Empirical evidence for Thailand surface air temperature change :
Possible causal attributions and impacts
Environmental Research and Training Center
Department of Environmental Quality Promotion
August 2004
1

CONTENTS
Abstract
บทสรุปส ำหรับภำษำไทย
1. Introduction
2. Analytical methods and data sources
2.1. Basic concepts of EOF analysis
2.2. Data sources i
1
5
10
13 2.3. EOF computation using the scatter matrix method
3. Results
3.1. Physical interpretation of EOF analysis
3.2. Temporal variability of EOF1 coefficient series and its relation
to ENSO signature
17
32
3.3. Linear trends in surface air temperature in Thailand
4. Discussion
51
4.1. Analytical methodology 54
4.2. Possible causal attribution of interannual and long-term changes 55
in surface air temperatures in Thailand
4.3. A reduction of diurnal temperature ranges
4.4. Changes in temperature extreme events
4.5. Possible biophysical and socio-economic impacts
5. Implications for future research
6. Acknowledgement
References
58
58
59
62
63
64
2

Abstract
This study attempts to investigate the dominant spatio-temporal structure of mean, maximum and minimum surface air temperatures ( T
B mean
B
, T
B max
B
, T
B min
B
), dewpoint temperature ( T
B dew
B
) and computed mean, maximum and minimum apparent temperatures
( T
B amean
B
, T
B amax
B
,
B B
T
B amin
B
) in Thailand. The atmospheric data used in this study are based on monthly data collected from 33 stations for period of 1951-2003. Empirical Orthogonal
Functions (EOF) analysis and other multivariate statistical techniques were used to reveal the dominant modes and temporal patterns.
An analysis indicates that the EOF1 mode of all temperature variables accounts for substantial amount of the total variance ranging from 61.2% to 71.3%. The EOF1 mode of all temperature variables is characterized by a monopole of spatial patterns, which correlations coefficients are positive and relatively high and about the same magnitude at all stations. Such a unique pattern implies a high intercorrelation and a relatively uniform variance distribution of surface air temperatures at all stations. Hence, the EOF1 mode is a robust representative of the dominant spatio-temporal structure of surface air temperatures in Thailand that probably share a common influence from the same origins.
On the basis of the EOF results, the time variability of the EOF1 mode of all temperature variables in Thailand has oscillated at three dominant timescales over the last 53 years: interannual/decadal timescales and long-term trends. The El Niño-Southern
Oscillation (ENSO) cycles are the most prominent timescale of interannual variability in surface air temperatures in Thailand. There is a significant indication that all temperature variables tend to be higher (lower) than normal during the El Niño (La Niña) years. The possible linking pathway between ENSO event and interannual changes in surface air temperature in Thailand may be through the “atmospheric teleconnections”, establishing by the shifts in the location of the organized rainfall in the tropics and the associated latent heat release.
The EOF1 coefficient series of T
B max
B
, T
B amax
B
, T
B min
B
and T
B amin
B
also exhibit salient decadal changes which are significantly related to the low-frequency component of
ENSO cycles. The overall warming trends of T
B max
B
, T
B amax
B
, T
B min
B
and T
B amin
B
since the late
1970s have been in phase with the persistent and exceptionally strong warm phase of
ENSO cycles. Furthermore, the EOF1 coefficient series of T
B min
B
and T
B amin
B
have monotonically increased at a faster rate than those of T
B max
B
, and T
B amax
B
since the mid 1950s that resemble the greenhouse warming fingerprint observed in instrumental records and predicted by some models. At this point, however, it is unclear whether the recent changes in T
B max
B
, T
B amax
B
, T
B min
B
and T
B amin
B
are in direct response to greenhouse gas forcing, or whether these changes are associated with the natural decadal timescale variation in the atmospheric circulation. Another conspicuous feather is that there is a significant narrowing for diurnal temperature ranges over most parts of Thailand, resulting from the differential changes in maximum and minimum temperatures.
The results from this study provide a vital clue of some key aspects of short-andlong term climate change in Thailand that has important implications for future prediction and environmental management. There is little doubt that climate change is an active and critical component of “our Earth System” as current and future threats for
3

human and environmental systems that is now happening even on regional/local scales and will likely continue or even intensify in the near future.
4

ควำมแปรปรวนหรือกำรเปลี่ยนแปลง
เป็นสิ่งปกติที่เกิดขึ้นในระบบภูมิอำกำศและสภำวะสมดุลแทบจะไม่เกิดขึ้นทุกค
ำบเวลำ (Timescale) หรือแม้กระทั่งคำบเวลำใดเวลำหนึ่งในระบบดังกล่ำว
หลักฐำนจำก palaeo-records ระบุชัดเจนว่ำ
ภูมิอำกำศของโลกมีกำรเปลี่ยนแปลงอย่ำงต่อเนื่องทุกคำบเวลำ
โดยสภำวะเฉลี่ยของโลกอยู่ภำยใต้ควำมแปรปรวนที่สูงของระบบภูมิอำกำศใน
ระดับภูมิภำค ควำมแปรปรวนของอุณหภูมิอำกำศ
จัดว่ำเป็นดัชนีที่ส ำคัญของกำรเปลี่ยนแปลงสภำพภูมิอำกำศของโลกที่มีกำรศึ
กษำวิจัยกันมำก
เนื่องจำกอุณหภูมิมีบทบำทที่ส ำคัญในกำรควบคุมและก ำหนดขบวนกำรระเหย
และกำรคำยน้ ำของพืช
ซึ่งมีกำรเชื่อมโยงโดยตรงกับวัฎจักรของน้ ำและสมดุลของควำมร้อนที่พื้นผิว
นอกจำกนี้ กำรเปลี่ยนแปลงอุณหภูมิทั้งอัตรำ
และควำมรุนแรงยังมีบทบำทและอิทธิพลอย่ำงสูงต่อหน้ำที่
และโครงสร้ำงของระบบนิเวศน์วิทยำ
ทิศทำง
พลวัตร
ตลอดจนสุขภำพและควำมสะดวกสบำยของมนุษย์ Intergovernmental
Panel on Climate Change (IPCC) รำยงำนไว้ในปี ค.ศ 2001 .
ว่ำอุณหภูมิเฉลี่ยของโลกในช่วงศตวรรษที่ 20 เพิ่มขึ้น 0.6
 0.2
องศำเซลเซียส
1990 IPCC
และอุณหภูมิเฉลี่ยของโลกเพิ่มขึ้นสูงสุดในช่วงทศวรรษที่
ยังยืนยันด้วยว่ำ "มีหลักฐำนที่เชื่อได้ว่ำ
กิจกรรมของมนุษย์ได้มีส่วนท ำให้ภูมิอำกำศโลกเปลี่ยนแปลงไปอย่ำงมำก
โดยเฉพำะอย่ำงยิ่งกำรปล่อยก๊ำซเรือนกระจก
ที่เกิดจำกกำรใช้เชื้อเพลิงฟอสซิลในภูมิภำคต่ำงๆ
5

ของโลกที่เพิ่มอย่ำงรวดเร็วในช่วงศตวรรษที่ผ่ำนมำ"
ได้มีกำรคำดกำรณ์กันไว้ว่ำ ในปี
อุณหภูมิเฉลี่ยของโลกจะสูงขึ้นประมำณ 1.4-5.8
ค.ศ 2100 .
องศำเซลเซียส
ซึ่งเป็นอัตรำกำรเพิ่มที่สูงสุดตั้งแต่สมัยสิ้นยุคโลกน้ ำแข็ง
P
นอกจำกนี้ยังมีหลักฐำนทำงวิทยำศำสตร์ที่บ่งชี้ว่ำ
กำรเปลี่ยนแปลงของอุณหภูมิอำกำศในระยะสั้นและระยะยำว
(ปีต่อปีถึงทศวรรษต่อทศวรรษ)
ทวีปอเมริกำเหนือ-ใต้
(Ice
ในหลำยภูมิภำคของโลก
Age)
เช่น
ทวีปเอเซียและทวีปแอฟริกำ
P
ยังได้รับผลกระทบจำกปรำกฎกำรณ์เอนโซ่ (El Niño-Southern Oscillation;
ENSO) หรือเอลนิโน่-ควำมผันแปรของระบบอำกำศในซีกโลกใต้
ซึ่งเป็นปรำกฎกำรณ์ธรรมชำติระดับโลกที่เกิดจำกกำรเชื่อมโยงระหว่ำงกำรเป
ลี่ยนแปลงที่ผิดปกติของอุณหภูมิผิวน้ ำทะเลบริเวณเส้นศูนย์สูตรทำงมหำสมุทร
แปซิฟิก และควำมผันแปรที่ผิดปกติของระบบอำกำศในซีกโลกใต้
ถึงแม้ว่ำหลักฐำนทำงวิทยำศำสตร์บ่งชี้อย่ำงชัดเจนถึงกำรเปลี่ยนแปลงสภำพภู
มิอำกำศของโลก
กำรพยำกรณ์ผลกระทบที่อำจจะเกิดขึ้นในอนำคตจำกกำรเปลี่ยนแปลงดังกล่ำ
ว ยังมีควำมไม่แน่นอนและมีข้อจ ำกัดค่อนข้ำงมำกในช่วงที่ผ่ำนมำ
เนื่องจำกยังขำดรู้ควำมเข้ำใจอย่ำงถ่องแท้ถึงกลไกกำรเชื่อมโยง
ปัจจัยภำยนอกที่บังคับ (Forcings) กำรตอบสนอง (Responses)
และผลที่ตำมมำ (Consequences)
ของกำรเปลี่ยนแปลงสภำพภูมิอำกำศของโลก
กำรศึกษำวิจัยเรื่องกำรเปลี่ยนแปลงของอุณหภูมิอำกำศ
ดังนั้น
โดยเฉพำะอย่ำงยิ่งกำรเปลี่ยนแปลงในระดับภูมิภำค
เป็นประเด็นที่ท้ำทำยและได้รับควำมสนใจจำกนักวิทยำศำสตร์เป็นจ ำนวนมำก
รวมทั้งเป็นวัตถุประสงค์หลักของโครงกำรวิจัยกำรเปลี่ยนแปลงของโลก
(Global Change Research) ส ำหรับประเทศไทย
ประเด็นดังกล่ำวยังไม่ค่อยได้รับควำมสนใจและมีกำรศึกษำมำกนัก
รวมทั้งไม่มีหลักฐำนที่แน่ชัดของกำรเปลี่ยนแปลงอุณหภูมิอำกำศทั้งระยะสั้นแ
ละระยะยำว ตลอดจนกระทบที่อำจจะเกิดขึ้น
ดังนั้น กำรศึกษำนี้มีวัตถุประสงค์หลัก เพื่อศึกษำ 1)
รูปแบบควำมแปรปรวนในเชิงพื้นที่และเชิงเวลำที่มีลักษณะโดดเด่นของอุณหภู
มิอำกำศในประเทศไทย 2)
6

กลไกกำรเชื่อมโยงทั้งในระยะสั้นและระยะยำวระหว่ำงควำมแปรปรวนดังกล่ำ
วกับพฤติกรรมของควำมแปรปรวนตำมธรรมชำติของสภำพภูมิอำกำศของโลก
หรือควำมผันแปรของสภำพภูมิอำกำศที่เกิดจำกกิจกรรมของมนุษย์ 3)
ผลกระทบที่อำจจะเกิดขึ้นต่อสภำพแวดล้อม นิเวศน์วิทยำ
สภำพเศรษฐกิจและสังคมรวมทั้งสุขภำพอนำมัยและควำมเป็นอยู่ของมนุษย์
ข้อมูลที่น ำมำวิเครำะห์ทำงสถิติในเชิงลึก ได้แก่
ข้อมูลรำยเดือนของอุณหภูมิอำกำศเฉลี่ย สูงสุด และต่ ำสุด ( T
B mean
B
, T
B max
B
, T
B min
B
)
และอุณหภูมิจุดน้ ำค้ำง ( T
B dew
B
) จำกกรมอุตุนิยมวิทยำ จ ำนวน 33 สถำนี
ซึ่งคลอบคลุมทั่วทุกภำคของประเทศไทย ในระหว่ำงปี ค.ศ. 1951-2003
ตลอดจนอุณหภูมิปรำกฎ (Apparent Temperature) เฉลี่ย สูงสุด และต่ ำสุด
( T
B amean
B
, T
B amax
B
, T
B amin
B
)
ซึ่งค ำนวณจำกข้อมูลอุณหภูมิอำกำศและอุณหภูมิจุดน้ ำค้ำงดังกล่ำวข้ำงต้น
เทคนิคทำงสถิติที่ใช้ในกำรวิเครำะห์ข้อมูล ประกอบด้วย Empirical
Orthogonal Functions (EOFs), ค่ำเฉลี่ยแบบเคลื่อนที่ (Moving Average),
กำรวิเครำะห์ควำมแปรปรวน (Variance Analysis), กำรวิเครำะห์สหสัมพันธ์
(Correlation Analysis), และกำรวิเครำะห์กำรถดถอยแบบเชิงเส้น (Linear
Regression Analysis) EOFs นับว่ำเป็นเทคนิคทำงสถิติเชิงตัวแปรพหุ
(Multivariate) ที่นิยมใช้กันอย่ำงแพร่หลำย
ในกำรวิเครำะห์ควำมแปรปรวนเชิงพื้นที่และเชิงเวลำของชุดข้อมูลที่มีขนำดใ
หญ่ โดยเฉพำะอย่ำงยิ่งตัวแปรที่เกี่ยวกับภูมิอำกำศ บรรยำกำศและมหำสมุทร
ที่มีจุดเก็บตัวอย่ำงเป็นจ ำนวนมำก
ควำมถี่ในกำรเก็บตัวอย่ำงสูงรวมทั้งระยะเวลำในกำรเก็บข้อมูลที่ยำวนำน
ซึ่งท ำให้มีชุดข้อมูลในเชิงพื้นที่และเชิงเวลำเป็นจ ำนวนมำกยำกต่อกำรจัดกำร
และวิเครำะห์โดยใช้เทคนิคอื่น ๆ
ในกรณีข้อมูลอุณหภูมิอำกำศในประเทศไทย
ที่ท ำกำรตรวจวัดทุกเดือนตลอดระยะเวลำ 53 ปี ณ 33 สถำนี
จัดว่ำเป็นชุดข้อมูลที่ค่อนข้ำงใหญ่ เนื่องจำกมีจ ำนวนข้อมูลทั้งหมดเท่ำกับ
20998 ชุดข้อมูล
ระเบียบวิธีของเทคนิค EOFs
อำศัยหลักกำรของกำรแปลงเชิงเส้นตรงของชุดข้อมูลเดิมที่มีขนำดใหญ่และมี
ตัวแปรจ ำนวนมำก
7

ไปสู่ชุดขนำดเล็กของตัวแปรแต่เป็นตัวแทนควำมแปรปรวนทั้งเชิงพื้นที่และเชิ
งเวลำส่วนใหญ่ของชุดข้อมูลเดิม โดยทั่วไปวิธีกำรวิเครำะห์ EOFs
จะค ำนวณจำกเมตริกซ์ควำมแปรปรวนร่วม (Covariance Matrix)
หรือเมตริกซ์ควำมสัมพันธ์ร่วม (Correlation Matrix) ของข้อมูล
เพื่อจ ำแนกข้อมูลเดิมออกเป็นค่ำ Eigenvalue, Eigenvector และอนุกรม
Time Coefficient (TC) โดย Eigenvector
คือชุดขนำดเล็กของตัวแปรที่ถูกแปลงมำจำกชุดข้อมูลเดิม ซึ่งแต่ละชุดของ
Eigenvector เรียกว่ำ EOF โหมด (EOF Mode) และจ ำนวน EOF
โหมดทั้งหมดจะเท่ำกับจ ำนวนตัวแปรในชุดข้อมูลเดิม ส ำหรับค่ำ Eigenvalue
โดยปกติจะเรียงล ำดับจำกมำกไปหำน้อยและแต่ละค่ำของ Eigenvalue
จะเป็นสัดส่วนกับเปอร์เซ็นต์ของควำมแปรปรวนในข้อมูลเดิมที่อธิบำยได้จำกแ
ต่ละ EOF โหมด ทั้งนี้กำรเปลี่ยนแปลงเชิงเวลำหรืออนุกรม TC
ของแต่ละชุดของ Eigenvector
สำมำรถค ำนวณได้จำกผลรวมทั้งหมดของข้อมูลเดิมฉำยำ (Projection)
บนแต่ละชุดของ Eigenvector หรือแต่ละ EOF โหมด ในแต่ละชุดของ
Eigenvector หรือแต่ละ EOF โหมด มีคุณสมบัติพิเศษคือ
เป็นอิสระหรือตั้งฉำกต่อกัน (Orthogonality) ในเชิงพื้นที่ เช่นเดียวกับอนุกรม
TC ของแต่ละ EOF โหมดมีคุณสมบัติเป็นอิสระหรือตั้งฉำกต่อกันในเชิงเวลำ
จำกคุณสมบัติดังกล่ำว ควำมแปรปรวนในข้อมูลเดิมที่อธิบำยได้จำกแต่ละ
EOF โหมด มีคุณสมบัติที่เป็นอิสระต่อกัน ดังนั้น
ผลรวมทั้งหมดของควำมแปรปรวนที่อธิบำยได้จำกทุก EOF โหมด
จะเท่ำกับผลรวมทั้งหมดของควำมแปรปรวนในข้อมูลเดิม โดยปกติ EOF
โหมดแรก ๆ เท่ำนั้น จะอธิบำยควำมแปรปรวนส่วนใหญ่ในข้อมูลเดิม ดังนั้น
ชุดข้อมูลใหม่ที่มีควำมแปรปรวนใกล้เคียงกับควำมแปรปรวนในข้อมูลเดิม
แต่มีจ ำนวนตัวแปรน้อยกว่ำมำกเมื่อเปรียบเทียบกับข้อมูลเดิม
สำมำรถสังเครำะห์ได้จำกผลรวมของอนุกรม TC คูณด้วย EOF
โหมดในโหมดแรก ๆ เท่ำนั้น
ผลลัพธ์ของกำรวิเครำะห์ EOFs ประกอบด้วย 1) ค่ำ Eigenvalue
รวมทั้งเปอร์เซ็นต์ของควำมแปรปรวนที่อธิบำยได้จำกแต่ละ EOF โหมด 2)
8

ชุดของ Eigenvector ส ำหรับแต่ละ EOF โหมด โดยแต่ละชุดของ
Eigenvector ประกอบด้วยค่ำที่เรียกว่ำ Component Loading
ซึ่งปกติจะแสดงในรูปค่ำสัมประสิทธิ์สหสัมพันธ์ (Correlation Coefficient)
และเป็นค่ำที่บ่งชี้ถึงอ ำนำจควำมสัมพันธ์ของอนุกรมข้อมูลในแต่ละสถำนีหรือจุ
ดของข้อมูลเดิมใน EOF โหมดที่ถูกแยกออกมำ และ 3) อนุกรม TC
ซึ่งแสดงกำรเปลี่ยนแปลงเชิงเวลำของแต่ละ EOF โหมด
ผลกำรวิเครำะห์ EOFs
ส ำหรับข้อมูลอุณหภูมิอำกำศในประเทศไทยในระหว่ำงปี ค.ศ. 1951-2003
ปรำกฎว่ำ EOF โหมดที่ 1 ของอุณหภูมิอำกำศทั้งเจ็ดตัวแปร ( T
B mean
B
, T
B max
B
,
T
B min
B
, T
B dew
B
, T
B amean
B
, T
B amax
B
และ T
B amin
B
)
สำมำรถอธิบำยควำมแปรปรวนในข้อมูลเดิมได้ถึงร้อยละ 61.2 % ถึง 71.3%
ส ำหรับ EOF โหมดที่เหลือ
สำมำรถอธิบำยควำมแปรปรวนของข้อมูลเดิมได้น้อยมำกเมื่อเปรียบเทียบกับ
EOFโหมดที่ 1 และร้อยละของควำมแปรปรวนที่อธิบำยได้ในแต่ละ EOF
โหมดมีค่ำใกล้เคียงกัน จำกลักษณะดังกล่ำว
แสดงว่ำควำมแปรปรวนโดยส่วนใหญ่ของข้อมูลเดิมสำมำรถอธิบำยได้จำก
EOF โหมดที่ 1 ส่วนควำมแปรปรวนที่เหลือส่วนน้อยที่ถูกแยกตำมสัดส่วนใน
EOF โหมดที่เหลืออำจจะเกิดจำก
หรือควำมแปรปรวนปลีกย่อยของแต่ละสถำนีในข้อมูลเดิม
ส ำหรับแต่ละตัวแปรของอุณหภูมิอำกำศ Component
“noise”
Loading
ซึ่งแสดงในรูปของค่ำสัมประสิทธิ์สหสัมพันธ์ระหว่ำงชุดข้อมูลในแต่ละสถำนีกั
บ EOF โหมดที่ 1 มีค่ำที่สูงและใกล้เคียงกันเกือบทุกสถำนี
ยกเว้นบำงสถำนีในภำคใต้ตอนล่ำงที่มีค่ำค่อนข้ำงต่ ำ นอกจำกนี้
ชุดข้อมูลทุกสถำนีมีควำมสัมพันธ์ทำงสถิติในเชิงบวกอย่ำงมีนัยส ำคัญกับ
EOF โหมดที่ 1 จำกผลดังกล่ำว สำมำรถสรุปได้ว่ำ
ควำมสัมพันธ์ของอุณหภูมิอำกำศระหว่ำงสถำนีมีค่ำสูงและควำมแปรปรวนของ
อุณหภูมิอำกำศทุกสถำนีมีกำรกระจำยตัวค่อนข้ำงสม่ ำเสมอ ดังนั้น
ควำมแปรปรวนเชิงพื้นที่ที่อธิบำยได้จำก EOF โหมดที่ 1
ไม่ได้เกิดจำกควำมแปรปรวนของข้อมูลอุณหภูมิอำกำศเฉพำะสถำนีใดสถำนีห
นึ่งหรือภำคใดภำคหนึ่งเท่ำนั้น
9

แต่เกิดจำกควำมแปรปรวนของข้อมูลอุณหภูมิอำกำศเกือบทุกสถำนีร่วมกัน
โดยควำมแปรปรวนดังกล่ำว
อำจจะเกิดจำกปรำกฎกำรณ์หรือแหล่งก ำเนิดเดียวกัน
ที่มีขนำดใหญ่เพียงพอที่จะสำมำรถมีอิทธิพลต่ออุณหภูมิอำกำศทั่วทุกภำคของ
ประเทศไทย ดังนั้น เพียง EOF โหมดที่ 1 สำมำรถใช้เป็นตัวแทนที่เหมำะสม
เพื่อน ำไปอธิบำยกำรเปลี่ยนแปลงเชิงพื้นที่และเชิงเวลำของอุณหภูมิอำกำศใน
ประเทศไทยโดยส่วนใหญ่และภำพรวมได้
กำรเปลี่ยนแปลงในเชิงเวลำของ EOF โหมดที่ 1 ดังแสดงในอนุกรม
TC จำกกำรสังเกต พบว่ำ EOF โหมดที่ 1 ของอุณหภูมิอำกำศทั้งเจ็ดตัวแปร
มีลักษณะกำรเปลี่ยนแปลงในเชิงเวลำที่ค่อนซับซ้อน
ระยะเวลำของกำรแกว่งไปมำระหว่ำงค่ำสูงสุดและค่ำต่ ำสุดไม่แน่นอนและไม่ส
ม่ ำเสมอ
โดยที่กำรเปลี่ยนแปลงคำบเดือนต่อเดือนหรือกำรเปลี่ยนแปลงที่ควำมถี่สูงปรำ
กฎโดดเด่นใน TC นอกจำกนี้ กำรเปลี่ยนแปลงที่ควำมถี่ปำนกลำงถึงต่ ำ
คือตั้งแต่ 2-3 ปี ถึงคำบ 10 ปี
ยังเป็นองค์ประกอบส ำคัญของกำรเปลี่ยนแปลงใน TC อีกด้วย เนื่องจำก TC
คือ กำรเปลี่ยนแปลงในเชิงเวลำของ EOF โหมดที่ 1 โดยภำพรวม
ซึ่งประกอบด้วยกำรเปลี่ยนแปลงของทุกคำบเวลำรวมกัน
ดังนั้นจึงไม่สำมำรถระบุได้ชัดเจนว่ำ EOF โหมดที่
มีลักษณะกำรเปลี่ยนแปลงที่ชัดเจนหรือโดดเด่นในช่วงคำบเวลำใดบ้ำง
1
นอกจำกนี้ กำรหำควำมสัมพันธ์หรือกำรเชื่อมโยงของ EOF โหมดที่ 1
กับควำมผันแปรของสภำพภูมิอำกำศของโลกทั้งในระยะสั้นหรือระยะยำว
อนุกรม TC ดังกล่ำวควรที่จะถูกจ ำแนกออกเป็น
ช่วงคำบเวลำของกำรเปลี่ยนแปลงที่ใกล้เคียงหรือสอดคล้องกับคำบเวลำที่โดด
เด่นของควำมผันแปรของสภำพภูมิอำกำศของโลก ด้วยเหตุผลดังกล่ำว
อนุกรม TC จึงถูกจ ำแนกออกเป็นสองคำบเวลำของกำรเปลี่ยนแปลง คือ
น้อยกว่ำ 5 ปีและมำกกว่ำ 5 ปี สำเหตุที่เลือกสองคำบเวลำดังกล่ำว
เพรำะคำบเวลำที่น้อยกว่ำ 5 ปี
แทนกำรเปลี่ยนแปลงระยะสั้นที่สอดคล้องกับวงจรของปรำกฎกำรณ์เอนโซ่ (El
Niño-Southern Oscillation;
ควำมผันแปรของระบบอำกำศในซีกโลกใต้
ENSO) หรือเอลนิโน่-
10

