, particularly for male of working age. It impacts on the family income and national economy directly and indirectly. The costs of medical care, funerals and loss of income due to mortality or disability can push a family into poverty (Nantulya and Reich, 2003). Therefore, the aim of this research was to model and forecast the transportation accident mortality rate in Thailand, using death certificate reports. Materials and Methods Data for registered deaths due to transportation accidents were obtained from the national vital registration database for the 10-year period from 2000 to 2009.The database is provided by the Ministry of Interior and coded as cause-of-death using ICD-10: B20-B24 by the Bureau of Policy and Strategy, Ministry of Public Health. Age, gender, residential area by region in Thailand and year were selected as the explanatory variables in studying the mortality rates of transportation accidents. Age was divided into nine groups (0-9,10-19,20-29,30-39,40-49,50-59,6069,70-79 and above 80 yrs). Various approaches have been developed to forecast morbidity and mortality rates. This study focuses upon the model proposed by Lee and Carter (1992) and Lee and Miller (2001) that initially used projections of the age-specific mortality rates in the United States. The Lee-Carter-based modeling framework is viewed in the current literature as among the most efficient and transparent methods of modeling and projecting mortality improvements (Butt and 10
W. Sriwattanapongse et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
∑
kt = 0 and where ε x,t is a set of random to be disturbances. Since some cells had no reported cases, to allow log-transformation, we replaced zero counts by a suitably chosen small constant, without changing any value of mx ,t greater than 0. The multivariate linear regression model takes into account correlation in the data between age groups.
Haberman, 2009). This method is also regarded as state-of-the-art in mortality forecasting and has become more and more popular for long-run forecasts of age-specific mortality rates. Data used in the current study are taken from the national vital registration database for the 10-year period from 2000 to 2009, provided by the Ministry of Interior and coded as cause-of-death using ICD-10. We decided to calculate mortality incidence rates based on the 9 age groups. Mortality rates are based on numbers of deaths registered in a country in a year divided by the size of the corresponding population. Deaths from transport accidents are classified to ICD-10 codes V01-V89 Since transportation accident dead counts based on small cells are often zero cases, it is necessary to make some adjustment to take transformation of 0, so we replaced zero counts by a suitably-chosen small constant greater than 0: the method we use is to define the mortality rate as:
mx ,t
( 0.5 + nx ,t ) = × 100,000 P
Results For each year these data were obtained from the national vital registration database for the 10-year period from 2000 to 2009, provided by the Ministry of Interior and coded as cause-of-death using ICD-10 by the Ministry of Public Health. The fields comprise characteristics of the subject and cause-of-death diagnosis, including dates of death and the subject’s age, gender, and address. The results from demographic variables show that out of 114,790 transport accident cases, 80.44% of transport accident deaths are male and 19.56% female; also, that 25.92% of transport accident deaths are aged 20-29 years. This study finds that the mortality rate of transport accident appears to be highest at 4.12 per 100,000 people in males, aged 20-29, Northern region, year 2003, and the lowest of 1.17 per 100,000 people in females, aged 0-9, North-East region, year 2009. This study finds that the highest average age-specific transport accident mortality rate is 3.11 per 100,000 per year in males, Southern Region, the lowest average age-specific transport accident mortality rate is 1.79 per 100,000 per year in females, North-East Region, and most of the transport accident fatalities are males in Central Region. The trends of transport accident mortality rates have decreased in females, but it is approximately decreased more than previously. The transport accident mortality decreased mostly in all age groups and regions, with the exception in the South. We replaced zeros by 0.5 before fitting the
1/ 3
,
(1)
where nx ,t is the number of transportation accident death cases in the cell, and P is the population at risk. For each region and gender combination, multivariate linear regression model was used to investigate and forecast transportation accident mortality by age group and year. The original principal component of the Lee-Carter model is expressed as:
log(mx, t ) = ax + bx kt + ε x,t
(2)
where mx ,t is the central death rate (per 100,000) in age group x and year t for the specified in each gender and geographical region. The factors ax and bx describe the level and annual increase, respectively of the age-specific mortality rate, kt is time of year where Lee - Carter chose constraints
11
Silpakorn U Science & Tech J Vol.7(1), 2013
Mortality Rate Model due to Transportation Accidents in Thailand
model. The two left panels of Figure 1 show the transport accident mortality rates plotted by age group for each year in each gender; the two right panels show the trends plotted by year (2000-2009) for each age group in each gender, together with the forecasts based on the model. Figure 1 shows that mortality rate in males
increases up to age group 20-29 years and decreases slightly before increasing in the age groups 7079, with the exception of the North-East region. However, the time trends shown in the two right panels indicate that the transport accident mortality rates decrease after 2000 over the seven year period in all age groups, except in the South.
Figure 1 Plot of transportation accident mortality rates by age group for each year (two left panels) and trends with forecasts for each age group (two right panels) for the four regions of Thailand.
12
W. Sriwattanapongse et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
Table 1 Results from multivariate linear models of mortality rates by region and gender for each age group Central Region
Age
a1
b1
R-squared
F-statistic(1, 8)
p-value
0
1.850421
-0.062667
0.8572
48.010
0.00012
and males
10
3.38182
-0.06977
0.5761
10.870
0.01090
20
3.87036
-0.08302
0.6371
14.050
0.00564
30
3.59456
-0.10047
0.875
55.980
7.05E-05
40
3.50233
-0.09051
0.8822
59.880
5.54E-05
50
3.43625
-0.0854
0.8714
54.220
7.89E-05
60
3.49581
-0.09414
0.8652
51.370
9.55E-05
70
3.26311
-0.04266
0.3604
4.507
0.06651
80
3.16646
-0.05977
0.2117
2.148
0.18090
0
0.6011
-0.01929
0.2395
2.519
0.15110
and females
10
0.66347
-0.04082
0.5535
9.918
0.01361
20
0.56559
-0.02948
0.3255
3.861
0.08501
30
0.86482
-0.06998
0.8318
39.560
0.00024
40
0.7074
-0.03446
0.4949
7.838
0.02321
50
0.81478
-0.05648
0.7888
29.870
0.00060
60
0.8073
-0.04493
0.4685
7.051
0.02901
70
0.84059
-0.0202
0.1794
1.749
0.22250
80
1.01488
-0.02869
0.1365
1.265
0.29340
NE Region
0
1.699949
-0.03863
0.7928
30.610
0.00055
and males
10
3.102011
-0.021898
0.4236
5.880
0.04153
20
4.027224
-0.09307
0.9209
93.090
1.11E-05
30
3.395589
-0.04004
0.8056
33.150
0.00043
40
3.155347
-0.041001
0.7016
18.810
0.00249
50
3.084323
-0.039349
0.8485
44.820
0.00016
60
2.95579
-0.02604
0.3544
4.391
0.06942
70
2.782492
-0.005427
0.01831
0.149
0.70940
80
2.77488
-0.05118
0.1927
1.909
0.20440
NE Region
0
1.450765
-0.02639
0.6362
13.990
0.00570
and females
10
1.836698
-0.008844
0.1393
1.295
0.28800
20
2.035431
-0.026235
0.7167
20.240
0.00200
30
1.907815
-0.013462
0.5069
8.225
0.02090
40
1.873416
-0.017265
0.6186
12.980
0.00696
50
1.917636
-0.0163
0.3763
4.827
0.05926
60
1.93559
-0.01157
0.07371
0.637
0.44800
70
2.03733
-0.0305
0.2406
2.534
0.15010
80
2.08128
-0.0553
0.2659
2.897
0.12710
Central Region
13
Silpakorn U Science & Tech J Vol.7(1), 2013
Mortality Rate Model due to Transportation Accidents in Thailand
Table 1 Results from multivariate linear models of mortality rates by region and gender for each age group (continues)
Age
a1
b1
R-squared
F-statistic(1, 8)
p-value
0
1.71152
-0.04043
0.533
9.132
0.016510
and males
10
3.206194
-0.01933
0.3464
4.239
0.073480
20
4.16056
-0.07712
0.7965
31.310
0.000513
30
3.638665
-0.05221
0.812
34.550
0.000371
40
3.48595
-0.05101
0.762
25.610
0.000976
50
3.315677
-0.03299
0.5767
10.900
0.010830
60
3.300696
-0.03535
0.7482
23.770
0.001231
70
3.24222
-0.03281
0.4057
5.461
0.047650
80
2.78955
-0.0214
0.05254
0.444
0.524100
0
0.8812
-0.04447
0.5948
11.740
0.008995
and Females
10
0.77932
-0.04452
0.5446
9.566
0.014820
20
0.95264
-0.06489
0.5882
11.430
0.009633
30
1.0217
-0.05898
0.4395
6.274
0.036670
40
0.93834
-0.05294
0.5556
10.000
0.013340
50
1.0556
-0.08177
0.8227
37.120
0.000292
60
0.8116
-0.02083
0.1059
0.948
0.358800
70
0.96557
-0.0446
0.3156
3.689
0.091020
80
1.07281
-0.03794
0.4125
5.618
0.045240
0
1.75978
-0.00992
0.04008
0.334
0.579200
and males
10
3.19195
-0.0136
0.1738
1.683
0.230700
20
3.73203
-0.03955
0.612
12.620
0.007483
30
3.296831
-0.01511
0.2905
3.275
0.107900
40
3.12518
-0.0066
0.05128
0.432
0.529300
50
3.195324
-0.00552
0.04325
0.362
0.564200
60
3.304793
0.007286
0.01997
0.163
0.697000
70
3.35089
0.03693
0.4578
6.754
0.031670
80
3.18859
0.01686
0.01576
0.128
0.729600
0
1.506459
0.011191
0.3945
5.212
0.051840
and Females
10
1.9225
0.01543
0.1571
1.491
0.256800
20
2.11064
-0.01862
0.2356
2.466
0.155000
30
2.135315
-0.01131
0.152
1.434
0.265400
40
2.20787
-0.01254
0.084
0.734
0.416600
50
2.225303
0.006165
0.0116
0.094
0.767100
60
2.308202
0.008418
0.02548
0.209
0.659600
70
2.29403
0.01015
0.04418
0.367
0.560000
80
2.267775
-0.00164
0.00036
0.003
0.958500
Northern Region
Northern Region
Southern Region
Southern Region
14
W. Sriwattanapongse et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
Discussion In this study, the multivariate linear regression was appropriate to model transport accident mortality in Thailand. In another way, the multiple regression and Poisson regression are commonly used for modeling the mortality rates and number of deaths in a specific population. However, Pocock et al. (1981) pointed out that unweighted multiple regression was not appropriate for modeling mortality rates in different areas which varied in population size. In addition, fully weighted regression was usually too extreme. Thus, they introduced an intermediate solution via maximum likelihood for modeling death rates. Tsauo et al. (1996) examined the effect of age, period of death and birth cohort in motor vehicle mortality in Taiwan from 1974 - 1992, using data from Vital Statistics. Log-linear regression was used for fitting the model to perform the effects of variables. In addition, Lix et al. (2004) used Poisson regression to investigate the relation of demographic, geographical, and temporal explanatory variables with mortality in difference regions of Manitoba, Canada between 1985 and 1999, using data from Vital Statistics records and the provincial health registry. Yang et al. (2005) used Poisson regression modeling to examine and compare age- and sexspecific mortality rates due to injuries in the Guangxi Province in South Western China in 2002, based on death certificate data. However, this study focused only on small areas. Congdon (2006) described a method for modeling mortality over area, age and time dimensions that took account of spatial correlation, interactions between dimensions, and cohort as well as age effects, by applying Poisson regression. Although Poisson regression are commonly used for modeling mortality due to road traffic accident death, it cannot be used for modeling when many of observed are zeros due to having small sample size. Lee et al. (2002) suggested that when many of observed were zeros, the zero-inflated Poisson
and the negative binomial were more appropriate than the standard Poisson model. In addition, a study of Lord (2006) exposed that low sample mean combined with a small sample size of crash database could affect the goodness of fit of the negative binomial model. Alternatively, Bride (1995) presented model of forecasting and monitoring the development in the number of fatalities in traffic. The model had been created through time series analysis covering the years 1977-1991. The model was simple, with the number of fatalities as the dependent variable and with time and traffic as the only predictors. The time factor described the cumulative effect of changes such as better roads, vehicles, drivers, etc. The model was multiplicative and permitted a non-proportional relationship with traffic volume. Taking into account the purely random fluctuations in the number of fatalities, the historical fit for the period 1977-1991 was very good. Also the forecasts for 1992 and 1993 had proved very accurate. The model would be revised as new annual data were received. At present, the model points to a favorable development. Neill and Mohan (2002) conducted studies to reduce the numbers of crash deaths and injuries, and countries need to adopt a broad array of research based measures in the US. Almost all the demonstrable gains produced by changing road user behavior have resulted from properly-enforced traffic safety laws. After 2000, the transport accident mortality rates declined in all age groups, as the Policy and Strategy of Government introduced strong control measures and prevent transport accidents in the previous 10 years. For this investigation, the data have been interpreted accurately and objectively. The Government can use the results of this study to decrease surveillance transport accidents, especially those occuring at festival times. Future study of transportation accident fatalities was shown by GIS. Jonesa et al. (2008), investigated data on road traffic fatalities, serious 15
Silpakorn U Science & Tech J Vol.7(1), 2013
Mortality Rate Model due to Transportation Accidents in Thailand
casualties and slight casualties in each local authority district in England and Wales for the period 19952000. District-level data were assembled for a large number of potential explanatory variables relating to population numbers and characteristics, traffic exposure, road length, curvature and junction density, land use, elevation and hilliness, and climate. Multilevel negative binomial regression models were used to identify combinations of risk factors that predicted variations in mortality. Statistically significant explanatory variables were the expected number of casualties derived from the size and age structure of the resident population, road length and traffic counts in the district, the percentage of roads classed as minor, average cars per capita, material deprivation, the percentage of roads through urban areas and the average curvature of roads. This study demonstrated that a geographical approach to road traffic crash analysis can identify contextual associations that conventional studies of individual road sections would miss. An urban-rural difference in traffic injuries has been recorded with a higher rural case-fatality rate. A number of known behavioral risk factors have been identified, i.e., drunk driving, speeding, substance abuse and failure to use helmets and seat belts (Suriyawongpaisal and Kanchanasut, 2003). The limitation of this research is that the data has many zero cases in small cells, 23,590 of 42,300 (55.77%) cells. In further study, we should use other methods of analysis, such as multiple imputations technique to solve this problem (Sterne et al., 2009).
