Evaluating the impacts of climate variability on deciduous forest using remote sensing

The archived information of vegetation phenology changes over large spatial and temporal extents and its response to climate variability will help to address many pressing issues such as

EVALUATING THE IMPACTS OF CLIMATE VARIABILITY ON DECIDUOUS FOREST

USING REMOTE SENSING

MRS PHAN KIEU DIEM ID: 57300800201

A THESIS SUBMITTED AS A PART OF THE REQUIREMENTS

FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

IN ENVIRONMENTAL TECHNOLOGY

THE JOINT GRADUATE SCHOOL OF ENERGY AND ENVIRONMENT

AT KING MONGKUT’S UNIVERSITY OF TECHNOLOGY THONBURI

2 ND SEMESTER 2017

COPYRIGHT OF THE JOINT GRADUATE SCHOOL OF ENERGY AND ENVIRONMENT

Evaluating the Impacts of Climate Variability on Deciduous Forest using Remote Sensing

Mrs Phan Kieu Diem ID: 57300800201

A Thesis Submitted as a Part of the Requirements for the Degree of Doctor of Philosophy

in Environmental Technology

The Joint Graduate School of Energy and Environment

at King Mongkut’s University of Technology Thonburi

Thesis Title: Evaluating the Impacts of Climate Variability on Deciduous Forest using

Remote Sensing

Student’s name, organization and telephone/fax numbers/email

Mrs Phan Kieu Diem

The Joint Graduate School of Energy and Environment (JGSEE)

King Mongkut’s University of Technology Thonburi (KMUTT)

126 Pracha Uthit Rd., Bangmod, Tungkru, Bangkok 10140 Thailand

Telephone: 0-9676-25068

Email: pkdiem@ctu.edu.vn

Supervisor’s name, organization and telephone/fax numbers/email

Assoc Prof Dr Amnat Chidthaisong

The Joint Graduate School of Energy and Environment (JGSEE)

King Mongkut’s University of Technology Thonburi (KMUTT)

126 Pracha Uthit Rd., Bangmod, Tungkru, Bangkok 10140 Thailand

Telephone: 0-8144-04925

Email: amnat@jgsee.kmutt.ac.th

Co-advisor’s name, organization and telephone/fax numbers/email

Dr Pariwate Varnakovida

Faculty of Department of Mathematics,

King Mongkut’s University of Technology Thonburi (KMUTT)

126 Pracha Uthit Rd., Bangmod, Tungkru, Bangkok 10140 Thailand

Telephone: 08-9866-5958

Email: pariwate@gmail.com

Topic: Evaluating the Impacts of Climate Variability on Deciduous Forest using Remote

in North of Thailand It was found that the phenological metrics of TDF (including dry dipterocarp forest (DDF) and mixed deciduous forest (MDF)) varied in responses to ENSO, affected by precipitation and temperature variability During the El Niño (Year 2010), significant delayed of SOSDDF and SOSMDF (20-26 days) and shorter of LOSDDF and LOSMDF(26-28 days) were found in most of areas whereas the larger difference was found in DDF During the La Niña (Year 2011), significant advanced of SOSDDF and SOSMDF (15-20 days) and longer of LOSDDF and LOSMDF (8-15 days) were found in most of areas whereas the larger was difference found in DDF SOS was positively correlated with maximum temperature and negatively correlated with precipitation in March-May EOS was found positively correlated with minimum temperature in December LOS varied according to the maximum temperature and precipitation during March to May The significant correlation

of phenological metrics to climate factors in this study implies that future variability in meteorological variables under climate change would affect to forest ecosystem functioning

In addition, Vegetation Photosynthesis model (VPM) was used to investigate the possibility

of scaling up flux-based measurement of GPP to satellite based estimate The simulation results of the VPM model have shown that the predicted GPP agreed well with the observed GPP of deciduous forest with r2=0.69 and r2=0.75 at Phayao site in 2015 and Ratchaburi site during 2010-2011, respectively The relationship between vegetation index and observed GPP at flux tower also demonstrated the better agreement of EVI with NDVI in terms of photosynthesis simulation The archived information of vegetation phenology changes over large spatial and temporal extents and its response to climate variability will help to address

many pressing issues such as global climate change, carbon budgets, and biodiversity

Keywords: phenology, NDVI, DEM, climate variability, meteorological data, MODIS

ACKNOWLEDGEMENTS

The success of my dissertation depended largely on the encouragement, support, and guidance of many important people I would like to express my deepest appreciation to my advisor, Dr Amnat Chidthaisong, who always offers advice and support to his students Working with him during my graduate study provides more valuable experiences and it helped me to develop my research ideas and intellectual skills He always provided very useful comments, remarks, and engagement through the learning process of this dissertation Without his guidance and persistent help, this dissertation would not have been possible

I give a special gratitude to my co-advisor, Dr Pariwate Varnakovida, and my committee members Dr Sirintornthep Towprayoon, Dr Sudarat Tripetchkul and Dr Yongyut Trisurat Their contribution in sharing invaluable knowledge, wisdom, advice, and comments, as well as precious time, with me, helped me go through the dissertation process

Mr Uday Pimple brought the technical and methodological perspectives to my research and

I would like to thank for his professional advice I am thankful for Miss Asamaporn Sitthi’s comments and suggestions on this research She helped me to improve my work from the very beginning of my study and she inspires me to learn and understand about remote sensing, which is very useful for my career I wish to thank Mr Sukan Pungkul and Mr Kumron Leadprathom for helping me during the field survey

I would also like to acknowledge with much appreciation to my scholarship sponsors United States Agency for International Development (USAID) and the National Science Foundation under the Partnership for Enhanced Engagement in Research (PEER) program (Grant number PGA-2000003836) for their financial support, which opened my bright vision on remote sensing and forest research and made me get this prestigious degree

I would like to extend my gratitude to the following people and organizations, the Royal Forest Department, Thailand, Dr N Yoshifuji (Forestry and Forest Products Research Institute), Dr Katsunori Tanaka, Dr Tanita Suepa (GISTDA), and the Thai Meteorological Department, for supporting the data and information for my research

Furthermore, I sincerely thank my friends in Thailand, especially my friends at Department of Land Resources, college of Environment and Natural Resources, Can Tho university for supporting and encouraging me during my study I am also grateful to all my friends at The Joint Graduate School of Energy and Environment at King Mongkut’s

University of Technology Thonburi and my Thai friends who made it enjoyable to do my research and helped me recover from homesickness

Finally, my deepest thank to my family, my mother, father, and sisters, for everything they have done to help me reach my goals They have supported me and encouraged me throughout entire process Their unending love and support made this achievement possible