ซึ่งวงจรกำรเกิดแต่ละครั้งจะมีช่วงระยะเวลำประมำณ 2 ถึง 6 ปี
ปรำกฎกำรณ์เอนโซ่เป็นปรำกฎกำรณ์ทำงธรรมชำติของควำมแปรปรวนของส
ภำพภูมิอำกำศของโลก
ที่เกิดจำกกำรเชื่อมโยงระหว่ำงกำรเปลี่ยนแปลงที่ผิดปกติของอุณหภูมิผิวน้ ำท
ะเลบริเวณเส้นศูนย์สูตรทำงมหำสมุทรแปซิฟิก
และควำมผันแปรที่ผิดปกติของระบบอำกำศในซีกโลกใต้
เป็นที่ทรำบกันดีว่ำปรำกฎกำรณ์เอนโซ่มีผลกระทบต่อสภำพภูมิอำกำศ
และสภำพแวดล้อมของโลกทั้งพื้นที่ใกล้เคียงและพื้นที่ห่ำงไกลในหลำยทวีป
โดยเฉพำะประเทศในเขตร้อน (Tropical) และกึ่งร้อน (Subtropical)
ส่วนคำบเวลำที่มำกกว่ำ 5 ปี แทนกำรเปลี่ยนแปลงในระยะยำว
(ทศวรรษต่อทศวรรษ)
ที่อำจจะมีควำมสัมพันธ์หรือเชื่อมโยงกับกำรเปลี่ยนแปลงภูมิอำกำศของโลกที่เ
กิดจำกกิจกรรมมนุษย์ เช่น
กำรเพิ่มขึ้นของปริมำณก๊ำซเรือนกระจกหรือเกิดจำกปรำกฎกำรณ์ทำงธรรมช
ำติอื่น ๆ เทคนิคที่ใช้ในกำรแยกคำบเวลำของกำรเปลี่ยนแปลงของอนุกรม
TC คือ ค่ำเฉลี่ยแบบเคลื่อนที่ (Moving Average) โดยใช้ 60 เดือน
อนุกรมเวลำ ส ำหรับคำบเวลำที่มำกกว่ำ
ส่วนกำรหำค่ำเฉลี่ยแบบเคลื่อนที่ที่น้อยกว่ำ 5
5
ท ำได้โดยน ำค่ำเฉลี่ยแบบเคลื่อนที่ที่มำกกว่ำ 5 ปี ลบด้วย อนุกรม TC เดิม
ปี
ปี
ซึ่งจะได้ผลลัพธ์คือ ค่ำผิดสภำพ (anomalies) ของอนุกรม TC
จำกอนุกรมของค่ำเฉลี่ยแบบเคลื่อนที่ที่มำกกว่ำ 5 ปี
หลังจำกนั้นน ำค่ำผิดสภำพดังกล่ำวไปค ำนวณหำค่ำเฉลี่ยแบบเคลื่อนโดยใช้
10 เดือน อนุกรมเวลำ
จำกผลกำรวิเครำะห์เพิ่มเติม ปรำกฏว่ำ อนุกรมของ TC
ของอุณหภูมิอำกำศทั้งเจ็ดตัวแปรส ำหรับค่ำเฉลี่ยแบบเคลื่อนที่ที่น้อยกว่ำ 5 ปี
แสดงกำรเปลี่ยนแปลงในระยะสั้นที่ชัดเจน
โดยระยะเวลำกำรแกว่งไปมำระหว่ำงค่ำสูงสุดและค่ำต่ ำสุดซึ่งอยู่ในช่วงประมำ
ณ 1 ถึง 4 ปี เป็นลักษณะเด่นของอนุกรมดังกล่ำว
ผลกำรวิเครำะห์ควำมแปรปรวน (Variance Analysis)
แสดงให้เห็นว่ำกำรเปลี่ยนแปลงที่น้อยกว่ำ 5
11

ปีของอุณหภูมิอำกำศทั้งเจ็ดตัวแปร มีเปอร์เซ็นต์ควำมแปรปรวนอยู่ในช่วงของ
17.6 ถึง 25.8 % ของควำมแปรปรวนทั้งหมดของอนุกรม TC
โดยเปอร์เซ็นต์ควำมแปรปรวนของอุณหภูมิอำกำศทั้งหกตัวแปร ยกเว้น T
B min
B
เป็นอันดับสองของควำมแปรปรวนทั้งหมดของอนุกรม TC รวมกัน
ลักษณะโดดเด่นอีกอย่ำงหนึ่งของอนุกรมค่ำเฉลี่ยแบบเคลื่อนที่ที่น้อยกว่ำ 5
ปีของ TC
รูปแบบกำรเปลี่ยนแปลงมีลักษณะคล้ำยกับดัชนีของปรำกฎกำรณ์เอนโซ่
คือ
(Multiple ENSO Index) โดยที่ค่ำผิดสภำพบวก (ลบ) ของอนุกรม TC
ส ำหรับค่ำเฉลี่ยแบบเคลื่อนที่ที่น้อยกว่ำ 5 ปี
ตรงกับหรือสอดคล้องกับปรำกฎกำรณ์เอลนีโญ (ลำนีญำ) โดยพบว่ำ
อุณหภูมิอำกำศในประเทศไทย สูง (ต่ ำ) กว่ำปกติ
ในระหว่ำงที่เกิดเหตุกำรณ์เอลนีโญ (ลำนีญำ) เช่น ในระหว่ำง 6
ครั้งที่เกิดปรำกฎกำรณ์เอลนีโญที่มีก ำลังรุนแรงที่สุดในรอบ 53 ปี
อุณหภูมิอำกำศในประเทศไทยสูงกว่ำปกติอย่ำงเด่นชัด
เช่นเดียวกับอุณหภูมิอำกำศในประเทศไทยต่ ำกว่ำปกติอย่ำงชัดเจนในระหว่ำง
8 ครั้งที่เกิดปรำกฎกำรณ์ลำนีญำที่มีก ำลังรุนแรงที่สุดในรอบ 53 ปี นอกจำกนี้
ในระหว่ำง ค.ศ. 1998-1998
อุณหภูมิอำกำศในประเทศไทยได้มีกำรแกว่งอยู่ในช่วงที่กว้ำงที่สุดในรอบ 53
ปี
ซึ่งสอดคล้องกับกำรเกิดปรำกฎกำรณ์เอลนีโญและลำนีญำที่มีก ำลังรุนแรงอย่ำ
งต่อเนื่องภำยในช่วงสองปีดังกล่ำว โดยปี ค.ศ. 1998
เป็นปีที่ร้อนที่สุดในประเทศไทยในรอบ 53 ปี ผลกำรวิเครำะห์สหสัมพันธ์
(Correlation Analysis) ยืนยันเพิ่มเติมว่ำ
อุณหภูมิอำกำศทั้งเจ็ดตัวแปรมีควำมสัมพันธ์ทำงสถิติในเชิงบวกอย่ำงมีนัยส ำคั
ญกับดัชนีของปรำกฎกำรณ์เอนโซ่ โดยเฉพำะอย่ำงยิ่ง T
B mean
B
, T
B max
B
, T
B amean
B
และ T
B amax
B
ที่มีค่ำสัมประสิทธิ์สหสัมพันธ์ค่อนข้ำงสูง (มำกกว่ำ 0.5)
ผลกำรศึกษำนี้แสดงให้เห็นว่ำ
ปรำกฎกำรณ์เอนโซ่เป็นปัจจัยที่ส ำคัญที่มีผลกระทบต่อกำรเปลี่ยนแปลงของอุ
ณหภูมิอำกำศในประเทศไทยในระยะสั้น และอำจสันนิษฐำนได้ว่ำ
ผลกระทบของปรำกฎกำรณ์เอนโซ่ต่อกำรเปลี่ยนแปลงปีต่อปีของอุณหภูมิอำก
12

ำศในประเทศไทย
น่ำจะมำจำกสำเหตุของกำรแผ่ขยำยกว้ำงไกลออกไปของปริมำณควำมร้อน
ที่เกิดจำกควำมผิดปกติของอุณหภูมิผิวน้ ำทะเล
และกำรเคลื่อนตัวของแอ่งน้ ำอุ่นในบริเวณเส้นศูนย์สูตรทำงมหำสมุทรแปซิฟิก
โดยกลไกกำรเชื่อมโยงน่ำจะผ่ำนทำง “Atmospheric Teleconnections”
นอกจำกนี้
ควำมผันแปรของระบบอำกำศโดยเฉพำะอย่ำงยิ่งกำรหมุนเวียนของอำกำศแบ
บวอคเกอร์ (Walker Circulation)
ที่เกิดจำกกำรเสียสมดุลของกำรแลกเปลี่ยนควำมร้อนระหว่ำงบรรยำยกำศและ
ทะเล
น่ำจะเป็นปัจจัยเสริมในกำรน ำพำควำมร้อนออกจำกบริเวณเส้นศูนย์สูตรของม
หำสมุทรแปซิฟิกมำสู่ประเทศไทย
นอกจำกนี้ กำรเปลี่ยนแปลงในระยะยำว (ทศวรรษต่อทศวรรษ)
ยังปรำกฎชัดเจนในอนุกรมค่ำเฉลี่ยแบบเคลื่อนที่ที่มำกกว่ำ 5 ปีของ TC ของ
T
B max
B
, T
B amax
B
, T
B min
B
และ T
B amin
โดยอุณหภูมิอำกำศทั้งสี่ตัวแปรนี้มีแนวโน้มเพิ่มขึ้นเรื่อย
หลังจำกปลำยทศวรรษที่
ซึ่งรูปแบบกำรเปลี่ยนแปลงดังกล่ำวสอดคล้องกับช่วงเวลำที่พฤติกรรมของ
ๆ
1970
B
ปรำกฎกำรณ์เอนโซ่มีแนวโน้มผิดปกติในคำบเวลำที่ยำวนำนมำกกว่ำ 10 ปี
โดยปรำกฎกำรณ์เอลนีโญเกิดขึ้นเป็นระยะเวลำที่ยำวนำนและบ่อยครั้งกว่ำปก
ติรวมทั้งมีก ำลังปำนถึงรุนแรง ซึ่งรู้จักกันดีในนำม “Climatic Regime Shift”
แต่ปรำกฎกำรณ์ลำนีญำแทบจะไม่เกิดขึ้นหรือเกิดขึ้นน้อยมำกเมื่อเปรียบเทียบ
กับปรำกฎกำรณ์เอลนีโญ หลังจำกปลำยทศวรรษที่ 1970
ผลกำรวิเครำะห์สหสัมพันธ์ ยังสนับสนุนควำมสอดคล้องดังกล่ำวข้ำงต้น
โดยพบว่ำอนุกรมค่ำเฉลี่ยแบบเคลื่อนที่ที่มำกกว่ำ 5 ปี ของ TC
ของอุณหภูมิอำกำศทั้งสี่ตัวแปรมีควำมสัมพันธ์ทำงสถิติในเชิงบวกอย่ำงมีนัย
ส ำคัญกับดัชนีของปรำกฎกำรณ์เอนโซ่
เช่นเดียวกับที่พบในกำรเปลี่ยนแปลงระยะสั้นข้ำงต้น ภำยหลังปี ค.ศ. 1990
เป็นช่วงทศวรรษที่อุณหภูมิอำกำศในประเทศไทยสูงที่สุดในรอบ 53 ปี
ซึ่งสอดคล้องกับอุณหภูมิเฉลี่ยของโลกที่สูงกว่ำค่ำปกติมำกในช่วงเวลำเดียวกั
น ผลกำรศึกษำนี้ แสดงให้เห็นว่ำปรำกฎกำรณ์เอนโซ่
อำจจะมีอิทธิพลต่อกำรเปลี่ยนแปลงของอุณหภูมิอำกำศในประเทศไทยในระย
13

ะยำวอีกด้วย
นอกจำกกำรเปลี่ยนแปลงของอุณหภูมิอำกำศในประเทศไทยที่สอดคล้องกับป
รำกฎกำรณ์เอนโซ่แล้ว ยังพบว่ำ T
B min
B
และ T
B amin
มีแนวโน้มเพิ่มขึ้นอย่ำงต่อเนื่องในลักษณะเชิงเส้นตรงตั้งแต่กลำงทศวรรษที่
1950 และอัตรำกำรเพิ่มขึ้นที่รวดเร็วและมำกกว่ำ T
B max
B
และ T
B amax
B
B
รูปแบบกำรเพิ่มขึ้นอย่ำงต่อเนื่องของ T
B min
B
และ T
B amin
B
ดังกล่ำว
มีลักษณะเหมือนกับอุณหภูมิเฉลี่ยผิวพื้นโลกที่เพิ่มสูงขึ้นในศตวรรษที่ 20
จำกกำรเพิ่มขึ้นของปริมำณก๊ำซเรือนกระจกผลสืบเนื่องมำจำกกิจกรรมมนุษย์
ซึ่งรู้จักกันดีในนำมสภำวะโลกร้อน (Global Warming) จำกกำรเปรียบเทียบ
พบว่ำ กำรเพิ่มขึ้นของ T
B min
B
และ T
B amin
B
ในประเทศไทย
มีอัตรำที่รวดเร็วและมำกกว่ำอุณหภูมิเฉลี่ยผิวพื้นโลก ดังนั้น กำรเพิ่มขึ้นของ
T
B min
B
และ T
B amin
B
ในประเทศไทย
น่ำจะมีส่วนส่งเสริมในแง่บวกที่ส่งผลให้อุณหภูมิเฉลี่ยผิวพื้นของซีกโลกเหนือ
รวมทั้งสภำวะโลกร้อนเพิ่มสูงขึ้น
ถึงแม้นกำรเปลี่ยนแปลงในระยะยำวของอุณหภูมิอำกำศในประเทศไทยซึ่งมีค
วำมสัมพันธ์กับปรำกฎกำรณ์เอนโซ่
และมีลักษณะที่คล้ำยคลึงกับสภำวะโลกร้อนอันเนื่องมำจำกกำรเพิ่มขึ้นของก๊ำ
ซเรือนกระจก ปรำกฎชัดเจนจำกผลกำรศึกษำนี้
ปัจจัยหลักที่ก่อให้เกิดกำรเปลี่ยนแปลงดังกล่ำว
ยังไม่สำมำรถแยกแยะหรือสรุปได้ชัดเจน
ว่ำเกิดจำกพฤติกรรมของควำมแปรปรวนตำมธรรมชำติของสภำพภูมิอำกำศ
เช่น ปรำกฎกำรณ์เอนโซ่
หรือผลกระทบโดยตรงจำกควำมผันแปรของสภำพภูมิอำกำศที่เกิดจำกกิจกรร
มของมนุษย์ โดยทั่วไป อำจจะเข้ำใจว่ำ
กำรเปลี่ยนแปลงที่เกิดจำกควำมผิดปกติของพฤติกรรมของควำมแปรปรวนตำ
มธรรมชำติของสภำพภูมิอำกำศ
น่ำจะมีรูปแบบหรือลักษณะในเชิงพื้นที่และเชิงเวลำที่แตกต่ำงจำกกำรเปลี่ยนแ
ปลงที่เกิดจำกกิจกรรมมนุษย์
แต่เมื่อพิจำรณำถึงพฤติกรรมของระบบภูมิอำกำศ
ที่มีรูปแบบกลไกกำรเชื่อมโยงที่ซับซ้อนและกำรตอบสนองต่อปัจจัยภำยนอกไ
ม่เป็นในลักษณะเชิงเส้นตรง (Nonlinear)
ซึ่งสำมำรถเปรียบเทียบได้กับค ำพังเพยที่ว่ำ
นักวิทยำศำสตร์หลำยท่ำน
“1 บวก 1 ไม่เท่ำกับ 2”
ได้เสนอแนะไว้ว่ำ
14

สภำวะโลกร้อนในช่วงไม่กี่ทศวรรษที่ผ่ำนมำ
อำจจะเกิดจำกพฤติกรรมของควำมแปรปรวนตำมธรรมชำติของสภำพภูมิอำก
ำศ ที่มีแนวโน้มผิดปกติทั้งในแง่ จ ำนวนครั้งที่เกิดขึ้น ทิศทำง
ระยะเวลำและควำมรุนแรง
โดยมีผลกระทบจำกกำรเพิ่มขึ้นของปริมำณก๊ำซเรือนกระจก
มำกกว่ำผลกระทบโดยตรงจำกปรำกฎกำรณ์เรือนกระจก
ตัวอย่ำงที่เห็นได้ชัดเจน ได้แก่
พฤติกรรมของปรำกฎกำรณ์เอนโซ่ที่มีแนวโน้มผิดปกติในคำบเวลำที่ยำวนำน
หลังจำกปลำยทศวรรษที่ 1970
ผลกำรศึกษำโดยใช้แบบจ ำลองทำงคณิตศำสตร์ ยังระบุว่ำ
กำรเพิ่มขึ้นของปริมำณก๊ำซเรือนกระจกจะท ำให้สภำวะเหมือนกับปรำกฎกำร
ณ์เอลนีโญ (El Niño-like) เกิดขึ้นบ่อยและระยะเวลำที่นำนขึ้นในอนำคต
ดังนั้น
ปัจจัยที่มีผลต่อกำรเปลี่ยนแปลงอุณหภูมิอำกำศในประเทศไทยในระยะยำว
จึงเป็นประเด็นที่ท้ำทำยที่ต้องศึกษำในรำยละเอียดต่อไป เพื่ออธิบำย
และสำมำรถแยกสัญญำณกำรเปลี่ยนแปลงที่เกิดจำกควำมผิดปกติของพฤติกร
รมของควำมแปรปรวนตำมธรรมชำติของสภำพภูมิอำกำศ
ออกจำกควำมผันแปรของสภำพภูมิอำกำศที่เกิดจำกกิจกรรมของมนุษย์
ถ้ำไม่ค ำนึงถึงสำเหตุที่ก่อให้เกิดกำรเปลี่ยนแปลง ผลจำกกำรศึกษำนี้
ได้แสดงอย่ำงชัดเจน ว่ำ T
B min
B
และ T
B amin
B
ในประเทศไทย ในช่วง 53
ปีที่ผ่ำนมำ เพิ่มขึ้นอย่ำงต่อเนื่องในอัตรำที่น่ำตกใจ
จำกลักษณะกำรเพิ่มขึ้นของ T
B min
B
และ T
B amin
B
ในอัตรำที่รวดเร็วและมำกกว่ำ T
B max
B
และ T
B amax
B
ส่งผลให้อุณหภูมิอำกำศต่ ำสุดทุกภำคของประเทศไทยขยับสูงขึ้นค่อนข้ำงมำก
อย่ำงมีนัยส ำคัญในอัตรำเฉลี่ย 1.35  C ภำยในระยะเวลำ 50 ปี
ตลอดจนช่วงของอุณหภูมิต่ ำสุดและอุณหภูมิสูงสุดรำยวัน (Diurnal
Temperature Range; DTR) ในประเทศไทยมีแนวโน้มที่แคบลงเรื่อย ๆ
อย่ำงมีนัยส ำคัญเช่นกันในอัตรำเฉลี่ย -0.99  C ภำยในระยะเวลำ 50 ปี
ปัจจัยเฉพำะแห่งที่มีผลกระทบต่อ DTR
อำจจะเกิดจำกกำรขยำยตัวของชุมชนเมือง
กำรขยำยตัวของพื้นที่ที่แห้งแล้งหรือทะเลทรำย
ระบบชลประทำน
และควำมแปรปรวนเที่เกิดจำกลักษณะกำรใช้ประโยชน์ของที่ดิน
15

โดยเฉพำะอย่ำงยิ่ง ในบริเวณชุมชนเขตเมือง ช่วงของ DTR จะแคบกว่ำปกติ
แต่อย่ำงไรก็ตำม
จำกผลกำรวิเครำะห์ผลกระทบของชุมชนเมืองต่อลักษณะกำรเปลี่ยนแปลงขอ
งอุณหภูมิอำกำศ พบว่ำ
ค่ำเฉลี่ยรำยปีของอุณหภูมิอำกำศต่ ำสุดและสูงสุดรวมทั้ง DTR
ของโลกและซีกโลกเหนือหรือใต้ที่ค ำนวณจำกสถำนีที่ไม่ตั้งอยู่ในชุมชนเมือง
แตกต่ำงเพียงเล็กน้อยเมื่อเปรียบเทียบกับค่ำดังกล่ำวข้ำงต้นที่ค ำนวณจำกสถำ
นีที่ตั้งอยู่ในเขตชุมชนเมือง นอกจำกนี้ ปัจจัยที่มีผลกระทบต่อ DTR
อำจจะเกิดจำกกำรเปลี่ยนแปลงของระบบหมุนเวียนของสภำพภูมิอำกำศโลก
ซึ่งประกอบด้วย กำรเพิ่มขึ้นของเมฆและละอองในชั้นบรรยำย
และกำรเพิ่มขึ้นของควำมเย็นพื้นผิวเนื่องมำจำกฝนและของปริมำณก๊ำซเรือน
กระจก ส ำหรับประเทศไทย DTR ที่มีแนวโน้มที่แคบลงเรื่อย ๆ
ซึ่งมีลักษณะกำรเปลี่ยนแปลงที่เหมือนและสอดคล้องกันทุกภำค
บ่งชี้ถึงแหล่งก ำเนิดของกำรเปลี่ยนแปลงของ DTR
ไม่น่ำจะเกิดจำกผลกระทบของปรำกฎกำรณ์เฉพำะแห่ง
แต่เป็นกำรสะท้อนให้เห็นถึง
กำรเพิ่มขึ้นค่อนข้ำงมำกของอุณหภูมิอำกำศต่ ำสุด
เนื่องจำกกำรตอบสนองต่อควำมผิดปกติของพฤติกรรมของควำมแปรปรวนตำ
มธรรมชำติของสภำพภูมิอำกำศโลก
หรือควำมผันแปรของสภำพภูมิอำกำศโลกที่เกิดจำกกิจกรรมของมนุษย์
กำรเปลี่ยนแปลงที่เหมือนกับผลดังกล่ำวข้ำงต้น
ไม่ได้เกิดขึ้นในประเทศไทยเท่ำนั้น ผลกำรศึกษำในหลำย ๆ พื้นที่ของโลก
ระบุถึง กำรเปลี่ยนแปลงของอุณหภูมิอำกำศต่ ำสุดและช่วงของ DTR
ที่ขยับสูงขึ้นอย่ำงต่อเนื่องและแคบลงอย่ำงมีนัยส ำคัญในศตวรรษที่ 20
ซึ่งส่งผลให้จ ำนวนวันหรือคืนที่อำกำศเย็นลดลงและช่วงของฤดูหนำวสั้นลงแต่
ฤดูใบไม้ผลิยำวขึ้น
เป็นที่ยอมรับกันโดยทั่วไปว่ำ
กำรเปลี่ยนแปลงของสภำพภูมิอำกำศโลกทั้งระยะสั้นและระยะยำว
ไม่ว่ำจะเกิดจำกควำมผิดปกติของพฤติกรรมของควำมแปรปรวนตำมธรรมชำ
ติหรือเกิดจำกกิจกรรมของมนุษย์
เป็นปัจจัยที่ส ำคัญที่ส่งผลกระทบอย่ำงกว้ำงขวำงและรุนแรงต่อสภำพแวดล้อม
ระบบนิเวศน์วิทยำ
สภำพเศรษฐกิจและสังคมรวมทั้งสุขภำพอนำมัยและควำมเป็นอยู่ของมนุษย์
16