such forecasts of transportation accident mortality rates, even if based purely on statistical data analysis, can provide a useful basis for allocation of resources for road accident fatalities prevention. Road accident in Thailand is a major social and economic problem, which causes a lot of losses in lives and injuries each year. Increasing transport service demands result in increasing the number of vehicle kilometers and road traffic accidents. Even though the trend of road accidents is increasing, the severity is tending to decrease. Fatality rate is a common use as the primary indicator in ranking the severity of the road safety situation in Thailand. Strategies for reducing the mortality rate due to transportation accidents involve not only gaining a better understanding of the risk factors for mortality but also finding measures to prevent transportation accidents as well. Acknowledgements We thank the Ministry of Interior for providing the data and Prof. Donald McNeil for supervising our research. References Bride, U. (1995) What is Happening to the Number of Fatalities in Road Accidents? A Model for Forecasts and Continuous Monitoring of Development up to the 2000. Accident Analysis and Prevention 27(3): 405-410. Butt, Z. and Haberman, S. ( 2009) A Collection of R Functions for Fitting a Class of LeeCarter Mortality Models using Iterative Fitting Algorithms, Sir John Cass Business School, London (UK). Congdon, P. ( 2006) A model framework for mortality and health data classified by age, area, and time. Biometrics 62: 269-278. Jonesa, A. P., Haynesa, R., Kennedy, V., Harveyb, I. M, Jewellc, T., and Lead,. D. (2008) Geographical variations in mortality and morbidity from road traffic accidents in
Conclusions The objective of this study was to model and forecast transportation accident mortality rates in Thailand using death certificate reports. The multivariate linear regression was used for modeling and forecasting age-specific transportation accident mortality rates in Thailand in order to prevent road accident fatalities by using suitable measures. Having a model that provides 16
W. Sriwattanapongse et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
England and Wales. Health Place 14(3): 519-35. Kardara, M. and Kondakis, X. (1997) Road traffic accidents in Greece: Recent trends (19811991). European Journal of Epidemiology 13(7): 765-70. Lee, R. D. and Carter, L. R. (1992) Modeling and Forecasting U.S. Mortality. American Statistics Association 87(419) : 659-671. Lee, R. D. and Miller, T. (2001) Evaluating the Performance of the Lee-Carter Method for Forecasting Mortality. Demography 38(4): 537-49. Lee, H. A., Stevenson, R. M., Wang, K., and Yau, K. K. ( 2002) Modeling young driver motor vehicle crashes: data with extra zeros. Accident Analysis and Prevention 34(4): 515-21. Lix, L. M., Ekuma, O., Brownell, M., and Roos, L. L. ( 2004.) A framework for modeling differences in regional mortality over time. Journal of Epidemiology and Community Health 58(5): 420-5. Lord, D. (2006) Modeling motor vehicle crashes using Poisson-gamma models: Examining the effects of low sample mean values and small sample size on the estimation of the fixed dispersion parameter. Accident Analysis and Prevention 38(4): 751-66. Nantulya, M. V. and Reich, R. M. (2003) Equity dimensions of road traffic injuries in low and middle-income countries. Injury Control and Safety Promotion 10(1-2): 13-20. National Statistics Office (2010) 100th anniversary of population censuses in Thailand: Population and housing census. [Online URL: www.popcensus.nso.go.th/] Neill B. O. and Mohan, D. (2002) Reducing motor vehicle crash deaths and injuries in newly motorizing countries. British Medical Journal 324 (7346): 1142-1145.
Pocock, J. S., Cook, G. D., and Beresford, A. S. (1981) Regression of area mortality rates on explanatory variables: What weighting is appropriate? Journal of the Royal Statistical Society, Series C Applied statistics 30(3): 286-95. Sriwattanapongse, W. and Khanabsakdi, S. (2011) Modeling of Mortality Rate due to Traffic Accident, Phayao Province, In Proceedings of the National Statistics and Applied Statistics Conference, Hat Yai, Songkhla, Thailand. Sterne, J. A. C., White, I. R., Carlin, J. B., Spratt, M., Royston, P., Kenward, M. G.,Wood, A. M., and Carpenter, J. R. (2009) Multiple imputations for missing data in epidemiological and clinical research: potential and pitfalls, British Medical Journal 338:b2393, DOI: 10.1136/bmj. b2393. Suriyawongpaisal, P. and Kanchanasut, S. (2003) Road traffic injuries in Thailand: Trends, selected underlying determinants and status of intervention. Injury Control and Safety Promotion 10(1-2): 95-104. Tanaboriboon, Y. (2004) ‘The Status of Road Safety in Thailand’, In the ADB-ASEAN Regional Safety Program, Country Report: CR09, Final Report, Thailand. Tsauo, J. Y., Lee, W. C., and Wang, J. D. (1996) Age-period-cohort analysis of motor vehicle mortality in Taiwan, 1974-1992. Accident Analysis and Prevention 28(5): 619-26. Wikipedia (2011) [Online URL: www.en.wikipedia. org/wiki/Provinces_of_Thailand] accessed on May 20, 2012. World Health Organization (2008) World Health Statistics, World Health Organization, Geneva. [Online URL: www.who.int/ whosis/whostat/EN_WHS08_Full.pdf] accessed on June 10, 2012. 17
Silpakorn U Science & Tech J Vol.7(1), 2013
Mortality Rate Model due to Transportation Accidents in Thailand
the Guangxi, China. Accident Analysis and Prevention 37(1): 137-141.
Yang, L., Lam, T. L., Lui, Y., Geng, K. W., and Lui, C. D. (2005) Epidemiological profile of mortality due to injuries in three cities in
18
Research Article Evaluation of the Shade Tolerance of Moth Bean (Vigna aconitifolia) and Two Tropical Legume Species Pantipa Na Chiangmai*, Yupa Pootaeng-on and Thanakrit Khewaram Faculty of Animal Sciences and Agricultural Technology, Silpakorn University, Phetchaburi IT Campus, Cha-Am, Phetchaburi, Thailand * Corresponding author. E-mail address: m_surin@yahoo.com Received March 1, 2012; Accepted October 17, 2012 Abstract This study aimed to evaluate the shading tolerance ability in moth bean (Vigna aconitifolia) in two experiments. In experiment 1, moth bean, Centrosema pascuorum cv. Cavalcade and Stylosanthes guianensis cv. Tha pra (Tha pra stylo) were grown under different sunlight shading levels by covering these plots with a black net. In Experiment 2, moth bean was intercropped with sunflower (Helianthus annuus) synthetic variety (Suranaree (S) 471). Both experiments were conducted in a field trial from May to August in 2011 at Agricultural Practice Farm of Faculty of Animal Science and Agricultural Technology, Silpakorn University, Phetchaburi Information Technology Campus, Phetchaburi, Thailand. In Experiment 1, the results showed that shading reduced almost all plant growth characteristics, except plant height. All of legume species had possessed different tolerant capacity to shading, but these plants were dead at 90% of the shading level. In Experiment 2, the agronomic traits of sunflower were not affected by intercropping with moth bean. Fresh and dry weight per plant of moth bean intercropping with sunflower was decreased, comparing with those of a moth bean monoculture. However, there was no significant difference in crude protein content of moth bean between monoculture and intercropping with sunflower. Key Words: Vigna; Moth bean; Shading; Intercropping; Forage crop Introduction The Vigna species, a diverse plant species which can grow under a wide range of climate and environment (Kharb et al., 1987) has been introduced worldwide. Some species such as V. radiata (mungbean) and V. unguiculata (cowpea) are used for human diet and for improving the soil fertility. Vigna species also have the nutritive value which is suitable for livestock production (Chujaroen et al., 2006; Kongcharoen et al., 2006;
Silpakorn U Science & Tech J 7 (1) : 19-31, 2013
Na Chiangmai et al., 2009a; Na Chiangmai et al., 2009b). Moth bean (V. aconitifolia) is one of Vigna species that possess a higher nutritional value than V. unguiculata line KVC7, V. radiata var. sublobata (TC1995), Centrosema pascuorum cv. Cavalcade and Stylosanthes guianensis cv. Tha pra stylo), both under normal and under water stress conditions (Na Chiangmai et al., 2009b). For this reason, moth bean is considered as an
Silpakorn U Science & Tech J Vol.7(1), 2013
Evaluation of the Shade Tolerance of Moth Bean
important forage crop in tropical region in the near future. This bean is thus a potential plant which may be used for intercropping with other economically important crops, such as coconut, para-rubber and oil palm. These crops are perennial trees which have been planted with a space between rows wide enough for growing other annual economic plants. As a result, the shading effect of the trees may have detrimental effect on the productivity of the intercropped annual plants. This study thus aimed to evaluate shading tolerance of moth bean in two situations both as monoculture and under intercropping system with sunflower. Shading tolerance of moth bean will be compared with other two legume species (Cavalcade and Tha pra stylo). The results of this study can be used in the breeding program to obtain better varieties of the beans. The output of this research may also be useful to farmers in choosing crops for mixed or intercropping.
Both experiments were carried out from May to August, 2011, at Phetchaburi, Thailand. The area has an elevation approximately 0.4 meter above sea level at N12°37.780´E099°51.067´. The test was conducted when it was the rainy season with an average rainfall of 191.5 mm in May and 108.5 mm in August. The averages of temperature were 27.7 °C in May and 27.3 °C in August. The averages of duration of sunshine were 5.5 hours/day in May and 4.5 hours/day in August, with a mean of a daylength at 12.60 hours. Growing conditions and Experimental design In experiment 1, two seedlings of each plant species (moth bean, Cavalcade and Tha pra stylo) were planted per hill, using the spacing 50 cm between rows and 25 cm between hills in 3 x 5.75 m2 plot size. Each block was divided into three plots. Each of three legume species was randomized for planting in each plot. However, the different on growth characteristic in these species made it difficult for comparison among them. Thus, the observation for the behavior and separated analysis in each species when was grown under different shading levels was designed in this experiment. Thus, Randomized Completely Block Design (RCBD) with four replications were conducted in this experiments in these legume species. Detail of plot layout and experiment design was given in Figure 1. The data were gathered with respect to number of days after planting when the seedlings emerged, plant height, number of branches and number of leaves, lodging score, fresh weight and dry weight per plant. The data was recorded at about 3 weeks after planting; this date was the first stage when the farmer controlled weed and applied fertilizer. And, the second record was replicated at about two months after planting; this date is the end of vegetative phase in various forage legume species.
Materials and Methods Genetic materials and seed preparing In experiment 1, moth bean (V. aconitifolia) was grown under different shading levels (0, 30, 70 and 90 percent of shading) by covering with black plastic nets to simulate various level of shading. The moth bean was tested in comparison with Centrosema pascuorum cv. Cavalcade and Stylosanthes guianensis cv. Tha pra (Tha pra stylo). In experiment 2, moth bean was grown between rows of a synthetic variety sunflower (Helianthus annuus), Suranaree (S) 471 in intercropping system. The moth bean and synthetic variety of sunflower were grown in the sandy clay, and low fertile and slightly acidic soil. Seeds were harvested from the field facility at Agricultural Practice Farm of the Faculty of Animal Science and Agricultural Technology, Silpakorn Unversity, IT Campus, Cha-am District, Phetchaburi Province, Thailand.
20
P. Na Chiangmai et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
Figure 1 Diagram for experimental set-up for Randomized Completely Block Design (RCBD) with four replications in each legume species.
In experiment 2, moth bean and sunflower were sown at spacing of 0.50 m x 0.25 m, 0.75 x 0.25 m between rows and hills, respectively, and plot sizes were 3 x 5.75 m2. For intercropping, two seeds of moth bean and one seed of sunflower were planted per hill. The intercropping was arranged alternately between rows among these two species (one row of sunflower was planted alternately with one row of moth bean). Thus, it has two treatments (monoand intercropping) which were arranged in RCBD with three replications. The data were collected as fresh and dry weight per plant of moth bean at 100 days after planting which was the same date for harvesting of sunflower seeds. At maturity (about 100 days after planting) of sunflower, six traits such as plant height, disk size, head dry weight, seed number per head, shelling percentage and 100 seeds weight were determined. In moth bean, the upper ground
part (or shoot part) was taken for dry matter and crude protein content analysis. In both experiments, the plants were irrigated twice a day and weed control was conducted manually once at 30 days after planting. No fertilizer was applied in the experiment. Results For experiment 1, the results were shown in Table 1, 2, 3 and 4. Different shading levels were found to affect all agronomic traits by decreasing the values of characteristics in moth bean, Cavalcade and Tha pra stylo, except branch number and lodging score in Tha pra stylo (Table 1-3). It was found that different shading levels affected all traits of moth bean, except lodging score and plant height between shading levels at 0% to 70% at 24 and 68 days after planting, respectively (Table 1).