I will be grateful forever for their love

2.1 Climate Variability Effects on Forest Growth and Productivity 11

2.2 Satellite Time-series Data and Vegetation Phenology 13

2.3 Relationship Between Phenology Metrics and Climate Factors 17

FOREST AS DRIVEN BY ENSO DURING 2001-2016: CASE

CONTENTS (Cont')

LIST OF TABLES TABLES TITLE PAGE

3.1 MODIS products and its spatial and temporal resolution for

3.2 Descriptions of the vegetation indices (VIs) used in this study 23

4.2 Results of phenological metrics difference during extreme

5.1 The different of climate variables between ENSO and neutral years 67 5.2 The number of stations with significant correlation between

5.3 The result of regression model between phenology metrics and

5.4 Threshold of temperature and precipitation at each station 80 5.5 Result of t-test of climate and DEM variables between two

groups significant and non-significant correlation to SOS of 19

5.6 Results of phenological metrics difference during extreme climate

5.7 Results of phenological metrics difference during extreme

5.8 Results of phenological metrics difference during El Nino year

2010 across TDF at different elevation levels in Northern of Thailand 86 5.9 Results of phenological metrics difference during La Nina year 2011

across TDF at different elevation levels in Northern of Thailand 87

LIST OF FIGURES FIGURE TITLE PAGE

2.2 A simple NDVI profile for a typical patch of vegetation 16 3.1 Administration and DEM maps of the study areas in Northern

3.2 Location of meteorological stations in Thailand, and

3.4 Study areas of Northern Thailand Forest map in 2007/2008 26 3.5 A flow chart of cloud removal methodology for MOD09Q1

3.7 The flow chart for assessing the forest phenology change 29 3.8 The concept of modeling gross primary production of forest 32

4.2 Location of training and validation points in Lampang 41 4.3 Daily temperature and precipitation records at Lampang

4.6 Reflectance value of 46 periods of band 1 MOD09Q1 data 2013 47 4.7 Checking result before and after smoothing in

4.8 Seasonal variations of monthly foliar litter biomass and NDVI 49 4.9 Time series of in-situ LAI against NDVI based on satellite 50 4.10 The variation in temporal of phenology derived from LAI and NDVI 51 4.11 Example of seasonal different in NDVI timing values of

4.12 Inter-annual variations of TDF phenology metrics in

LIST OF FIGURES (Cont’) FIGURE TITLE PAGE

4.14 The spatial difference of SOSDDF and SOSMDF

4.15 The spatial different of LOSDDF and LOSMDF between

5.1 Distribution of deciduous forest in Northern of Thailand 62 5.2 Example of climate variation at one meteorological station at Phayao 66 5.3 The average and standard deviation of seasonal variation of yearly

maximum temperature and precipitation of all 19 climate stations 67 5.4 The profile of MODIS NDVI obtained from random pixel

5.6 Average temporal variation of phenology metrics of DF 70 5.7 Spatial pattern of phenological metric period 2001-2013 71

5.9 The number of stations with significant correlation between

5.10 Location of nineteen meteorology stations Cyan circle show

the stations have strong correlation between phenology metrics

5.11 Seasonal variation in mean NDVI of TDF estimated for each class of

5.13 Inter-annual variations of TDF phenology metrics in different

5.14 Location of 500 points located randomly across the DF pixels 84 6.1 The season dynamic of observed gross primary production

6.2 The season dynamic of photosynthestically active radiation (PAR)

6.3 The season dynamic of gross primary production (GPPObs)

and vegetation indices (EVI, NDVI) at Phayao forest in 2015 97

LIST OF FIGURES (Cont’) FIGURE TITLE PAGE

6.4 Simple linear regression between observed gross primary

production (GPP) and vegetation indices (NDVI, EVI) at

Phayao forest in 2015 97 6.5 The season dynamics of MODIS-derived Land Surface Water

Index (LSWI) at Phayao forest in 2015 98 6.6 The season dynamic of gross primary production (GPP) from

VPM model and observed GPP at Phayao forest in 2015 98 6.7 Simple linear regression analysis between predicted gross

primary production (GPP) from VPM model and observed

GPP at Phayao forest in 2015 99 6.8 The season dynamic of gross primary production (GPP) from

observed GPP vegetation indices (EVI) at Ratchaburi forest

period 2010-2011 99 6.9 The season dynamic of photosynthestically active radiation (PAR)

and mean air temperature at Ratchaburi forest during 2010-2011 100

6.10 Simple linear regression analysis between observed gross

primary production (GPP) and vegetation indices (EVI) at

Ratchaburi forest during 2010-2011 100

6.11 The season dynamic of MODIS-derived Land Surface

Water Index (LSWI) at Ratchaburi forest during 2010-2011 101

6.12 The season dynamic of gross primary production (GPP) from

VPM model and observed GPP at Ratchaburi forest during 2010-2011 101

6.13 Simple linear regression analysis between predicted gross

primary production (GPP) from VPM model and observed GPP at

CHAPTER 1 INTRODUCTION

1.1 Problem Statement

Forest phenology is the study of the timing of seasonal growth stages in life cycles

of plant including the start, the end and the duration of growing season [1] The forest phenology varies among the different species and it depends on the changing in photoperiod, timing of rainfall, change in temperature and soil moisture [2-7] The variation in phenology could result in significant impacts on forest ecosystem such as leaf area, capacity of photosynthesis, species composition, and ecosystem functions [8]

In the tropics, tropical deciduous forests (TDF) are one of the main forest types, occupying about 43% of forest area in the tropical belt with great diversity of species [9] Water stress

is one of the major factors driving the sequence of phenological events in tropical forest [10] Timing of leaf flushing and leaf senescence in TDF has significantly responded to climate variation both by maximum temperature and precipitation [7] The delaying in the starting

of season (SOS) coincides with reducing productivity [11] Yoshifuji et al [5] reported that the inter-annual variations of canopy duration in tropical deciduous forest spanned between

40 and 60 days This was much larger than the inter-annual variations reported previously in temperate deciduous forests, implying a profound potential impact of such variations on surface energy balance and canopy-atmosphere water and carbon exchange on an annual time scale On the other hand, Kushwaha, 2011 concluded that one of the major constraints with respect to understand the effect of climate change on tropical tree phenology is incomplete understanding of annual cycle of TDF He also sugguest that phenological research in tropical deciduous forest trees needs critical consideration to understand the possible impact of climate change on phenology of tropical trees [2] There is a need for improving the understanding of the spatial and temporal variation in growing season length over tropical monsoon regions in responses to climate change and variability [5]