เนื่องจำกควำมซับซ้อนของระบบสภำพแวดล้อมและนิเวศน์วิทยำรวมทั้งควำม
ไหวต่อปัจจัยภำยนอกของระบบดังกล่ำว
กำรเปลี่ยนแปลงของสภำพภูมิอำกำศโลก
อำจก่อให้เกิดผลกระทบในลักษณะที่ไม่เป็นในเชิงเส้นตรง (Nonlinear)
ดังค ำพังเพยที่ว่ำ “the straw that breaks the camel’s back”
ซึ่งจะส่งผลให้กำรตอบสนองที่ค่อนข้ำงรุนแรงของระบบสภำพแวดล้อมและนิเ
วศน์วิทยำ ต่อกำรเปลี่ยนแปลงเพียงเล็กน้อยของสภำพภูมิอำกำศโลก
ตัวอย่ำงที่เห็นได้ชัดเจน ได้แก่
กำรตอบสนองอย่ำงรุนแรงและกว้ำงขวำงของระบบสภำพแวดล้อมและนิเวศน์
วิทยำทั่วภูมิภำคของโลก
ต่อกำรเปลี่ยนแปลงของอุณหภูมิอำกำศของโลกเพียงแค่ 0.6
 0.2
องศำเซลเซียส ในช่วงศตวรรษที่ 20
หลักฐำนทำงวิทยำศำสตร์เท่ำที่มีในปัจจุบัน ระบุชัดเจนว่ำ
กำรเปลี่ยนแปลงของสภำพภูมิอำกำศทั้งในระดับภูมิภำคและระดับโลก
โดยเฉพำะอย่ำงยิ่ง
กำรเพิ่มขึ้นของอุณหภูมิอำกำศในช่วงไม่กี่ทศวรรษที่ผ่ำนมำ
ได้ส่งผลกระทบต่อระบบสภำพแวดล้อมและนิเวศน์วิทยำต่ำง ๆ
ในหลำยภูมิภำคของโลก ตัวอย่ำงของกำรเปลี่ยนแปลงที่ได้ค้นพบ ได้แก่
กำรละลำยของภูเขำน้ ำแข็งในบริเวณขั้วโลกเหนือและใต้ กำรละลำยของหิมะ
ระดับน้ ำทะเลสูงขึ้น
กำรเคลื่อนตัวสู่ขั้วโลกของพื้นที่ที่สำมำรถด ำรงชีวิตของพืชและสัตว์บำงชนิดใ
นเขตร้อน (Tropical) และกึ่งร้อน (Subtropical)
กำรเพิ่มหรือลดลงของจ ำนวนประชำกรของพืชและสัตว์บำงชนิด
ฤดูกำรเจริญเติบโตของพืชและสัตว์ในบริเวณ mid-to-high latitude ยำวขึ้น
พืชและสัตว์ออกดอกและผสมพันธ์เร็วขึ้น
ควำมสัมพันธ์ระหว่ำงกำรเปลี่ยนแปลงของอุณภูมิในระดับภูมิภำค
นอกจำกนี้
และกำรเปลี่ยนแปลงของระบบสภำพแวดล้อมและนิเวศน์วิทยำ
ยังปรำกฎชัดเจนในมหำสมุทร
กำรเปลี่ยนแปลงของกำรกระจำยตัวและจ ำนวนประชำกรของแพลงตอนพืชแ
ละสัตว์ในบริเวณชำยฝั่งของรัฐแคลิฟอร์เนีย
ต่อกำรเปลี่ยนแปลงทั้งในระยะสั้นและระยะยำวของอุณหภูมิผิวน้ ำทะเล
ที่เกิดจำกปรำกฎกำรณ์เอนโซ่และสภำวะโลกร้อน
เป็นที่ทรำบกันดีในช่วงไม่กี่ทศวรรษที่ผ่ำนมำ
17

นิเวศน์วิทยำชำยฝั่งที่มีคุณค่ำมหำศำลทำงเศรฐศำสตร์และควำมหลำกหลำยท
ำงชีวภำพ รวมทั้งเป็นแหล่งอำหำรที่ส ำคัญของมนุษย์ โดยเฉพำะแนวปะกำรัง
ก ำลังได้รับภัยคุกคำมจำกกำรเพิ่มขึ้นของอุณหภูมิทะเล
กำรเพิ่มขึ้นของปรำกฎกำรณ์ฟอกขำวของปะกำรัง (Coral Reef Bleaching)
อำจจะเกิดจำกกำรเพิ่มขึ้นของอุณหภูมิทะเลโลก หลักฐำนทำงวิทยำศำสตร์
ระบุว่ำ ได้เกิดปรำกฎกำรณ์ฟอกขำวของปะกำรังที่รุนแรง จ ำนวน 6 ครั้ง
ตั้งแต่ปี ค.ศ .1979
และควำมรุนแรงรวมทั้งจ ำนวนครั้งมีแนวโน้มเพิ่มขึ้นตั้งแต่นั้นมำ
ซึ่งปรำกฎกำรณ์ฟอกขำวของปะกำรังที่รุนแรงที่สุด
เกิดขึ้นในช่วงที่ตรงกับหรือสอดคล้องกับกำรเกิดปรำกฎกำรณ์เอลนีโญในระห
ว่ำงปี ค.ศ. 1997-1998 โดยทั้ง 10
แนวเขตปะกำรังที่มีขนำดใหญ่ของโลกได้รับผลกระทบอย่ำงรุนแรง
นอกจำกนี้ กำรเปลี่ยนแปลงของอุณหภูมิ
ยังเป็นปัจจัยเสี่ยงต่อกำรสูญพันธ์ของพืชและสัตว์ โดยเฉพำะอย่ำงยิ่ง
ตระกูลหรือชนิดที่ได้รับภัยคุกคำมจำกกำรเปลี่ยนแปลงของสิ่งแวดล้อมแล้วใน
ปัจจุบัน
กำรเพิ่มขึ้นของระดับน้ ำทะเลจำกสภำวะโลกร้อน
เป็นประเด็นที่ได้รับควำมสนใจอย่ำงกว้ำงขวำง
และส่งผลกระทบต่อหลำยประเทศที่มีพื้นที่เป็นเกำะและอำณำเขตติดกับทะเล
กำรเพิ่มขึ้นของระดับน้ ำทะเลยังมีอัตรำที่ไม่แน่นอน
แต่จำกกำรประมำณครั้งล่ำสุดของ IPCC พบว่ำอยู่ในช่วง 10-94 เซนติเมตร
ภำยในปี ค.ศ .2100
เนื่องจำกกำรเคลื่อนตัวของควำมร้อนเกิดขึ้นช้ำในมหำสมุทร
กำรเพิ่มขึ้นของระดับน้ ำทะเลจะเกิดขึ้นอย่ำงต่อเนื่องนำนกว่ำกำรเปลี่ยนแปลง
ของอุณหภูมิ
ถึงแม้ปริมำณกำรปล่อยก๊ำซเรือนกระจกจะถูกควบคุมให้อยู่ในระดับคงที่ทันที
ทันใดในปัจจุบันก็ตำม
กำรเพิ่มขึ้นของระดับน้ ำทะเลจะเกิดขึ้นอย่ำงต่อเนื่องเป็นระยะเวลำที่ยำวนำนร่
วมศตวรรษ ซึ่งจะส่งผลกระทบอย่ำงรุนแรงต่อประชำกรของโลกเป็นล้ำน ๆ
คน ประเทศในแถบตะวันตกเฉียงเหนือของมหำสมุทรแปซิฟิก
หมู่เกำะในมหำสมุทรแปซิฟิก
และในแถบตะวันออกของทวีปเอเซียรวมทั้งประเทศไทย
มีโอกำสสูงที่จะได้รับผลกระทบจำกกำรเพิ่มขึ้นของระดับน้ ำทะเล
18

เนื่องจำกพื้นที่ดังกล่ำวมีลักษณะสูงกว่ำระดับน้ ำทะเลเพียงเล็กน้อย
จำกผลกำรประมำณอัตรำกำรเพิ่มขึ้นของระดับน้ ำทะเลในบริเวณอ่ำวไทยตอ
นบน โดยใช้แบบจ ำลองทำงคณิตศำสตร์ พบว่ำ
ระดับน้ ำทะเลจะเพิ่มขึ้นในช่วง 1-3 เมตร ภำยในปี ค.ศ .2100
ระดับน้ ำทะเลที่เพิ่มขึ้นในอัตรำดังกล่ำว
จะท ำให้พื้นที่ที่มีควำมสูงเฉลี่ยประมำณ 1 เมตร
ซึ่งส่วนใหญ่จะเป็นพื้นที่ป่ำชำยเลนและพื้นที่ชำยฝั่ง
จะจมอยู่ใต้น้ ำในศตวรรษหน้ำ โดยกำรเพิ่มขึ้นของระดับน้ ำทะเล
อำจจะมีอัตรำที่เร็วขึ้นกว่ำที่คำดกำรณ์ไว้
เนื่องจำกกำรทรุดตัวของแผ่นดินในบริเวณกรุงเทพฯและปริมณฑล
ผลสืบเนื่องจำกกำรสูบน้ ำบำดำลมำใช้
ผลกระทบจำกกำรเพิ่มขึ้นระดับน้ ำทะเล
จะท ำให้ปัญหำสิ่งแวดล้อมที่เกิดขึ้นแล้วในปัจจุบันในบริเวณดังกล่ำว เช่น
กำรกัดเซำะของชำยฝั่ง กำรรุกล้ ำของน้ ำเค็ม
ควำมเสื่อมโทรมของทรัพยำกรธรรมชำติและมลพิษและน้ ำท่วม
มีทวีควำมรุนแรงและเลวร้ำยเพิ่มขึ้น
จำกกำรประยุกต์ใช้เทคนิคทำงสถิติตัวแปรพหุ โดยเฉพำะอย่ำงยิ่ง
EOFs ในกำรวิเครำะห์ข้อมูลอุณหภูมิอำกำศ
มีส่วนช่วยให้เข้ำใจถึงแง่มุมที่ส ำคัญบำงประกำรของกำรเปลี่ยนแปลงของสภำ
พภูมิอำกำศในประเทศไทย โดยหลักฐำนจำกกำรศึกษำนี้
จะมีประโยชน์อย่ำงยิ่งต่อกำรศึกษำวิจัยเพิ่มเติมในรำยละเอียดของเรื่องกลไก
กำรเปลี่ยนแปลงของสภำพภูมิอำกำศ กำรพยำกรณ์ผลกระทบที่อำจจะเกิดขึ้น
รวมทั้งกำรอนุรักษ์และกำรจัดกำรสิ่งแวดล้อมในระดับภูมิภำคในอนำคต
นอกจำกนี้ หลักฐำนดังกล่ำว ยังเป็นข้อมูลพื้นฐำนที่ส ำคัญในกำรสร้ำง พัฒนำ
และปรับเทียบแบบจ ำลองทำงคณิตศำสตร์
ซึ่งเป็นงำนที่ท้ำทำยของกำรเปลี่ยนแปลงสภำพภูมิอำกำศของโลกในอนำคตอี
กด้วย อย่ำงไรก็ตำม
ยังมีหลำยประเด็นที่ส ำคัญที่ต้องศึกษำในรำยละเอียดและควรที่จะมุ่งเน้นในกำ
รศึกษำวิจัยในอนำคตอันใกล้ ประเด็นที่ส ำคัญอันดับต้น ๆ คือ
“ผลกระทบที่อำจจะเกิดขึ้นต่อสภำพแวดล้อม นิเวศน์วิทยำ
สภำพเศรษฐกิจและสังคมรวมทั้งสุขภำพอนำมัยและควำมเป็นอยู่ของมนุษย์
จำกสภำวะโลกร้อนที่เพิ่มขึ้นอย่ำงต่อเนื่องและแนวโน้มผิดปกติทั้งในแง่
จ ำนวนครั้งที่เกิดขึ้น ทิศทำง
ระยะเวลำและควำมรุนแรงของปรำกฎกำรณ์เอนโซ่” กำรตอบค ำถำมนี้
19

นับว่ำเป็นขั้นตอนเบื้องที่ส ำคัญในกำรก ำหนดยุทธศำสตร์กำรตั้งรับและกำรปรั
บตัวเข้ำกับสภำพกำรเปลี่ยนแปลงที่อำจจะเกิดขึ้น
เพื่อลดควำมรุนแรงแต่แสวงหำผลประโยชน์สูงสุดจำกผลกระทบดังกล่ำว
เนื่องจำกหลักฐำนได้เสนอแนะว่ำ
กำรเปลี่ยนแปลงสภำพภูมิอำกำศทั้งในระดับภูมิภำคและระดับโลก
มีแนวโน้มเกิดขึ้นอย่ำงต่อเนื่องและอำจจะทวีควำมรุนแรงภำยในระยะเวลำ
50-100 ปีข้ำงหน้ำ ดังนั้นข้อมูลทำงวิทยำศำสตร์ดังกล่ำวข้ำงต้น
นับว่ำมีควำมส ำคัญอย่ำงยิ่งยวดในกำรลดควำมไม่แน่นอนในกำรประเมินผลก
ระทบ
เพื่อให้ผู้บริหำรระดับนโยบำยมีควำมมั่นใจในกำรตอบสนองต่อผลกระทบ
ที่อำจจะเกิดขึ้นจำกกำรเปลี่ยนแปลงของสภำพภูมิอำกำศโลก
จะเห็นได้ว่ำ กำรเปลี่ยนแปลงของสภำพภูมิอำกำศของโลก
เป็นองค์ประกอบที่ส ำคัญและวิกฤติ ของระบบ “Integrated Earth System”
ซึ่งเป็นปัจจัยคุกคำมในปัจจุบันและอนำคตต่อระบบสภำพแวดล้อมและนิเวศน์วิ
ทยำในหลำยพื้นที่รวมทั้งประเทศไทยด้วย
กำรเตรียมพร้อมที่จะเผชิญกับสภำพภูมิอำกำศเปลี่ยนแปลงนับว่ำเป็นประเด็น
ที่ส ำคัญซึ่ง หำกไม่รีบกระท ำตอนนี้
เมื่อกำรเปลี่ยนแปลงมำถึงเรำอำจไม่สำมำรถปรับตัวเข้ำกับกำรเปลี่ยนแปลงได้
หรืออำจส่งผลเสียหำยมำกกว่ำที่ควรจะเป็น ดังนั้น
กำรป้องกันน่ำจะดีกว่ำกำรแก้ไข (Better to be safe than sorry)
แน่นอนปัญหำของสภำพภูมิอำกำศเปลี่ยนแปลงไม่ใช่เรื่องปัจจุบันทันด่วนที่รัฐ
บำลจะต้องแก้ไขทันที แต่จะเพิกเฉยโดยไม่ให้ควำมส ำคัญไม่ได้
ควรมีกำรก ำหนดนโยบำยในกำรศึกษำด้ำนนี้อย่ำงชัดเจน
และจัดสรรงบประมำณแทรกซึมไว้ในกระทรวงหรือหน่วยงำนที่เกี่ยวข้อง
เพื่อสำมำรถด ำเนินกำรศึกษำด้ำนนี้ได้
กำรเชื่อมโยงข้อมูลของแต่ละหน่วยงำนเป็นสิ่งจ ำเป็นอย่ำงยิ่งในกำรศึกษำวิจั
ยเพื่อสำมำรถมองภำพรวมได้ถูกต้อง ในกำรสร้ำงนโยบำยระดับประเทศนั้น
ต้องมองให้เห็นภำพรวมทั้งในและต่ำงประเทศให้ชัดเจน
เพื่อแจกแจงแยกแยะจัดล ำดับประเด็นที่ส ำคัญและจ ำเป็นต่อประเทศชำติเป็น
ล ำดับแรก นอกจำกนี้ ควรศึกษำแบบครบวงจร
เนื่องจำกผลกระทบไม่ได้เกิดขึ้นแต่เพียงภำคใดภำคหนึ่งเท่ำนั้น จะเห็นได้ว่ำ
กำรเปลี่ยนแปลงสภำพภูมิอำกำศนั้น เกี่ยวข้องกับหลำยหน่วยงำน
ศำสตร์หลำยด้ำน งำนวิจัยหลำยสำขำ ดังนั้น แนวทำงในกำรด ำเนินงำำน
เพื่อกำรเตรียมควำมพร้อม อำจแบ่งได้เป็นประเด็นด้ำนโครงสร้ำงพื้นฐำน
20

ส ำหรับหน่วยงำนและบุคลำกรเพื่อรองรับกำรท ำงำนและกำรศึกษำด้ำนสภำพ
ภูมิอำกำศเปลี่ยนแปลง
ประเด็นด้ำนนโยบำยซึ่งรัฐบำลต้องให้ควำมส ำคัญและสนับสนุนทั้งด้ำนกำรวำ
งแผนระยะยำวและด้ำนงบประมำณ เพรำะหำกไม่มีงบประมำณสนับสนุน
กำรศึกษำวิจัย อำจไม่เป็นไปตำมควำมคำดหมำย
และประเด็นกำรศึกษำวิจัยซึ่งควรพิจำรณำให้ควำมส ำคัญเฉพำะส่วนที่ส่งกระ
ทบโดยตรงกับประเทศไทยเท่ำนั้น งำนวิจัยที่ควรส่งเสริมได้แก่
กำรศึกษำแบบจ ำลองท ำนำยกำรเปลี่ยนแปลงสภำพภูมิอำกำศในประเทศไทย
กำรศึกษำผลกระทบ กำรปรับตัวและพื้นที่ล่อแหลมต่อกำรเปลี่ยนแปลง
21

1. Introduction
Variability and change are realities of the climate system, and static, so-called equilibrium, conditions are unlikely to be a part of the system on almost any time scale.
The palaeo-records clearly show that the global climate has varied continuously on all time scales, with global mean condition masked by immense variations in regional responses (Kasting, 1993; Petit et al., 1999; IGBP, 2001a,b; IPCC, 2001a). Fluctuations of surface temperature are the most obvious and probably well-documented key indicator of global climate change ( e.g., Hurrell, 1995,1996; Easterling et al., 1997; Enfield and
Mestas-Nuñez, 1999; Mann et al., 1999; Easterling et al., 2000b; IPCC, 2001a;
Trenberth, 2001). Surface temperature plays a crucial role in regulating evaporation and transpiration processes and so have direct connections to both the hydrological cycle and surface energy budget. Because temperature significantly affects biological processes and metabolic rates at almost every trophic levels (Hughes, 2000; McCarty, 2001;
Ottersen et al., 2001; Walther et al., 2002), ecosystem functioning and dynamics, as well as human health and comfort are all inevitably influenced by changes in both magnitude and rate of surface temperature through a variety of mechanisms. The additional stress of surface temperature changes will interact in different ways across regions that may reduce the ability of some environmental systems to provide, on sustained basis, key goods and services needed for successful economic and social development. However, there are many uncertainties in determining their impacts and predicting probable climate scenarios for the future, due to our incomplete understanding of interlinks of global climate system, forcings, responses and consequences (IPCC, 2001a). Studies of global and regional surface temperature variations and their impacts have, therefore, undergone a quantum jump and are one of the fundamental aims of global change research (IGBP, 2001a,b; IPCC, 2001a).
There is now growing evidence that human activities have increasingly influenced the global climate through the enhanced greenhouse effect, by past and continuing emission of carbon dioxide (CO
B
2
B
) and other gases which will cause the temperature of the Earth’s surface to increase –popularly termed the “global warming”(IGBP, 2001a,b; IPCC, 2001a; Trenberth, 2001). For a thousand years prior to the industrial revolution, abundance of the greenhouse gases was relatively constant.
However, as the world’s population increased, as the world became more industrialized and as agriculture developed, abundance of the greenhouse gases increased markedly.
The amount of CO
B
2
B
in the atmosphere has increased by about 31 percent since 1750
(IPCC, 2001a). The modern instrumental records indicate that surface temperature changed in a similar sense to atmospheric CO
B
2
B
concentrations, with a global mean warming of 0.6

0.2

C over the past 100 years and the 1990s being the warmest decade
22

on record (IPCC, 2001a). Synthesis of information from tree rings, corals ice cores, and historical data further indicates that the 1990s were the warmest decade in at least the past 1,000 years (IGBP, 2001a,b; IPCC, 2001a). In the light of new and stronger evidence and taking into account the remaining uncertainties, the Intergovernmental
Panel on Climate Change (IPCC) concluded in 2001 that most of the warming observed over last 50 years is attributable to the increase in atmospheric greenhouse gases due to human activities, and that global warming was indeed happening faster, and the consequences looked more severe than predicted.
On interannual and decadal timescales, there is also good evidence that fluctuations of regional surface temperatures are somewhat closely linked to changes in large-scale atmospheric and ocean circulations, as well as deep ocean heat content ( e.g.,
Loon and Rogers, 1978; Rogers, 1984; Li, 1990; Leathers et al., 1991; Yasunari and Seki,
1992; Trenberth and Hurrell, 1994; Hurrell, 1995,1996; Mantua et al., 1997; Zhang et al.,
1997; Qian and Zhu, 2001). Persistent large-scale atmospheric patterns tend to be wavelike so that regional changes of atmospheric heating, if powerful and persistent enough, can give rise to a sequence of remote atmospheric teleconnections (Horel and
Wallace, 1981; Wallace and Gutzler, 1981; Trenberth, 1990; Zahn, 2003). Thus a number of well-separated areas of anomalous temperature with opposite character may be produced. The strongest teleconnection pattern which has well documented within the earth’s climate on seasonal to decadal timescales is the set of processes known as the El
Niño-Southern Oscillation (ENSO). This phenomenon is the strongest natural mode and involves a set of complex interactions between the tropical oceans and the atmosphere centered on the Pacific and Indian Ocean basins with the life-cycle typically lasting from
2-7 years ( e.g., Horel and Wallace, 1981; Philander, 1990; McPhaden, 1999). The ENSO is now known to be at the root of many of the disastrous interannual climate fluctuations affecting tropical and subtropical countries (Rasmusson and Wallace, 1983; Hawana et al., 1989b; Philander, 1990; Li, 1990; Wang and Li, 1990; Janicot et al., 1996; Ware and
Thomson, 2000; Barlow et al., 2002; Hoerling and Kumar, 2003; Huber and Caballero,
2003). Moreover, warming over the large continental areas and cooling over the North
Pacific and North Atlantic in the winter during the past three decades is another example of more complex consequences of interconnected climate networks and interplay of different climate modes (IPCC, 2001a). This cold ocean-warm land pattern has been linked to changes in the atmospheric circulation over the northern hemisphere, in particular, to the tendency in the past few decades for the North Atlantic Oscillation
(NAO) to be in its positive phase (Hurrell, 1995; Hurrell and Loon, 1997). Similarly, the
Pacific-North American (PNA) teleconnection pattern has been in a positive phase in association with the tendency for favoring more the warm El Niño phase of ENSO phenomenon following the 1976/77 climatic regime shift (Nitta and Yamada, 1989;
Trenberth, 1990; Hurrell, 1996; Zhang et al., 1997).
Although global temperature has increased in the past century, its pattern was not spatial uniform or temporal monotonic, with large regional differences (Chapman and
Walsh, 1993; Schlesinger and Ramankutty, 1994). For example, the winter temperature in northern Europe has increased during the past 30 years, whereas northeastern America and Greenland have experienced increasingly colder winters in the same period (Hurrell and Loon, 1997). Much of this variation in regional winter climate conditions in the northern hemisphere can be attributed to variations in the natural climate pattern over the
North Atlantic or NAO (Hurrell and Loon, 1997). The climate of a given region is
23

determined by the interaction of forcings and circulations that occur at the planetary, regional and local spatial scales, and at a wide range of temporal scales (IGBP, 2001a,b).
Planetary scale forcings regulate the general circulation of the global atmosphere. This in turn determines the sequence and characteristics of weather events and weather regimes that characterize the climate of a region. Embedded within the planetary scale circulation regimes, regional and local forcings and mesoscale circulations modulate the spatial and temporal structure of the regional climate signal, with an effect that can in turn influence planetary scale circulation features. Because of their complex interaction, there is increasing need to better understand the processes that determine regional climate, along with the teleconnection effects of regional forcing anomalies (IPCC, 2001a).
The most highly developed tool which has currently used to predict future climate is known as coupled general circulation models (GCMs). These models are based upon sound, well-established physical principles and use descriptions in simplified physical terms of atmosphere, ocean and land processes. The predictive powers of a model can be tested by running the model with known forcing from the past through it and then comparing the results to actual climate records. Although models are exceedingly useful tools for carrying out numerical climate experiments, they do have limitations and must be used carefully (Trenberth, 2001). The latest models have been able to reproduces the major large-scale features of atmosphere, ocean and land processes in the past century or so with increasing accuracy (IPCC, 2001a). However, on regional scales (2000 km or less), there are significant errors in all models (Mearns et al.,
1995; IPCC, 2001a). This is mainly due to the complexity and scale of the physics involved and difficulties in relating the area-mean GCM output to the point or station scale (Osborn, 1997; Osborn and Hulme, 1997; Boyle, 1998). Moreover, our climate models so far are of relatively coarse resolution, and simplified versions of the real world
(IPCC, 2001a; Trenberth, 2001). Given the unproven reliability of GCMs at small scales especially in simulating surface temperature, it is desirable to search for signals of surface temperature changes in the observational records.
A surface temperature signal or any other climatic variables at any fixed location/region will typically consists of a complex mixture of variation, resulting from interactions among physical processes within the atmosphere-ocean-cryosphere system that operate on a wide range of spatial and temporal scales. Interactions within the components of the climate system usually include positive and negative feedbacks.
When these feedbacks combine properly and balance each other, they can give rise to irregular but can be separated and identified as trends, periodic and random oscillations
(Jassby and Powell, 1990; Ware and Thomson, 2000). The motivation for exploratory methods of data analysis in climate comes from the need to separate the climate “signal” from the background climate variability or “noise”. This decomposition of the data is done with the hope of identifying the physical processes responsible for the generation of the signal (Emery and Thomson, 1997). A fundamental characteristic of the statistical methods for signal detection is their ability to represent spatially distributed data in a compressed way such that the physical processes behind the data, or their effects, can best be visualized (Venegas, 2001)
.
As summarized by Emery and Thomson (1997) and
Venegas (2001), signal detection in climate is useful to achieve four main goals in climate research:
24