21
Silpakorn U Science & Tech J Vol.7(1), 2013
Evaluation of the Shade Tolerance of Moth Bean
Table 1 Characteristics of moth bean grew under different levels of sunlight shading (mean ± SE) and value percentage from control values (in the parentheses in lower line) at 24 and 68 days after planting. Sunlight shading levels
Germination day
Plant height
Leaf number
(DAP†)
(cm)
(no./plant)
4.5±0.5a
5.03±0.71c
3.25±0.03a
3.85±0.15
(100%)
(100%)
(100%)
(100%)
3.0±0.0b
11.25±0.99b
2.90±0.06b
3.40±0.16
(67%)
(224%)
(89%)
(88%)
3.0±0.0b
14.75±0.75a
2.25±0.09c
3.25±0.18
(67%)
(293%)
(69%)
(84%)
3.0±0.0b
NA§
NA
NA
Lodging score£
at 24 DAP 0% (control) 30% 70% 90%
(67%) Mean
3.4
10.34
2.80
3.50
F-test
**
**
**
NS‡
LSD (0.05)
0.80
2.69
0.25
0.54
Sunlight shading
Branch number
Plant height
Leaf number
Lodging score
(no./plant)
(cm)
(no./plant)
11.75±0.85a
101.50±14.91
85.00±6.77a
4.25±0.25a
(100%)
(100%)
(100%)
(100%)
3.25±1.38b
94.00±2.68
32.00±3.19b
1.75±0.25b
(28%)
(93%)
(38%)
(41%)
0.00±0.00c
117.75±5.12
8.75±0.48c
1.00±0.00b
(0%)
(116%)
(10%)
(24%)
90%
Dead
Dead
Dead
Dead
Mean
5.00
105
41.92
2.33
F-test
**
NS
**
**
LSD (0.05)
2.95
32.41
11.21
0.87
levels at 68 DAP 0% (control) 30% 70%
DAP, days after planting. NS, not significant at the 0.05 level of probability. ** significant at the 0.01 level of probability. § NA, not applicable. £ Lodging score, 4 = erect; 3 = curve not over 45 degree; 2 = curve over 45 degree; 1 = flat on floor. † ‡
22
P. Na Chiangmai et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
At 24 days after planting of moth bean, number of days to emergence, plant height and leaf number per plant were affected, in response to the increased shading level between 0%-70% (Table 1). Leaf number per plant decreased, when compared with the control treatment. Date of germination also decreased as the shading levels had increased. Plant height, however, had responded differently from other traits to shading level in that this trait increased by 224% and 293% from that of the control treatment at 30 and 70% shading levels, respectively. At 68 days after planting of moth bean, there were significant difference in number of branches, number of leaves and lodging score. At high shading levels, there were the decreased in a number of branches, number of leaves and lodging score, comparing with those of the control treatment (as 28, 38 and 41 percent at 30% shading level, and as 0, 10 and 24 percent at 70% shading level, respectively). There was significant difference in plant height between the different levels of shading. All the plants of three legume species died after exposure to 90% of shading level. The Cavalcade was affected by low shading level on all traits both at 24 and 68 days after planting (Table 2). At 24 days after planting, high shading level showed clear effect with the reduction in the leaf number per plant. At 68 days after planting, at 70% shading level, branch number and leaf number per plant decreased substantially, comparing with those at normal light level (control treatment) (Table 2) (38 and 27 percents, respectively). However, at 90% shading level, all plants stopped growing and plants had wilt symptom and died. For Tha pra stylo, at 68 days after planting, different levels of light shading had an effect on plant height and leaf number. The reduced values of these traits from control levels were 46 and 57%, respectively at 70% shading level (Table
3). Branch number and lodging score were not significantly different at different levels of shading in this stage of growing. Fresh and dry weight of all legume species decreased as the level of shading increased (Table 4). The reduction of both fresh weight and dry weight per plant at 30% shading of Cavalcade was less than moth bean and Tha pra stylo, comparing with the control treatment (Table 4). The values of fresh and dry weight per plant at 30 % shading were 39, 17 and 3 percent of control treatment value, and were 38, 17 and 4 percent of control treatment value in Cavalcade, moth bean and Tha pra stylo, respectively. However, at 70% of shading level, both fresh weight and dry weight per plant values were lower than 10%, comparing with the control treatment in all species. In experiment 2, the results were shown in Table 5 and 6. Crude protein content was not affected by the different growing practices. The crude protein content of moth bean grown in monoculture was 10.88%, while that of the moth bean grown together with sunflower in a mixedculture was 13.10% (Table 5). Percent dry matter of moth bean grown under monoculture and mixedculture were not significant different. Only dry weight per plant of moth bean was affected by the growing practice (Taber 6). In monoculture, dry weight per plant of moth bean was higher (20.70 g/plant) than growing as intercrop with sunflower (10.53 g/plant). As for sunflower, all of the six agronomic traits were not affected by the different in growing practices; either in monoculture or in intercrop with moth bean (Table 6). Discussion In Experiment 1, the decreasing values of growth characteristics in moth bean, Cavalcade and Tha pra stylo caused by high shading level reflected the importance of light for plant growth (Table 1, 2 and 3). Although some characteristic 23
Silpakorn U Science & Tech J Vol.7(1), 2013
Evaluation of the Shade Tolerance of Moth Bean
Table 2 Characteristics of Cavalcade grew under different levels of sunlight shading (mean ± SE) and value percentage from control values (in the parentheses in lower line) at 24 and 68 days after planting. Sunlight shading levels
Plant height
Leaf number
(cm)
(no./plant)
Lodging score
at 24 DAP† 0% (control)
NA§
NA
NA
30%
11.39±0.99b
2.65±0.13a
3.95±0.05a
70%
14.33±0.50a
2.02±0.06b
3.63±0.04b
90%
8.64±1.09c
0.62±0.22c
3.89±0.08a
Mean
11.45
1.76
3.82
F-test
**
**
*
LSD (0.05)
2.17
0.44
0.22
Branch number
Plant height
Leaf number
(no./plant)
(cm)
(no./plant)
6.50±0.29a
146.60±6.51a
51.25±2.69a
4.00±0.00a
(100%)
(100%)
(100%)
(100%)
3.50±0.29b
172.13±18.69a
24.75±4.09b
2.00±0.00b
(54%)
(117%)
(48%)
(50%)
2.50±0.65b
103.00±2.08b
14.00±1.78b
2.00±0.00b
(38%)
(70%)
(27%)
(50%)
Dead
Dead
Dead
Dead
Sunlight shading levels
Lodging score£
at 68 DAP 0% (control) 30% 70% 90% Mean
4.17
140.58
21.25
2.67
F-test
**
**
**
**
LSD (0.05)
2.95
32.41
11.21
0.87
DAP = Days after planting. significant at the 0.05 level of probability.
† *
** § £
significant at the 0.01 level of probability. NA = Not applicable. Lodging score, 4 = erect; 3 = curve not over 45 degree; 2 = curve over 45 degree; 1 = flat on floor.
24
P. Na Chiangmai et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
Table 3 Characteristics of Tha pra stylo grew under different levels of sunlight shading (mean ± SE) and value percentage from control values (in the parentheses in lower line) at 68 days after planting. Sunlight shading
Branch number
Plant height
Leaf number
(no./plant)
(cm)
(no./plant)
0.75±0.75
43.38±0.63a
12.25±2.39a
4.00±0.00
(100%)
(100%)
(100%)
(100%)
0.00±0.00
10.75±1.11c
4.75±0.85b
4.00±0.00
(0%)
(25%)
(39%)
(100%)
0.00±0.00
20.00±1.74b
7.00±0.71b
3.50±0.29
(0%)
(46%)
(57%)
(88%)
90%
Dead
Dead
Dead
Dead
Mean
0.25
24.71
8.00
3.83
F-test
NS‡
**
*
NS
LSD (0.05)
1.50
4.17
4.36
0.58
levels 0% (control) 30% 70%
Lodging score£
NS, not significant at the 0.05 level of probability. significant at the 0.05 level of probability. ** significant at the 0.01 level of probability. £ Lodging score, 4 = erect; 3 = curve not over 45 degree; 2 = curve over 45 degree; 1 = flat on floor. ‡ *
such as branch number and lodging score in Tha pra stylo were not significant different between shading levels at 0%-70%, all plants of the three legume species were wilted and died at 90% of shading level. This evidence showed that high shading level severely affected all three legumes. For moth bean, all plants wilted and died at 90% shading level at 68 days after planting (Table 1). This means that it can not withstand high level of shading at this age. However, moth bean could survive lower levels of shading at 24 days after planting (Table 1). Plants are exposed to some degree of shade during their growth and development (Valladares and Niinemets, 2008) and this caused a differential tolerance on shading at different growing stages of plant. Lodging scores from different shading were not different at 24 days after planting in moth bean.
This was because plants were still at the early stage of development. At 68 days after planting, plant height was the only one trait which was not affected by the shading level between of 0% to 70% (Table 1). This was because plant height had the negative direction for value changing, comparing with other traits. Three traits (such as date of germination, plant height and leaf number per plant) were highly affected by shading level in moth bean at 24 days after planting (Table 1). This result suggests that difference in shading tolerance could be evaluated by observing the characteristics of these traits. The reduced values of two traits (such as date of germination and leaf number per plant) were found at higher level of shading when comparing with the control treatment. At higher shading level, moth bean showed decreasing value in date of germination and leaf number per plant
25
Silpakorn U Science & Tech J Vol.7(1), 2013
Evaluation of the Shade Tolerance of Moth Bean
Table 4 Fresh and dry weight per plant of legume species grew under different levels of sunlight shading (mean ± SE) and value percentage from control values (in the parentheses in lower line) at 68 days after planting. Traits
Sunlight shading levels 0% (control)
30%
70%
90%
Mean
F-test
LSD
69.81±4.47a
11.55±1.46b
3.14±0.45b
Dead
28.17
**
8.55
(100%)
(17%)
(4%)
13.42±0.89a
2.22±0.32b
0.61±0.07b
Dead
5.42
**
1.68
(100%)
(17%)
(5%)
22.54±6.93a
8.78±2.46ab
1.64±0.02b
Dead
10.99
*
15.5
(100%)
(39%)
(7%)
5.67±1.76a
2.16±0.63ab
0.44±0.02b
Dead
2.76
*
3.94
(100%)
(38%)
(8%)
3.45±1.00a
0.12±0.04b
0.03±0.00b
Dead
1.20
**
2.01
(100%)
(3%)
(1%)
0.85±0.26a
0.03±0.01b
0.01±0.00b
Dead
0.30
*
0.52
(100%)
(4%)
(1%)
Moth bean Fresh weight Dry weight
Cavalcade Fresh weight Dry weight
Tha pra stylo Fresh weight Dry weight
*
significant at the 0.05 level of probability. significant at the 0.01 level of probability.
**
Table 5 Characteristics of crude protein and dry matter percentage of moth bean grew under mono- and mixed culture (with sunflower). Cultures
‡
% crude protein (on dry basis)
% dry matter
Monoculture
10.88±3.13
10.05±0.28
Mix culture
13.10±1.34
10.28±0.30
Mean
11.99
10.16
F-test
NS‡
NS
NS, not significant at the 0.05 level of probability.
26
P. Na Chiangmai et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
Table 6 Characteristics of moth bean and sunflower grew under monoculture and mix culture. Cultures
Plant height
Fresh plant weight
Dry plant weight
(cm)
(g/plant)
(g/plant)
Monoculture
161.97±7.43
137.97±28.59
20.70±2.76a
Mix culture
151.50±12.12
55.47±6.80
10.53±0.91b
Mean
156.74
96.72
15.62
F-test
NS‡
NS
*
LSD (0.05)
39.89
93.76
8.23
Plant height
Head diameter
Head dry weight
(cm)
(cm)
(g)
Monoculture
169.53±5.10
12.53±0.38
18.90±2.17
Mix culture
171.37±6.03
13.27±0.79
20.80±2.73
Mean
170.45
12.90
19.85
F-test
NS
NS
NS
LSD (0.05)
5.69
4.14
2.62
Seed number
% shelling
100 seed weight
Moth bean
Cultures Sunflower
Cultures
(no./plant)
(g)
Sunflower
‡ *
Monoculture
620±102
51.83±3.69
3.57±0.25
Mix culture
595±60
50.77±0.66
3.33±0.13
Mean
608
51.30
3.45
F-test
NS
NS
NS
LSD (0.05)
396.58
13.24
1.50
NS, not significant at the 0.05 level of probability. significant at the 0.05 level of probability.