In Southeast Asia (SEA), El Niño–Southern Oscillation (ENSO) is the primary driver

of seasonal variation in the spatial distribution of temperature and precipitation [12] Several studies have shown that El Niño events have negative impacts on TDF ecosystems, such as significant increasing level of tree mortality, changing in plant phenology and carbon flux

[13-20] In addition, it has been shown that different tropical forest types exhibit asynchronous responses to seasonal and El Nino-driven drought [17] In Thailand, the site studies of tropical deciduous forest (TDF) show that decreasing precipitation and unusually high temperature in relation to severe drought (El Niño) results in the variation of leaf expansion and significant reduction of the CO2 uptake [21] Since climate change/variability usually cover an extended periods and large spatial scales, such rare available in-situ observations may not be able to capture the patterns reflecting the response of representing TDF in the regions [2, 22] The increase of Pacific sea surface temperatures (SSTs) has led

to a significant reduction of summer monsoon rainfall over Thailand in recent decades [23]

In addition, TDF occupies 52.9% of the total forest area in Thailand [24] As mentioned above, TDF is one of the important forest ecosystems in the tropics and SEA, but the information on TDF responses to climatic variability is sparse The impacts of extreme climate events to inter-annual variation in leaf phenology of tropical deciduous forest canopy

in spatial scale have not yet reported Accordingly, information covering the temporal and spatial phenology pattern of TDF and its variation in relation to ENSO is needed to aid our understanding of future ecosystem dynamics and quantify the effects of climate change on terrestrial ecosystems

Forest phenology is an important indicator for monitoring the response of vegetation

to climate change [25] However, phenological data are rare in some regions of the world

In addition, ecological models require large spatial scales information of phenology While the inventory data focus on specific plant species and mostly conducted at point observations As a result, inferring phenological characteristics of vegetation based remote sensing data is increasingly because it is one of key to understanding large area seasonal phenology [26] Satellite imagery provides consistent and repeatable measurements at a spatial and temporal scale for dynamic vegetation studies [27-30] It addresses spatial

limitation associated with in situ observation Normalized Difference Vegetation Index

(NDVI) is one of the most widely used index for monitoring the spatial and temporal pattern

of vegetation phenology [28, 31] because it used to measure the amount of green leaf and biomass [32] and the strength of plant activity [33] However, Testa et al [34] indicated that satellite-based phenology may lead to misinterpretation of climate and forest ecosystem interactions due to complexity within ecosystem and limitation of available satellite observations He also suggests that the ground based phenology observation and reliable ground sampling could improve the interpretation of spatial variations in forest phenology

The variations of gross primary production (GPP) and ecosystem respiration (R) through the course of the year determine the net ecosystem exchange (NEE) of CO2 between the atmosphere and forest ecosystems In recent years, the measurements of CO2 at flux tower sites have provided more detail information of the photosynthetically active period and GPP of forest ecosystems [35] It is thought that even modest shifting in length of growing season could result in large changes in annual GPP of forest One of the useful approach to scale up the site specific measurements of GPP is to use process based biogeochemical models, whereas the climate, soil factors and vegetation types are need to consider CO2 flux data collecting from the towers are useful for parameterization and validation of biogeochemical models For regional analysis, satellite observations and climate data are most useful parameters in the biogeochemical models [36] A number of satellite based modeling studies have been used to effectively estimate the GPP or NPP at the large spatial scales [37] The focus of this study is alsoto find out whether the changing

of phenological derived-satellite imagery agree to rare in-situ observations for phenology

monitoring and to evaluate the response to climate extreme events of phenological metrics

at spatial scale The insight understanding of the interaction between canopy phenology and climate extreme event in related to GPP variation will improve prediction accuracy of vegetation dynamic under climate change scenarios

1.2 Literature Review

1.2.1 Interactions between climate change and plant phenology

Phenology is a study of the timing of seasonal activities of plants and animals such

as flowering or breeding and can be influenced by climate change [38] Phenology is a dominant and often overlooked aspect of plant ecology from the scale of individuals to whole ecosystems The timing of growth onset and senescence also determine growing season length, thus driving annual carbon uptake in terrestrial ecosystems [39]

Global climate change could significantly affect plant phenology because of the influences of temperature to timing of plant development [4] Thermometer records over the past 30 years show that, average surface temperatures in global scale had increased by 0.280C per decade [40] In addition, numerous studies have found that the changes of frost dates, length of growing season are consistent with climate warming [41] In tropical ecosystems, plant phenology might be less sensitive to temperature and photoperiod, but may more

affected by shifting in precipitation These shifts are expected to occur in concert with rising global temperatures, which could vary regionally in both the direction and magnitude of change [42]

Vegetation cover is usually considered as an indicator of the response to climate and plays an important role in hydrological and biochemical cycles, due to driving energy and mass exchanges between the atmosphere and land surface [43] For example, Cubasch et al [44] concluded that in Amazon forests the growth and productivity are being stimulated by widespread environmental changes Other study at Costa Rica found that growth rates in six

canopy species was decreased during 1984–2000 [45] The large reductions in tree growth

resulted the releases of CO2 to the atmosphere during the record in1997–1998 El Nino These conclusion are consistent with decreased NEE in tropical forests in the warmer years during the last two decades [13] Dong et al [46] indicated that the variability of solar radiation is

an important environmental factors forcing the dynamics of tropical tree growth [47] In the other hand, Kariyeva et al [48] has studied about effect of climate gradients on the vegetation phenology The results reveal that vegetation phenology responded to climate variation in Central Asia, whereas phenology metrics were found positively correlated to precipitation and negatively correlated to temperature, the length of growing seasons was longer with increasing temperature and less precipitation [43] These difference responses

of phenology metrics can be explained by local and region climate pattern of various temporal scales Climate change impacts would vary among tropical forest types due to the influence of water balance at sites [25]

1.2.2 Phenological changes as observed by remote sensing

The knowledge of land cover changes over large spatial and temporal extents are useful to managers, policy makers and researchers in order to address many pressing issues such as climate change, carbon budgets, and biodiversity [49] Detecting the phenology change over time is the first step for identifying the drivers of the change and understanding

of its mechanism Each system used for detecting change has its capacity for accounting the variability at different scale There are three main changes of the ecosystem: seasonal change caused by annual temperature and rainfall changes, gradual change in land management, and abrupt change caused by disturbances such as deforestation, urbanization, floods, and fires [50]