1. to recognize the patterns of natural climate variability and distinguish them from presumed anthropogenic or other external effects,
2. to use the physical mechanisms inferred from the detection signals to construct numerical climate models,
3. to validate numerical climate models by comparing the fundamental characteristics of the modeled data with those of the observed data, and
4. to use the signals themselves to forecast the behavior of the system in the future.
The complicated behavior and the non-linear character of the climate system provide a real challenge to the exploratory data analysis methods (IPCC, 2001a). Climate variations on different time scales, for example, may be connected with one another by nonlinear mechanisms. Some episodic phenomena, such as the periodic seasonal changes in surface temperatures, are better suited to be analyzed in the frequency domain. For certain phenomena it is not clear whether an oscillatory or episodic picture is most appropriate. Also, a number of signals, such as ENSO, exhibit a mixture of time-domain or “event” characteristics and frequency-domain or “oscillatory” characteristics (Emery and Thomson, 1997). Such quasi-oscillatory signals are characterized by a dominant timescale of variation, and are often combined with frequency modulation and episodic large-amplitude events. The choice of the appropriate analysis method is of extreme importance when the objective is to search for specific signals in time, space, or time and space combined, within large multivariate data sets (Venegas, 2001).
It is usual in climate studies to be presented with a large data set consisting of time series over a grid of stations which we wish to compress into a smaller number of independent pieces of information. Typically it is necessary to deal with an ensemble of instantaneous samples (maps) of geophysical fields (for example, surface temperature) defined at a number of points (stations). In such cases, the data are in the form of simultaneous time series records from a grid on a horizontal plane: x
B i
B
(t) , y
B i
B
(t) . The grid points may be regularly spaced (such as model-generated data or grid observation) or irregularly spaced (such as locations of meteorological stations). Analyses of data sets with the described characteristics, that is, consisting of a number of spatially distributed time series are known as multivariate statistical procedure. The method of Empirical
Orthogonal Functions (EOFs) is a particularly useful technique for compressing the variability in this type of data sets and is most widely applied to the problem of spatiotemporal signal detection in climatic data sets (Lagerloef and Bernstein, 1988;
Preisendorfer, 1988; Emery and Thomson, 1997). This method is also known as
Principal Component Analysis (PCA). The EOF procedure is equivalent to a data reduction method widely used in the social sciences known as factor analysis. An advantage of EOF analysis is that it provides a compact description of the spatial and temporal variability of data series in terms of orthogonal functions, or statistical modes.
In this study, the EOF analysis as well as other multivariate statistical techniques were applied. The primary objective is to identify the dominant spatio-temporal patterns of surface air temperature in Thailand, which the time evolution of their leading modes can further be investigated :
1. interannual and multi-decadal variability as well as long-term trends,
2.
its relation to the ENSO and anthropogenic-induced climate changes and the possible linking mechanisms, and
25

3.
its possible biophysical and socio-economic impacts.
The paper is organized as follows. An analytical method and data sources are outlined in the next section. Also reviewed in this section will be the basic concepts of
EOF analysis and EOF computation using the scatter matrix method. Physical interpretation of EOF analysis and temporal structures of the EOF1 coefficient series and their relations to large-scale climate signals are presented in section 3. The final section goes on discussing advantage/disadvantage of analytical technique, possible causes/effects of surface air temperature changes, and implication for future research.
2. Analytical methods and data sources
2.1. Basic concepts of EOF analysis
EOFs as used by meteorologists and oceanographers are a statistical technique for analysis of the spatial or temporal variability of physical fields. For example, a situation benefiting from such analysis occurs when a succession of snapshots of the surface temperature field over any given region of the globe is made at monthly times during ten years or longer. When these snapshots are viewed in rapid succession, it becomes apparent to the eye where the areas of great variability of surface temperature are. In order to succinctly represent and think about such complex variations, scientists in both meteorology and oceanography have learned over recent decades to use and develop the concept of EOF analysis, a tool arising originally in biology and psychometey, to resolve the complex variance patterns of physical fields. Thus, EOF analysis is simply a method for portioning the variance of a spatially distributed group of concurrent time series. Its goal is to replace the spatial and temporal variability of original data series by a smaller number of new variables, linear combination of the original variables, that capture most of the total original variance but are uncorrected with each other (Davis, 1976; Lagerloef and Bernstein, 1988; Preisendorfer, 1988; Dunteman, 1989; Jassby and Powell, 1990;
Emery and Thomson, 1997). The new variables are called orthogonal functions and are arranged in descending order according to the amount of the original variance they
26

reproduce. Usually, most of the variance of a spatially distributed series is in the first few orthogonal functions whose patterns may then be linked to possible dynamical mechanisms. The theory behind EOF computation is straightforward (see, for example,
Preisendorfer, 1988, Dunteman, 1989; Emery and Thomson, 1997 for a particularly compact and lucid description). There are two approaches for computing EOFs for a number of time series. The first constructs the covariance matrix of the data series and then decomposes it into eigenvalues and eigenvectors. The second uses the Singular
Value Decomposition (SVD) of the data matrix to obtain all the components of the EOFs
(eigenvalues, eigenvectors, and time-dependent amplitudes) without computation of the covariance matrix. The EOFs determined by the two methods are identical. The differences are mainly the greater degree of sophistication, computational speed, and computational stability of the SVD approach. Only the EOF computation using the scatter matrix will be described below. Details of the SVD method can be found in
Preisendorfer (1988) and Emery and Thomson (1997). Note that the readers who are unfamiliar with matrix algebra and eigenvalue-eigenvector problems should review their basic concepts which can be found in many basic mathematical textbooks.
The algebratic essentials of EOF analysis can be described as follows. Let z(t,x) be surface temperature or other climatic variables at point x in any given area of the globe at time t . Suppose this measurement be taken over the set of locations x =
1,…, p at times t
=
1,…, n
. Thus the snapshots referred to above are collections { z(t,x) : x =
1,…, p } of reading z(t,x) taken at each of the n times t , and are centered on their time averages. It can be thought these collection as p x 1 ( i.e., column) vectors
U z
U
(t) = [ z(t,1),…, z(t,p)
]
P
T
P forming a swarm of points about the origin of a p -dimensional euclidian space E
B p
B
. The symbol “T” denotes the transpose operation. These collections can also be placed into an n x p matrix:
Location

Z z
B
1
B
(1,1)
.
. . .
. . . z
B
1
B
(1,p)
.

= z
B n
B
(n,1) . . . z
B n
B
(n,p)
.
(1)
.
The first step in the EOF analysis of Z is to center the values z(t,x) on their averages over the t index. Thus, for each x = 1,…, p
, t -centered values can be written: z ( x ) 
1 n t n

 1 z ( t , x ) and form the anomalies or departure from the record mean z  ( t , x )  z ( t , x )  z ( x ).
(2)
(3)
This procedure ensures that analysis is not dominated by the variance from any given locations (all locations are given a relatively uniform distribution of variance over the
27

different spatial locations). Using these t -centered values z

(t,x) , a new n x p matrix Z
 can be formed in the manner of (1) :
Location
 z

B
1
B
(1,1)
Z

.
= z

B n
B
(n,1)
. . .
. . .
. . . z z


B
1
B
B n
B
(1,p)
.
(n,p)

.
Time
(4)
B
Z

Z

B of the field z

(t,x)
P
T
P
)
P
R
B
Z

Z

B
(5)
Expending the product of matrices :
= Z

P
T
P

Z

.
 z

B
1
B z

B
1
B
  z

B
1
B z

B
2
B

. . .
 z

B
1
B z

B n
B

. . . . . . . . . . . .
(6)
R
B
Z

Z

B
=
 z

B n
B z

B
1
B
  z

B n
B z

B
2
B

. . .
 z
 where
 z

B i
B z

B j
B
 is the covariance between time series z

B n
B z

B n
B

B i
B
and z

B j
B
( z

B B
at locations i and j ) defined as : z  i z  j
 z  j z i
 
1 n t n

 1 z i
 ( t ) z  j
( t ),
(7) where i, j = 1,…, p
. The matrix product R
B
Z

Z

B is symmetric and square, even if Z

itself is not square. A dimension of R
B
Z

Z

B
is p x p . It should be noted that some authors define the data matrix Z

as the transpose of that defined in equation (4), that is, with n columns corresponding to time steps and p rows corresponding to locations. In such case, the determination of the spatial covariance matrix should be done as
R
B
Z

Z
B
 = Z
 
Z

P
T
P
. (8)
The rest of the procedure, however, is identical to what is described here.
Once the covariance matrix has been calculated from the data, the EOF analysis can be done by solving eigenvalue-eigenvector problems which R
B
Z

Z
B
 is decomposed into matrices L and E :
R
B
Z

Z

B

E = E

L or
( R
B
Z

Z

B
- L )

E = 0. (9)
28

L is the p x p diagonal matrix containing eigenvalues

B k
B
( k = 1, …, p ) of R
B
Z

Z

,
B and the off-diagonal elements of R
B
Z

Z

B are all zero:
L =

B
1
B
. . .

0 . . . 0
B
2
B
. . . . . . (10)
0 0 . . .

B p
B
.
The square matrix E also has dimension p x p . Its column vectors e
B k
B
(k = 1,…, p) are the eigenvectors of R
B
Z

Z

B corresponding to eigenvalues

B k
B
: e
B
1
B
(1) e
B
2
B
(1) . . e
B p
B
(1)
E =
. . . . . .
.
. . . e
B
1
B
(p) e
B
2
B
(p)
. e
B p
B
(p)
.
.
 e
1
 e
2
 e p

Eigenvector e
B k
PB
(11)
The equation in (9) governs the required direction e
B k
B of extremal scatter. Non-trivial solutions ( i.e., e
B k
B

0 ) of this set of p linear algebraic equations for the components of e
B k
B
= [ e
B k
B
(1), . . ., e
B k
B
(p) ]
P
T
P
occur only for special values of

B k
B
. In theory of linear algebra, it is shown (Wilkinson, 1965; Franklin, 1968) that a symmetric matrix such as R
B
Z

Z

B
in (6) generally has p eigenvectors e
B k
B
= [ e
B k
B
(1),…, e negative eigenvalues

B k
B such that
B k
B
(p) ]
P
T
P in E
B p
B and p associated real, non-
R
B
Z

Z

B
 e
B k
B
=

B k
B

e
B k
B
, k = 1 … p.
(12)
The eigenvalue-eigenvector problems in (9) correspond to the series of linear system equations:
[
 z

B
1
B z

B
1
B

-

B
1
B
]e
B
1
B
+
 z

B
1
B z

B
2
B
 e
B
2
B
+ ,…,
 z

B
1
B z

B p
B
 e
B p
B
=0
 z

B
2
B z

B
1
B

e
B
1
B
+ [
 z

B
2
B z

B
2
B

-

B
2
B
] e
B
2
B
+ ,…,
 z

B
2
B z

B p
B
 e
B p
B
=0
(13)
… … …
 z

B n
B z

B
1
B

e
B
1
B
+
 z

B n
B z

B
2
B

e
B
2
B
+ ,…,[
 z

B n
B z

B p
B

-

B p
B
]e
B p
B
=0 .
Since the data matrix Z

is real, the covariance matrix R
B
Z

Z
B
 is positive definite, which means that all its eigenvalues are greater or equal to zero. Each non-zero eigenvalue

B k
B in matrix L is associated with a column eigenvector e
B k
B in matrix E . The eigenvector matrix E has the property that
E

E
P
T
P
= E
P
T
P

E = I
B p
B
, (14)
29

P

E = I
B p
B
simply indicates that the cross products of any two eigenvectors are 0 and the sum of squares of the elements for a given eigenvector are equal to 1. This means that eigenvectors are uncorrected over space, that is, they are orthogonal to one another. Each eigenvector E
B k
B represents the spatial EOF pattern of mode k (it has dimension p , that is, the number of locations in the original data).
From matrix e
B k
B
, time-dependent amplitudes, a
B k
B
(t) , of the data set can be derived by projecting the original data series z

(t,x) onto eigenvector e
B k
B
and summing over all locations p : a k
( t )  x p

 1 z  ( t , x ) e k
( x ) ,
(15) where x = 1
,…, p counts the location, t = 1
,…, n counts the time steps and k = 1
,…,
p counts the EOF modes. These a
B k
B
(t) , thought of as time series { a
B k
B
(t) : t = 1,…, n
}, have the important property of temporal uncorrelatedness, and they carry information about the variance of the data set along the direction e
B k
B
. In matrix notation, matrix A which has dimension n x p is obtained by multiplying matrices Z

and E:
A = Z
 
E. (16)
Just as the spatial patterns E
B k
B
are orthogonal in space, the a
B k
B
(t) are orthogonal in time.
This means that the time-averaged covariance of the amplitudes satisfies a i
( t ) a j
( t ) i, j = 1,…, p,
  i
 ij where

B ij
B is the Kronecker delta :
(uncorrected time variability), (17)
 ij
 {
0
1
,
, j j  i
 .i
The overbar in (17) denotes the time-averaged value and
B
(18)
 i
 a i
( t ) 2 
1 n j n

 1
[ a i
( t j
) 2 ]
(19) is the variance in each EOF mode. The matrix version of this is
A
P
T
P

A = L.
(20)
The eigenvalues in L are usually sorted in decreasing order according to its corresponding eigenvector, so that

B
1
B
>

B
2
B
> …

B p
B
. Each eigenvalues

B k
B
is proportional to the percentage of the variance of the original data that is accounted for by mode k .
This percentage is calculated as :
% variance mode k 

 k i p
 1
 i
* 100
(21)
30

The first mode contains the highest percentage of total variance,

B
1
B
; of the remaining variance, the greatest percentage is in the second mode,

B
2
B
, and so on.
Since EOF modes and their time-dependent amplitudes are uncorrected over time and space, each one makes an independent contribution to accounting for the variance of the original data set.
If we add up the total variance in all the time series, we get x p

 1
1 n t n

 1
[ z x
( t )]
2
 k p

 1
 k
(22)
Sum of variances in original data = sum of variances in eigenvalues.
Finally and most important, the original centered data set can be totally represented in the form: z  ( t , x )  j p

 1 a j
( t ) e j
( x )
t = 1,…,n ; x = 1,…, p.
In matrix notation :
Z

(24)
P
=A
(23)

E
P
T
P
.
As noted above, if the eigenvalues are ordered by size (that is, by fraction of the variance explained by the corresponding eigenvector), it is usually found that only the first few empirical modes account for a very fraction of the variance. The reconstruction of an approximate, compressed and less noisy version
U z

U
(t,x) of the original z

(t,x) , using only the first few modes ( k ) with k << p , can be represented meaningful physical processes, which are associated with fundamental characteristic spatial and temporal variability in a very large data set. This leads to a significant reduction of the amount of
U data while retaining most of the variance of variables. In addition, the synthetic version z

U
(t,x) can produce a lower total mean-square error, because sum of variances in eigenvalues (right term in (22))is close to sum of variances in original data (left term in the (22)).
2.2. Data sources
A set of surface weather observations for a 53-year period (1951-2003) collected at
34 stations in Thailand (Table 1) forms the basis for the EOF analysis. The data set obtained from the Meteorological Department of Thailand consists of monthly averaged mean, maximum, minimum temperatures ( T
B mean
B
, T
B max
B
, T
B min
B
), and monthly averaged mean dewpoint temperature ( T
B dew
B
) which are all derived from daily observations. The 34 site records used here were chosen on the basis of record length and completeness, the requirement that there were no significant effects from station relocation during the period, and to provide a reasonable spatial coverage over much of Thailand. Monthly averaged mean, maximum and minimum apparent temperatures ( T
B amean
B
, T
B amax
B
,
B B
T
B amin
B
),
31

which combine temperature and humidity effects on the human body, were further calculated by using Steadman’s (1984) regression equation
T
B x
B
= -1.3+ 0.92*t +2.2*e, (25) where T
B x
B is T
B amean
B
, T
B amax
B
,
B B or T
B amin
B
, t is T
B mean
B
, T
B max
B or T
B min
B and e is water-vapor pressure
(kilopascals). The effects of wind and radiation are ignored in this equation, and e were calculated from T
B dew
B as: e  0 .
6108 exp



T
17 .
dew
27

T dew
237 .
3



.
(26)
32

Table 1. Listing of weather stations used in this analysis. Asterisks indicate stations which data are available from 1952 to 2003.
Station number
1
2
3
4
5
6
7
8
21
22
23
24
25
17
18
19
20
13
14
15
16
9
10
11
12
Station Code
303201
327501
330201
331201
351201
376202
378201
379201
354201
356201
357201
381201
383201
405201
407501
431201
432201
400201
425201
426201
450201
455201
455601
440201
459204
Province
CHIANG RAI
CHIANG MAI
PHARE*
NAN
UTTARADIT
MAE SOT
PHITSANULOK
PHETCHABUN*
UDON THANI
SAKON NAKHON*
NAKHON PRANOM*
KHON KAEN
MUKDAHAN
ROI ET
UBON RATCHATHANI
NAKON RATCHASIMA
SURIN
NAKON SAWAN
SUPHAN BURI*
LOP BURI
KANCHANA BURI
BANGKOK METROPOLIS
DON MUANG AIRPORT
ARANYA PRATHET
SATTATHIP
Location (latitude; N, longitude; E)
19

55

, 99

50

18

47

, 98

59

18

10

, 100

10

18

47

, 100

47

17

37

, 100

06

16

40

, 98

33

16

47

, 100

16

16

26

, 101

09

17

23

, 102

48

17

09

, 104

08

17

25

, 104

47

16

26

, 102

50

16

32

, 104

45

16

03

, 103

41

15

15

, 104

52

14

58

, 102

05

14

53

, 103

30

15

48

, 100

10

14

28

, 100

08

14

48

, 100

37

14

01

, 99

32

13

44

, 100

34

13

55

, 100

36

13

42

, 102

35

12

41

, 101

01

24

Table 1. (continued)
Station number
26
31
32
33
34
27
28
29
30
Station Code Province
480201
500201
517201
552201
532201
564201
567201
568501
583201
CHANTHA BURI
PRACHUAP KHIRIKHAN
CHUMPHON
NAKONSI THAMMARAT
RANONG
PHUKET
TRANG AIRPORT
SONGKHLA
NARATHIWAT
Location (latitude; N, longitude; E)
12

37

, 102

07

11

50

, 99

50

10

29

, 99

11

8

28

, 99

58

9

59

, 98

37

7

58

, 98

16

7

31

, 99

32

7

12

, 100

36

6

25

, 100

49

25

A relatively stable network of stations and consistency in operational procedures ensure that the quality of surface weather observations in Thailand, for climate studies, is somewhat good (Ouprasitwong, 2002; Brikshavana and Ouprasitwong, 2002). However, the obtained data set was subject to further quality control procedures. Two types of erroneous data were identified. The first was an abrupt shift in mean values, associated with station relocations. According to station history information, there were five stations (CHIANG RAI, KHON KAEN, BANGKOK METROPOLIS, SURAT THANI and NAKHONSI THAMMARAT) that locations were changed during 1990-2000
(Ouprasitwong, 2002). Ouprasitwong (2002) used Multiple Analysis of Series for
Homogenization (MASH) to examine homogeneity of rainfall data, and found that inhomegeneity of rainfall data at stations BANGKOK METROPOLIS and NAKHONSI
THAMMARAT coincided with the years of station relocations. The MASH program can, however, provide the reliable results only for rainfall data (Ouprasitwong, 2002). Thus, the non-parametric Mann-Whitney statistic ( U
B k
B
) for testing that two samples (x
B
1
B
, x
B
2
B
, …, x
B p
B
) and (x
B p+1
B
, x
B p+2
B
, …, x
B p+n
B
) come from the same population is alternatively suitable for examining the occurrence of an abrupt change in other climatic variables (Petitt, 1979;
Demaree and Nicolis, 1990). An analysis can be done by partitioning the series into two sub-period, before and after site moves, and calculated U
B k
B from: where M
B i
B
is the rank of the i th observation when the values x
B
1
B
, x
B
2
B
, …, x
B
N
B
in the series are arranged in ascending order. The results of two-sided Mann-Whitney test indicate that
U k
 2 i k

 1
M i
 k ( N  1 )
(27)
(Table 2) only temperature value at station BANGKOK METROPOLIS was significantly different between before and after station relocation, and this data record was then excluded for the EOF analysis. Moreover, the remaining station records were visually inspected for any abrupt shift, but there was no evidence for such changes. The geographical distribution of 33 selected weather stations is shown in Fig. 1. The second type of erroneous data involved outlier data that may have been introduced either due to data-entry, data observing or transmitting procedure biases were identified and excluded according to statistical criteria. An objective approach eliminated apparent statistical outliers, which exceeded specified acceptable range, was arbitrarily set 3 standard deviation from monthly mean (mean

3SD) (Limsakul et al., 2001). Any existing errors, which could not be detected by the statistical methods were usually random and equivalent to “noise”.
From 1951 to 2003, each of the station records is, on average, 98% complete, and the overall dataset has only 1.7% missing values. There were missing data in some years and months particularly during 1951-1955 when more data are missing (Fig. 2). However, small amounts of random missing data should not introduce significant biases in temporal trends, since data used to the EOF analysis consist of many stations. To further prevent missing data from introducing any bias, monthly climatological means calculated from entire record were used for missing values.
2.3. EOF computation using the scatter matrix method
24

Details of the covariance matrix approach can be found in Preisendorfer (1988) and
Emery and Thomson (1997). This recipe, which is only one of several possible procedures that can be applied, involves the preparation of the data and the solution of equation (9) as follows:
1. Construct the n x p matrix Z in equation (1), by organizing the n rows (times) and p columns (locations) of the original data. In case of surface air temperature data used here that were collected at 33 stations and from 1951 through 2003 (53-year period), n
= 1, …, 636
(53 years x 12 months) and p
= 1, …, 33
. Care should be taken to ensure that the start and end times for all p time series of length n are identical.
2. Compute the climatological monthly means using equation (2) and subtract them from original data, z(t,x ), in the equation (3). The new n x p matrix Z

can be formed in the manner of equation (1), but consists of the anomalies or departure from climatological monthly means (4). Note that the anomalies of all missing values are equal to 0, since climatological monthly means are used for those missing data.
3. Construct the spatial covariance matrix R
B zz

B by using equations (5), (6) and (7).
4. Solve eigenvalue-eigenvector problems which R
B zz

B is decomposed into eigenvalues and eigenvectors in the equations (9), (10) and (11).
5. Compute time-dependent amplitude, a
B k
B
(t), by projecting the original data series onto eigenvector in equation (15).
6. Calculate the percentages of the variance explained by each mode, using equation
(21).
25

Table 2. Results of the two-sided Mann-Whitney test for monthly averaged mean temperature before and after site moves for 4 stations.
Station Period before/after site move
CHIANG RAI
KHON KAEN
BANGKOK METROPOLIS
NAKHONSI THAMMARAT
1951-1991
1992-2003
1951-1997
1998-2003
1951-1993
1993-2003
1951-1997
1998-2003
N – number of observations; Uk – Mann-Whitney statistic
N
492
143
564
72
515
120
501
72
Median
(

C)
25.6
25.6
27.4
27.2
27.9
29.0
27.4
27.4
Uk
-0.10
-0.95
-7.63
-0.73 p -value
0.92
0.34
<0.001
0.47
26