27
Silpakorn U Science & Tech J Vol.7(1), 2013
Evaluation of the Shade Tolerance of Moth Bean
(Table 1). Thus, the date of germination was the first parameter to indicate the stress of shading for the growing of moth bean. Although leaf number per plant in moth bean did not constantly decrease, this trait had higher shading tolerance when comparing with date of germination which remained at higher percentage at 30% and 70%, comparing with at 0% of shading level (Table 1). The response of plants in increasing height occurred because they competed for light for plant photosynthesis (Porter, 1937; Bunce et al., 1977; Nagashima et al., 1995; Schmitt et al., 1999). Thus, the stem extension reached to more than 2 times of control treatment. The significant effect of shading level was found on lodging score only at 68 days after planting. At 30% shading level, plant lodging was also observed and the score was less than 50% when comparing with the control treatment. Plants grown in the shading typically have low root/shoot ratio, possess slender stems, and have low branching (Smith, 1982; Weiner et al., 1990; Sparkes et al., 2008). Thus, they were easily lodged because the stem diameter, stem stiffness, and root anchoring were reduced (Niklas, 1998). Although light is an important resource for photosynthesis, both insufficient and excess sunlight can limit the performance of plants (Grubb, 1998; Valladares and Niinemets, 2008). The elongation of plant shoot in low light is one of the trait that shows phenotypic plasticity, which tends to be high in non shade-tolerant species. While phenotypic plasticity tends to be low in shade-tolerant species (e.g., scant elongation in low light), plasticity for certain traits, particularly for morphological features such as shoot elongation will optimizes light capturing (Valladares and Niinemets, 2008). Lodging was found in plant receiving more shading level than normal light condition. There were several reports that lodging had limited yield
in many plants such as barley (Day, 1957), grain sorghum (Larson and Maranville, 1977) and rice (Basak et al., 1962; Setter et al., 1997). The changing in all traits under different shading levels could be observed in Cavalcade (Table 2). Number of leaves per plant was clearly decreased at different levels of shading (Table 2). This indicates that shading stress can be observed at the early stage of plant growth in Cavalcade. Number of branches per plant could also be an indicator of insufficiency of light for Cavalcade at the later growth stage (at 68 days after germination). Tha pra stylo is a perennial forage legume species, and its initial growth is slower than other two legume species in this study (Table 3). Number of leaves per plant and plant height could clearly be an indicator of the unavailable of light for Tha pra stylo at this growth stage. Plant height of moth bean and cavalcade, however, increased at high shading level, except in Tha pra stylo (Table 3). This was because legume species were short-lived perennial legume (2 to 3 years) (Phengsavanh and Inger, 2003), making it to grow slowly at the early planting stage, compared with moth bean and Cavalcade which were annual plant species. Fresh weight and dry weight per plant were decreased with the increased shading level (Table 4) and this resulted to the reduction in yield, biomass production and accumulation (Ephrath et al., 1993; Henry and Thomas, 2002; Liu et al., 2010). Thus yield of three legumes species was affected by higher shading level. At 30% shading level, the percentage of fresh and dry weight per plant of Cavalcade was higher than those of moth bean and Tha pra stylo (Table 4). Moth bean has higher vegetative yield than Cavalcade in normal sunlight level, under both normal water application and water deficiency (Na Chiangmai et al., 2009a). Thus, moth bean was sensitive to shading, and this showed that it had less shading tolerance than Cavalcade. For Tha pra 28
P. Na Chiangmai et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
stylo, due to it is slow growth at initial stage, the degree of shading tolerance cannot be compared with that of the other species. In Experiment 2, there was no significant different on crude protein content from vegetative part between mono- and mixed-culture of moth bean with sunflower (Table 5). This evidence showed that crude protein content of moth bean was not affected under intercropping with sunflower. The crude protein content under intercropping of moth bean with sunflower was lower than that reported by Na Chiangmai et al. (2009b), a result from the pot condition study. Although moth bean grown under intercropping condition had lower value of dry plant weight than that of monoculture (Table 6), the benefit of intercropping was that this practice could control weed proliferation when comparing with that of the sunflower monoculture. In competition for light, growth of smaller plants has been affected by shading more than larger plants (Casper and Jackson, 1997). The competition begins when a single necessary factor (as light in this case) is the common demands of the plants (Went, 1973). Moreover, growth reduction also occurred as a result of growing plants at closer proximity (Clements et al., 1929). Both belowground and aboveground competition occurred when the available soil resources were limit (Casper and Jackson, 1997). For aboveground, light is an influencing factor for plant growth and survival and results to competitive interactions in the community (Canham et al., 1990; Valladares, 2003). Thus, the minimum light required for survival and shade tolerance plays a major role in plant community dynamics (Valladares and Niinemets, 2008). Under mixed-culture, dry plant weight of moth bean was lower than that of monoculture at about twofold. However, there was no significant difference on yield between mono- and mixedculture in traits of sunflower (Table 6). Thus, moth
bean can be grown as intercrop with sunflowers. Conclusion This study emphasized the importance of shading level for legumes production, particularly at the early growth stage in three legumes species (moth bean, Cavalcade and Tha pra stylo). Shading effect reduced the value of all characteristics, except plant height which increased due to shoot elongation. All of three legume species had different shading tolerance ability, except at 90% shading level that they did not survive. Although many traits were affected by shading in all of three legume species, leaf number per plant and dry shoot weight per plant were mostly affected. Thus, both of these traits are suitable to use as an indicator for shading tolerance. By comparing dry weight per plant, moth bean showed lower shading tolerance than Cavalcade but higher than Tha pra stylo. In mixed-culture, performance of sunflower was not affected by intercropping. But, intercropped moth bean had lower fresh and dry weight per plant about twofold, comparing with the monoculture. However, crude protein content (based on dry basis) was not different between mono- and mixedculture. This suggest that moth bean can be used for planting between rows of sunflower for weed control and forage crop production. Acknowledgements We gratefully thank the Faculty of Animal Sciences and Agricultural Technology, Silpakorn University, Thailand for funding and supports during the course of this research. References Bunce, J., Patterson, D. T., and Peet, M. M. (1977) Light acclimation during and after leaf expansion in soybean. Plant Physiology 60: 255-258. 29
Silpakorn U Science & Tech J Vol.7(1), 2013
Evaluation of the Shade Tolerance of Moth Bean
Basak, M. N., Sen, S. K., and Bhattacharjee, P. K. (1962) Effects of high nitrogen fertilization and lodging on rice yields. Agronomy Journal 54: 477-480. Canham, C. D, Denslow, J. S, Platt, W. J., Runkle, J. R., Spies, T. A., and White, P. S. (1990) Light regimes beneath closed canopies and tree-fall gaps in temperate and tropical forests. Canadian Journal of Forest Research 20(5): 620-631. Casper, B. B. and Jackson, R. B. (1997) Plant competition underground. Annual Review of Ecology and Systematics 28: 545-570. Chujaroen, S., Kongcharoen, A., and Na Chiangmai, P. (2006) Characterization of yield and yield components in Vigna spp. Kamphaengsaen Academic Journal 4: 733-739. Clements, F. E., Weaver, J. E., and Hanson, H. C. (1929) Plant competition analysis of community functions, Carnegie Institution of Washington publication. Day, A. D. (1957) Effect of loding on yield, test weight and other seed characteristics of spring barley grown under flood irrigation as a winter annual. Agronomy Journal 49: 536-539. Ephrath, J. E., Wang, R. F., Terashima, K., Hesketh, J. D., Huck, M. G., and Hummel, J. W. (1993) Shading effects on soybean and corn. Biotronics 22: 15-24. Grubb, P. J. (1998) A reassessment of the strategies of plants which cope with shortages of resources. Perspectives in Plant Ecology, Evolution and Systematics 1: 3-31. Henry, H. A. and Thomas, S. C. (2002). Interactive effects of lateral shade and wind on stem allometry, biomass allocation, and mechnical stability in Abutilon Theophrasti (Malvaceae). American Journal of Botany 89(10): 1609-1615. Kharb, R. P. S., Singh, V. P., and Tomar, Y. S. (1987) Moth bean (Vigna aconitifolia Jacq.
(Mareehal)). A review Forage Research Journal. Kongcharoen, A., Chujaroen, S., Nilprapruck, P., Pummarin, P., and Na Chiangmai, P. (2006) Comparison of nutritional values in various species of Vigna spp. Kampaengsaen Academic Journal 4: 740-746. Larson, J. C. and Maranville, J. W. (1977) Alternations of yield, test weight, and protein in lodged grain sorghum. Agronomy Journal 69: 629-630. Liu, B., Liu, X. B., Wang, C., Li, Y. S., Jin, J., and Herbert, S. J. (2010) Soybean yield and yield component distribution across the main axis in response to light enrichment and shading under different densities. Plant Soil Environment 56(8): 384-392. Na Chiangmai, P., Nanongtoom, S., and Arunkeereewat, S. (2009a) The effect of drought manipulation on seed yield and seed yield component characters in Vigna spp. and Centrosema pascuorum cv. Cavalcade in the field. In Proceedings: Second International Conference on Suctainable Animal Agriculture for Developing Countries (SAADC 2009), Kuala Lumpur, Malaysis. Na Chiangmai, P., Chansem, T., and Bootnoi, S. (2009b) Drought manipulation: Effects on nutritive values of legume species; Vigna spp., Centrosema pascuorum cv. Cavalcade and Stylosanthes guianensis cv. Tha pra. In Proceedings: Second Interactional Conference on Suctainable Animal Agriculture for Developing Countries (SAADC 2009), Kuala Lumpur, Malaysis. Nagashima, H., Terashima, I., and Katoh, S. (1995) Effects of plant density on frequency distributions of plant height in Chenopodium album stands: analysis based on continuous monitoring of height growth of individual plants. Annals of Botany 75: 173-180. Niklas, K. J. (1998) The influence of gravity and 30
P. Na Chiangmai et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
wind on land plant evolution. Review of Paleobotany and Palynology 102: 1-14. Phengsavanh, P. and Inger, L. (2003) Effect of Stylo 184 (Stylosanthes guianensis CIAT 184) and Gamba grass (Andropogon gayanus cv. Kent) in diets for growing goats; Livestock Research for Rural Development 15(10). [Online URL:www.lrrd.org/lrrd15/10/ seut1510.htm]. Accessed on September 27, 2012. Porter, A. M. (1937) Effect of light intersity on the photosynthetic efficiency of tomato plants. Plant Physiology 12(2): 225-252. Schmitt, J., Dudley, S. A., and Pigliucci, M. (1999) Manipulative approaches to testing adaptive plasticity: phytochrome-mediated shadeavoidance responses in plants. American Naturalist 154 (suppl.): S43-S54. Setter, T. L., Laureles, E. V., and Mazaredo, A. M. (1997) Lodging reduces yield of rice by self-shading and reductions in canopy photosynthesis. Field Crops Research 49: 95-106. Smith, H. (1982) Light quality, photoreception, and plant strategy. Annual Review of Plant
Physiology 33: 481-518. Sparkes, D. L., Berry, P., and King, M. (2008) Effects of shade on root characters associated with lodging in wheat (Triticum aestivum). The Annuals of Applied Biology 158(3): 389-395. Valladares, F. (2003) Light heterogeneity and plants: from ecophysiology to species coexistence and biodiversity. In Progress in Botany (Esser, K., Luttge, U., Beyschlag, W., and Hellwig, F., eds.), pp. 439-471. SpringerVerlag, Heidelberg. Valladares, F. and Ăœlo Niinemets. (2008) Shade tolerance, a key plant feature of complex nature and consequences. Annual Review of Ecology, Evolution, and Systematics 39: 237-257. Weiner, J., Berntson, G. M., and Thomas, S. C. (1990) Competition and growth form in a woodland annual. Journal of Ecology 78: 459-469. Went, F. W. (1973) Competition Among Plants. In Proceedings of the National Academy of Science of the United States of America 70(2): 585-590.
31
Research Article Odds Prediction of Drought Category Using Loglinear Models Based on SPI in the Northeast of Thailand Wisoot Salee and Veeranun Pongsapukdee* Department of Statistics, Faculty of Science, Silpakorn University, Nakhon Pathom, Thailand * Corresponding author. E-mail address: veeranun@su.ac.th; veeranun@hotmail.com. Received October 8, 2012; Accepted November 13, 2012 Abstract The prediction of drought category is performed through Loglinear modeling for three dimensional contingency tables. The frequencies of drought categories are evaluated from a 12-month time scale by means of Standardized Precipitation Index (SPI), using the raw data obtained from 19 rain gauge stations in 19 provinces in the Northeast of Thailand. SPI monthly values were computed in a 12 month time scale for the period from January 1962 to December 2009, within 48 years or 576 months in each of 19 rainfall stations. The results show that the selected quasi-association loglinear model is an adequate model and appropriate tool to fit the data from 19 stations (P-value = 0.927). The values of various drought class transitions are estimated in three consecutive months. The predicted odds ratios and the corresponding confidence intervals are evaluated to predict the drought classes’ transitions. Even if, most of the results display for more normal drought classes than those of moderate or severity drought classes; however, there are still some areas of investigation that drought could be developed. Therefore, future monitoring of drought watch system is needed to be set the pace in these areas to keep up with them. Furthermore, an appropriate statistical model to predict drought phenomenon in each area is necessary and is probably able to provide more effectiveness in the administrative management of hazard from drought. Key Words: Odds ratio; Drought class transitions; Three-dimensional loglinear models Introduction Thailand has approximately all year warm climate and attractively is an agricultural and rice export country in the world. Drought is particularly important in affecting both agriculture and climate; especially, in the northeast of the country. Prediction or forecasting of drought initiation and ending are both essential for timely and appropriate implementation of measures to cope with drought as well as to make drought warning is possible.