Vegetation phenology is the study of the timing of seasonal events (leaf bud bursts and leaf senescence) that are considered to be the results of adaptive responses to climatic

constraints The understanding of phenology changes will bring important insights into both climate and vegetation interactions and their impacts on energy exchange processes at local, regional and global scales [51] Because field phenological observations are work-intensive and cannot be easily generalized, remote sensing tools were developed to track Earth surface changes

Satellite imagery provides consistent and repeatable measurements in the spatial scale, which is suitable for capturing the causes of natural and anthropogenic disturbances [52] Satellite observations play an important role in the study of phenological and ecological responses to environmental changes over space and time Among vegetation indices, the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI), representation of “greenness” are often used as the proxy for plant production EVI improves discrimination of bare soil from vegetation over NDVI [27] Phenological derived from MODIS time series were compared to field measurements at the main deciduous forest stands across France The results show that the NDVI temporal profile fitted by asymmetric double sigmoid is a good for investigate the onset of green up in deciduous stands with the bias about 7.5 days [53] DeFries et al [48] studied the long term land surface phenology recorded as timing start, length of growing season and vegetation productivity derived from NDVI, and to explore the relationship between the phenology and precipitation and temperature variables in Central Asia The results show that the phenology metric was mostly influenced by fall and spring temperature and winter precipitation, whereas the vegetation explored longer season length with decreasing precipitation and increasing temperatures [43]

In another study, a time series of MODIS 250 met enhanced vegetation index (EVI) data were used to characterize the foliage condition of the three distinctive sub forests before, during, and after a severe drought in 2009 in indigenous forests in South Africa [54] EVI anomalies during the drought and post drought periods were calculated using annual median EVI values The results show that maximum foliage loss occurred one year after the driest year, indicating the cumulative effects of drought stress on forest production and retention

of foliage Other case study, phenology of deciduous broadleaf forest in Northern Iran evaluated by using ground observation and MODIS images [55] Ground observations of deciduous forest growth process were performed both visually and by measuring leaf chlorophyll concentration (chlorophyll meter SPAD-502) from January to December 2004 for each 7-15 days interval Regression analysis showed that relationship between NDVI and

SPAD measurements (chlorophyll content) was a positive There is a negative relationship between air temperature and anthesis, so a temperature increases leads to anthesis appear earlier in the year [55]

Hmimina et al [51] used MODIS daily and 16-day composite products for monitoring the seasonal dynamics of different types of vegetation cover, which are representative of the major terrestrial biomes, including temperate deciduous forests, evergreen forests, African savan, and crops This study conducted to compare the temporal

patterns of phenological metrics derived from in situ NDVI and derived MODIS daily and

16-day composite products The effects of residual noise and the influence of data gaps in MODIS NDVI time series to the identification the vegetation phenology metrics was also evaluated The results showed that the inflexion points of a model fitted to a MODIS NDVI time series allow accurate estimates of the onset of greenness in the spring and the onset of yellowing in the autumn in deciduous forests (RMSE ≤ one week)

Based on the time series of Enhanced Vegetation Index (EVI)-derived MODIS data (2000– 2009), Yu et al [56] extracted forest phenological variables in Northeast China used

a threshold based method, which included the start of the growing season (SOS), end of the growing season (EOS), and length of the growing season (LOS) The result shows that SOS was delayed at the rate of less than 1.5 days per year in Northeast China The SOS had a negative relationship with precipitation in the warm temperate deciduous broadleaf forest region The LOS of the temperate steppe region and temperate mixed forest region increased due to increased temperature in spring, and the LOS was positively correlated with the mean temperature of summer in the cool temperate needle leaf forest region [56]

Wang et al [57] tested the feasibility of the frequently observed MODIS NDVI time series (500-meter, 8-day) for examining alpine phenology in the Tibetan Plateau from 2000

to 2010 The correlation analysis in this study indicates that precipitation performed a decreasing trend in the west and increasing trend in the east Precipitation may serve as the primary driver of the onset and peak dates of greenness The delay of the end of season in the west could be related to higher late season temperature in this case study [57]

1.2.3 Vegetation Photosynthesis Model (VPM) for estimating seasonal dynamics

of GPP of deciduous forests

In recent years, continuous CO2 flux measurements at flux tower sites have provided more detailed information of the photosynthetically active period (leaf phenology) and gross primary production (GPP) of forest ecosystems [35] It is thought that even modest changes

in the length or magnitude of the plant-growing season could result in large changes in the annual GPP of forests The CO2 flux measurements over footprints with sizes and shapes depends on tower height, canopy physical characteristics and wind velocity The upscaling

of those CO2 flux measurements is a challenging task because of the large spatial heterogeneity and temporal dynamics of complex landscapes and regions [36]

Light use efficiency (LUE) is one of important variables for evaluating the carbon cycle and climate change research Wu and Z Niu [58] have presented a LUE model incorporation of vegetation indices (VIs) derived from MODIS in a case study at Harvard Forest Three indices, including the normalized difference vegetation index (NDVI), enhanced vegetation index (EVI) and the soil-adjusted vegetation index (SAVI), were selected as indicators of forest canopy greenness This research showed that a single VI provided moderate estimates of LUE with coefficients of determination (R2) of 0.62, 0.70 and 0.75 for NDVI, EVI and SAVI, respectively The significant relationship was found between Land Surface Temperature (LST) and both air temperature (R2 = 0.88) and vapor pressure deficit (VPD) (R2 = 0.69) The model with vegetation indices (VI) × (Scaled (LST)) provided improved of LUE estimation with R2 of 0.73, 0.76 and 0.79 for NDVI, EVI and SAVI, respectively [58]

Sasai et al [59] used a biosphere model integrating eco physiological and mechanistic approaches based on satellite data (BEAMS) and observations with 1 km resolution to estimate terrestrial carbon fluxes in the central Far East Asia region The model estimations showed reasonable seasonal and annual patterns The total GPP and NPP were determined to be about 2.1 and 0.9 PgC/year, respectively The total NEP estimation was about +5.6 TgC/year, showing that this area played a role as a carbon sink from 2001 to 2006 [59]

Tang et al [60] proposed a model for estimating the carbon sequestration of temperate deciduous forest based on MODIS data, including land surface temperature (LST), enhanced vegetation index (EVI), land surface water index (LSWI), fraction of absorbed photosynthetically active radiation (FPAR), and leaf area index (LAI) The study aimed to improve the accuracy of predicting NEE according to the growth NEE are strongly correlated with both photosynthesis and respiration during the growing season (p<0.01) The study also implies the potential of vegetation phenology for modeling carbon dioxide fluxes using remotely sensed data [60]