20
18
16
14
12
CHIANG RAI
CHIANG MAI NAN
PHARE
UTTARADIT UDON THANI NAKHON PRANOM
SAKON NAKHON
PHITSANULOK
MAE SOT
PHETCHABUN KHON KAEN
MUKDAHAN
ROIET
NAKHON SAWAN
NAKHON RATCHASIMA
LOP BURI
UBON RATCHATHANI
SURIN
SUPHAN BURI
KANCHANA BURI
BANGKOK ARANYA PRATHET
SATTATHIP
CHANTHA BURI
PRACHUAP KHIRIKHAN
CHUMPHON
RANONG
10
NAKHONSI THAMMARAT
8
PHUKET
TRANG
SONGKHLA
NARATHIWAT
6
98 99 100 101 102 103 104 105
Latitude
Fig. 1. Geographical distribution of 33 selected weather stations used in this study. The atmospheric variables used to analysis are monthly averaged mean, maximum and minimum temperatures and dewpoint temperature collected from 1951 to 2003.
27

100
80
60
100
T dew
80
60
100
T min
80
60
100
T max
80
T mean
60
1950 1960 1970 1980 1990 2000
Year
Fig. 2. Monthly percentage of data coverage for 33 stations used in this study.
28

3. Results
3.1. Physical interpretation of EOF analysis
An EOF analysis generates three types of output. First the eigenvalues together with the percentage of the total variance of each EOF mode are given. In the EOFs these equal the variance accounted by each mode. Also provided are the eigenvectors for each
EOF mode. Each eigenvector is composed of values called the component loadings for that mode. These loadings are usually presented as correlation coefficients between each time series and the associated time-dependent amplitudes, and may be considered a measure of the relative importance of each time series in the extracted EOF mode. The sum of the squared correlations for each EOF mode equals the associated eigenvalue. If, for example, surface air temperatures at stations 1, 2 and 3 have high positive loadings on the leading EOF mode and that at station 4 has a high negative loading, this means that the largest proportion of the variance in the original data can be accounted for by the trends in these four stations. The different signs indicate that surface air temperature at station 4 has high values in a certain set of time steps, whereas surface air temperatures at stations 1, 2 and 3 have high values in a completely different set of time steps. The third set of results is a matrix of time-dependent amplitudes or component scores. These series describe the evolution of the EOF’s with time. One set of timedependent amplitudes is provided for each of EOF mode, and each time-dependent amplitude corresponds to one time step. This is computed by simply multiplying the component loadings by the original data.
According to Peixoto and Oort (1992), one way to understand the basic idea behind EOFs is to imagine that each of the n time series as a vector f
B n
B in the p dimensional space, such that : f
B n
B
= {f
B n1
B
, f
B n2
B
, …, f
B np
B
} at time t = n .
Each vector f
B n
B
includes the values of field f at all location x = 1, …, p
for a given time n .
Each of n data vectors is directed from origin to a point in the p space (Fig. 3). If there exists some correlation between the n vectors, we expect that their extremities will be organized in cluster or along some preferred directions. The problem we want to solve with the EOF decomposition is to find an orthogonal basis { e
B
1
B
, e
B
2
B
, …, e
B p
B
} in the p dimensional space, instead of the original basis, such that vector e
B
1
B best represents the largest cluster of the original data vectors, e
B
2
B
best represents the second largest cluster of the original data vectors, and so on. In other words, e
B
1
B
accounts for the largest portion of the data variance, e
B
2
B
for the second largest portion, and so on (Fig. 3). This is equivalent to find a set of p vector e
B p
B
whose orientation is such that the sum of the squares of the projections of all the n observation vectors f
B n
B
onto each e
B p
B
is maximized sequentially. The vectors e
B x
B
, x = 1, …, p are mutually orthogonal and they are what we called the EOFs. If all possible EOF modes are used, then they define a space which has the same dimension as the original variable space and, hence, completely account for the variance in the original data. However, there is no advantage in retaining all of the EOF modes since there would have as many EOF modes as original variable and, thus, would not have simplified matters.
The first step is to decide on how many EOF modes are needed to adequately describe the dominant spatio-temporal characteristics of surface air temperatures, dewpoint temperature and apparent temperatures in Thailand. The scree plot proposed
29

Fig. 3. Example of a possible configuration of the data vectors f
B n
B
(n =1… N denote the time steps) and the empirical orthogonal vectors e
B m
B
, m = 1…M
. From Peixoto and Oort
(1992).
30

by Cattell (1966) is one of the good rules of thumb, providing a means of assessing how many EOF modes to retain. It is constructed by simply plotting either eigenvalues or percentage of the total variance of each EOF mode in descending order to produce a line graph. Since the leading EOF mode has the largest eigenvalue or the largest percentage of the total variance, and the following ones are in descending value, this produces a line graph that slopes down to the right. By looking for the point where a pronounced change in the slope occurs, how many EOF modes to retain can thus be decided. It should be noted that this criterion is somewhat arbitrary, and how many and which EOF modes to retain depend, in part, upon the goals of the analysis.
The scree plots for individual and cumulative percentages of the total variance of monthly averaged mean, maximum and minimum temperatures ( T
B mean
B
, T
B max
B
, T
B min
B
), monthly averaged mean dewpoint temperature ( T
B dew
B
) and monthly averaged mean, maximum and minimum apparent temperatures ( T
B amean
B
, T
B amax
B
,
B B
T
B amin
B
) are presented in
Figs. 4, 5 and 6. What stands out from Figs 4, 5 and 6 is that steep slopes are evident from the first to the second EOF modes and the remaining EOF modes can be fitted fairly well by a straight line of negligible slope. The EOF1 mode of all seven temperature variables accounts for substantial amount of the total variance ranging from
61.2% to 71.3%, whereas the remaining modes explain considerably less. These patterns of scree plots indicate that only the first mode is physically meaningful in determining the dominant mode of variability, and higher order modes are potentially mixed and non-interpretable due largely to climatic noise associated with high-frequency variability in the climate system. Consequently, by using Cattell’s scree criterion, only the EOF1 mode was retained to describe spatio-temporal characteristics of all temperature variables.
The loadings on the EOF1 mode of all seven temperature variables are graphically illustrated in Figs. 7-13. A visual examination reveals that the EOF1 mode of each temperature variable has positive correlations with all stations, and correlation coefficients are relatively high and are about the same magnitude, excepting for a few stations in the south. High (r >0.5) and low (r <0.5) loadings on the leading modes range from 75% to 90% and from 10% to 25%, respectively. These loading patterns strongly indicate that temperature data at all stations in Thailand are highly intercorrelated and nearly equally important in defining the EOF1 mode. Thus, it can be appropriately viewed that the EOF1 mode is a robust representative of the dominant spatio-temporal structures of T
B mean
B
, T
B max
B
, T
B min
B
, T
B dew
B
,
B B
T
B amean
B
, T
B amax
B and
B B
T
B amin
B
in Thailand.
Time evolution of the leading mode of all seven temperature variables is shown in the time series of their coefficients (Figs. 14, 15 and 16). Note that units are arbitrary, because of EOF calculation based on covariance matrix. As can be seen, all series exhibit irregular oscillations, due to a mixture of several signals of variability contained in the time series. As indicated by the integral timescales, all series form complex longterm patterns and are rather noisy, which month-to-month variations are prominent, superimposed on much lower-frequency variations with timescales of a few years or longer. Some underlying periodic oscillations and trends also seem to be present in the time series. A visual inspection further reveals that two series of T
B min
B
and T
B amin
B do appear to contain the dominant long-term trends. However, inferences about the dominant temporal characteristics of the leading modes from unsmoothed series are relatively obfuscating, due to the presence of various scales of motion in time series. In
31

order to gain insight on the particular signals blended together and hidden inside a noisy time
100
80
60
40
20
0
100
80
60
40
20
T mean
T max
T min
0
0 5 10 15 20 25 30
Mode number
Fig. 4. Scree plot for EOF analysis of monthly averaged mean, maximum, minimum temperatures ( T
B mean
B
, T
B max
B
, T
B min
B
) collected during 1951-2003 and at 33 stations in
Thailand.
100
80
60
40
20
100
80
60
40
20
0 0
0 5 10 15 20 25 30
Mode number
Fig. 5. Scree plot for EOF analysis of monthly averaged mean dewpoint temperature
( T
B dew
B
) collected during 1951-2003 and at 30 stations in Thailand.
32

100
80
60
40
20
0
100
80
60
40
20
T aa
T amax
T amin
0
0 5 10 15 20 25 30
Mode number
Fig. 6. Scree plot for EOF analysis of monthly averaged mean, maximum and minimum apparent temperatures ( T
B aa
B
, T
B amax
B
,
B B
T
B amin
B
) calculated by using surface air temperature and dewpoint temperature collected during 1951-2003 and at 29 stations in Thailand.
33

20
18
16
14
12
10 r = 1.0
r = 0.7
r = 0.4
r = 0.1
8
6
98 100 102 104
Latitude
Fig. 7. Loadings on the EOF1 mode of monthly averaged mean temperature. The loadings are correlation coefficients between each time series and the first time-
34

dependent amplitudes. The sizes of blue cycles are proportional to correlation coefficients.
20
18
16
14
12
10
8 r = 1.0
r = 0.7
r = 0.4
r = 0.1
6
98 100 102
Latitude
104
35

Fig. 8. Loadings on the EOF1 mode of monthly averaged maximum temperature. The loadings are correlation coefficients between each time series and the first timedependent amplitudes. The sizes of blue cycles are proportional to correlation coefficients.
20
18
16
14
12
10 r = 1.0
r = 0.7
r = 0.4
r = 0.1
8
6
98 100 102
Latitude
36
104

Fig. 9. Loadings on the EOF1 mode of monthly averaged minimum temperature. The loadings are correlation coefficients between each time series and the first timedependent amplitudes. The sizes of blue cycles are proportional to correlation coefficients.
20
18
16
14
12
10 r = 1.0
r = 0.7
r = 0.4
r = 0.1
8
6
98 100 102
Longitude
37
104

Fig. 10. Loadings on the EOF1 mode of monthly averaged dewpoint temperature. The loadings are correlation coefficients between each time series and the first timedependent amplitudes. The sizes of blue cycles are proportional to correlation coefficients.
20
18
16
14
12
10 r = 1.0
r = 0.7
r = 0.4
r = 0.1
8
6
98 99 100 101 102 103 104 105
Longitude
38

Fig. 11. Loadings on the EOF1 mode of monthly averaged mean apparent temperature.
The loadings are correlation coefficients between each time series and the first timedependent amplitudes. The sizes of blue cycles are proportional to correlation coefficients.
20
18
16
14
12
10 r = 1.0
r = 0.7
r = 0.4
r = 0.1
8
6
98 100 102
Longitude
39
104

Fig. 12. Loadings on the EOF1 mode of monthly averaged maximum apparent temperature. The loadings are correlation coefficients between each time series and the first time-dependent amplitudes. The sizes of blue cycles are proportional to correlation coefficients.
20
18
16
14
12
10 r = 1.0
r = 0.7
r = 0.4
r = 0.1
8
6
98 100 102
Longitude
40
104

Fig. 13. Loadings on the EOF1 mode of monthly averaged minimum apparent temperature. The loadings are correlation coefficients between each time series and the first time-dependent amplitudes. The sizes of blue cycles are proportional to correlation coefficients.
1
T min
0.5
0
-0.5
-1
1
0.5
0
-0.5
-1
1
0.5
0
-0.5
-1
1950
T max
T mean
1960 1970 1980
Year
1990 2000
41

Fig. 14. Coefficient time series of the EOF1 mode. Units are relative.
1
T amin
0.5
0
0
-0.5
-1
1
0.5
-0.5
-1
1
0.5
0
-0.5
-1
1950
Tamax
T amean
1960 1970 1980
Year
1990 2000
42

Fig. 15. Same as in Fig. 14.
1
T dew
0.5
0
-0.5
-1
1950 1960
Fig. 16. Same as in Fig. 14.
1970 1980
Year
43
1990 2000

series, undesired scales of variability must first be removed. The interesting signals can then be examined in isolation without the complications of the other features. The trend and other components can be separated by a smoothing technique (Chatfield, 1989;
Jassby and Powell, 1990; Emery and Thomson, 1997). Smoothing always involves some form of local averaging of data such that the nonsystematic components of individual observations cancel each other out. The most common method is moving average smoothing which replaces each element of the series by simple average of n adjacent element, where n is the width of smoothing window (Chatfield, 1989; Jassby and Powell,
1990; Emery and Thomson, 1997; Eskridge et al., 1997). By modifying the window size, the filtering of different scales of motion can be controlled.
To further investigate and compare the standing temporal evolution of the leading mode with the well-known modes of global climate variability, each unfiltered series was decomposed into interannual (1-5 years) and decadal (longer than 5 years) timescales. The two different timescales were chosen on the basis that the separated interannual variability corresponds to the ENSO cycles, while decadal fluctuations represent long-term behavior of variables. Decomposition of the coefficient time series can be done by first applying a centered 60-term moving average. The resulting filtered series represent long-term variations. Interannual variability was subsequently estimated by subtracting the smoothed from the original series, forming the residual series. A centered 10-term moving average was then employed to the residual series to eliminate the variability less than ten months.
3.2. Temporal variability of EOF1 coefficients series and its relation to ENSO signature
3.2.1. Fundamental mechanisms of ENSO and the commonly used indices
The term El Niño is widely used to refer to a phenomenon associated with the unusually warm water that occasionally forms across much of the tropical eastern and central Pacific (Fig. 17). The time between successive El Niño events is irregular but they typically tend to recur every 2 to 7 years ( e.g., Horel and Wallace, 1981; Philander,
1990; Trenberth, 1984, 1997; McPhaden, 1999). A La Niña is the counterpart to an El
Niño and is characterized by cooler than normal Sea Surface Temperatures (SSTs) across much of the equatorial eastern and central Pacific (Fig. 18). A La Niña event often, but not always, follows an El Niño and vice versa. Once developed, both El Niño and La Niña events tend to last for roughly a year although occasionally they may persist for 18 months or more. El Niño and La Niña are both a normal part of the earth’s climate and there is recorded evidence for their occurrence for hundreds of years
(Trenberth, 1997; Gill and Rasmusson, 1983; Gu and Philander, 1995; Huber and
Caballero, 2003; Fedorov and Philander, 2000).
Although El Niño and La Niña events are characterized by warmer or cooler than average SSTs in the tropical Pacific, they are also associated with changes in patterns of wind, pressure, rainfall, air temperature and total cloudiness fraction of the
44

sky (Horel and Wallace, 1981; Philander, 1983, 1990; McPhaden, 1999). Schematic views of the links between SSTs and other atmospheric components are illustrated in
Fig. 19. In normal conditions, the trade winds blow towards the west across the tropical
Pacific. These winds pile up warm surface water in the western Pacific, so that the sea surface is about 0.5 meter higher at Indonesia than at Ecuador. The SST is about 8

C higher in the west, with cool temperature off South America, due to an upwelling of cold water from deeper levels (Fig. 19a). Rainfall is found in rising air over the warmest water, and the east Pacific is relatively dry (Fig. 19a). During El Niño events, the trade winds relax in the central and western Pacific leading to strong countercurrent which carries warm water across the equatorial region, and a depression of the thermocline in the eastern Pacific and an elevation of the thermocline in the west (Fig. 19b). This reduces the efficiency of upwelling to cool the surface, resulting in a dramatic rise in
SST off South America (Fig. 19b). Rainfall follows the warm water eastward, with associated flooding in Peru and drought in Indonesia and Australia (Fig. 19b). The eastward displacement of the atmospheric heat source overlying the warmest water results in large changes in the global atmospheric circulation, which in turn force changes in weather in remote regions far from the tropical Pacific (Troup, 1965; Horel and Wallace, 1981; Wallace and Gutzler, 1981). While, La Niña conditions could be thought of as an enhancement of normal condition. During these events the trade winds strengthen, colder than average ocean water extends westward to the central Pacific, and the warmer than average SSTs in the western Pacific are accompanied by heavier than usual rainfall (Fig. 19c).
While the tropical ocean affects the atmosphere above it, so too does the atmosphere influence the ocean below it. In fact, the interaction of the atmosphere and ocean is an essential part of El Niño and La Niña events (the term coupled system is often used to describe the mutual interaction between the ocean and atmosphere).
During an El Niño, sea level pressure tends to be lower in the eastern Pacific and higher in the western Pacific, while the opposite tends to occur during a La Niña. This see-saw in atmospheric pressure between the eastern and western tropical Pacific is called the
Southern Oscillation (SO). The main centers of action of the SO are situated around
Darwin (12.4

S 130.9

E) in the northern Australia and Tahiti (17.5

S 149.6

W) in the
South Pacific (Fig. 20). Therefore, the difference in sea level pressures between the points has been long used as a standard measure of the SO (Troup, 1965). Since El Niño and the Southern Oscillation are related, the two terms are often combined into single phrase the El Niño-Southern Oscillation, or ENSO (Troup, 1965; Trenberth, 1984,
1997).
Several indices have been used to monitor ENSO. They have conventionally been calculated based only on sea level pressures at a combination of a few stations situated primarily near the main center of action of ENSO. These usually only involve those at Darwin and Tahiti (Troup, 1965; Trenberth, 1984, 1997; Ropelewski and Jones,
1987; Kiladis and van Loon, 1988). A drawback of this index is that it is based on the pressures at two points and therefore can easily be affected by local weather disturbances, making it somewhat “noisy” when viewed on a month-to-month basis. In recent decades, the indices based on SSTs have come into common usage because satellite and an observing network of buoys in the equatorial Pacific now allow for collection real time, high quality data. Indices based on SSTs (or, more often, its departure from long-term average) are those obtained by simply taking the average value over some specified region of the ocean (Wang, 1995; Trenberth and Hoar, 1996;
45

Trenberth, 1997). There are several regions of the tropical Pacific Ocean that have been highlighted as being important for monitoring and identifying El Niño and La Niña. The most common ones are illustrated in Fig. 21. For widespread global climate variability,
NINO3.4 is generally preferred, because the SST variability in this region has the strongest effect on shifting rainfall in the western Pacific. This in turn leads to shift the location of rainfall from the western to central Pacific which modifies greatly where the location of the heating that drives the majority of the global atmospheric circulation.
Newly generated version of index is the Multivariate ENSO Index (MEI) calculated as the first unrotated Principal Component of all six observed variables over the tropical
Pacific (Wolter and Timlin, 1993, 1998). These six variables are: sea level pressure, zonal and meridional components of the surface wind, sea surface temperature, surface air temperature and total cloudiness fraction the sky. The MEI is computed separately for each of twelve sliding bi-monthly seasons (Dec/Jan, Jan/Fec , …, Nov/Dec).
Negative values of the MEI represent the cold ENSO phase (La Niña), while positive
MEI values represent the warm ENSO phase (El Niño). Since the MEI integrates more information than other indices, it fully reflects the nature of the coupled oceanatmosphere system, and thereby is better for monitoring ENSO phenomenon, including, for instance, world-wide correlations with surface temperatures and rainfall than the SOI or SST-based indices (Wolter and Timlin, 1993, 1998). To make the MEI comparable with the monthly coefficient time series of EOF1 mode of T
B mean
B
, T
B max
B
, T
B min
B
, T
B dew
B
,
B B
T
B amean
B
,
T
B amax
B and
B B
T
B amin
B
, the MEI values of month (i-1) and month (i) were averaged for the value of month (i), and the MEI series was then decomposed into interannual and decadel timescales.
3.2.2.
Interannual variability of EOF1 coefficient series and its relationship with
ENSO events
Residual EOF1 coefficient series of all seven temperature variables, after the fluctuations less than ten months were removed, exhibit a salient mode of interannual variability (Figs. 22, 23, 24, 25, 26, 27 and 28). All series show strong negative and positive signs, and the oscillations between maxima and minima with period of about 1-
4 years stand out as reasonably clear signals above the otherwise noisy background of short-term climatic fluctuations. The results from variance analysis reveal that variability on interannual timescale for all seven temperature variables ranges from 17.6 to 25.8 % of the total variance (Table 3). A closer examination of the data indicates that interannual variability of all but T
B min
B
accounts for the second source of the total variance
(Table 3). A noteworthy feature emerged from Figs. 22-29 is that the interannual variability patterns of all seven temperature variables resemble that of MEI, and the anomalously positive/negative time-varying amplitudes of the EOF1 mode of them appear to be in phase with the warm/cold phase of ENSO (positive/negative MEI).
There is a clear indication that T
B mean
B
, T
B max
B
, T
B min
B
, T
B dew
B
,
B B
T
B amean
B
, T
B amax
B and
B B
T
B amin
B
in Thailand tend to warmer (colder) than normal during El Niño (La Niña) phase of ENSO. During the 6 strongest historic El Niño events, for example, all these variables were prominently higher than normal (Table 4), while they were anomalously lower than average during the 8 strongest historic La Niña events (Table 5). Moreover, the EOF1 coefficient series of all seven temperature variables underwent largest interannual variability during the recent extreme phase reversals of ENSO, when the 1997-98 El
Niño, by some measures the strongest on record, was followed by the strong 1998-2000
La Niña. A nonparametric Spearman rank correlation test provides further evidence that there were significant positive correlations between each smoothed EOF1 coefficient
46

series of surface temperature variables and the 10-term smoothed MEI series, for the 47year period (Table 6). It is readily seen that the 10-term smoothed EOF1 coefficient series of T
B mean
B
, T
B max
B
, T
B amean
B
and T
B amax
B have high correlations with that of the MEI, with correlation coefficients higher than 0.5 (Table 6). The similar but less pronounced
Fig. 17. An example of departure of sea surface temperature from the long-term average for an El Niño event during December 1991. Yellow shading indicates warmer than average temperatures. Units are

C and contours are drawn at 0.5

C intervals. Note that this picture was obtained from website of International Research Institute for climate prediction (http://iri.columbia.edu/climate/ENSO).
Fig. 18. An example of departure of sea surface temperature from the long-term average for a La Niña event during December 1998. Blue shading indicates colder than average
47

(a)
(b)
(c) temperatures. Units are

C and contours are drawn at 0.5

C intervals. Note that this picture was obtained from website of International Research Institute for climate prediction (http://iri.columbia.edu/climate/ENSO).
48

Fig. 19. Schematic views of the links between SSTs and tropical atmospheric variables in the equatorial Pacific ocean during normal (a), El Niño (b) and La Niña conditions (c).
SSTs are shaded: blue-cold and orange-warm. The dark arrows indicate the direction of air movement in the atmosphere: upward arrows are associated with clouds and rainfall and downward-pointing arrows are associated with a general lack of rainfall. Note that this picture was obtained from website of International Research Institute for climate prediction (http://iri.columbia.edu/climate/ENSO).
Fig. 20. The main centers of action of the Southern Oscillation. Tahiti and Darwin are at opposite ends of the Southern Oscillation’s seesaw, and so the difference in pressure between them is used to measure the Southern Oscillation. The figure shows that pressure variation at Tahiti is as closely related to Darwin as are locations near to
Darwin, but with the opposite sign. Note that this picture was obtained from website of International Research Institute for climate prediction
(http://iri.columbia.edu/climate/ENSO).
49

Fig. 21. The NINO regions. The thin grey line in the center is the equator. Note that this picture was obtained from website of International Research Institute for climate prediction (http://iri.columbia.edu/climate/ENSO).
0.6
(a)
0.3
0
-0.3
-0.6
(b)
0.2
0
-0.2
1950 1960 1970 1980 1990 2000
Year
Fig. 22. The EOF1 coefficient series of T
B mean
B
residuals (original series minus 60-term smoothed series) (a) and 10-term smoothed T
B mean
B
residuals (b).
0.8
(a)
0.4
0
-0.4
-0.8
0.4
(b)
0.2
0
-0.2
1950 1960 1990 2000 1970 1980
Year
50

Fig. 23. The EOF1 coefficient series of T
B max
B
residuals (original series minus 60-term
-0.4
-0.8
0.2
0.8
0.4
0
(a)
(b)
0
-0.2
1950 1960 1970 1980 1990 2000
Year smoothed series) (a) and 10-term smoothed T
B max
B
residuals (b).
Fig. 24. The EOF1 coefficient series of T
B min
B
residuals (original series minus 60-term smoothed series) (a) and 10-term smoothed T
B min
B
residuals (b).
1
(a)
0.5
0
-0.5
-1
0.2
(b)
0
-0.2
1950 1960 1970 1980
Year
51
1990 2000

Fig. 25. The EOF1 coefficient series of T
B dew
B
residuals (original series minus 60-term
0.8
0.4
0
-0.4
-0.8
(a)
(b)
0.2
0
-0.2
1950 1960 1970 1980 1990 2000
Year smoothed series) (a) and 10-term smoothed T
B dew
B
residuals (b).
Fig. 26. The EOF1 coefficient series of T
B amean
B
residuals (original series minus 60-term smoothed series) (a) and 10-term smoothed T
B amean
B
residuals (b).
0.8
(a)
0.4
0
-0.4
-0.8
0.4
(b)
0.2
0
-0.2
1950 1960 1990 2000 1970 1980
Year
52