Silpakorn U Science & Tech J 7 (1) : 32-40, 2013
Frequency and severity make drought both a hazard and a disaster: a hazard because it is a natural accident of unpredictable occurrence but of recognizable recurrence; a disaster because it corresponds to the failure of the precipitation regime, causing the disruption of the water supply to natural and agriculture ecosystems and other human activities (Moreira et al., 2006; Sharma, 1997). The hazard and disaster nature of droughts makes it important to develop prediction tools, including statistical
W. Salee and V. Pongsapukdee
Silpakorn U Science & Tech J Vol.7(1), 2013
modeling using loglinear models and Markov chains models through the analysis of Standardized Precipitation Index (SPI) drought class transitions (Moreira et al., 2008; McKee, et al., 1993; Paulo, et al. 2005, 2006b; Sivakuma.and Wilhite, 2002; Nebraska-Lincoin University, 2002). The SPI was developed by McKee et al. (1993) and it is widely used as it allows a reliable comparison between different location and climate for the identification of drought events and to evaluate its severity. It is also often that drought events are becoming more frequent and/ or more severe due to the climate change. Loglinear models (Nelder, 1974; Agresti, 1990, 2002) are of use primary when at least two variables are response variables. The models specify how the expected count in contingency table depends on levels of the categorical variables for that cell as well as associations and interactions among those variables. For the basic theory or an associated sampling distribution, consider an (I × J) contingency table that cross-classifies a multinomial sample of n subjects on two categorical responses. The cell probabilities are {Pij} and the expected frequencies are {nPij}. In addition, loglinear models formulas generally apply with Poisson sampling for independent cell counts. Conditional on the sum n of cell counts, Poisson loglinear models for expected counts become multinomial models for cell probabilities. The main purposes of loglinear modeling are not only the analysis of association and interaction patterns but also its predictions. Loglinear models for expected counts in three-way tables such that for higher dimensions are tested and analyzed in this paper. They are considered to be a more adequate tool to analyze the drought class because they have shown to be adequate to
perform a monthly prediction of SPI drought class transitions and/ or contingency data (Agresti, 2002, Paulo, et al. 2005, and Fienberg, 2000). Therefore, the SPI data of drought events with the 12-month time scale are computed and analyzed through the adjusting loglinear models to the probabilities of transitions between the SPI drought classes in the form with 3-dimensional contingency table (Nelder, 1974; Paulo, et al., 2003; Pereira, et al., 2002). Then, odds prediction of drought category using loglinear models based on SPI for the whole northeast of Thailand and that for sub-area, Khonkaen province are performed and investigated to interpret through the odds ratios predictions. Material and Methods The general methodology of 3-dimensions loglinear models describes the association patterns among categorical variables which are performed for the cell counts in contingency tables (Agresti, 1990). A Poisson sampling model for counts is usually used for counts in contingency tables and assumes that they are independent Poisson random variable. Let the 3-dimension criterions, A, B and C with levels i, j, and k (i=1, 2,……, 4) , (j=1,…., 4) and (k=1,…, 4), respectively. The categorical variables A, B and C refer to drought classes at month’s t-2, t-1 and t, respectively. The levels 1, 2, 3, 4 are associated to the drought classes: 1 to the nondrought class, 2 to the near normal drought class, 3 to the moderate drought class, and 4 to the severe/ extreme drought class. The severity drought classes which are defined by McKee et al., 1993 (McKee, et al.) and modified by Moreira et al., 2006 are shown in Table 1.
33
Silpakorn U Science & Tech J Vol.7(1), 2013
Odds Prediction of Drought Category Using Loglinear
Table 1 Drought classes classified by Standardized Precipitation Index (SPI) values Code
Drought classes
SPI values
1
Non-drought class
SPI ≥ 0
2
Near normal drought class
-1 < SPI < 0
3
Moderate drought class
-1.5 < SPI ≤ -1
4
Severe/extreme drought class
SPI ≤ -1.5
o
months in drought class 1 (or non-drought class). Therefore, the loglinear modeling provides the expected frequencies of drought class transitions corresponding to a 2-month step transition from drought class i to drought class j (i.e., from t-2 to t-1) and from class j to class k (i.e., from t-1 to t). The contingency table for such data is given in Table 2.
The observations frequencies ( ijk ) are the response variable for the analysis using loglinear methods and it’s data refer to the observed number of transitions between the drought classes i at month’s t-2 (A), the drought classes j at month’s t-1 (B), and the drought classes k at month’s t (C). For example, the observation 111 is the number of times that a given site stays for three consecutive
o
Table 2 Three-dimensional contingency table (I × J × K) or (4 × 4 × 4) for two consecutive transitions between drought class i at month t-2 and class j at month t-1 and class k at month t corresponding to a 2month step transition from drought class i to class j (t-2 → t-1) and from class j to class k (t-1 → t) Drought
Drought class month t (k=1, 2, 3, 4)
class
1
2
3
4
month
Drought class month t-1
Drought class month t–1
Drought class month t–1
Drought class month t–1
t–2
(j=1, 2, 3, 4)
(j=1, 2, 3, 4)
(j=1, 2, 3, 4)
(j=1, 2, 3, 4)
1
2
3
4
1
2
3
4
1
2
3
4
1
2
3
4
1
O111
O121
O131
O141
O112
O122
O132
O142
O113
O123
O133
O143
O114
O124
O134
O144
2
O211
O221
O231
O241
O212
O222
O232
O242
O213
O223
O233
O243
O214
O224
O234
O244
3
O311
O321
O331
O341
O312
O322
O332
O342
O313
O323
O333
O343
O314
O324
O334
O344
4
O411
O421
O431
O441
O412
O422
O432
O442
O413
O423
O433
O443
O414
O424
O434
O444
(i=1,2,3,4)
1: Non-drought; 2: Near normal; 3: Moderate; 4: Severe/extreme
For selection of models’ goodness-of-fit to the three dimensional contingency table, we obtained that for overall area in northeastern of Thailand, the quasi-association model is the one that has adequately best fitted (P-value = 0.927) the
data of which the true model shown in (1). log (Eijk) = λ+λ Ai +λ Bj +λ Ck +β u i v j +α u i w k +η
v j w k + τu i v j w k +δ 1i I(i=j)+δ 2 i I(i=k)+δ 3 j I(j=k)+ δ 4 i I(i=j=k) 34
(1)
W. Salee and V. Pongsapukdee
Silpakorn U Science & Tech J Vol.7(1), 2013
where, λ is constant, λ Ai denotes the ith level for A, λ Bj denotes the jth level for B, and λ Ck denotes the kth level for C, and β, α, η, τ are model parameters, u i , v j , and w k are, respectively, the ith, jth, kth level scores for A, B, C with i, j, k ∈ {1, 2, 3, 4}, and δ 1i , δ 2 i , δ 4 i are parameters associated to the ith diagonal element of A, and δ 3 j the jth diagonal element of B. I is the indicator function defined as usual by I=0 if condition is true, and I=1 if condition is false. The expected counts are then obtained from the specified models. However, for each sub-area/province, we simplify (1) into other simpler good fit models; such as, in Khonkaen province, the true model is given by (2) (with goodness-of-fit P-value = 0.854).
Figure 1 Selected regions in 19 provinces (19 rainfall stations) in northeast of Thailand method using SAS. The residual deviances that have approximate chi-square distribution with degrees of freedom equal to the number cells in the contingency table minus the number of linearly independent estimated model parameters (Nelder, 1974, Agresti, 1990) were obtained for goodnessof-fit tests.
log (Eijk) = λ+λ Ai +λ Bj +λ Ck +β u i v j +η v j w k +δ 1i I(i=j)+δ 2 i I(i=k)
(2)
For other provinces in the northeast of Thailand we obtain similar models as given in (2) with only some different parameters (not shown here), and all of them are the sub-models of (1). All selected regions in 19 provinces (19 rainfall stations) are in northeast of Thailand, each of which the boundaries are given in Figure 1. All raw data used in this investigation were from the Institute for Water, Thailand. We collected the data and organized them using the SPI monthly values. Then data were computed in a 12 month time scale for the period from January 1962 to December 2009, all together, 576 months or 48 years in each 19 rainfall stations in the northeast of Thailand. All research works are processed and corporate using the program run with SAS® version 9.1. Several loglinear models for threedimensional contingency tables were fitted and tested. In loglinear models with Poisson sampling, the error is Poisson random variable. The models’ parameter estimation for loglinear models is estimated with the iterative maximum likelihood
Results The results for odds prediction of drought category using loglinear models based on SPI in the northeast of Thailand are processed for both the whole areas in northeast of Thailand and that for a sub-area such that Khonkaen province in the northeast of Thailand. The Results for Northeast of Thailand The null hypothesis tested is: the model fits data well. The null hypothesis is not rejected for those models having a residual deviance not exceeding the chi-square quantile for a probability 1 - a = 0.95 with the corresponding degrees of freedom. The backward elimination method (Agresti, 2002) was applied to each complete QA model adjusted to the northeast of Thailand data set to reduce the number of model parameters without significant loss of information
35
Silpakorn U Science & Tech J Vol.7(1), 2013
Odds Prediction of Drought Category Using Loglinear
to allow the selection of an alternative sub-model eliminating the less significant parameters of the QA model. The observed frequencies of drought class
transitions from month t-2 to month t-1 to month t: for the northeast of Thailand SPI 12 month’s time scale are presented in Table 3.
Table 3 Observed values of drought class transitions from month t-2 to month t-1 to month t: for the Northeast of Thailand SPI 12 month’s time scale Drought
Drought class month t
class
1
2
3
4
month
Drought class month
Drought class month
Drought class month
Drought class month
t–2
t–1
t–1
t–1
t-1
1
2
3
4
1
2
3
4
1
2
3
4
1
2
3
4
Observed values 1
216
3
0
0
21
23
0
0
0
1
0
0
0
3
3
3
2
19
21
1
1
6
127
11
0
0
18
8
0
0
1
3
5
3
1
1
0
0
0
16
6
0
0
2
9
1
0
0
0
26
4
1
0
0
0
0
3
3
5
0
2
1
3
0
4
6
34
Drought classes: 1: Non-drought; 2: Near normal; 3: Moderate; 4: Severe/extreme.
the corresponding confidence intervals, which are used to analyze and predict the drought classes’ transitions under the QA model, are investigated. The predicted values of odds ratios/estimated values of various drought class transitions in three consecutive months are summarized in Table 4.
It is found that the Quasi Association loglinear model (or QA) in (1) is selected from several loglinear models performed to be the most adequate for all the whole areas in northeast of Thailand of 19 provinces at the significance level of a = 0.05 (P-value = 0.927). The odds ratios prediction and
36
W. Salee and V. Pongsapukdee
Silpakorn U Science & Tech J Vol.7(1), 2013
Table 4 Estimates of the Odds ( Ω34|ij ) and correspondent confidence interval referring to estimates for month t when the drought class are known for months t-1 and t-2 for the Northeast of Thailand Drought class month t-1
Ω34|ij Drought class month t-2 1
2
3
4
1
2
3
4
24.561
0.403
0.007
0.003
(12.482 , 65.561)
(0.222 , 1.684)
(0.001 , 1.642)
(0.001 , 1.332)
0.608
11.043
2.475
0.038
(0.021 , 1.568)
(2.135 , 25.211)
(0.978 , 5.644)
(0.021 , 1.245)
0.190
70.333
30.094
0.060
(0.002 , 1.854)
(20.314 , 94.254)
(15.622 , 56.421)
(0.029 , 1.558)
0.188
0.507
0.190
0.103
(0.011 , 1.255)
(0.111 , 1.894)
(0.023 , 2.546)
(0.002 , 1.698)
The upper value in each cell denotes the odds estimate and the lower one refers to the oddsconfidence interval.
class and the estimate of odds ratio was Ω34|44 =
From Table 4, an estimated odds is a ratio
E
(( Ω kl | ij )) of expected frequencies ( ijk ) predicted from loglinear model, ranging from 0 to + ∞ , and represented the number of times that is more or less, or equally probable the occurrence of a certain event instead of another. Where, the odds ratios for three– dimensional loglinear models are given by
0.103 with the confidence interval (0.002, 1.698). Since the value 1 is included in that interval, then it is concluded that it was equally probable that by March this site would be in severe/extreme drought (k=4) instead of being in moderate drought (k=3) given that in January and February it was in severe/ extreme drought (i, j =4).