Light use efficiency (LUE) models have been widely used to estimate terrestrial gross primary production (GPP) because of their theoretical soundness and practical conveniences Zhang et al [61] evaluated the performances of the four major LUE models including Vegetation Photosynthesis Model (VPM), Carnegie Ames Stanford approach (CASA), Global Production Efficiency Model (GLO-PEM), and Eddy Covariance Light Use Efficiency (EC-LUE) The results showed that the GLO-PEM, VPM, and EC-LUE exhibited the similar capabilities in simulating GPP (68%) and overall performed better than CASA (58%) Two model EC-LUE and VPM were the optimal options because both required less model input parameters and result in better performance [61]

Summary

In general, previous research has shown that MODIS data can be used successfully for monitoring the vegetation phenology, as well as the forest phenology, around the world Different research studies showed different responses of forest to climate variability From the above studies, we can see that, forest growth has positive correlation with temperature

in some region while the others show the opposite trend The different responses can be explained by local and region climate pattern of various temporal scales In addition, climate change impacts would vary among tropical forest types, according to the degree to which the site water balance is influenced However, the response of forest (phenology and growth

of forest) in South East Asia is still poorly understood [62]

Most studies, mainly using MODIS data, focused on two general issues, which are developing methods for reconstructing time series datasets and evaluating the use of satellites to derive phenological metrics for in-depth study of terrestrial ecosystems dynamics at different spatial and temporal scales This requires VI time series with good temporal resolution, over homogeneous area, cloud free and not affected by atmospheric, geometric and variations in sensor characteristics

In terms of seasonal dynamics of GPP estimation of deciduous forests, all the above results show that VIs can be used as input parameters for successful estimated annual changes in forest carbon Different light use efficiency (LUE) models have been widely used

to estimate terrestrial gross primary production (GPP) with theoretical soundness and practical conveniences Among of them VPM is the optimal one because it requires less model inputs and shows the lowest uncertainty

1.3 Research Objectives

The goal is to improve the understanding of the phenological characteristics of TDF

in Thailand under climate variations and extreme climate events This includes;

 Detecting the deciduous forests phenology and its change in North Thailand

 Finding out the main climate drivers (temperature and precipitation) of deciduous forest phenology changes in the study areas

 Estimating and assessing seasonal dynamics of GPP on deciduous forests by using Vegetation Photosynthesis Model (VPM) and linking with the variation of forest phenology

Figure 1.1Concept of study

This study focused on the tropical deciduous forest because of its well-defined seasonality in phenology, which serves as a good indicator for influences of climate change

on forest ecosystems The concept and scope of study is outlined in Figure 1.1

1.4 Scope of Research Work

 This study focused on deciduous forest in North of Thailand

 Temperature and precipitation were considered in climate driver assessment

 Phenology of deciduous forest focused on timing of the start and length of the growing season

 MODIS imagery was used for detecting the phenology of forest and an input in VPM for detecting the GPP changes in deciduous forest in North of Thailand

CHAPTER 2 THEORIES

2.1 Climate Variability Effects on Forest Growth and Productivity

There are 2 distinguishable impacts of climate change on forests Firstly, the long term impacts of changing temperature and precipitation regimes could lead to change in species composition Secondly, shorter term impacts of extreme weather events can direct damage to trees and modifications to forest land management In general, climate change has both directly and indirectly affects the growth and productivity of forests The direct impacts mainly results to the changes in atmospheric carbon dioxide and indirect impacts are through complex interactions between forest ecosystems and external drivers including climatic factors [63]

Many different aspects of projected climate change will likely affect forest growth and productivity Three main causes are the increase in carbon dioxide (CO2), increase in temperature, and change in precipitation

2.1.1 Effects of increasing atmospheric CO 2 concentrations

Forests use CO2 and water as substrates for the process of photosynthesis, leading to the production of biomass and oxygen as the end products of the process Forests are also sources of CO2 through the processes of respiration, decomposition of organic matter, and when there are forest fires and other disturbances [64] (Figure 2.1)

Figure 2.1 Forest carbon cycle [64]

Rising concentrations of CO2 in the atmosphere result in increased rates of photosynthesis for some vegetation species Rates of photosynthesis are then affected by the ratio of atmospheric CO2 and O2 Increase of the CO2:O2 ratio within the plant tissues results

in suppression of respiration and enhancement of photosynthesis, which consequently leads

to the increase of the net photosynthesis However, the rates of photosynthesis changes according to the duration of the period, the plant nitrogen status, and species Changes in

CO2 levels also result in changing the plant structure such as changing leaf area, numbers of leaves, and diameter stems and branches [64]

2.1.2 Effects of Changes in Temperature and Precipitation

Changes in temperatures and precipitation regimes will directly affect tree growth and survival Warming temperatures could changing the length of the growing season and shift the geographic ranges of some tree species Some species may be at risk if the current geographic range have no longer suitable for them [65] For example, some species may die out as they cannot shift to a higher altitude because of climate warming

Climate change is likely to increase the risk of drought, extreme precipitation and flooding in some areas Drought affect to ability of sap producing of tree, which protects from destructive insects such as pine beetles [66] Availability of moisture in forests is influenced by changes in both temperature and precipitation Higher temperatures lead to increased water losses though out the process of evaporation and evapotranspiration and affect to water use efficiency of plants Prolong higher temperature during the growing seasons can resulte severe moisture stress, which lead to reduction of the growth and health

of trees However, these impacts vary according to the characteristics, age class structure and soil depth and forest types [67] For instance, productivity decline and dieback were found in quaking aspen in Western Canada after a severe drought in 2001-2003 [68]

Over the last century, climatic variability affected differentially tree mortality and canopy disturbance along the West East precipitation gradient with possible important implications for forest dynamics In the other hand, climate controls the distribution of many plants, therefore, increasingly of extreme climatic events are the important drivers of vegetation change and species distributions [69]

2.2 Satellite Time Series Data and Vegetation Phenology

2.2.1 Remote sensing approaches

Plant phenology is the study of the timing of different stages of the vegetation seasonal cycle such as leaf unfolding, first bloom, and leaf fall It helps to improve the understanding of periodic biological events in plants, which are influenced by both the environment and climate [70] Phenology is consider as the key for tracking the changes in the ecology of species in response to climate change [71] In recent year, the remote sensing data are widely used in phenology topic because it could monitor in both temporal and spatial characteristics, which are very important for the research of plant phenology Moreover, the up-scaling the observation of plant phenology from points at sites to coverage larger spatial enhance our understanding of phenology across regions, countries, continents [72]