Fig. 27. The EOF1 coefficient series of T
B amax
B
residuals (original series minus 60-term
1
0.5
0
-0.5
-1
(a)
(b)
0.2
0
-0.2
1950 1960 1970 1980 1990 2000
Year smoothed series) (a) and 10-term smoothed T
B amax
B
residuals (b).
Fig. 28. The EOF1 coefficient series of T
B amin
B
residuals (original series minus 60-term smoothed series) (a) and 10-term smoothed T
B amin
B
residuals (b).
3
(a)
2
-1
-2
-2
3
2
1
0
1
0
-1
(b)
1950 1960 1970 1980
Year
53
1990 2000

Fig. 29. Time series of Multivariate ENSO Index (MEI) residuals (original series minus
60-term smoothed series) (a) and 10-term smoothed MEI residuals (b).
54

Table 3. Decomposition of the coefficient time series of the first EOF mode. Total variance was calculated from the original coefficient series, whereas the variance for long-term and interannual changes was calculated from 60-term and 10-term smoothed series, respectively. The remaining variance (total – (long-term +interannual)) would account for variability less than 10 months or seasonal change. The values in the parentheses denote percentage of the total variance.
Source
Long-term change
T
B mean
B
0.004
(16.7)
T
B max
B
0.005
(16.1)
T
B min
B
0.009
26.5
T
B dew
B
0.004
(12.5)
T
B amean
B
0.005
(15.6)
T
B amax
B
0.005
(14.3)
T
B amin
B
0.008
(19.0)
MEI
0.182
(19.0)
Interannual change 0.006
(25.0)
0.008
(25.8)
0.006
(17.6)
0.007
(21.9)
0.008
(25.0)
0.009
(25.7)
0.008
(19.0)
0.637
(66.4)
Seasonal change
Total
0.014
(58.3)
0.024
(100)
0.018
(58.1)
0.031
(100)
0.019
(55.9)
0.034
(100)
0.021
(65.6)
0.032
(100)
0.019
(59.4)
0.032
(100)
0.021
(60.0)
0.035
(100)
0.026
(62.0)
0.42
(100)
0.141
(14.6)
0.96
(100)
54

Table 4. Listing of the 6 strongest El Niño events after 1954 as defined by MEI exceeding 0.5 for at least 6 consecutive months and a maximum
MEI of at least 1. The starting and ending months of each is given along with the duration in months. T
B mean
B
, T
B max ,
B
T
B min
B
, T
B dew
B
, T
B amean
B
, T
B amax
B and
T
B amin
B values are mean averaged for the entire duration.
Begin
Apr 1957
End
Aug 1958
Duration
17
Peak value of MEI
1.36
T
B mean
B
0.06
T
B max
B
0.07
T
B min
B
0.07
T
B dew
B
0.07
T
B amean
B
0.09
T
B amax
B
0.1
T
B amin
B
0.09
May 1965
Mar 1972
Apr 1966
Apr 1973
12
14
1.32
2.13
0.07
0.09
0.07
0.12
0.08
0.09
0.09
0.08
0.09
0.11
0.09
0.14
0.1
0.1
Jul 1982
Aug 1986
Apr 1997
Aug 1983
Jan 1988
Jun 1998
14
18
15
2.03
1.59
2.36
0.05 0.07 0.01 0
0.06 0.07 0.05 0.03
0.13 0.18 0.08 0.06
0.04
0.07
0.13
0.07 0.01
0.07 0.05
0.18 0.09
55

Table 5. Listing of the 7 strongest La Niña events after 1954 as defined by MEI exceeding -0.5 for at least 6 consecutive months and a minimum
MEI of at least -1. The starting and ending months of each is given along with the duration in months. T
B mean
B
, T
B max ,
B
T
B min
B
, T
B dew
B
, T
B amean
B
, T
B amax
B and
T
B amin
B values are mean averaged for the entire duration.
Begin
Nov 1954
End
Oct 1956
Duration
24
Peak value of MEI
-1.25
T
B mean
B
T
B max
B
T
B min
B
T
B dew
B
T
B amean
B
T
B amax
B
T
B amin
B
-0.06 -0.06 -0.07 -0.06 -0.08 -0.07 -0.08
May 1964 Dec 1964 8 -1.0 -0.05 -0.06 -0.03 -0.01 -0.05 -0.06 -0.03
Jul 1970
Feb 1975
Nov 1971
Feb 1976
17
13
-1.28
-1.09
-0.08 -0.11 -0.07 -0.05 -0.09 -0.11 -0.08
-0.03 -0.05 -0.02 -0.01 -0.04 -0.05 -0.02
Feb 1984
May 1988
Jul 1995
Oct 1998
Dec 1985
Sep 1989
Dec 1996
Apr 2000
23
17
18
20
-1.33
-1.53
-1.09
-1.03
-0.04 -0.04 -0.01 -0.01 -0.04 -0.04 -0.02
-0.05 -0.06 -0.04 -1.1 -0.05 -0.06 -0.04
-0.07
-0.07
-0.1
-0.11
-0.06
-0.04
-0.02
-0.04
-0.07
-0.08
-0.09
-0.11
-0.05
-0.04
56

relationships can also be observed in T
B min
B
, T
B amin
B
and T
B dew
B series (Table 6). To further illustrate the nature of seasonal changes association with ENSO, month-by-month correlations between the 10-term smoothed MEI series and the smoothed EOF1 coefficient series of all seven temperature variables were constructed and shown in Fig.
30. As noticeable from Fig. 30, there is strong element of seasonal cycle in the relationships between ENSO and surface air temperatures in Thailand with the exception of T
B dew
B
that is physically consistent with the seasonal evolution and development of ENSO. Weak correlations between them occur during transition months
(March-July), corresponding to the change of phase of ENSO (Fig. 30). However, the corrections increase and become more statically significant for the period from August through February (Fig. 30), when either phase of ENSO progressively develops and becomes mature. These results imply that links between ENSO and surface air temperature changes in Thailand are usually stronger during progressive development and mature phase of ENSO, because large shifts in the atmospheric heat source in the tropical Pacific. This in turn leads to intensify the global hydrological cycle and atmospheric circulation, and therefore “enhanced atmospheric teleconnections” during this period that strongly influence the seasonal temperature patterns over much of the tropics and sub-tropics, and some mid-latitude areas. It should be noted that, among temperature variables, T
B max
B and T
B amax
B show strongest connections with ENSO. Based on the observed evidence, it is reasonable to suggest that changes in phases and intensities of ENSO phenomenon are responsible for a large fraction of interannual variability of surface air temperatures in Thailand.
3.2.3. Decadal/interdecadal changes in ENSO and EOF1 coefficient series
The MEI series based on centered 60-term moving average (Fig. 31b) clearly reveals low-frequency variability with timescales greater than 5 years that accounts for 19% of the total variance (Table 3). A break or “regime shift which is defined as anomalies tending to have the same sign for a decade or more” after about 1976, when the MEI was at its most persistent and extreme positive phase, is the most outstanding feather of
Fig. 31b. This indicates the tendency for more El Niño and fewer La Niña events since the late 1970s. A protracted period of persistent positive MEI since the late 1970s, during which several strong to weak El Niño events occurred with no intervening La
Niña events, was previously mentioned by some studies (Quinn and Neal, 1984, 1985;
Trenberth, 1990; Graham, 1994; Trenberth and Hurrell, 1994; Zhang et al., 1997) to be unprecedented in the observational record of the previous 113 years and to be statistically very rare (Trenberth and Hoar, 1996). In addition, prominent patterns of decadal/interdecadal oscillations are visually evident in the 60-term smoothed EOF1 coefficient series of all seven temperature variables (Figs. 31a, 32, 33) that explain variability ranging from 12.5 to 26.5 % of total variance (Table 3). The smoothed EOF1 coefficient series of T
B mean
B
, T
B max
B
, T
B amean
B
, and T
B amax
B
(Figs. 32b, 32c, 33b, 33c) show gradual decreases from the early 1950s to the mid 1970s. Following this period, there were overall rises, which the steady warming trends became more pronounced in T
B max
B and T
B amax
B series that were closely associated with the persistent and exceptionally positive phase of the MEI. These results agree with the study of Rungdilokroaja and
Nimma (1990), illustrating the annual mean temperature in the 1980s increasing gradually in all parts of Thailand. Correlation analysis provides further evidence that there were significant positive correlations between the smoothed series of MEI and the smoothed EOF1 coefficient series of T
B max
B
( r
B s
B
=0.69, p<0.05, N
B eff
B
=17, N=576 ) and T
B amax
B
( r
B s
B
=0.51, p<0.025, N
B eff
B
=18, N=576 ). However, no significant correlations between
54

smoothed series of MEI and the smoothed EOF1 coefficient series of T
B mean
B
, T
B amean
B
and
T
B dew
B
were found. Whereas, the smoothed EOF1 coefficient series of T
B min
B
and T
B amin
B
(Figs.
32a, 33a) exhibit distinct long-term trends which have monotonically and sharply increased at a faster rate than those of T
B mean
B
, T
B max
B
, T
B amean
B
, and T
B amax
B
since the mid 1950s.
Another conspicuous feather emerged from Figs. 32a and 33b is that increasing trends of
B B
T
B min
B
and T
B amin
B
in Thailand strongly resemble the greenhouse warming fingerprint observed in modern instrumental records and predicted by some general circulation models (IPCC, 2001a). Moreover, the timing of the zero-crossing of the smoothed EOF1 coefficient series of
B B
T
B min
B
and T
B amin
B
was synchronous with an abrupt transition from fewer El Niño and more La Niña to more El Niño and fewer La Niña events in the late
1970s. Despite their somewhat different visual appearances, the smoothed series of MEI highly correlated with those of T
B min
B
( r
B s
B
=0.71, p<0.05, N
B eff
B
=16, N=576 ) and T
B amin
B
( r
B s
B
=0.72, p<0.05, N
B eff
B
=17, N=576 ). These results indicate that decadal/interdecadal changes in T
B max
B
, T
B amax
B
,
B B
T
B min
B
and T
B amin
B
in Thailand could in part be a signal of anthropogenic warming, or underwent synchronized changes in close response to a regime shift in the ENSO phenomenon.
3.3. Linear trends in surface air temperatures in Thailand
The EOF analysis clearly shows that T
B min
B
and T
B amin
B
increased monotonically at faster rate than T
B max
B
, and T
B amax ,
B leading to a narrowing of temperature range. However, coefficient time series of EOFs can not provide the actual rate of change, since EOF calculation is based on covariance matrix resulting in their units being arbitrary. Thus, it is of interest to further evaluate the rate of changes in T
B min
B
, T
B max
B and temperature ranges at each station. This expanded analysis can be done by using nonparametric method of the median of pairwise slopes regression ( e.g., Lanzante, 1996). This method is resistant to outliers in the time series and robust to nonnormal data distribution (Wilks, 1995). It involves computing the slopes of lines connecting each pair of annual averaged maximum and minimum temperatures and differences between annual averaged maximum and minimum temperatures, and finding the median value of those slopes.
This method avoids placing undue emphasis on large anomalies near the ends of time series. Two-tailed t tests of the robust and resistant Spearman rank-order correlation coefficient (between temperature variables and time) were used to evaluate the statistical significance of the trends.
Calculated linear trends indicate a clear warming of annual averaged maximum and minimum temperatures at most stations (Figs. 34a, 34b). These results are fairly consistent with what was previously seen from the coefficient time series of EOF analysis. There is an indication that maximum temperature increased only slightly with an overall trend of 0.59

C per 50 years, whereas minimum temperature significantly increased at faster rate with overall trend of 1.35

C per 50 years. At almost stations, upward trends in annual averaged minimum temperature overwhelmingly exceeded those in annual averaged maximum temperature, which the largest and most statistically significant increasing trends occurred in all parts of Thailand (Figs. 34a, 34b). As a consequence of differential increases in maximum and minimum temperatures, there was a significant and substantial reduction of temperature ranges over almost parts of
Thailand (Fig. 34c) with decreasing rates ranging from -0.1 to -2.2

C per 50 years.
Since annual averaged maximum and minimum temperatures were calculated originally from daily observations, such an observed decrease would result from a reduction in the diurnal temperature range. This finding agrees well with the previous study, illustrating
55

that the diurnal temperature range is continuing to decrease in most parts of the world, because minimum temperatures are increasing at about twice at the rate of maximum
(Easterling et al., 1997). As a consequence, the freeze-free periods in most mid- and high- latitude regions are lengthening and satellite data reveal a 10% decrease in snow cover and ice extent since the late 1960s (IPCC, 2001a).
56

Table 6. Nonparametric Spearman correlation coefficients ( r
B s
B
) between the 10-term smoothed temperature series and the 10-term smoothed MEI series. N and N
B eff
B are the number of data points in each series and the effective number of independent observation, respectively.
Variables
T
B mean
B
& MEI
N
566
N
B eff
116
B r
B s
B
0.64 p-value
<0.001
T
T
B max
B
& MEI
T
T
B dew
B
& MEI
T
B amax
B
& MEI
T
B
B min amean
B amin
B
B
& MEI
B
& MEI
& MEI
566
566
566
566
566
566
116
122
129
118
117
123
0.70
0.49
0.38
0.60
0.68
0.47
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
The advantage of nonparametric method is free of the assumption that the data being analyzed have normal distribution with equal variances, and so does not emphasize the extreme values that are often present in the data. For the smoothed time series, autocorrections, which are introduced through moving averages and/or are often present within each time series, can give rise to spurious correlations due to overestimation of the number of degree of freedom that do not present actual mechanistic relationships. To circumvent this problem, significance levels of computed r
B s
B
-values were estimated by calculating the effective number of independent observations ( N
B eff
B
) following Davis
(1976), Chelton (1983), Trenberth (1984) and Emery and Thomson (1997):
N eff
=
(1
 r
1 r
1
 
N r
2 r
2
 
...
) where N is the number of data points in each of the two smoothed series, r and
1 the lag-one autocorrelations of the two smoothed series, r
2 and r
2
 r
1
 are are the lag-two autocorrections, and so on. The calculation was terminated after a lag of 10 months.
57

0.4
0.2
0
0.8
0.6
0.4
0.2
0
0.8
0.6
0.8
(a)
0.6
(b)
(c)
T amax
T amean
T amin
T dew
0.4
0.2
T max
T mean
T min
0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month
Fig. 30. Correlograms between the smoothed MEI series and the 10-term smoothed
EOF1 coefficient series of seven temperature variables. The number of data points for each month and the effective number of independent observations ( N
B eff
B
) for all temperature series are 47 and 11, respectively. The N
B eff
B
calculation was terminated after a lag of 5 months. The dash lines denote correlation coefficients at the p-value = 0.05 and 0.025 levels of significance for N
B eff
B
= 11.
58

0
-0.1
0.2
(a)
0.1
-0.2
0.2
(b)
0.1
0
-0.1
-0.2
0.2
(c)
0.1
0
-0.1
-0.2
1950 1960 1970 1980 1990 2000
Fig. 32. The 60-term smoothed EOF1 coefficient series of T
B min
B
(a), T
B max
B
(b) and T
B mean
B
(c).
59

0.2
(a)
0.1
0
-0.1
-0.2
0.2
(b)
0.1
0
-0.1
-0.2
0.2
(c)
0.1
0
-0.1
-0.2
1950 1960 1970 1980 1990 2000
Fig. 33. The 60-term smoothed EOF1 coefficient series of T
B amin
B
(a), T
B amax
B
(b) and T
B amean
B
(c).
60

0.2
(a)
0.1
0
-0.1
-0.2
1
(b)
0.5
0
-0.5
-1
1950 1960 1970 1980 1990 2000
Fig. 31. The 60-term smoothed EOF1 coefficient series of T
B dew
B
(a), and the 60-term smoothed MEI (b).
61

2
(c)
0
-2
(b)
2
0
-2
2
0
-2
(a)
North Northeast Central East South
1=CHIANG RAI
2=CHIANG MAI
3=PHITSANULOK
4=UTTARADIT
5=NAN
6=MAE SOT
7=PHARE
8=PHETCHABUN
9=UDONTHANI
10=SAKON NAKHON
11=NAKHON PRANOM
12=KHON KAEN
13=MUKDAHAN
14=ROI ET
15=NAKON RATCHASIMA
16=UBON RATCHATHANI
17=SURIN
18=NAKON SAWAN
19=LOP BURI
20=DON MUANG AIRPORT
21=SUPHAN BURI
22=KANCHANA BURI
23=ARANYA PRATHET
24=SATTATHIP
25=CHANTHA BURI
26=PRACHUAP KHIRIKHAN
27=CHUMPHON
28=NAKONSI THAMMARAT
29=TRANG AIRPORT
30=SONGKHLA
31=NARATHIWAT
32=RANONG
33=PHUKET
0 3 6 9 12 15 18 21 24 27 30 33
Station number significant at the 0.05 level
Not significant at the 0.05 level
Fig. 34. Linear trends (

C per 50 yrs) for annual averaged maximum temperature (a), annual averaged minimum temperature (b), and the differences between annual averaged maximum and minimum (c) from 1951 to 2003 .
62

4. Discussion
4.1. Analytical methodology
In this study, EOFs, as an excellent statistical tool which has been conventionally applied to oceanographic and meteorological datasets, were used to decompose and describe the dominant spatial and temporal structures of surface air temperature fields in Thailand. Representation of a very large dataset in terms of EOFs provides a seemingly plausible way of eliminating data variables, namely deleting those modes which contribute little to the variance of the data field. As outlined by Davis
(1976), the principal virtues of EOFs are 1) they provide the most efficient method of compressing data, 2) they may be regarded as uncorrelated ( i.e.
orthogonal) modes of variability of the data field, and 3) they simplify understanding the procedures of minimum mean square error estimation. Often it can be advisable to employ the EOFs as a filter to eliminate unwanted scales of variability. In such a case, a limited number of the first few EOFs (those with large eigenvalues) are used to reconstruct a good approximation of the original dataset, thereby eliminating those scales of variability not coherent over the dataset and therefore less energetic in their contribution to the data variance. An EOF analysis can then be made of the filtered dataset to provide a new apportionment of the variance for those scales associated with most of the variability in the original dataset. In this application, EOF analysis is very similar to the standard
Fourier analysis used to filter out scales of unwanted variability (Davis, 1976; Emery and Thomson, 1997). In fact, for homogenous time series sampled at evenly spaced increments, it can be shown that the EOFs are Fourier trigonometric functions.
In interpreting the meaning of EOFs, it needs to keep in mind that, while EOFs offer the most efficient statistical compression of the data filed, they do not necessarily correspond to true dynamical modes or modes of physical behavior. It is often the case that a single physical process is spread over several EOF modes. In other cases, more than one physical process may contribute to the variance contained in a single EOF.
The statistical construct derived from this procedure must be considered in light of accepted physical mechanisms rather than as physical modes themselves (Preisendorfer,
1988; Jassby and Powell, 1990; Emery and Thomson, 1997; Venegas, 2001)
.
It is often likely that the strong variability associated with the dominant modes is attributable to recognizable physical mechanisms. A clue to the physical mechanisms associated with the EOF mode can be found in the time-series coefficients. Something may be known about the temporal variability of a process that might resemble the time series of the
EOF coefficients, which would then suggest a causal relationship not readily apparent in the spatial structure of the EOF (Emery and Thomson, 1997; Venegas, 2001)
.
Although an EOF analysis has the outstanding advantage over other statistical techniques, the physical interpretation of EOFs is inevitably limited by some fundamental constraints: the imposed spatial orthogonality of the EOF patterns and the resulting temporal independence of the time coefficients. While it is often possible to associate the first EOF mode with a known physical process, this is much more difficult with the second (and higher order) EOF because it is constrained to be orthogonal to the first EOF. The orthogonality constraint implies that the EOF modes can only represent physical processes that operate independently and with orthogonal patterns. Real world processes, however, do not need to have orthogonal or uncorrelated patterns. In contrast,
63

they are in general interrelated. Another constraint of a traditional EOF analysis is that its procedure can only detect standing waves, but can not capture progressive waves which two time-series coefficients of consecutive EOF modes vary coherently and are
90 degrees out of phase with one another. It is advisable to use lagged covariance matrix
(Weare and Nasstrom, 1982), or a complex EOF analysis in the frequency domain
(Wallace and Dickinson, 1972; Horel, 1984) to detect propagating wave phenomena.
4.2. Possible causal attributions of interannual and long-term changes in surface air temperatures in Thailand
With surface air temperature data collected at 34 stations in Thailand, the EOF technique was very useful to extract the dominant spatial and temporal structures, which the first EOF mode could be accounted for most of the variance of each surface air temperature variable for the period 1951-2003. The EOF1 modes of all seven temperature variables was characterized by a monopole of spatial patterns, which correlations coefficients were relatively high and about the same magnitude, and had the same sign at all stations. Such a unique pattern implies a high intercorrelation and a relatively uniform variance distribution of surface air temperatures at all selected stations, in combination with the “ characteristic spatial scale”
of the surface air temperature field which is comparable to, or larger than, the spatial domain considered here. This is simply because the system will have tendency to create anomalies of the same sign in the entire domain when the typical scale of the filed is as large as the domain of study. Hence, the coherent variance carried by the EOF1 mode well represents the dominant spatial structure of surface air temperatures integrated from all stations in Thailand that probably share a common influence from the same origins, which are not physically local.
On the basis of the EOF results, the time variability of the leading EOFs of all seven temperature variables in Thailand has oscillated at three dominant timescales over the last 53 years: interannual and decadal/interdecadal timescales and long-term trends.
The ENSO cycle has been the most prominent timescale of the interannual variability of surface air temperatures in Thailand. There was a significant indication that all T
B mean
B
,
T
B max
B
, T
B min
B
, T
B dew
B
,
B B
T
B amean
B
, T
B amax
B and
B B
T
B amin
B
in Thailand tended to higher (lower) than normal during El Niño (La Niña) years. These results are consistent well with an enormous body of previous evidence, illustrating that the ENSO has been related to variations of precipitation and temperature over much of the tropics and sub-tropics, and some midlatitude areas ( e.g., Prospero and Nees, 1986; Hanawa et al., 1989a, b; Li, 1990;
Philander, 1990; Trenberth, 1990; Yamagata and Masumoto, 1992; Janicot et al., 1996;
Sun and Trenberth, 1998; McPhaden, 1999; Fedorov and Philander, 2000; Hoerling and
Kumar, 2003). The climate impacts observed during and following El Niño events include higher global mean temperature as a result of the ocean transfer heat to the atmosphere (IPCC, 2001a), increased heat export to the extratropics (Sun and Trenberth,
1998), and continental warmth over parts of Asia and North America (Hoerling and
Kumar, 2003). In association with the strongest 1997/98 El Niño event, for example, the global mean temperature experienced the warmest anomaly (0.55