Ωkl | ij =
Eijk /Eijl , k ≠ l meaning that, 1 month from now
(t), it is Ω kl | ij times more, less, or equally probable that a specific site is in class k instead of class l, given that at present it is in class j, and 1 month before it was in class i, with i, j, k, and l ∈ {1, 2, 3, 4} and k ≠ l . For example, the results for the estimates
The Results for Khonkaen province in the northeast of Thailand For each sub-area in northeast of Thailand, we obtain for each province quite similar models as given in (2) and that all of them are sub-models of (1) with only some different parameters. Consequently, a selected province, Khonkaen, is demonstrated. The observed frequencies of drought class transitions from month t-2 to month t-1 to month t: for Khonkaen province’s SPI 12 month’s
E /E
of Ω34| ij = ij3 ij4 and respectively confidence intervals in Table 4. They mean that, taking the case for the drought class in January and February, which was 4 (i=j=4) or the severe/extreme drought
37
Silpakorn U Science & Tech J Vol.7(1), 2013
Odds Prediction of Drought Category Using Loglinear
the drought classes’ transitions under the QA loglinear model. The predicted values of odds ratios for various drought class transitions in three consecutive months are shown in Table 6.
time scale are presented in Table 5. The odds ratios prediction and the corresponding confidence intervals are performed and evaluated in order to investigate and predict
Table 5 Observed values of drought class transitions from month t-2 to month t-1 to month t: for the Khonkaen province in Thailand Drought
Drought class month t-1
class
1
2
3
4
month t-2
Drought class month
Drought class month
Drought class month
Drought class month
t–1
t–1
t–1
t-1
1
2
3
4
1
2
3
4
1
2
3
4
1
2
3
4
Observed values 1
239
13
0
0
27
19
2
0
2
2
1
0
1
0
0
1
2
27
20
2
0
6
88
4
0
1
7
3
1
0
3
2
2
3
2
1
1
0
1
11
4
0
0
1
11
1
0
0
4
10
4
0
0
0
0
0
1
3
2
0
1
6
10
0
0
3
27
Drought classes: 1: Non-drought; 2: Near normal; 3: Moderate; 4: Severe/extreme.
Table 6 Estimates of the odds ( Ω34|ij ) and correspondent confidence interval referring to estimates for month t when the drought class are known for months t-1 and t-2 for Khonkaen province of Thailand Drought class month t-1
Ω34|ij Drought class month t-2 1 2 3 4
1
2
3
4
1.591
127.121
102.231
0.008
(0.987 , 3.245)
(89.754 , 154.643)
(80.647 , 130.254)
(0.002 , 0.664)
40.667
2.545
1.212
0.478
(25.254 , 68.974)
(0.215 , 3.654)
(0.654 , 3.264)
(0.3254 , 1.264)
2.546
13.875
2.427
0.102
(0.326 , 5.647)
(5.647 , 25.699)
(0.984 , 3.654)
(0.025 , 1.365)
0.254
17.564
1.549
0.425
(0.125 , 1.658)
(8.947 , 25.644)
(0.687 , 3.2654)
(0.058 , 1.254)
The upper value in each cell denotes the odds estimate and the lower one refers to the odds confidence interval.
38
W. Salee and V. Pongsapukdee
Silpakorn U Science & Tech J Vol.7(1), 2013
In Table 6, for Khonkaen province, the model given by (2) is accepted adequately of fit with the deviance = 11.23, P-value = 0.854, α = .05. The predicted values of the odds ( Ω34| ij ) for Khonkaen province referring to estimates for month t when the drought class are known for months t-1 and t-2 are evaluated. The results for odds estimates Ω34| ij = ij3 ij4 and respectively confidence intervals for the Khonkaen province are also reported. Taking the case for the drought class in January and February, which was 3 (moderate drought) and 2 (near normal drought) (i = 3, j = 2), respectively and the estimate of odds Ω34| ij was 13.875, with the confidence intervals (5.647, 25.699). It indicates that the value 1 is not included in that interval which means that if it was given in January and February, which was 3 (medium) and 2 (near normal) it was 13.875 times more likely that by March this province would be in moderate drought (k=3) instead of being in severe/extreme drought (l=4). For more details and information are given in Table 6.
and that could be attributed to the climate change. Therefore, future monitoring of drought watch system is needed to be set the pace in these areas in order to keep up with them as well as more analyzing data from an extended time period is also needed to detect a possible long term climate change. Especially, an appropriate statistical model used to predict drought phenomenon in each area is necessary and is probably able to provide more effectiveness in the administrative management of hazard from drought. Furthermore, future research works concerning building and extending loglinear models and/or logit models (Nelder, 1974, Agresti, 1990, Pongsapukdee, 2012) including model selection for prediction of drought category are also strongly recommended.
Discussion and Conclusion This research shows that the application of loglinear modeling to the drought data allowed the comparison of the three periods in terms of odds ratios or that in terms of probabilities of transition between drought classes. We have investigated and built the loglinear models using the backward method in SAS to compare and select the adequacy goodness-of-fit of the fitted models. Results from the odds ratios with the confidence intervals are analyzed and used to predict the drought classes together with their corresponding estimated probability values of various drought class transitions in three consecutive months. Most of the results are consistent with the existence of a long-term normal natural periodicity. Even if, most of the results display for more normal drought classes than those of moderate or severity drought classes; however, there are still some areas of investigation that drought could be developed
References Agresti, A. (1990) Categorical Data Analysis. John Wiley & Sons, New York. Agresti, A. (2002) Categorical Data Analysis. 2nd ed., John Wiley & Sons, New York. Fienberg, S. E. (2000) Contingency Tables and Log-Linear Models: Basic Results and New Developments. Journal of the American Statistical Association 95: 643-647. Moreira Elsa E., Ana A. Paulo, Luis S. Pereira, and Joao T. Mexia. (2006) Analysis of SPI drought class transitions using loglinear models. Journal of the Science Direct 331: 349-359. Moreira Elsa E., Carlos A. Coelho, Ana A. Paulo, Luis S. Pereira, and Joao T. Mexia. (2008) SPI-based drought category prediction using loglinear models. Journal of the Science Direct 354: 116-130.
E /E
Acknowledgements All raw data used in this investigation are from the Institute for Water, Thailand. We would also like to thank the editors and the referees for their very helpful comments.
39
Silpakorn U Science & Tech J Vol.7(1), 2013
Odds Prediction of Drought Category Using Loglinear
Mckee, T. B., Doesken, N. J., and Kleist, J. (1993) The relationship of drought frequency and duration to time scales. In Proceedings of the Eighth Conference on Applied Climatolgy. American Meteorological Society. Boston: 233-236. Nelder, J. A. (1974) Loglinear models for contingency tables: a generalization of classical least squares. Applied Statistics. John Wiley & Sons, New York. Nebraska-Lincoin University. â&#x20AC;&#x153;SPI.â&#x20AC;? [Online URL: www.drought.unl.edu] accessed on July 10, 2009. Paulo, A. A., Ferreira, E., Coelho, C., and Pereira, L. S. (2005) Drought class transition analysis through Markov and Loglinear models, an approach to early warning. Agricultural Water Management 77:59-81. Paulo, A. A. and Pereira, L. S. (2006b) Prediction of SPI drought class transitions using Markov chains. Water International 31 (in press). Paulo, A. A., Pereira, L. S., and Matias, P. G. (2003) Analysis of local and regional droughts in
southern Portugal using the theory of runs and the Standardized. Water International 28: 65-68. Pereira, L. S., Cordery, I., and Iacovides, I. (2002) Coping with Water Scarcity. UNESCO IHP VI, Technical Documents in Hydrology No. 58, UNESCO, Paris, 267 pp. [Online URL: www.unescdoc.unesco.org/images/ 0012/001278/127846.pdf]. Pongsapukdee, V. (2012) Analysis of Categorical Data: Theories and Applications with GLIM, SPSS, SAS and MTB. 3rd ed., Silapakorn University Press, Nakhon-Pathom. Sharma, T. C. (1997) Estimation of drought severity on independent and dependent hydrologic series. Water Resources Management 11: 35-50. Sivakumar, M. V. K. and Wilhite, D. A. (2002) Drought preparedness and drought management. In: Drought Mitigation and Prevention of Land Desertion (Proceedings of the International Conference, Bled, Slovenia), UNESCO and Slov. Nat. Com. ICID, Ljubljana, CD-ROM Paper 2.
40
Research Article Principal Component Analysis Coupled with Artificial Neural Networks for Therapeutic Indication Prediction of Thai Herbal Formulae Lawan Sratthaphut1*, Samart Jamrus1, Suthikarn Woothianusorn1 and Onoomar Toyama2 Department of Health-Related Informatics, 2Department of Pharmaceutical Chemistry, Faculty of Pharmacy, Silpakorn University, Nakhon Pathom, Thailand * Corresponding author E-mail address: lawan.s@su.ac.th
1
Received October 8, 2012; Accepted November 26, 2012 Abstract This study illustrated the principal component analysis coupled with artificial neural networks (PCANN) as a useful tool in therapeutic indication prediction of Thai herbal formulae official in the National List of Essential Medicine 2011 and the National Traditional Household Remedies. A set of 71 herbal formulae from the National List of Essential Medicine 2011 and the National Traditional Household Remedies and 19 formulae without therapeutic indication was used as a training set, a monitoring set and a validation set. The performance of the model was measured by the percentage of â&#x20AC;&#x153;correctly classifiedâ&#x20AC;?, True Positive rate and False Positive rate of the PC-ANN model. The results suggested that principal component analysis technique could condense all of the variables in which there were interrelated, into a few principal components, while retaining as much variation presented in the data set as possible. The use of a PC-ANN technique provided a good prediction of therapeutic indication of these herbal formulae as well as distinguishing these formulae from the one without therapeutic indication. Key Words: Artificial neural network; Principal component analysis; Thai herbal formula Introduction Natural resources are plentiful in Thailand, and the country has its own traditional herbal medicine. Historically, the use of traditional herbal medicine has been a part of life since Sukhothai Period (1238 - 1377) (Subcharoen and Chuthaputti, 2006). With the entrance of western medicine, the role of traditional herbal medicine was declined due to lacking of scientific basis. However, high cost of new drugs, increased side effects and microbial resistance are some reasons for renewed interest in traditional herbal medicine. The development of traditional and herbal remedies supported by Thai government policy was launched through the
Silpakorn U Science & Tech J 7 (1) : 41-48, 2013
implementation of the Fourth National Economic and Social Development Plan (1977-1981). Some items of herbal preparations, which have been used traditionally and widely by the people from time immemorial, were placed on the National List of Essential Medicines (Herbal Medicine List) and the National Traditional Household Remedies as part of an effort to promote the use of herbal preparations and provide diversity of alternatives for health care since 1999 (Chokevivat et al., 2005). Hence, the public interest in therapy based on traditional herbal medicine has been growing and the increasing effort has been directed towards scientific proof, clinical evaluation, and recipe analysis. Nowadays,
Silpakorn U Science & Tech J Vol.7(1), 2013
Principal Component Analysis Coupled with Artificial Neural Networks
artificial intelligence (AI) has become an important field of study with a wide spread of applications in several fields. It also has been successfully applied to traditional medicine research area (Chen et al., 2006; Ung et al., 2007; Cao et al., 2009). In AI field, one of the most used branches is artificial neural networks (ANN). ANN is a computational model or mathematical model inspired in the natural biological neural networks (Zupan and Gasteiger, 1999). ANN method can simulate learning and generalize behavior of the human brain through data modeling and pattern recognition. The difference between an ANN model and a statistical model is that the ANN allows one to estimate relationships between one or several input variables called independent variables and one or several output variables called dependent variables without a specific mathematical function. Hence, an ANN works well for solving complicated non-linear problems of multivariate systems. In this work, we used ANN to determine the appropriate model for the prediction of therapeutic indication of herbal formulae. Since the ratio between samples and variables in the ANN should be kept as high as possible, the principal component analysis (PCA) was used to trim down the number of variables in a sample data matrix. The combined use of PCA and ANN usually also improves the training speed, enhances the robustness of the model and reduces model errors.
Data Sources and Datasets A set of 71 (47+24) herbal formulae official in the National List of Essential Medicine (NLEM) 2011 (the current version) and the National Traditional Household Remedies (NTHR) and 19 formulae without therapeutic indication (nontherapeutic formulae, NF) was used as a data set. Formulae without therapeutic indication were constructed by random combination of herbal items and by modification of existing herbal formulae. The 71 herbal formulae were classified into 7 categories based on therapeutic indication, e.g. cardiovascular system (CS, 10 recipes), gastrointestinal tract (GI, 23 recipes), obstetrics and gynecology (OG, 8 recipes), antipyretic (AP, 11 recipes), respiratory system (RS, 9 recipes), bone and muscle (BM, 8 recipes), and tonic (TN, 2 recipes). The list of 71 herbal formulae in NLEM and NTHR is shown in Table 1. The total number of Thai herbal items is 199. The data set of 90 herbal formulae was randomly divided into a training set (57 recipes), a monitoring set (10 recipes) and a validation set (23 recipes). A monitoring set was used to optimize the model parameters and a validation set was employed for testing the accuracy and precision of the models. To make a standard for the amounts of herbs in the preparations, the amount of each herb in all preparations was normalized as the percentage of each herb to all herbs in the same preparations.
Materials and Methods Apparatus and Software All input datasets were formatted as CSV files and stored for analysis by Microsoft Excel 2007. These data were processed by Intel速 CoreTM i3 computer having 2GB for RAM (Windows 7 operating system). The principal component analysis was performed by Scilab version 5.3.1 and artificial neural networks were implemented in WEKA version 3.7.5 using multilayer perceptron algorithm. Both programs are open source software.
Results and Discussion Muti-herb Formulae The average number of herbs in formula for CS, GI, OG, AP, RS, BM and TN groups were 38.5, 16.5, 11.8, 12.8, 8.6, 10.1 and 6.0, respectively. The distribution of constituent herbs, in 71 herbal formulae, used in data set is illustrated in Figure 1. Most of these herbal formulae contain four or twelve herbal items.