Land surface phenology (LSP) is defined as the seasonal patterns of the spatial and temporal variation of vegetation on the land surfaces, which can be monitored by using remote sensing [73] The traditional definition, LSP is specific life cycle events throughout

the validation from in situ measurements of individual plants [74] Vegetation indices

derived satellite time-series data usually used now to present a global coverage Xiao et al [75] revealed that LSP is very sensitive to climate change and useful for improving our understanding of vegetation responses to climate change thought out the model of carbon cycle, water cycle and energy fluxes Changing of phenology can be observed in both the length and magnitude of the plant growing season, which could result the changes in annual gross primary production The variability of inter-annual of phenology, particularly during severe climate years could help to understand how ecosystem is susceptible to future climate change [76]

Forest phenology is study of the timing of recurrent biological events with regards to biotic and abiotic forces, as well as their interrelations [25] The understanding of tropical deciduous forest (TDF) phenology is important because of its relation to various processes and factors such growth periodicity, flowering and fruiting, plant water stress, leaf gas exchange, leaf area index and ecosystem properties [8, 77] Richardson et al [25] reported that the global researches on forest phenology have increased, but noted that these studies provided incontrovertible evidence of some ecological impacts caused by changing in climate variables Several factors such as changes in photoperiod, timing of rainfall, change

in temperature and soil moisture can cause phenology variations [3-7, 78] Hwang et al [79]

reported that not only microclimatic factors, but also topography plays an important role in controlling the forest phenology It is well known that the phenology varies among the different forest types and either same forest type could perform variation of phenology in spatial distribution because the wide geographic distribution and site characteristics are highly variable [2, 10, 80] Therefore, considering variation in topography and diversity in forest type it is important to understand phenological processes at the individual forest level

Satellite imagery provides consistent and repeatable measurements at a spatial and temporal scale for dynamic vegetation studies [28-29, 34] and addresses the spatial limitation

of observations from phenological networks For examples, Advanced Very High Resolution Radiometer (AVHRR) data with frequent temporal coverage have been used for monitoring vegetation with five spectral bands and resolution more than one kilometer It was resulting

in difficulties of assessing spatial variations in vegetation in term of amount and condition [81] It is noted that the AVHRR sensor was firstly designed for monitoring the weather and climate, and aimed to support data for evaluating the clouds, snow and ice extent, temperature of radiating surface and sea surface temperature Therefore, there are some limitations such as lack of calibration, poor geometry, and high level of noise due to large pixel size when application of vegetation studies [75] The other Vegetation Instrument (VGT) Sensors of SPOT4 and 5 and LANDSAT can also provide phenological characteristics However, the spatial resolution of VGT is coarse and LANDSAT imagery are limited for their temporal resolution and availability in a large scale and long term study [82]

In contrast, MODIS, launched in December 1999, provides 8-day and 16-day datasets

at spatial resolution of 250-meter atmospherically corrected and screened for clouds MODIS has 36 spectral bands, seven of which are designed for the study of vegetation and land surfaces, has become the standard product for phenological studies, providing an improved basis for monitoring global ecosystem dynamics [72] The vegetation indices show the combination between absorbing and non-absorbing leaf spectral features to capture the phenomena of vegetation In the reflected energy, spectral blue band (470 nm) Red band (670 nm) is absorbed the most by chlorophyll and non-absorbed by the spectral of near-infrared band (NIR), therefore the contrasting between these spectral bands are sensitive and usually used to detect the canopy greenness or the canopy photosynthetic activity The maximum contrast is showing the healthy vegetation and the minimum contrast presented the stressed or the senesced of leaves At the canopy observation level, the maximum contrast

between red and near infrared band present for the dense of leaves expansion, whereas the lowest contrast present the open canopy or the sparse canopy [83] Among the vegetation indices, Normalized Difference Vegetation Index (NDVI) is one of the most widely used indexes for monitoring the spatial and temporal patterns of vegetation phenology [28, 31] because it is related to the amount of green leaf and biomass [32] and the strength of plant activity [33] For these reasons, MODIS is considering to use in tropical region such as SEA However, Falge et al [34] indicated that satellite based phenology may lead to misinterpretation of climate and forest ecosystem interactions due to complexity within ecosystem and limitation of available satellite observations He also suggests that the ground based phenology observation and reliable ground sampling could improve the interpretation

of spatial variations in forest phenology However, there is limited in-situ measurement for

validation the long-term variation of forest phenology in the Monsoon Southeast Asia [2, 62, 84]

2.2.2 Phenological metrics derived from satellite time-series

It is necessary to identify the appropriate phenological metrics to extract useful information about the vegetation growing seasons [85, 86] were among the first who reported on key phenological metrics using the time series of NDVI They proposed a set of phenological metrics including the timing of green-up and senescence, maximum rates of green-up and senescence, and the amplitude of maximum NDVI Zhang et al [87] proposed four key transition dates: green up, maturity, senescence, and dormancy for monitoring global vegetation phenology in the United States These proposed metrics can improve the understanding of vegetation dynamic over large areas, and these were focused more on the identifying phenology behavior characteristics Jönsson and L Eklundh [88] had clearly explained the method for extracting the seasonal parameters which further improving the understanding of the seasonal curve of phenology

The important phenological parameters that are usually investigated in the past are

as follows the start of a season, the end of the season, the length of the growing season, the base level is given as the average of the left and right minimum values, the middle of a season, the maximum of fitted data, the amplitude of the season, the left derivative, the right derivative, the large integral, the small integral

Figure 2.2 A simple NDVI profile for extracting the seasonality parameters of

vegetation [88]

There are various methods that have been developed to reduce satellite noise of series data, from simple linear smoothing methods to complicated curve function methods Among them, Fourier Transform has become one of the principal methods of phenological analysis because of strong seasonal cycles of phenology [89] However, Fourier Transform methods could result in bias, when applying to irregular or asymmetric vegetation index time series data because it depends on symmetric sine and cosine functions [90]

time-A high-order spline fitting model is also one of the methods for fitting the curve and asymmetrically weights all data points above and below the average to fit the upper envelope

of the data Although exponential weighting is suitable for fitting rapidly green-up, however

it also fits data errors (spikes) that fall above the time series [91]

Asymmetric Gaussian Functions is used to fit the data at intervals around maxima and minima of time series It was used as a function for describing the annual cycles of vegetation [88] Beck et al [92] revealed that Asymmetric Gaussian Functions is appropriate for presenting the vegetation dynamics at high latitudes However, it may be a problematic

to identify a reasonable and consistent set of maxima and minima to which the local functions can be fitted, especially for noisy data or for time series of data with no clear seasonality in land cover [93]

The Double Logistic Function performs based on the minimum and maximum value

of vegetation index, two inflection points, and parameters related to the rate of increase or decrease in vegetation index [94] Similar to Asymmetric Gaussian Functions, Double