C) in the instrumental record since 1861 (IPCC, 2001a). Such a exceptionally positive anomaly associated with the intense 1997/98 El Niño event prominently stands out in the amplitude time series of the EOF1 mode of all seven surface air temperatures in
Thailand, as an unprecedented event on record (Figs. 22-29). The possible linking pathways, whereby the ENSO phenomenon exerts its influence on the climate
64

conditions outside the tropical Pacific region, may mediate via the intensified atmospheric teleconnections which are established by the shifts in the location of the organized rainfall in the tropics and the associated latent heat release (Horel and
Wallace, 1981; Wallace and Gutzler, 1981; Gill and Rasmusson, 1983; Philander, 1983;
Trenberth, 1990; Graham, 1994; Trenberth and Hurrell, 1994; McPhaden, 1999; Zahn,
2003). These anomalous processes, in turn, alter the heating patterns of the atmosphere which forces large-scale waves in the atmosphere (Horel and Wallace, 1981; Wallace and Gutzler, 1981; Trenberth and Hurrell, 1994; Fedorov and Philander, 2000; Huber and Caballero, 2003). The atmospheric teleconnections are an important driver because they induce anomalies in the circulation and associated anomalies in temperature and precipitation in remote regions (Wallace and Gutzler, 1981; Trenberth and Hurrell,
1994). The likely effects of the Mount Pinatubo eruption were also seen in the EOF1 coefficient series of all seven surface air temperatures, which showed noticeable drops during 1992-93 (Figs. 22-28), despite the moderate El Niño event. This is due to the fact that a great deal of aerosols introduced into the atmosphere during the eruption of Mt.
Pinatubo in the Philippines in June 1991 blocked enough radiation for two years to cause observable cooling in many regions (Trenberth, 2001).
It has been recognized for several decades that ENSO phenomenon is non-stationary
( e.g., Troup 1965; Philander, 1983; Wang, 1995; Fedorov and Philander, 2000). Climate proxy records (for instance, ice cores, tree rings and corals) suggest that this phenomenon has occurred for millennia (Ware and Thomson, 2000; GBP, 2001b; Huber and Caballero, 2003). The amplitude, activity and frequency of ENSO have exhibited notable secular variation since 1871 with considerable irregularity in time, as revealed by wavelet analysis of the NINO3 SST index (Gu and Philander, 1995) and the SOI
(Wang and Wang, 1996). Wang (1995) and Wang and An (2002) pointed out that not only the amplitude and dominant period of ENSO cycles have increased, but also the onset and the structure and development characteristics of this episode have changed during the 1980s and 1990s. La Niña episodes were very weak or practically absent during those decades, whereas El Niño episodes attained unprecedented amplitudes in
1982 and 1997 and were unusually prolonged in 1992 (Trenberth and Hoar, 1996;
Fedorov and Philander, 2000). Hence, in addition to the dominant interannual timescale of 2 to 6 years, the decadal/interdacal variability is one of fundamental characteristics of
ENSO cycles, as clearly observed in the smoothed MEI series (Fig. 29). It has been well documented that the tendency for more frequent El Niño events and fewer La Niña events since the late 1970s has been linked to decadal changes in climate throughout the
Pacific basin (Trenberth, 1990; Trenberth and Hurrell, 1994; Kumar et al., 1994;
Graham, 1994; Trenberth and Hoar, 1996; Zhang et al., 1997). The EOF1 coefficient series of some surface air temperature variables in Thailand (Figs. 32a, 32b, 33a, 33b) do indicate similar decadal/interdecadal changes which were significantly related to the low-frequency component of ENSO cycles. The overall warming trends of T
B max
B
, T
B amax
B
,
T
B min
B
and T
B amin
B
in the 1980s and 1990s, with the1990s being their warmest decade on record, were consistent with the persistent and exceptionally strong warm phase of
ENSO cycles. These findings, therefore, provide a notation that the ENSO-induced variations in the quasi-stationary planetary wave in the atmosphere, among other factors, could produce spatially coherent patterns of surface air temperature anomalies in
Thailand, not only on interannual but also decadal/interdecadal timescales. Hurrell
(1996) also showed that the warm phase of ENSO since the late 1970s has been associated with widespread continental warming, particularly over North America and
65

parts of Siberia, resulting in a particular surface temperature anomaly pattern which has amplified the hemispheric-averaged warming.
On a long-term perspective, there is evidence that T
B min
B
and T
B amin
B
in Thailand have been rising continuously at unprecedented rate since the early 1950s to what is obviously linear trends superimposed on strong interannual variability (Figs. 32a, 33a).
Similar long-term changes in the past half century have been also found over various regions in the world, for example, the United States (Easterling et al., 1997; Gaffen and
Ross, 1998; 1999; Easterling et al., 2000b), China (Zhai et al., 1999; Qian and Zhu,
2001; Wang and Gaffen, 2001), New Zealand (Easterling et al., 1997), most parts of
Europe (Easterling et al., 1997). The secular changes in T
B min
B
and T
B amin
B
in Thailand were basically consistent with patterns of globally and hemispherically averaged air temperatures in the 20
P th
P
century (IPCC, 2001a), but their amplitudes were larger.
Therefore, T
B min
B
and T
B amin
B
rising must have a positive contributor to the change of the northern hemisphere temperature and to the global warming, and could in part be a signal of a direct forcing by the anthropogenic increase in greenhouse gases. However, the observed long-term warming of T
B min
B
and T
B amin
B
in Thailand may be a manifestation of the ENSO-related multidecadal variation, though this itself might have an anthropogenic-driven component, because abrupt increases in T
B min
B
and T
B amin
B since the late 1970s (Figs. 32a, 33a) have been significantly associated with what has been described as a strong shift to the warm phase of ENSO episode (Fig. 31b). This makes it complicating to assess whether the observed changes are in direct response to greenhouse gas forcing, or whether the changes are part of a natural decadal timescale variation in the atmospheric circulation. One may think that the climate response to anthropogenic forcing should be distinct from the patterns of natural climate variability.
In view of nonlinear dynamical complex behavior of climate system, however, it has been argued that the spatial patterns of the response to anthropogenic forcing may in fact project principally into modes of natural variability. This interpretation is supported by many studies, suggesting that human-induced emission of greenhouse gases might be responsible for the recent change in the behavior of natural modes of large-scale atmospheric fluctuations such as NAO, ENSO and PNA, and that the recent warming may be related to increasing tropical ocean temperature leading to an enhancement of the tropical hydrological cycle (Trenberth and Hoar, 1996; Timmerman et al., 1999;
Fedorov and Philander, 2000; Hoerling and Kumar, 2003; Huber and Caballero, 2003).
Several climate models do indicate a more El Niño-like climate (greater warming in the tropical east Pacific ocean and an eastward shift in convective activity in the Pacific) with increased greenhouse gases ( e.g., Meehl and Washiton, 1996; Knutson and Manabe,
1995; Knustson et al., 1997; Timmerman et al., 1999). Therefore, it has been suggested that recent observed warming trends in various parts of the world might be projected onto changes in the natural modes of climate variability, through changes in their frequency and preferred sign, and more directly related to the thermal structure of atmospheric circulation than to any anthropogenic forcing pattern itself (Corti et al.,
1999). From this point of view, the actual causes for such long-term increasing trends in
T
B min
B
and T
B amin
B
in Thailand are not clearly identified here, and therefore, remaining as a crucial question which needs to be given more detailed study. Regardless of causes, the results from this study do indicate that T
B min
B
and T
B amin
B
in Thailand have progressively increased at alarming rate over the last 53 years.
4.3. A reduction of diurnal temperature ranges
66

Apart from climate-related temporal changes in surface air temperature, there was a widespread narrowing of diurnal temperature ranges (DTR) over most parts of
Thailand (Fig. 34). This probably results from the differential changes in maximum and minimum temperatures. Local effects such as urban growth, irrigation, desertification, and variations in local land use can all affect the DTR (Easterling et al., 1997), especially urbanized areas often show a narrower DTR (Gallo et al., 1996). However, an analysis of urban effects based on a metadata set shows that globally and hemispherically averaged time series for the annual maximum and minimum temperatures and the DTR calculated using only non-urban stations slightly differ from those calculated using all available stations (Easterling et al., 1997; IPCC, 2001a).
Large-scale climatic effects on the DTR include increases in cloud cover, surface evaporative cooling from precipitation, greenhouse gases, and tropospheric aerosols
(IPCC, 2001a). Easterling et al. (1997) found that the DTR changes were associated with circulation changes during the northern hemispheric winter. A reduction of the
DTR in Thailand may be partly associated with increased urbanization. If the spatial extent of urban heating has been growing, weather stations near large cities might experience high temperatures more frequently. However, the regional consistency of the
DTR decreases (Fig. 34) suggests that their origins are not strictly local, but it appears to be the manifestation of a strong increase in minimum temperature in response to either anthropogenic or natural modes of climate change.
4.4. Changes in temperature extreme events
Given a change in mean state of temperature, there is likely to be an amplified change in what are called extreme events, such as a very low or very high daily temperatures and heat and cold waves. Recent evidence has suggested that temperature extreme events have apparently increased in frequency (Cooter and LeDuc, 1995;
Changnon et al., 1996; DeGaetano, 1996; Heino et al., 1999; Plummer, et al., 1999; Zhai et al., 1999; Easterling et al., 2000a, b). In the northeastern United States, significant trends to fewer extreme cold days, but also trends to fewer warm maximum temperatures were observed (DeGaetano, 1996). In other parts of the world different trends prevail. In both Australia and New Zealand, the frequency of days below freezing decreased in phase with warming in daily minimum temperatures (Plummer, et al.,
1999). Similarly, a number of frost days in northern and central Europe have decreased since the 1930s, which appear to be associated with strong increases in winter minimum temperatures (Heino et al., 1999). This leads to significantly lengthen the freeze-free season in many mid-and high latitude regions. Based on an apparent temperature index, an important measure for human comfort, Gaffen and Ross (1998) and Wang and
Gaffen (2001) found that the frequency of extreme heat-stress events in the United
States and China, caused by extremely hot and humid days as well as by heat waves in summertime, has increased over the period from 1951 to 1994. Short-duration episodes of extreme heat or cold are often responsible for the major impacts on health, as evidenced by the 1995 heat wave in the midwestern United States that caused hundred of fatalities in the Chicago area (Changnon et al., 1996; Karl and Knight, 1997).
In Thailand, evidence has also been found, supporting the notion that changes in temperature extreme events are likely to occur in concert with the observed warming trends in maximum and minimum temperatures. In an analysis of 46 synoptic stations in
Thailand, Ouprasitwong (2002) pointed out that trends in the number of days with the maximum temperature equal to or higher than the 99
P th
P
percentile threshold have been
67

significantly increased since 1961. Whereas, trends in the number of days with the minimum temperature equal to or less than the 1
P st
P
percentile have been substantially decreased during the same period (Ouprasitwong, 2002). Along with these changes, apparent temperature extremes, which was not yet examined by Ouprasitwong (2002), may have increased in frequency, particularly during summertime. As noted by Gaffen and Rose (1998), hot weather can cause heat stress in humans, when humidity is high.
Therefore, these temperature extreme events may pose a public health problem, particularly as there are increasing numbers of elderly people, who are most vulnerable to heat-related sickness and mortality (Gaffen and Rose, 1998). Given the warming of surface temperature observed in this study continue in the near future, it can be expected that temperature extreme events such as those mentioned earlier will become more prevalent in Thailand. However, time will tell whether this expectation is correct, perhaps within a few decades when high-quality, long-term climate data with the time resolution suitable for analyzing a full range of temperature extremes become increasingly available.
The changes in temperature extremes observed to date have only recently been compared to the changes projected by models (Zwiers and Kharin, 1998; Easterling et al., 2000b; IPCC, 2001a). A number of changes in future weather and climate extremes from climate model have already observed in instrumental records in various parts of the world (Zwiers and Kharin, 1998; Easterling et al., 2000b; IPCC, 2001; Trenberth, 2001).
This includes warmer mean temperature associated with more extreme high temperatures and heat waves but fewer extreme low temperatures and a number of frost days and cold waves, along with reduced diurnal temperature range. Increases of daily minimum temperatures are projected to occur over the most land areas and are generally larger where snow and ice retreat, and a decreased number of frost days and cold wave are likely (Easterling et al., 2000b; Trenberth, 2001). Some current models have shown the future mean Pacific climate base state could more resemble an El Niño-like state, with increased greenhouse gases (e.g., Meehl and Washiton, 1996; Knutson and Manabe,
1995; Knustson et al., 1997; Timmerman et al., 1999). For such an El Niño-like state, or even for a uniform future warming of SSTs across the tropical Pacific, future seasonal temperature extreme associated with a given El Niño would be more intense in various parts of the world than present, due to the nonlinear relation between SST and the processes of evaporation and heat fluxes.
4.5. Possible biophysical and socio- economic impacts
Ecosystems, human health, and economy are all particularly sensitive to shot- and-long term changes in climate, including both the magnitude and rate of climate change. It is widely accepted that climate change represents an important determinant for natural and human systems, and have far-reaching consequences for a wide range of ecological systems (forests, grasslands, wetlands, rivers, lakes and marine environments) and human systems (agriculture, water resources, coastal resources, human health, financial institutions and human settlements). Biophysical systems often respond nonlinearly to climate forcings, owing to their complexity and sensitivity to external influences (Taylor et al., 2002). This leads to an exceptionally strong response of biophysical systems to even a weak, subtle climatic signal, as clearly evidenced from large and widespread responses across natural systems to relatively low average rates of global temperature increase (Walther et al., 2002; Parmesan and Yohe, 2003).
68

Available observational evidence indicates that regional changes in climate, particularly a recently prolonged increase in temperature, have already affected a diverse set of biophysical systems in many parts of the world. Examples of observed changes include shrinkage of glaciers, thawing of permafrost, sea level rise, freezing and earlier breakup of ice on rivers and lakes, lengthening of mid-to-high latitude growing seasons, poleward and altitudinal shifts of plant and animal ranges, increases/declines of some plant and animal populations and earlier flowing/breeding, emergence of insects and egg-laying in birds ( e.g., Grabherr et al., 1994; Barry et al., 1995; Anderson et al.,
1996; Parmesan, 1996; Shindler et al., 1996; Alward et al., 1999; Hughes, 2000;
McCarty, 2001; IPCC, 2001b; Walther et al., 2002; Parmesan and Yohe, 2003).
Furthermore, associations between changes in regional temperatures and observed changes in biophysical systems have been widely found in many aquatic and marine environments ( e.g., Strub et al., 1985; Goldman et al., 1989; Roemmich and McGowan,
1995; Feeland et al., 1997; McGowan et al., 1998; Reid et al., 1998; Chavez et al., 1999;
Gerten and Adrian, 2000; Levitus et al., 2000; Edwards et al., 2001). Changes in the distribution and abundance of several mobile marine organisms off the coast of
California in response to short-and-long term variations in sea surface temperature accompany El Niño events and global warming have been well particularly documented over the past few decades ( e.g., Chelton et al., 1982; Barber and Chavez, 1983;
Roemmich and McGowan, 1995; McGowan et al., 1998; Rebstock, 2002). The surface waters of the California Current warmed 1.2-1.6

C between 1951 and 1993; this warming was accompanied by a 70% decline in zooplankton abundance, possibly because of increased surface temperatures reducing the upwelling of cold, nutrient-rich waters to the surface (Roemmich and McGowan, 1995). Several highly vulnerable, diverse and productive coastal ecosystems such as coral reefs are seriously endangered by increased sea temperature (Glynn, 1991; Brown, 1997; Hoegh-Guldberg, 1999;
Spencer et al., 2000). The increasing incidence of coral reef bleaching may be a consequence of recent rise in global ocean temperature (Glynn, 1991; Brown, 1997;
Hoegh-Guldberg, 1999; Goreau et al., 2000; Spencer et al., 2000). Six periods of mass coral bleaching have occurred since 1979, and the incidence of mass coral bleaching is increasing in both frequency and intensity (Hoegh-Guldberg, 1999; Goreau et al., 2000).
The most significant mass bleaching was associated with the 1997-98 ENSO event, when all ten reef provinces of the world were affected (Hoegh-Guldberg, 1999; Goreau et al., 2000; Spencer et al., 2000). In some areas, most notably the Indian Ocean, this event was followed by mass mortality where up to 90% of all the corals died over thousands of square kilometers (Goreau et al., 2000; Spencer et al., 2000). Moreover, ongoing temperature change is additional source of stress for species already threatened by local and global environmental changes, increasing the risk of extinction (McCarty,
2001). On the basis of the collective evidence which reviews large-scale, consistent changes in the biophysical systems across diverse localities and/or regions with the expected of regional changes in temperature, and taking into account the effects of other factors such as land-use changes and pollution, IPCC concluded with high confidence that recent regional changes in temperature have had discernible impacts on many biophysical systems.
Human systems that are sensitive to climate change include mainly water resources, agriculture, human settlements and health, coastal zones and energy. The vulnerability of these systems varies with geographic location, time and social, economic and environmental conditions. Nevertheless, all regions are likely to experience some adverse effects of climate change. Based on models and other studies,
69

projected adverse impacts of increased temperature include: 1) a general reduction in potential crop yields in most tropical and sub-tropical regions, 2) an increase in heat stress mortality, 3) a widespread increase in the risk of flooding for many human settlements from sea level rise, and 4) increased energy demand for space cooling due to higher summer temperature (IPCC, 2001b). The ability of human systems to adapt to and cope with global warming depends upon factors such as wealth, technology, education, information, skills, infrastructure, assess to resources, and management capabilities. Population and communities are highly variable in their endowments with these attributes, and the developing countries, especially the least developed countries, are generally poorest in this regards. As a consequence, they have lesser capacity to adapt and are more vulnerable to global warming damages, just as they are more sensitive to other stresses. This condition is most extreme among the poorest people. In light of the above, the effects of global warming are expected to be greatest in developing countries, particularly those in Africa and Asia, in terms of loss of life and relative effects on investment and the economy.
The anticipated sea level rise resulting from global warming may threaten many coastal settlements and islands. The extent of a sea level rise is still uncertain, but the latest estimates are in the range 10-94 cm by the year 2100 (IPCC, 2001a). Because heat penetrates slowly into the voluminous oceans, a sea level rise is expected to be manifest over a longer period of time than temperature change. Even if anthropogenic greenhouse gases emissions are stabilized immediately, sea level will continue rise for centuries with serious consequences for millions of people. Countries in the Northwestern Pacific and East Asia sub-regions and the Pacific Island countries including Thailand may be particularly vulnerable to sea level rise because many of their human settlements and industrial facilities are located in coastal or lowland areas (IPCC, 2001a,b). Initial analysis suggests that a sea level rise of even several inches will result in aggravation of existing coastal erosion, loss of low-lying coastal lands, depletion of mangrove swamps, and a high risk of fresh water contamination due to saline intrusion (IPCC, 2001a,b).
These changes in turn would affect agriculture, tourism, fishery production, urban systems, industrial complexes and other kinds of infrastructure on the coastline. Based the scenario of Hoffman (1984), which assumes continued increases in greenhouse gases at low, medium and high rates, sea level of the Chao Praya Delta in the Gulf of
Thailand was predicted to rise in the range of 1-3 m by the year 2100 (Jarupongsakul,
1998). The alarming rise projected by the model indicates that the present tidal zone, which has an average elevation of about one meter and is occupied by mangrove swamp, would be submerged by the next century; and submergence will be accelerated by land subsidence due largely to ground water extraction (Jarupongsakul, 1998). These impacts are likely to exacerbate the problems already threatened in the Gulf of Thailand, including erosion and inundation of land, environmental deterioration and pollution, seawater intrusion and flooding from storm surges.
Potential changes in the frequency, intensity and persistent of temperature extreme events are also emerging as another key determents of natural and human systems. Recent years have seen a number of weather events including temperature extremes cause large losses of life as well as a substantial increase in economic loss from weather hazards (Karl and Knight, 1997; Pielke and Landsea, 1998). Heat waves are one of weather hazards that cause an increase in mortality (Changnon et al., 1996).
In the United States, for instance, heat-wave deaths were exceptionally high in 1980,
1988, and 1995 (Changnon et al., 1996). Projected climate change will be accompanied
70

by an increase in heat waves, often exacerbated by increased humidity and urban air pollution, which would cause an increase in heat-related deaths and illness episodes
(IPCC, 2001b). The evidence indicates with high confidence that the impacts would be greatest in urban populations, affecting particularly the elderly, sick, and those without access to air-conditioning (IPCC, 2001b). In addition, temperature extremes could be tremendously devastating to ecosystems. Several apparently gradual biological changes are linked to response to a few, brief, temperature extreme events (Easterling et al.,
2000b). Although some changes arising from extreme weather may be benign or even beneficial, the economic effects of these short-lived events will be substantial and clearly warrant attention in policy debates.
5. Implications for future research
There is still an obvious disparity in scientific knowledge to contribute the better understanding of climate change and its impacts between developed and developing countries. Advances have been made in most of the industrial countries to detect changes in biophysical systems, and steps have been taken to improve the public understanding of adaptive capacity, vulnerability to climate change, and other critical impact-related issue. In many developing countries where human, financial, natural resources as well as institutional and technological capability are relatively limited, on the other hand, climate change is not regarded as the same light. This situation is particularly true for Thailand where climate studies are rare, and therefore very little is known about climate change and its consequences. Hence, this study provides a vital clue to better understand some key aspects of short-and-long term climate change in
Thailand, and builds more bridges to narrow the great gaps of scientific knowledge that have important implications for climate research and prediction, and environmental management and conservation on regional/local scales in the future. This study can also contribute to preparation for future challenges by providing valuable data for model construction and validation. However, there are several key issues that need to be addressed in details and should be emphasized in the future research studies. “What are the potential impacts for natural and human systems of continued global warming and intense and frequent ENSO event” stands out as one of several key questions.
Answering this question is a necessary step for establishing adaptation strategy to reduce adverse impacts of climate change and to enhance beneficial impacts. Since the evidence indicates that changes in regional/global climate will likely continue and even accelerate over the next 50-100 years (IPCC, 2001a), such knowledge is a special need in the near future to strengthen future assessments and to reduce uncertainties to assure that sufficient information is available for policymaking about responses to possible consequences of climate change.
Climate scientific research, by nature, encompasses many disciplines, and is not, and can not possibly be the responsibility of one single institution, ministry or agency.
Climate-related studies in Thailand, for instance, have been or are being conducted by scientists in many related but different disciplines in many agencies such as, Ministry of
National Resources and Environment, Ministry of Agriculture and Cooperatives,
Ministry of Information and Communication Technology, universities, and Southeast
Asia Regional Center. Advances for what is not well known and what is uncertainties about climate change and its regional impacts, therefore, at best a compromise achieved through a concerted effort of all these sectors involved. This situation calls for strong national cooperation and coordination between government agencies and universities
71

and other sectors, together with international collaboration, with a view to attaining consensus on how regional assessment of impacts, vulnerability and adaptation to achieve programmatic and logistical efficiencies and effectiveness.
Most of climate-related studies generally require high-quality, long-term atmospheric/biological data with the time resolution appropriate for analyzing and evaluating the processes and mechanisms in question. However, a lack of such necessary data is common obstacles for climate change research. Only few historical data sets which have collected for a long period exist in Thailand, but it is difficult to access them. This is probably because these data have been collected originally through monitoring/research programs of individual scientists and/or a variety of organizations, and most of them are not in digital forms. Moreover, to date, there are no central access points where one can retrieve those available data. Since data gathering processes are costly and often conducted in a piece-meal fashion, implementing agencies should promote the free exchange of data in order to maximize the benefit from them. The
IOC/WESTPAC initiative to provide free access to different collections of oceanographic data from the Gulf of Thailand region is a good example of what can be achieved (Snidvongs, 1997). Because the historical data sets are relatively rare, establishment of new baseline monitoring programs for some key environments will also be important.
A last important consideration is the potential value of a number of tools and techniques such as those introduced in this study that have high capabilities for handling metadata and separating climatic signals blended together or hidden inside the background climate variability or “noise”.
All in all, there is little doubt that climate change is an active and critical component of “our integrated Earth System” as current and future threats for human and environmental systems that is now happening on global, regional and local scales, and will likely to continue or even intensify in the near future. The challenge of mitigating and adapting its potential risks as well as ensuring a sustainable future is daunting and it is immediate. It can be met, but only with a new or even vigorous integrated, multidisciplinary approach to studying a full range of climate change and its consequences and uncertainties.
6. Acknowledgements
I am especially grateful to the Meteorological Department of Thailand for generously providing high-quality, long-term atmospheric data. To my colleagues at the
Water Research and Technology Development Section, The Environmental Research and Training Center, for providing me an unlimited supply of their help, warmth and friendship and a really excellent working atmosphere, I express my sincere thanks.
Finally, I delicate this work to my beloved parents and wife.
72

References
T
Alward, R.D., Detling, J.K., Milchunas, D.G., 1999. Gassland vegetation changes and nocturnal global warming. Science, 283, 229-231.
Anderson, W.L., Robertson, D.M., Magnuson, J.J., 1996. Evidence of recent warming and El Niño-related variations in ice breakup of Wisconsin lakes. Limnology and
Oceanography, 41, 815-821.
Barber, R.T., Chavez, F.P., 1983. Biological consequences of El Niño. Science, 222,
1203-1210.
Barlow, M., Cullen, H., Lyon, B., 2002. Drought in central and southwest Asia: La Niña, the warm pool, and Indian ocean precipitation. Journal of Climate, 15, 697-700.
Barry, J.P., Baxter, C.H., Sangarin, R.D., Gilman, S.E., 1995. Climate-related, longterm faunal changes in the California rocky intertidal community. Science, 267, 672-
675.
Boyle, J.S., 1998. Evaluation of the annual cycle of precipitation over the United States in GCMs : AMIP simulations. Journal of Climate, 11, 1041-1055.
Brishavana, M., Ouprasitwong, N., 2002. Rainfall and temperature patterns in Thailand during ENSO events and links with the Southern Oscillation Index and sea surface temperature in the tropical Pacific ocean (in Thai with English abstract). Technical
Document No. 551.524-02-2002, Meteorological Department of Thailand, 142 pp.
Brown, B.E., 1997. Coral bleaching: Causes and consequences. Coral Reefs, 16, 129-
138.
Cattell, R.B., 1966. The scree test for the number of factors. Multivariate Behavioral
Research, 1, 245-276.
Changnon, S.A., Kunkel, K.E., Reinke, B.C., 1996. Impacts and responses to the 1995 wave: A call to action. Bulletin of the American Meteorological Society, 77, 1497-
1506.
Changnon, S.A., Pielke, R.A., Changnon, D., Sylves, R.T., Pulwarty, R., 2000. Human factors explain the increased losses from weather and climate extremes. Bulletin of the American Meteorological Society, 81, 437-442.
Chapman, W.L., Walsh, J.E., 1993. Recent variations of sea-ice and air temperatures in high latitudes. Bulletin of the American Meteorological Society, 74, 33-47.
Chatfield, C., 1989. The analysis of time series: An introduction. Chapman and Hall,
London, England.
Chavez, F.P., Strutton, P.G., Friederich, G.E., Feely, R.A., Feldman, G.C., Foley, D.G.,
McPhaden, M.J., 1999. Biological and chemical response of the equatorial Pacific
Ocean to the 1997-98 El Niño. Science, 286, 2126-2131.
Chelton, D.B., 1983. Effects of sampling errors in statistical estimation.
Deep-Sea
Research I, 30, 1083-1103.
Chelton, D.B., Bernal, P.A., McGowan, J.A., 1982. Large-scale interannual physical and biological interaction in the California Current. Journal of Marine Research, 40,
1095-1125.
73