42
L.Sratthaphut et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
Table 1 List of 71 herbal formulae selected from NLEM and NTHR Thai herbal formulae
Therapeutic Data indication
Thai herbal formulae
sources
Therapeutic Data indication
categories
sources
categories
Ya-hom Tip-osod 1
CS
NLEM
Ya-Prasaplai 1
OG
NLEM
Ya-hom Tip-osod 2
CS
NTHR
Ya-Prasaplai 2
OG
NTHR
Ya-hom Tapepajit 1
CS
NLEM
Ya-Faipralaigun 1
OG
NLEM
Ya-hom Tapepajit 2
CS
NTHR
Ya-Faipralaigun 2
OG
NTHR
Ya-hom Navagote 1
CS
NLEM
Ya-Faihaagong 1
OG
NLEM
Ya-hom Navagote 2
CS
NTHR
Ya-Faihaagong 2
OG
NTHR
Ya-hom Intachak 1
CS
NLEM
Ya-Luadngam
OG
NLEM
Ya-hom Intachak 2
CS
NTHR
Ya-Satreelungklod
OG
NLEM
Ya-hom Kaelomwingwian
CS
NLEM
Ya-Kiewhom 1
AP
NLEM
Ya-Bumrung-lohit
CS
NTHR
Ya-Kiewhom 2
AP
NTHR
Ya-thad Bunjob 1
GI
NLEM
Ya-Chantaleela 1
AP
NLEM
Ya-thad Bunjob 2
GI
NTHR
Ya-Chantaleela 2
AP
NTHR
Ya-thad Obchoey
GI
NLEM
Ya-Prasachandaeng 1
AP
NLEM
Ya-Prasakaprao 1
GI
NLEM
Ya-Prasachandaeng 2
AP
NTHR
Ya-Prasakaprao 2
GI
NTHR
Ya-Prasaproayai 1
AP
NLEM
Ya-Prasakanplu 1
GI
NLEM
Ya-Prasaproayai 2
AP
NTHR
Ya-Prasakanplu 2
GI
NTHR
Ya-Mahanintangtong 1
AP
NLEM
Ya-Prasajetpangkee 1
GI
NLEM
Ya-Mahanintangtong 2
AP
NTHR
Ya-Prasajetpangkee 2
GI
NTHR
Ya-Haaraag
AP
NLEM
Ya-Muntathad 1
GI
NLEM
Ya-Ammareukwatee 1
RS
NLEM
Ya-Muntathad 2
GI
NTHR
Ya-Ammareukwatee 2
RS
NTHR
Ya-Wisaampayayai 1
GI
NLEM
Ya-Prasamawaeng 1
RS
NLEM
Ya-Wisaampayayai 2
GI
NTHR
Ya-Prasamawaeng 2
RS
NTHR
Ya-Mahachakyai 1
GI
NLEM
Ya-kae-ai-Pasomkanplu
RS
NLEM
Ya-Mahachakyai 2
GI
NTHR
Ya-kae-ai-Pasommanowdong
RS
NLEM
Ya-Luangpidsamut 1
GI
NLEM
Ya-kae-ai-Peunbaan-E-san
RS
NLEM
Ya-Luangpidsamut 2
GI
NTHR
Ya-Treepala
RS
NLEM
Ya-Apaisalee
GI
NLEM
Ya-Prabchompootaweep
RS
NLEM
Yatye-deekluafarang
GI
NLEM
Ya-Kasaisaen
BM
NLEM
Ya-Pasompetsangkart 1
GI
NLEM
Ya-kaelom-Ammappruek
BM
NLEM
Ya-Pasompetsangkart 2
GI
NLEM
Ya-Pasomkoklan 1
BM
NLEM
Ya-Ridseeduangmahakan
GI
NLEM
Ya-Pasomkoklan 2
BM
NLEM
Ya-tye
GI
NTHR
Ya-Pasomkoklan 3
BM
NLEM
Ya-Treegaysornmas
TN
NLEM
Ya-pasom-Taowanpriang 1
BM
NLEM
Ya-Treepigut
TN
NLEM
Ya-pasom-Taowanpriang 2
BM
NLEM
Ya-Sahadtara
BM
NLEM
43
Silpakorn U Science & Tech J Vol.7(1), 2013
Principal Component Analysis Coupled with Artificial Neural Networks
Figure 1 Distribution of herbal formulae with respect to the number of constituent herbs
Dimension Reduction with PCA The performance of an ANN usually depends on data representation and one important characteristic of data representation is not interrelated. This is because correlated data reduce the distinctiveness of data representation and introduce confusion to the ANN model during the learning process (Mohamad-Saleh and Hoyle, 2008). The eliminating correlation in the sample data before they are submitted into an ANN is necessary. In this paper, an input data set consists of 90x199 (formulae x herb items). PCA was done onto this input data set prior to the ANN training process. PCA technique is one of the multivariate data analysis approach based on the latent variable decomposition and is widely used for dimension reduction. PCA was first introduced by Pearson
(1901), and independently developed by Hotelling (1933) (Jolliffe, 2002). This technique can compress all of the variables in which they are correlated, into a few principal components (PC), which are ordered by decreasing variability. The last of these components can be removed with minimum loss of real data. Thus, dimension of a sample data set can be reduced. The first PC defines the combination of variables that explains the greatest amount of variation and the second PC (independent to the first PC) indicates the next largest amount of variation and so on. The new variables, which are uncorrelated and called the principal components, are formed by taking linear combinations of the original variables. The principal component can be written as
zij = ai1 x1 j + ai 2 x2 j + ... + aim xmj
44
(1)
L.Sratthaphut et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
where z is the component score, a is the component loading, x is the measured value of variable, i is the principal component number, j is the sample number and m is the total number of variables (Cornish, 2007). In applying PCA, the input data matrix was reduced as loading and score matrices were gathered. The first 49 components were observed to represent 98.23% of total variance. Once explored the data by PCA, a classification model was performed by the ANN into the next step.
f ( x) =
o j = f [∑ ( si wij )]
(3)
The log-sigmoid hidden layer is critical as it allows the network to learn non-linear relationships between inputs and outputs. The learning process was carried out through the back-propagation algorithm. The back-propagation network learns by calculating an error between desired and actual output and propagating this error information back to each node in the network. This back-propagation error is used to drive the learning at each node. The process of changing the weight of the connections to achieve some desired result is called learning or adaptation (Daniel et al., 1997). To optimize ANN parameters (the number of neurons in the hidden layer, momentum and learning rate), the 10 recipes were constructed and used as monitoring set and all parameters were optimized by the Change One Separate factor at a Time (COST) technique. At this point, the mean square error (MSE) was calculated. Each time a new node was added to the hidden layer at arbitrary learning rate, momentum and the number of iterations. The number of neurons at the hidden layer, which had the minimum MSE value, was selected as the optimum number. After this step, the learning rate was varied from 0.1 to 0.9 at the optimum number of neurons at the hidden layer, arbitrary momentum and the number of iterations. The learning rate, which had the minimum MSE value, was selected as the optimum number. In the same way, the optimum number of momentum was defined. In order to avoid the overtraining and choosing the suitable number of epochs, the network was terminated before it learned idiosyncrasies present in the training data by searching the minimum MSE for the monitoring set. Finally, the number of the neurons at the hidden layer with the use of optimized momentum and learning rate was determined. Figure 2 illustrates the MSE value of the network with different learning rate and momentum and Figure 3 shows the variation
ANN Modeling In this application, the set of 57 herbal formulae (with 49 components) was employed as a training set and the ANN model was established by WEKA program with “multilayerPerceptron” algorithm in “classify” tab. This ANN model consisted of three layers of neurons, which were the basic computing units: the input layer with a number of active neurons corresponding to PC, one hidden layer with a number of active neurons, and the output layer with eight active neurons corresponding to the categories of therapeutic indication. The neurons were fully connected in a hierarchical manner. i.e. the outputs of one layer of nodes were used as inputs for the next layer and so on. The nodes in the input layer transfer the input data to all nodes in hidden layer. These nodes calculate a weighed sum of the inputs that is subsequently subjected to a non-linear transformation: I
1 1 + e− x
(2)
i =1
where si is the input to node i in the input layer, I is the number of nodes in the input layer, wij (weights) are the connections between each node i in the input layer and each node j in hidden layer, and oj is the output of node j in hidden layer, and f is a non-linear function called log-sigmoid function (Eq. (3)).
45
Silpakorn U Science & Tech J Vol.7(1), 2013
Principal Component Analysis Coupled with Artificial Neural Networks
Figure 2 The relationship between learning rate (
) and momentum (.....) versus MSE
Figure 3 The relationship between number of neurons in hidden layer versus MSE
of MSE values of the network when the numbers of neurons in hidden layer was changed. The summary specifications for the network created for ANN models were listed in Table 2. To study the ability of established ANN in the prediction of therapeutic indication for herbal
formulae, 23 test herbal formulae were analyzed using the proposed method. The predicted class in each preparation was then compared with the known class in the respective preparation. The predicted results of ANN can be visualized by a confusion matrix (Table 3) of the desired and predicted
46
L.Sratthaphut et al.
Silpakorn U Science & Tech J Vol.7(1), 2013
network output and by the performance indexes of the network. A confusion matrix is a table that displays information about actual classification (target output) on the rows and predicted classification (network output) on the columns. The ideal classification result is to have large numbers down the main diagonal and small, ideally zero, off-diagonal elements of the matrix (Witten et al., 2011). The confusion matrix in Table 3 shows perfect classification for ANN model which the percentage of â&#x20AC;&#x153;correctly
classifiedâ&#x20AC;? of models is 100.00%. Furthermore, the performance indexes of such model was evaluated using data presented in the matrix and defined in terms of True Positive (TP) rate, False Positive (FP) rate and the percentage of accuracy. TP rate is the proportion of positive cases that are correctly classified, FP rate is the proportion of negative cases that are incorrectly classified, and the percentage of accuracy is the percentage of total number of predictions that are correct. The performance is confirmed by TP of 1, FP of 0 and the percentage of accuracy of 100.00%. This study indicates that combination of herbs and herbal proportions in Thai herbal formulae were rationally formulated and structurally defined. However, this ANN model has some limitation in that it is used to predict class of therapeutic indications of herbal formulae with herbal constituents modified around those from 71 formulae in NLEM or NTHR.
Table 2 Artificial neural network specifications and parameters Parameter
ANN
Input nodes
49
Hidden nodes
16
Output nodes
8
Learning rate
0.2
Momentum
0.7
Hidden layer transfer function
log-sigmoid
Optimum number of iterations
4000
Conclusion The capability of a hybrid methodology that merges PCA techniques and an ANN modeling technique is presented and evaluated in this paper. In accordance with the achieved results, the suggested model that used PC as input variables could predict the therapeutic indication group. The performance
Table 3 A confusion matrix for 23 herbal formulae when applying ANN model to the validation set
47
Silpakorn U Science & Tech J Vol.7(1), 2013
Principal Component Analysis Coupled with Artificial Neural Networks
of this model was improved with extraction of the factors that poorly contributed in the therapeutic indications and simplified the training procedure of the ANN by the inclusion of only the significant PC in the model. Furthermore, the PC-ANN model is useful for not only predicting class of therapeutic indications but also validating herbal formulae, distinguishing the formulae from the formulae without therapeutic indication.
pdf] accessed on September 24, 2012. Daniel, S., Kvasnicka, V., and Pospichal, J. (1997) Introduction to multi-layer feed-forward neural networks. Chemometrics and Intelligent Laboratory Systems 30: 43-62. Jolliffe, I. T. (2002) Pricipal Component Analysis 2nd ed., Springer-Verlag, USA. Mohamad-Saleh, J. and Hoyle, B. S. (2008) Improved Neural Network Performance Using Principal Component Analysis on Matlab. International Journal of The Computer, the Internet and Management, 16(2): 1-8. Subcharoen, P. and Chuthaputti, A. (2006) Thai Traditional Medicine Kingdom of Thailand. In WHO Global atlas of traditional, complementary and alternative medicine, pp.103-106. The WHO Centre for Health Development, Kobe. Ung, C. Y., et al. (2007) Are herb-pairs of traditional Chinese medicine distinguishable from others? Pattern analysis and artificial intelligence classification study of traditionally defined herbal properties. Journal of Ethopharmacology 111: 371-377. Witten, I. H., Frank, E., and Hall, M. A. (2011) Data Mining Practical Machine Learning Tools and Techniques 3rd ed., pp.164. Morgan Kaufmann Publishers, Elsevier, USA. Zupan, J. and Gasteiger, J. (1999) Neural Networks in Chemistry and Drug Design, 2nd ed., Wiley-VCH Publishing Company, Weinheim.
References Cao, T., Kamei, K., and Dang T. L. (2009) Visualization System of Herbal Prescription Effects in Oriental Medicine by SelfOrganizing Map. Biomedical Soft Computing and Human Sciences 14(1): 101-108. Chen, X., et al. (2006) Database of traditional Chinese medicine and its application to studies of mechanism and to prescription validation. British Journal of Pharmacology 149: 1092-1103. Chokevivat, V., Chuthaputti, A. and Khumtrakul, P. (2005) The Use of Traditional Medicine in the Thai Health Care System. In Regional Consultation on Development of Traditional Medicine in the South East Asia Region (World Health Organization Regional Office for South East Asia), Document no.9. Pyongyang, DPR Korea. Cornish, R. (2007) Principal Component Analysis. In Statistics, [Online URL: www.mlsc.lboro. ac.uk/resources/statistics/3.2PrincipleCom.