Logistic Function can work well with negative noise and approximate winter conditions [95] However, similarly to Asymmetric Gaussian Functions, Double Logistic Function depend

on the linear combination of local and independent intra-annual functions; so it may fail to match the global waveform of numerous time series in some cases [96]

Among many different functions, Savitzky-Golay Filtering is one of the potential for fitted curve of time-series data Savitzky-Golay Filtering performs least-squares polynomial regression on each point to determine the smoothed value of the vegetation index time-series [88] By applying the iterative weighted moving average to time series this approach preserves features of the distribution and capture rapid phenology changes [93] Tottrup et

al [97] claimed that Savitzky-Golay Filtering provided good results in smoothing data and extracted phenological metrics for mapping forest cover in Southeast Asia Tottrup et al [97] also confirmed that the Savitzky-Golay Filtering approach is applicable in this monsoon climate with multiple annual seasons during a year

Among the methods for extracting the phenology metrics, threshold-base phenology

is considered as the most effective technique This technique uses a defined or a relative reference value for defining starting of growing season (SOS) and ending of growing season (EOS) [98] Various methods of threshold based methods have been developed to extract phenological metrics such as thresholds based on long-term mean vegetation index, baseline year, reference value, and vegetation index ratios [99]

These study thresholds based on vegetation index (VI) were applied to studying temporal variations in phenological metrics This study, the ratio of the seasonal amplitude measured from the left and right minimum values was used to define the starting and ending

of growing season, respectively [88, 90] This method can reduce the effect of soil background at lower vegetation signals In addition, users can define the appropriate threshold depending on spatial local climate and VI characteristics; therefore, this technique

is flexible and can be applied in various regions [88]

2.3 Relationship Between Phenology Metrics and Climate Factors

Vegetation phenology provides information for monitoring ecosystem dynamics, biodiversity, and land use changes Vegetation phenological metric is an indicator for detecting and monitoring global environmental change which driven mainly by temperature and precipitation

The tropical forest owns the largest annual rate of carbon sequestration [84] They are also strongly involved in regulating the climate system through physical, chemical and biological processes that affect various hydrological cycles and interactions of ecosystem carbon [10, 84] Unlike in temperate forests where temperatures fluctuate widely during the course of a year, variations of temperatures in tropical forests are small, with trees adapted

to grow in a relatively narrow temperature range Hence, the relative impact of climate warming will likely be greater in the tropics than in other regions because predicted changes

in temperature are large compared to normal inter-annual variations [71] In addition, changing precipitation patterns such as a shift towards more extreme events and extended droughts under climate change potentially could result in a loss of tropical forest and in large amount of CO2 released to the atmosphere [100] Cavaleri et al [101] reported a decline in the carbon sink of tropical forest during El Niño because of reducing photosynthesis and increasing respiration rates [21, 101] Rolim et al [102] studied changes in Atlantic tropical moist forest in a long-term experimental plot also indicated a rapid biomass decline associated with El Niño events Several studies have reported that the extreme El Niño events have negative impacts on forest ecosystems, which could result in significantly increasing levels of tree mortality, shifting plant phenology and change in carbon flux [13-15, 17, 19-20] Among various types of forests in the world, the tropical deciduous forest (TDF) occupies about 43% of forested area in the tropical belt with a large diversity of forest species [9] In tropical zone, the Amazon and Southeastern Asia tropical evergreen-wet and semi-evergreen forests were found suffering from droughts during ENSO events (El Niño – La Niña) In the short term of drought, tropical forests may be resilient to drought However, this may be compensate by increased vulnerability to fire after long-term droughts or strong

El Niño In tropical evergreen-wet forests, mortality may increase during strong El Niño years due to severe drought, while seasonal semi-evergreen forests may experience relatively little change Nevertheless, the effect of various climate variable to tropical deciduous forest dynamics have been poorly understood

2.4 Carbon Storage and Flux in Tropical Forests

Photosynthesis is the process through which plants assimilate carbon in the form of carbon dioxide Specifically, photosynthesis requires CO2, sunlight, and water as inputs to produce glucose (carbohydrates), oxygen, and water If the carbon uptake of photosynthesis

exceeds the carbon efflux of respiration, intact forests are thought to remain a carbon sink [103] The photosynthesis controlled by a stoma, whose behavior can be determined in term

of stomatal conductance The opening of stomata is regulated by the intercellular CO2concentration The ability to control stomatal aperture allows plants response to an environment change because plants have to optimize two contrasting needs at the same time; maximizing the amount of CO2 uptake by opening stomatal as long as possible and conserving water by closing the stomata as fast as possible Normally, higher atmospheric

CO2 can enhance tree growth by increasing photosynthesis and intrinsic Water Use Efficiency (WUE) Understanding the response of tropical forests to water stress is important because effects of canopy to air vapor deficits and stomatal feed-back could determine how tropical forest responds to future climate change

Tropical forests have high contributions to the total terrestrial gross primary productivity (GPP) with occupying about 8% of the total atmospheric CO2 cycles annually There are about 40% of the carbon stored in tropical forest (estimated at 428 Gt of carbon), whereas the vegetation and soil accounting for 58% and 41%, respectively [104] This ratio varies greatly throughout different tropical forest types

There are only a few studies that have investigated on interactions in tropical forest ecosystems Among forest systems, tropical dry deciduous and montane forests almost remain unknown because little research has been done on these forest types Many climate scenarios predict that soil drying in Amazon leads to reduce capacity of the ecosystem to take up carbon However, modelling carbon flux and productivity in tropical forests is difficult because of their structural and age complexity and species composition Therefore,

it is essential to understand how increasing temperatures could affect these processes and the balance between them [104]

CHAPTER 3 GENERAL METHODOLOGY

3.1 Study Areas

Since the majority of the deciduous forest in Thailand is distributed in the northern part, this study was focused on this area The sudy areas cover 15 provinces including Chiang Rai, Mae Hong Son, Chiang Mai, Phayao, Lamphun, Lampang, Phrae, Nan, Uttaradit, Phitsanulok, Sukhothai, Tak, Phichit, Kamphaeng Phet and Phetchabun [105] The total area

is about 169.644 square kilometers Most areas of the part are hilly and mountainous The highest mountain is about 2,572 meters high above mean sea level (Figure 3.1) The climate

of this area is influent by monsoon winds of seasonal character such as southwest monsoon and northeast monsoon The southwest monsoon starts in May bringing a stream of warm moist air from the Indian Ocean towards Thailand and causing abundant rain over the country, especially the windward side of the mountains The northeast monsoon starts in October bringing the cold and dry air from the anticyclone in China mainland over major parts of Thailand, especially the Northern and Northeastern parts Southwest monsoon usually starts in mid-May and ends in mid-October while northeast monsoon normally starts

in mid-October and ends in mid-February [105]