Cooter, E., LeDuc, S., 1995. Recent frost date trends in the northern United States.
International Journal of Climatology, 15, 65-75.
Corti, S., Molten, F., Palmer, T.N., 1999. Signature of recent climate change in frequencies of natural atmospheric circulation regimes. Nature, 398, 799-802.
Davis, R.E., 1976. Predictability of sea surface temperature and sea level pressure anomalies over the North Pacific Ocean. Journal of Physical Oceanography, 6, 249-
266.
DeGaetano, A., 1996. Recent trends in maximum and minimum temperature threshold exceedences in the northern United States. Journal of Climate, 9, 1646-1657.
Demaree, G.R., Nicolis, C., 1990. Onset of Sehelian drought viewed as a fluctuationinduced transition. Quarterly Journal Royal Meteorology Society, 116, 221-238.
Dunteman, G.H., 1989. Principal components analysis. Sage Publication Inc, California,
USA, 96 pp.
Easterling, D.R., Evans, J.L., Groisman, P.Y., Karl, T.R., Kunkel, K.E., Ambenje, P.,
2000a. Observed variability and trends in extreme climate events: A brief review.
Bulletin of the American Meteorological Society, 81, 417-425.
Easterling, D.R., Horton, B., Jones, P.D., Peterson, T.C., Karl, T.R., Parker, D.E.,
Salinger, M.J., Razuvayev, V., Plummer, N., Jamason, P., Folland, C.K., 1997.
Maximum and minimum temperature trends for the globe. Science, 277, 364-367.
Easterling, D.R., Meehl, G.A., Parmesan, C., Changnon, S.A., Karl, T.R., Mearns, L.O.,
2000b. Climate extremes: Observations, modeling, and impacts. Science, 289, 2068-
2074.
Edwards, M., Reid, P., Planque, B., 2001. Long-term and regional variability of phytoplankton biomass in the Northeast Atlantic (1960-1995). ICES Journal of
Marine Science, 58, 39-49.
Emery, W.J., Thomson, R.E., 1997. Data analysis methods in physical oceanography.
Pergamon Press, New York, USA, 634 pp.
Enfield, D.B., Mestas-Nuñez, A.M., 1999. Multiscale variabilities in global sea surface temperatures and their relationships with tropospheric climate patterns. Journal of
Climate, 12, 2719-2733.
Eskridge, R.E., Ku, J.Y., Rao, S.T., Porter, P.S., Zurbenko, I.G., 1997. Separating different scales of motion in time series of meteorological variables. Bulletin of the
American Meteorological Society, 78, 1473-1483.
Fedorov, A.V., Philander, S.G., 2000. Is El Niño changing? Science, 288, 1997-
2002.
Fields, P.A., Graham, J.B., Rosenblatt, R.H., Somero, G.N., 1993. Effects of expected global climate change on marine faunas. Trends in Ecology and Evolution, 8, 361-
367.
Franklin, J.N., 1968. Matrix theory. Prentice-Hall, Englewood Cliffs, New Jerseys, USA.
Freeland, H., Denman, K., Wong, C.S., Whitney, F., Jacques, R., 1997. Evidence of change in the winter mixed layer in the Northeast Pacific Ocean. Deep-Sea Research
I, 44, 2117-2129.
74

Gaffen, D.J., Ross, R.J., 1998. Increased summertime heat stress in the US. Nature, 396,
529-530.
Gaffen, D.J., Ross, R.J., 1999. Climatology and trends of U.S. surface humidity and temperature. Journal of Climate, 12, 811-828.
Gallo, K.P., Easterling, D.R., Peterson, T.C., 1996. The influence of land use/land cover on climatological values of the diurnal temperature range. Journal of Climate, 9,
2941–2944.
Gerten, D., Adrian, R., 2000. Climate-driven changes in spring plankton dynamics and the sensitivity of shallow polymictic lakes to the North Atlantic Oscillation.
Limnology and Oceanography, 45, 1058-1066.
Gill, A.E., Rasmusson, E.M., 1983. The 1982-83 climate anomaly in the equatorial
Pacific. Nature, 306, 229-234.
Goldman, C.R., Jassby, A., Powell, T., 1989. Interannual fluctuations in primary production: Meteorological forcing at two subalpine lakes. Limnology and
Oceanography, 43, 310-323.
Goreau, T., McClanahan, T., Hayes, R., Strong, A.E., 2000. Conservation of coral reefs after the 1998 global bleaching event. Conservation Biology, 14, 5-15.
Grabherr, G., Gottfried, M., Pauli, H., 1994. Climate effects on mountain plants. Nature,
369, 448.
Graham, N. E., 1994. Decadal–scale climate variability in the tropical and North Pacific during the 1970s and 1980s: Observations and model results. Climate Dynamics, 10,
135-162.
Gu, D., Philander, S.G.H., 1995. Secular changes of annual and interannual variability in the tropics during the past century. Journal of Climate, 8, 864-876.
Gylnn, P.W., 1991. Coral reef bleaching in the 1980s and possible connection with global warming. Trends in Ecology and Evolution, 6, 175-179.
Hanawa, K., Yoshikawa, Y., Watanabe, T., 1989a. Composite analyses of wintertime wind stress vector fields with respect to SST anomalies in the western North Pacific and the ENSO events : Part I SST composite. Journal of Meteorological Society of
Japan, 67, 385-399.
Hanawa, K., Yoshikawa, Y., Watanabe, T., 1989b. Composite analyses of wintertime wind stress vector fields with respect to SST anomalies in the western North Pacific and the ENSO events: Part II ENSO composite. Journal of Meteorological Society of
Japan, 67, 833-845.
Heino, R., Brázdil, R., Førland, E., Tuomenvirta, H., Alexandersson, H., Beniston, M.,
Pfister, C., Rebetez, M., Rosenhagen, G., Rösner, S., Wibig, J., 1998
. Progress in the study of climatic extremes in Northern and Central Europe. Climatic Change, 42,
151-181.
Hoegh-Guldberg, O., 1999. Climate change, coral bleaching and the future of the world’s coral reefs. Marine and Freshwater Research, 50, 839-866.
Hoerling, M., Kumar, A., 2003. The perfect ocean for drought. Science, 299, 691-694.
Hoffman, J.S., 1984. Estimate of future sea-level rise. In Greenhouse effect and sealevel rise, Barth and Titus (eds.), New York, USA, 79-103 pp.
75

Horel, J.D., 1984. Complex principal component analysis: Theory and example. Journal of Applied Meteorology, 23, 1660-1673.
Horel, J.D., Wallace, J.M., 1981. Planetary-scale atmospheric phenomena associated with the southern oscillation. Monthly Weather Review, 109, 813-829.
Huber, M., Caballero, R., 2003. Eocene El Niño: Evidence for robust tropical dynamics in the “Hothouse”. Science, 299, 877-881.
Hughes, L., 2000. Biological consequences of global warming: Is the signal already.
Trends in Ecology and Evolution, 15, 56-61.
Hurrell, J.W., 1995. Decadal trends in the North Atlantic Oscillations: Regional temperatures and precipitation. Science, 269, 676-679.
Hurrell, J.W., 1996. Influence of variations in extratropical teleconnections on northern hemisphere temperature. Geophysical Research Letters, 23, 665-668.
Hurrell, J.W., Loon, H.V., 1997. Decadal variations in climate associated with the North
Atlantic Oscillation. Climatic Change, 36, 301-326.
IGBP, 2001a. IGBP Science No. 4, Global change and the Earth system: A planet under pressure. Steffen, W., Tyson, P., Eds., 32 pp.
IGBP, 2001b. IGBP Science No. 3, Environmental variability and climate change. Eliott,
S., Ed., 31 pp.
International Research Institute for Climate Prediction. Overview of the ENSO system.
Web site linking science to society, http://iri.columbia.edu/climate/ENSO.
IPCC, Climate change 2001a: The scientific basis. Houghton, J.T., Ding, Y., Griggs,
D.J., Noguer, M., van der Linden, P.J., Dai, x., Maskell, K., Johnson, C.A., Eds., 944 pp.
IPCC, Climate change 2001b: Impacts, adaptation and Vulnerability. McCarthy, J.J.,
Canziani, O.F., Leary, N.A., Dokken, D.J., White, K.S., Eds, 1005 pp.
Janicot, S., Moron, V., Fontaine, B., 1996. Sahel droughts and ENSO dynamics.
Geophysical Research Letters, 23, 515-518.
Jarupongsakul, T., 1998. The implications of sea-levelrise and flooding for the Gulf of
Thailand region. In SEAPOL integrated studies of the Gulf of Thailand, Volume 1,
Southeast Asian Program in Ocean and Law, Policy and Management, Johnston,
D,M., (Ed.), 97-129.
Jassby, A.J., Powell, T.M., 1990. Detecting changes in ecological time series. Ecology,
71, 2044-2052.
Karl, T.R., Knight, R.W., 1997. The 1995 Chicago heat wave: How likely is a recurrence? Bulletin of the American Meteological Society, 78, 1107-1119.
Karl, T.R., Knight, R.W., 1998. Secular trends in precipitation amount, frequency, and intensity in the United States. Bulletin of the American Meteorological Society, 79,
231-241.
Kasting, J.F., 1993. Earth’s early atmosphere. Science, 259, 920-926.
Kiladis, G.H., Loon, H.V., 1988. The southern oscillation: Part VII Meteorological anomalies over the Indian and Pacific sectors associated with the extreme of the oscillation. Monthly Weather Review, 116, 120-136.
76

Knutson, T.R., Manabe, S., 1995. Time-mean response over the tropical Pacific to increased CO
B
2
B
in a coupled ocean-atmosphere model. Journal of Climate, 8, 2181-
2199.
Knutson, T.R., Manabe, S., Gu, D., 1997. Simulated ENSO in a global coupled oceanatmosphere model: multidecadal amplitude modulation and CO
B
2
B
sensitivity. Journal of Climate, 10, 138-161.
Kumar, A., Leetmaa, A., Ji, M., Simulations of atmospheric variability induced by sea surface temperatures and implications for global warming. Science, 266, 632-634.
Lagerloef, G.S.E., Bernstein, R.L., 1988. Empirical Orthogonal Function analysis of
Advanced Very High Resolution Radiometer Surface Temperature patterns in Santa
Barbara channel. Journal of Geophysical Research, 93, 6863-6873.
Lanzante, J.R., 1996. Resistant, robust and non-parametric techniques for the analysis of climate data: Theory and examples, including applications to historical radiosonde station data. International Journal Climatology, 16, 1197-1226.
Leathers, D.J., Yarnal, B., Palecki, M.A., 1991. The Pacific/North American teleconnection pattern and United States climate: Part I Regional temperature and precipitation associations. Journal of Climate, 4, 517-528.
Levitus, S., Antonov, J.I., Boyer, T.P., Stephens, C., 2000. Warming of the world ocean.
Science, 287, 2225-2229.
Li, C., 1990. Interaction between anomalous winter monsoon in East Asia and El Niño events. Advances in Atmospheric Sciences, 7, 36-46.
Li, C., 1990. Interaction between anomalous winter monsoon in East Asia and El Niño events. Advances in Atmospheric Sciences, 7, 36-46.
Limsakul, A., Saino, T., Midorikawa, T., Goes, J.I., 2001. Temporal variations in lower trophic level biological environments in the northwestern North Pacific Subtropical
Gyre from 1950 to 1997. Progress in Oceanography, 49, 129-149.
Loon, H.V, Roger, J.C., 1978. The seesaw in winter temperatures between Greenland and northern Europe: Part I General description. Monthly Weather Review, 106, 296-
310.
Mann, M.E., Bradley, R.S., Hughes, M.K., 1999. Northern hemisphere temperatures during the past millennium: Inferences, uncertainties, and limitations. Geophysical
Research Letters, 26, 759-762.
Mantua, N. J., Hare, S. R., Zhang, Y., Wallace, J. M., Francis, R. C., 1997. A Pacific
Interdecadal Climate Oscillation with impacts on salmon production.
Bulletin of the
American Meteorological Society, 78, 1069-1079.
McCarty, J.P., 2001. Ecological consequences of recent climate change. Conservation
Biology, 15, 320-331.
McGowan, J. A., Cayan, D. R., Dorman, L. M., 1998. Climate-Ocean variability and ecosystem response in the Northeast Pacific. Science, 281, 210-217.
McPhaden, M.J., 1999. Genesis and evolution of the 1997-98 El Niño. Science, 283,
950-954.
77

Mearns, L.O., Giorgi, F., McDaniel, L., Shields, C., 1995. Analysis of daily variability of precipitation in a nested regional climate model: Comparison with observations and doubled CO
B
2
B
results. Global Planetary Change, 10, 55-78.
Meehl, G.A., Washington, W.M., 1996. El Niño-like climate change in a model with increased atmospheric CO
B
2
B
concentration. Nature, 382, 56-60.
Nitta, T., Yamada, S., 1989. Recent warming of tropical sea surface temperature and its relationship to the northern hemisphere circulation, Journal of Meteorological Society of Japan, 67, 375-383.
Osborn, T.L., 1997. Areal and point precipitation intensity changes: Implication for the application of climate model. Geophysical Research Letters, 24, 2829-2832.
Osborn, T.L., Hulme, M., 1997. Development of a relationship between station and grid-box rainday frequencies for climate model evaluation. Journal of Climate, 10,
1885-1908.
Ottersen, G., Planque, B., Belgrano, A., Post, E., Reid, P.C., Stenseth, N.C., 2001.
Ecological effects of the North Atlantic Oscillation. Oecologia, 128, 1-14.
Ouprasitwong, N., 2002. Indices and trends in extreme rainfall and temperature in
Thailand (in Thai with English abstract). Technical Document No. 551. 577.3-01-
2002, Meteorological Department of Thailand, 44 pp.
Parmesan, C., 1996. Climate and species’ range. Nature, 382, 765-766.
Parmesan, C., Yohe, G., 2003. A globally coherent fingerprint of climate change impacts across natural systems. Nature, 421, 37-42.
Peixoto, J.P., Oort, A.H., 1992. Physics of climate. American Institute of Physics, New
York, USA, 520 pp.
Petit, J.R., Jouzel, J., Raynaud, D., Barkov, N.I., Barnola, J.-M., Basile, I., Bender, M.,
Chappellaz, J., Davis, M., Delaygue, G., Delmotte, M., Kotlyakov, V.M., Legrand,
M., Lipenkov, V.Y., Lorius, C., Pépin, L., Ritz, C., Saltzman, E., Stievenard, M.,
1999. Climate and atmospheric history of the past 420,000 years from the Vostok ice core, Antarctica. Nature, 399, 429-436.
Petitt, A.N., 1979. A non-parametric approach to the change point problem. Applied
Statistics, 28, 126-135.
Philander, S.G., 1990. El Niño, La Niña and the Southern Oscillation. Academic Press,
New York, 293 pp.
Philander, S.G.H., 1983. El Niño Southern Oscillation phenomena. Nature, 302,
295-301.
Pielke, R.J., Landsea, C., 1998. Normalized Atlantic hurricane damage: 1925-1995.
Weather Forecasting, 12, 621-631.
Plummer, N., Salinger, M.J., Nicholls, N., Suppiah, R., Hennessy, K.J., Leighton, R.M.,
Trewin, B., Page, C.M., Lough, J.M., 1999. Changes in climate extremes over the
Australian region and New Zealand during the twentieth century. Climatic Change,
42, 183-202.
Preisendorfer, R.W., 1988. Principal component analysis in meteorology and oceanography. Development in Atmospheric Science 17, Elsevier, New York, USA,
419 pp.
78

Prospero, J.M., Nees, R.T., 1986. Impact of the North African drought and El Niño on mineral dust in the Barbados trade winds. Nature, 320, 735-738.
Qian, W., Zhu, Y., 2001. Climate change in China from 1880 to 1998 and its impact on the environmental condition. Climatic Change, 50, 419-444.
Quinn, W.H., Neal, V.T., 1984. Recent climate change and the 1982-83 El Niño. In proceeding of the 8
P th
P
Annual Climate Diagnosis Workshop, Downsvile, ON, NOAA,
Canada, 148-154.
Quinn, W.H., Neal, V.T., 1985. Recent long-term climate change over the eastern tropical and subtropical Pacific and its ramifications. In proceeding of the 9
P th
P
Annual
Climate Diagnosis Workshop, Corvallis, OR, NOAA, 101-109.
Rasmusson, E.M., Wallace, J.M., 1983. Meteorological aspects of the El Niño/Southern
Oscillation. Science, 222, 1195-1202.
Rebstock, G.A., 2002. Climatic regime shifts and decadal-scale variability in calanoid copepod populations off southern California. Global Change Biology, 8, 71-89.
Reid, P.C., Edwards, M., Hunt, H.G., Warner, A.J., 1998. Phytoplanton change in the
North Atlantic. Nature, 391, 546.
Roemmich, D., McGowan, J.A., 1995. Climatic warming and the decline of zooplankton in the California Current. Science, 267, 1324-1326.
Ropelewski, C.F., Jones, P.D., 1987. An extension of the Tahiti-Darwin southern oscillation index. Monthly Weather Review, 115, 2161-2165.
Rungdilokroaja, V., Nimma, S., 1990. The variation of annual rainfall and surface temperature over Thailand. In proceeding of the Workshop on the Global Changes,
National Research Council of Thailand, Bangkok, Thailand.
Schindler, D.W., Bayley, S.E., Parker, B.R., Beaty, K.G., Cruikshank, D.R., Fee, E.H.,
Schindler, E.U., Stainton, M.P., 1996. The effects of climatic warming on the properties of boreal lakes and streams at the Experimental lakes area. Limnology and
Oceanography, 41, 1004-1017.
Schlesinger, M.E., Ramankutty, N., 1994. An oscillation in the global climate system of period 65-70 years. Nature, 367, 723-726.
Snidvongs, A., 1997. The oceanography of the Gulf of Thailand: Research and management priorities. Paper presented at the SEAPOL Gulf of Thailand Project,
First Meeting of Experts, Hua Hin, Thailand.
Spencer, T., Teleki. T.A., Bradshaw, C., Spalding, M., 2000. Coral bleaching in the southern Seychelles during the 1997-1998 Indian Ocean warm event. Marine
Pollution Bulletin, 40, 569-586.
Steadman, R.G., 1984. A universal scale of apparent temperature. Journal of Applied
Meteorology, 23, 1674-1687.
Strub, P.T., Powell, T., Goldman, C.R., 1985. Climatic forcing: Effects of El Niño on small, temperate lake. Science, 227, 55-57.
Sun, D.Z., Trenberth, K.E., 1998. Coordinated heat removal from the equatorial Pacific during the 1986-87 El Niño. Gophysical Research Letters, 25, 2659-2662.
Suppiah, R., Hennessy, K.J., 1998. Trends in total rainfall, heavy rain events and number of dry days in Australia. International Journal of Climatology, 10, 1141-1164.
79

Taylor, A.H., Allen, J.I., Clark, P.A., 2002. Extraction of a weak climatic signal by an ecosystem. Nature, 416, 629-632.
Timmermann, A., Oberhuber, J., Bacher, A., Esch, M., Latif, M., Roeckner, E., 1999.
Increased El Niño frequency in a climate model forced by future greenhouse warming. Nature, 398, 694-697.
Trenberth, K. E., 1990. Recent observed interdecadal climate changes in the northern hemisphere. Bulletin of the American Meteorological Society, 71, 988-993.
Trenberth, K. E., Hurrell, J. W., 1994. Decadal atmosphere-ocean variations in the
Pacific. Climate Dynamics, 9, 303-319.
Trenberth, K.E., 1984. Signal versus noise in the southern oscillation. Monthly Weather
Review, 112, 326-332.
Trenberth, K.E., 1997. The definition of El Niño. Bulletin of the American
Meteorological Society, 78, 2771-2777.
Trenberth, K.E., 2001. Strong evidence of human influences on climate: The 2001 IPCC assessment. Environment, 43, 8-19.
Trenberth, K.E., Hoar, T.J., 1996. The 1990-1995 El Niño-Southern Oscillation event:
Longest on record. Geophysical Research Letters, 23, 57-60.
Troup, A.J., 1965. The southern oscillation. Quarterly Journal Royal Meteorology
Society, 91, 490-506.
Venegas, S.A., 2001. Statistical methods for signal detection in climate, Technical report, 96 pp.
Wallace, J. M., Gutzler, D. S., 1981. Teleconnections in the geopotential height field during the northern hemisphere winter. Monthly Weather Review, 109 , 784-811.
Wallace, J.M., Dickinson, R.E., 1972. Empirical orthogonal representation of time series in the frequency domain: Part II Application to the study of tropical wave disturbance.
Journal of Applied Meteorology, 11, 893-900.
Walther, G.-R., Post, E., Convey, P., Menzel, A., Parmesan, C., Beebee, T.J.C.,
Fromentin, J.-M., Hoegh-Guldberg, O., Bairlein, F., 2002. Ecological responses to recent climate change. Nature, 416, 389-395.
Wang, B., 1995. Interdecadal changes in El Niño onset in the last four decades. Journal of Climate, 8, 267-285.
Wang, B., An, S.I., 2002. A mechanism for decadal changes of ENSO behavior: Roles of background wind changes. Climate Dynamics, 18, 475-486.
Wang, B., Wang, Y., 1996. Temporal structure of the Southern Oscillation as revealed by waveform and wavelet analysis. Journal of Climate, 9, 1585-1598.
Wang, J.X.L., Gaffen, D.J., 2001. Late-twentieth-century climatology and trends of surface humidity and temperature in China. Journal of Climate, 14, 2833-2845.
Wang, W., Li, K., 1990. Precipitation fluctuation over semiarid region in Northern
China and the relationship with El Niño/Southern Oscillation. Journal of Climate, 3,
769-783.
Ware, D.M., Thomson, R.E., 2000. Interannual to multidecadal timescale climate variations in the Northeast Pacific. Journal of Climate, 13, 3209-3220.
80

Wearn, B.C., Nasstrom, J.S., 1982. Examples of extended empirical orthogonal function analyses. Monthly Weather Review, 110, 481-485.
Wilkison, J.H., 1965. The algebraic eigenvalue problem. Clarendon Press, Oxford,
England.
Wilks, D.S., 1995. Statistical methods in atmosphere science: An introduction.
Academic Press, 467 pp.
Wolter, K., Timlin, M.S., 1993. Monitoring ENSO in COADS with a seasonally adjusted principal component index. In proceeding of the 17
P th
P
Climate Diagnosis
Workshop, OK, NOAA/N MC/CAC, NSSL, Oklahoma Climate Survey. CIMMS and the School of Meteorology, University of Oklahoma, 52-57.
Wolter, K., Timlin, M.S., 1998. Measuring the strength of ENSO – how does 1997/98 rank? Weather, 53, 315-324.
Yamagata, T., Masumoto, Y., 1992. Interdecadal natural climate variability in the western Pacific and its implication in global warming. Journal of Meteorological
Society of Japan, 70, 167-175.
Yasunari, T., Seki, Y., 1992. Role of the Asian monsoon on the interannual variability of the global climate system. Journal of the Meteorological Society of Japan, 70, 177-
189.
Zahn, R., 2003. Monsoon linkages. Nature, 421, 324-325.
Zhai, P., Sun, A., Ren, F., Liu, X., Gao, B., Zhang, Q., 1999. Changes of climate extremes in China. Climatic Change, 42, 203-218.
Zhang, Y., Wallace, J.M., Battisti, D.S., 1997. ENSO-like interdecadal variability:
1900-93. Journal of Climate, 10, 1004-1020.
Zwiers, F.W., Kharin, V. V., 1998. Changes in the extremes of the climate simulated by
CCC GCM2 under CO
B
2
B
doubling. Journal of Climate, 11, 2200–2222.
81