48
Short Communications Usage of Chitosan in Thai Pharmaceutical and Cosmetic Industries Rapeepun Chalongsuk* and Namfon Sribundit Department of Community Pharmacy, Faculty of Pharmacy, Silpakorn University, Nakhon Pathom, Thailand * Corresponding author. E-mail address: rapee@su.ac.th Received October 4, 2012; Accepted December 12, 2012 Abstract Chitosan has varieties of applications in both pharmaceutical and cosmetic products. To explore the situation of chitosan awareness and usage in Thailand is an essential information for product development from chitosan. Entrepreneurs or their representatives from GMP certified pharmaceutical and cosmetic manufacturers were interviewed during June 2007 to February 2008. One hundred and five from 225 factories (47%) participated in this study. Product awareness was very high (93.3%). For pharmaceutical industry, the most familiar applications of chitosan were anti-obesity and cholesterol lowering agent. In cosmetic industry, value of chitosan was perceived as moisturising agent, emollient, and film former. Current usage of chitosan was found in 41.4% of cosmetic industry, mainly for skin care products and 8% of pharmaceutical industry used chitosan mainly for anti-obesity food supplement. Although chitosan use was highly perceived, actual usage was moderate and low in cosmetic industry and pharmaceutical industry, respectively. Key Words: Chitosan; Cosmetic manufacturer; Pharmaceutical manufacturer Introduction Pharmaceutical and cosmetic industries in Thailand are downstream businesses which are dependent on imported raw materials, both active ingredients and excipients. To develop sustainability of these industries, researches and developments for local production of raw material are essential (Kuanpoth, 2007; Umprayn, 2000; Chantarasakul et al., 1999). Chitosan is a linear copolymer of -(1â&#x20AC;&#x201C;4) linked 2-acetamido-2-deoxy-d-glucopyranose and 2-amino-2-deoxy-d-glycopyranose. It is obtained by deacetylation of its parent polymer chitin, a polysaccharide widely distributed in nature (e.g.
Silpakorn U Science & Tech J 7 (1) : 49-53, 2013
crustaceans, insects and certain fungi) (Muzzarelli; 1977). Chitosan is one of the interesting substances because of its extensive applications. Various uses of chitosan in pharmaceutical industries were investigated. It was used for a process of direct compression, controlled drug release and drug delivery (Dodane and Vilivalam, 1998; Ilium, 1998). For cosmetic production, emollient and film former were the major roles of chitosan (Lang and Clausen, eds., 1989). Chitosan can be extracted from crustacean shells e.g. shrimp, crab and squid (Muzzarelli; 1977). Fortunately, there is an abundance of natural sources of chitosan from fishery industry waste in Thailand (FAO, 2006). Many Thai researchers
Silpakorn U Science & Tech J Vol.7(1), 2013
Usage of Chitosan in Thai Pharmaceutical and Cosmetic Industries
have focused on the development of chitosan production and its applications (Petchsangsaia M., 2011; Khunawattanakula W., 2011; Nunthanid J., 2008; Phaechamud T.,2008). The exploration of the situation of chitosan awareness and chitosan usage in the pharmaceutical and cosmetic industries in Thailand will provide crucial information for product development from chitosan.
rates of pharmaceutical and cosmetic industries were 46% and 48.3%, respectively. Thirty one point four percents of respondents were R&D managers. One-fourth of all were production managers. The rests were quality control (QC) managers or pharmacists who worked for production or research and development department. Mean of respondents’ experiences in the industries were 8.67± 8.43 years. Chitosan Awareness and Perceived Value All respondents were asked whether they knew “chitosan” to check for the awareness of the substance. It was found that 93.3% of them knew chitosan. Perceived value of chitosan or basic knowledge about its application of the respondents are reported in Table 1. Anti-obesity agent was the most perceived value of chitosan (50%). The second perceived value was cholesterol lowering agent (25%). Drug delivery application ranked third (20%).
Materials and Methods Population and Sample Population in this study was 225 GMP certified manufacturers in Thailand. Sample size was calculated by using proportion of chitosan usage 31% (Center of Biotech chitin-chitosan, 2003) ±10% at 0.05 significant level. Total sample size needed for this study was 82 manufacturers. Data Collection Constructed interviews were conducted by five trained persons during June 2007 – February 2008. Production managers or research and development (R&D) managers or their representatives were interviewed by phone on general information of factory, main products, respondents’ chitosan awareness and perceive value. Chitosan usage of each manufacturer was also in the surveyed topics. Data Analysis Obtained data were analyzed employing descriptive statistics; frequency, percentage. Factors of characteristics of manufacturers in both industries which have influence on chitosan perceive value and usage were explored with chi-square statistics.
Table 1 Perceived value of chitosan applications (first 10 ranking) Chitosan applications
% perceived value ( n=105)
Results and Discussions Response Rate and Respondent’s Characteristics Of 255 manufacturers, 105 (46.6%) participated in this study, 76 from pharmaceutical industry and 29 from cosmetic industry. Response
50
Anti-obesity agent
49.5
Cholesterol lowering agent
24.8
Drug delivery agent
20.0
Agricultural applications
17.1
Diluent
10.5
Moisturising agent
10.5
Sustain release agent
9.5
Film former
8.6
Coating agent
8.6
Binder
5.7
Wound healing agent
5.7
R. Chalongsuk and N. Sribundit
Silpakorn U Science & Tech J Vol.7(1), 2013
Chitos an values perceived f r o m pharmaceutical respondents and cosmetic respondents were significantly different, especially the applications as anti-obesity agent, moisturising agent, emollient and film former. Perceived value of chitosan as moisturising agent, emollient and film former from respondents in cosmetic industry is more than those from pharmaceutical area (p<.0001, =0.029 and <0.0001, respectively). On the other hand, application as an anti-obesity agent was more perceived by respondents from pharmaceutical industry (p = 0.034).
For those who have never used chitosan for their productions, the major reasons were lacking of scientific supporting data (21%) and their products were simple dosage forms, not advance dosage forms (22.2%). The top 5 reasons were displayed in Table 3. Table 3 Top 5 reasons of chitosan never users Reason of chitosan never users
Produce only simple dosage form Lack of scientific supporting data Lack of knowledge about its applications Strictly follow OEM* contract
Chitosan Usage Current usage of chitosan was investigated in 105 manufacturers. Of 105, 18 (17.1%) were chitosan current users. Current usage of chitosan was found in 41.4% and 8% of cosmetic industry and of pharmaceutical industry, respectively.
Lack of information regarding price and application *
Table 2 Chitosan usage (classified by industry) Chitosan current usage
Pharmaceutical Industry
YES
Both industries
6 (7.9%)
12 (41.4%)
18 (17.1%)
NO
70 (92.1%)
17 (58.6%)
87 (82.9%)
Total
76 (100%)
29 (100%)
105 (100%)
18 (22.2) 17 (21.0) 8 (9.9) 6 (7.4) 6 (7.4)
OEM = Original Equipment Manufacturer
Besides these reasons, a few manufacturers experienced problems from chitosan. The problems were reported as follows: - Unstable viscosity of cosmetic product when using chitosan as a thickening agent - Bad smell of glacial acetic acid (from chitosan preparation) in final product. - More time consuming for gel swelling when compared with other celluloses
Number of manufacturers (%) Cosmetic Industry
Frequency (n=81) (% )
Most of chitosan users in pharmaceutical industries applied the value of anti-obesity to produce food supplement for weight reduction. The remained pharmaceutical products that used chitosan were blood static pad, disinfectant soap and artificial tear. For cosmetic manufacturers, chitosan was mainly used for skin care cosmetic production, particularly moisturising cream or lotion. Food supplement, hair care and firming gel productions used chitosan as active ingredient or excipients as well.
Discussions Although this survey revealed high awareness of chitosan and its values were high perceived, actual usage was moderate and low in cosmetic industry and pharmaceutical industry, respectively. To promote usage of local chitosan production, more information of its applications should be widely provided. Particularly for Thai pharmaceutical industry, value of chitosan was limited as an active ingredient in anti-obesity food supplement
51
Silpakorn U Science & Tech J Vol.7(1), 2013
Usage of Chitosan in Thai Pharmaceutical and Cosmetic Industries
Acknowledgement The authors would like to thank the funding support provided through the Silpakorn University Research and Development Institute (SURDI).
or cholesterol reducing product. The application in drug delivery or sustain release product might be far from the manufacturersâ&#x20AC;&#x2122; interest because approximately 75% of the pharmaceutical factories did not develop sustain release drug or controlled release drug. Therefore, more information regarding excipient applications e.g. binder, disintegrant, filler, and diluent should be presented to potential users. (Kato Y, 2003) Due to continuing high rising price of lactose, chitosan might has opportunity in diluent market.
References Center of Biotech chitin-chitosan. (2003) C. U. A Study of Chitin-Chitosan Production and Chitin-Chitosan Market. [Online URL: www.mtec.or.th/th/images/pdf/chitinchitosan/chapter6.pdf.] accessed on Aug 7, 2006 Chantarasakul J., Pongchareonsuk P., and Sakulbumrungsil R. (1999) A Study of Pharmaceutical Raw Materials with Potentials for Local Production. Office of the National Research Counsil of Thailand: Bangkok. Dodane, V. and Vilivalam V. D. (1998) Pharmaceutical applications of chitosan. Pharmaceutical Science & Technology Today, 1(6): 246-253. FAO. (2006) Fisheries global information system, F. FIGIS time series query on Aquaculture: THAILAND 1994-2004. [Online URL: www.fao.org/figis/servlet/SQSerlet?file=/ usr/local/tomcat/FI/5.5.9/fi5/webapps/figis/ temp/hqp_42599.xml&outtype=html.] accessed on May 23, 2006 Ilium, L. (1998) Chitosan and Its Use as a Pharmaceutical Excipient. Pharmaceutical Research, 15: 1326-1331. Kato Y., Onishi H., and Machida Y. (2003) Application of chitin and chitosan derivatives in the pharmaceutical field. Curr Pharm Biotechnol. 4(5):303-309. Khunawattanakula W., Puttipipatkhachornb S., Radesc T., and Pongjanyakula T. (2011)
Conclusion The present survey explored chitosan awareness and actual usage from 105 manufacturers. Product awareness was very high (93.3%). For pharmaceutical industry, the most familiar applications of chitosan were anti-obesity, cholesterol lowering agent and drug delivery agent. Chitosan value perceived from pharmaceutical respondents and cosmetic respondents were significantly different. Respondents from cosmetic industry perceived value of chitosan as moisturiser, emollient and film former more than those from pharmaceutical area. On the other hand, anti-obesity effect was more perceived by respondents from pharmaceutical industry. Current usage of chitosan was found in 17.1% of all manufacturers. Cosmetic industry were the major chitosan current users (41.4%) while only 8 % of pharmaceutical industry used chitosan. The main reasons of never users were lacking of scientific supporting data regarding to its applications and their products were simple dosage forms. To promote more usage of chitosan, more information of its applications should be widely available, and attractive price also encourages new chitosan users.
52
R. Chalongsuk and N. Sribundit
Silpakorn U Science & Tech J Vol.7(1), 2013
drug delivery using spray-dried chitosan acetate and hydroxypropyl methylcellulose. European Journal of Pharmaceutics and Biopharmaceutics 68: 253-259. Petchsangsaia M., Sajomsanga W., Gonila P., Nuchuchuaa O., Sutapunb B., Puttipipatkhachornc S., and Ruktanonchaia U. R. (2011) A water-soluble methylated N-(4-N,N-dimethylaminocinnamyl) chitosan chloride as novel mucoadhesive polymeric nanocomplex platform for sustained-release drug delivery. Carbohydrate Polymers. 83: 1263-1273. Phaechamud T. (2008) Effect of Particle Size of Chitosan on Drug Release from Layered
Novel chitosanâ&#x2C6;&#x2019;magnesium aluminum silicate nanocomposite film coatings for modified-release tablets. International Journal of Pharmaceutics 407: 132â&#x20AC;&#x201C;141. Kuanpoth J. (2007) Intellectual Property Rights and Pharmaceuticals:A Thai Perspective on Prices and Technological Capability. Thammasat Economic Journal. 25(4): 1-46. Lang G. and Clausen T., eds. (1989) The Use of Chitosan in Cosmetics Chitin and Chitosan: Sources, Chemistry, Biochemistry, Physical Properties and Applications (Skjak-Braek G., Anthonsen T., and S. P., eds.), pp.139147. Elsevier Sciences Publishers, London, New York. Muzzarelli RAA.(ed. (1977) Industrial Production and Application Chitin, pp.207-265. Pergamon Press, New York. Nunthanid J., Huanbutta K., Luangtana-anan M., Sriamornsak P., Limmatvapirat S., and Puttipipatkhachorn S. (2008) Development of time-, pH-, and enzyme-controlled colonic
Matrix System Comprising Chitosan and Xanthan Gum. Thai Pharm Health Sci J. 3(1):1-11. Umprayn K. (2000) An Analysis of Thai pharmaceutical structure and potentials of pharmaceutical export in ASEAN countries Chulalongkorn Review, 13(49): 60-76.
53