Deciduous forests are widely distributed in the northern most part of Thailand, about 52.9 percent of total forest areas in Thailand [106] This study was targeted on deciduous forest in North of Thailand because of the following reasons; (1) the majority of deciduous forest is located in Northern Thailand which is facing scatter changing without clearing causes [107]; (2) the previous analyses showed that annual growth variability in tropical trees is determined by a combination of both temperature and precipitation variability The predominantly negative relationship between temperature and growth may imply decreasing

growth rates of tropical trees as a result of global warming [108]; (3) there is a long term in situ observation of leaf areas index (LAI) during 2001-2011 in Mea Mo, Lampang province

where validation of phenology metric patterns derived from satellite imagery could be made

Figure 3.1 Administration and DEM maps of the study areas in Northern Thailand

3.2 Data

3.2.1 MODIS data

The Moderate Resolution Imaging Spectral radiometer (MODIS) instrument is operating on both the Terra and Aqua spacecraft It has a viewing swath width of 2,330 km and views the entire surface of the earth every one to two days [109]

Resolution of MODIS data

Spatial resolution: Band 1-2: 250m; Band 3-7: 500m; Band 8 - 36: 1,000m Radiometric resolution: 12 bit

Temporal resolution: The MODIS Land products are produced at various

temporal resolutions based on the instruments' orbital cycle It includes daily, 8-day, 16-day, and monthly images

The MODIS has total of 36 bands and this study mainly used the normalized difference vegetation index (NDVI) and the enhanced vegetation index (EVI) as a surrogate measure of the regional vegetation growth (Table 3.1)

Table 3.1 MODIS products and its spatial and temporal resolutions for NDVI and EVI

extraction

Index name Spatial

resolution

Temporal resolution

MODIS Production Product

Enhanced

Vegetation

Index (EVI)

250m 16 day MOD13Q1, MYD13Q1 SD product

Normalized

Difference

Vegetation

Index (NDVI)

SD product: Standard product; SR product: surface reflectance product

The MODIS VI products are currently produced at 250-met, 500-met, 1-kilomet and 5,600-met spatial resolutions For production purposes, MODIS VIs are mapped in the Sinusoidal (SIN) grid projection They are processed, with the aim of reducing processing and disk space requirements According to the Table 3.1, MOD13 with 16-day temporal resolution and 250-met spatial resolution is the first option because it is a standard product with the highest spatial resolution The second option is MOD09 with 8-day temporal

resolution and 500-met spatial resolution MOD09 provides surface reflectance data, which need to do reprocessing such as cloud removal and replacement This study used the MODIS product MOD09Q1 and MOD09A1 to generate NDVI at 250-met, 8-day resolution Commonly used VIs include the normalized difference vegetation index (NDVI) and the enhanced vegetation index (EVI) The vegetation indices were calculated as the equations shown in Table 3.2

Table 3.2 Descriptions of the vegetation indices (VIs) used in this study

Bands refer to the MODIS standard setting

Where ρ x represents the reflectance at wavelength band x; Band 1 (620-670 nm), Band 2 (841-876 nm), Band 3 (459-479 nm), Band 5 (1230-1250 nm)

Satellite imagery provides consistent and repeatable measurements at a spatial and temporal scale for dynamic vegetation studies [27-28,112] It addresses spatial limitations

associated with in situ observations from phenological networks Among remotely sensed

surface parameters, Normalized Difference Vegetation Index (NDVI), Enhance vegetation indices (EVI) is the most widely used index for monitoring of long term deciduous forest phenology [113-116] Li et al [117] found that NDVI has a better agreement with field observations as compared to the Enhanced Vegetation Index (EVI) when detecting phenology of deciduous broadleaf forest The prediction from EVI could have more accurate results but more outliers due to soil adjustment factors are more sensitive to topographic conditions Moreover, NDVI shows the highest correlation coefficient for the relationship

between the starting day of the growing season as observed with MODIS and in-situ

observations [31] Thus, NDVI was considered for detecting long-term phenology which

could help to minimize the effect of topographic to dynamics of phenology evaluation TDF

in this study

3.2.2 Climate data

The climate data used in this study were collected from the Thai Meteorological Department There are 33 meteorological stations in Northern of Thailand and some stations are located in the vicinity of TDF

Figure 3.2 (a) Location of meteorological stations in Thailand, (b) Deciduous forest map

and location of nineteen meteorological stations (yellow circle), which included forests within 10 km in Northern Thailand

Among them, the stations were selected if they satisfied the following criteria; (1) climate data missing is less than 20% (2) there is extended forest cover within 10 km from climate stations The internal consistency and temporal outliers check were performed for the climate data set The missing values and outlier values on data set were checked and removed [118-119] The daily temperatures and precipitation were then averaged and aggregated for yearly periods

3.2.3 Flux tower data

Flux tower data were collected from the ThaiFlux network There are nine flux towers a cross all of Thailand at the Kog Ma, Mae Moh, Tak, Ratchaburi, Mae Klong, Sakaerat, Buengkhan and Pasot Among these towers, the data at 2 towers (Ratchaburi and Phayao) were collected for calibrating and verifying the VPM model

Table 3.3 List of flux towers in Thailand

Figure 3.3 Location and information of flux towers in Thailand [120]

3.2.4 Forest map and digital elevation map

Figure 3.4 Study areas of Northern Thailand Forest map in 2007/2008 (Source: Thai

Royal Forest Department, 2007/2008)

Land use map in Northern Thailand was collected from the Thai Royal Forest Department (2007/2008) The major forest types include mixed deciduous forest, dry dipterocarp forest, pine forest, swamp forest, evergreen forest and bamboo forest This current study considered only the deciduous forest (Figure 3.4)

3.3 Methodology

3.3.1 Assessment of the impact of climate variability on forest phenology

This study used the surface reflectance of MOD09Q1 from MODIS, which provided band 1 (Red) and band 2 (NIR) at 250-met resolution, and MOD09A1 at 500-met resolution imagery captured in an 8-day period The data were downloaded from the EROS Data Center, US Geological Survey [121] Cloud cover was present in MOD09Q1 images, which limits the potential of images for ground information extraction Removing and replacing cloud-contaminated pixels is necessary in phenology extraction The methodology for cloud coverage removal developed by Hoan and Tateishi [122] was applied to all 644 images from