Shifts in Growing Season of Tropical Deciduous Forests As Driven by El Niño and La Niña During 2001–2016

Article Shifts in Growing Season of Tropical Deciduous Forests as Driven by El Niño and La Niña during 2001–2016

Phan Kieu Diem 1, Uday Pimple 1 ID , Asamaporn Sitthi 2, Pariwate Varnakovida 3, Katsunori Tanaka 4, Sukan Pungkul 5, Kumron Leadprathom 5, Monique Y. LeClerc 6 and Amnat Chidthaisong 1,*

1 The Joint Graduate School of Energy and Environment and Center of Excellence on Energy Technology and Environment, King Mongkut’s University of Technology Thonburi, Bangkok 10140, Thailand; [email protected] (P.K.D.); [email protected] (U.P.) 2 Department of Geography, Faculty of Social Science, Srinakharinwirot University, Bangkok 10110, Thailand; [email protected] 3 Department of Mathematics, King Mongkut’s University of Technology Thonburi, King Mongkut’s University of Technology Thonburi Geospatial Engineering and Innovation Center, Bangkok 10140, Thailand; [email protected] 4 GRID INC. Ao Building 6F 3-11-7 Kita Aoyama, Minato-ku, Tokyo 107-0061, Japan; [email protected] or [email protected] 5 Royal Forest Department, 61 Phaholyothin Road, Chatuchak, Bangkok 10900, Thailand; [email protected] (S.P.); [email protected] (K.L.) 6 College of Agricultural and Environmental Sciences, The University of Georgia, Grifﬁn Campus, Grifﬁn, GA 30223, USA; [email protected] * Correspondence: [email protected] or [email protected]; Tel.: +66-2470-8309 (ext. 10)

Received: 22 May 2018; Accepted: 20 July 2018; Published: 25 July 2018

Abstract: This study investigated the spatiotemporal dynamics of tropical deciduous forest including dry dipterocarp forest (DDF) and mixed deciduous forest (MDF) and its phenological changes in responses to El Niño and La Niña during 2001–2016. Based on time series of Normalized Difference Vegetation Index (NDVI) extracted from Moderate Resolution Imaging Spectroradiometer (MODIS), the start of growing season (SOS), the end of growing season (EOS), and length of growing season (LOS) were derived. In absence of climatic ﬂuctuation, the SOS of DDF commonly started on 106 ± 7 DOY, delayed to 132 DOY in El Niño year (2010) and advanced to 87 DOY in La Niña year (2011). Thus, there was a delay of about 19 to 33 days in El Niño and an earlier onset of about 13 to 27 days in La Niña year. The SOS of MDF started almost same time as of DDF on the 107 ± 7 DOY during the neutral years and delayed to 127 DOY during El Niño, advanced to 92 DOY in La Niña year. The SOS of MDF was delayed by about 12 to 28 days in El Niño and was earlier about 8 to 22 days in La Niña. Corresponding to these shifts in SOS and LOS of both DDF and MDF were also induced by the El Niño–Southern Oscillation (ENSO).

Keywords: deciduous forest; phenology; normalized difference vegetation index (NDVI); El Niño and La Niña; MODIS; savitzky-golay

1. Introduction Tropical forests contain about 25% of the carbon in the terrestrial biosphere and account for 34% of Earth’s gross primary production [1]. Unlike temperate forests where temperatures ﬂuctuate widely during the course of a year, their variation in tropical forest is modest. The trees of tropical forest are

Forests 2018, 9, 448; doi:10.3390/f9080448 www.mdpi.com/journal/forests Forests 2018, 9, 448 2 of 20 thus adapted to grow in a relatively narrow temperature range. Hence, the relative impact of climate warming is likely greater in the tropics than in other regions because predicted changes in temperature are large compared to normal inter-annual variations [2]. In addition, changes in precipitation patterns such as a shift toward more extreme events and extended droughts under climate change may result in loss of tropical forest and in large amount of CO2 released to the atmosphere [3,4]. Among various types of tropical forests, tropical deciduous forest occupies about 43% of the forest area in the tropical belt with great diversity of species [5]. It provides valuable services involving biodiversity, water resources and carbon sinks. However, like other forest ecosystems, these services are being affected by climate change and variability. Cavaleri [6] for example, reported a decline in the carbon sink of tropical rainforest during El Niño because of reduced photosynthesis and increased respiration rates. Strong climatic disturbances can severely reduce forest biomass, and if the frequency and intensity of these events increases beyond historical averages, these changing disturbance regimes have the capacity to significantly reduce forest biomass, resulting in a net source of carbon to the atmosphere. The study on Atlantic tropical moist forest in a long-term experimental plot also indicated a rapid biomass decline associated with El Niño events [7]. Many others studies have indicated that strong El Niño events have negative impacts on forest ecosystems, which could result in significant increasing level of tree mortality, changing plant phenology and carbon flux [3,4,8–12]. Forest phenology including the start, end and duration of growing season is an important indicator of vegetation response to climate change [13]. However, the information on tropical deciduous forest response to such climate extremes is sparse. Preliminary assessment over Southeast Asia suggests that decreasing precipitation and unusual high temperature in relation to severe drought (El Niño event) results in a significant reduction the CO2 uptake [14]. Such reduced CO2 uptake in tropical forests may be caused by reduced photosynthetic activity, shortened growing season, or a combination of both [6,15]. Further improvements in our understanding of tropical forests responding to climatic drivers and its links to ecosystem functions is thus needed. In this study we applied the Normalized Difference Vegetation Index (NDVI), one of the most widely used indices, to quantify the interannual variation in canopy phenology of a tropical deciduous forest in Northern Thailand and to evaluate canopy response to extreme climate events. The indices derived from remote sensing were compared against in situ observations of canopy phenology. The insight gained through such observations will help to improve our understanding of key feedback mechanisms and our ability to predict vegetation dynamics under climate changes scenarios.

2. Materials and Methods

2.1. Study Area Lampang Province is located in Northern Thailand with an area of 12,534 square kilometers (Figure1). The Asian monsoon with two distinct seasons governs the climate in Lampang; a wet (May–October) and a dry season (November–April). The mean annual precipitation during the study period 2001–2016 is 1231 mm. The mean annual temperature is 25 ◦C, with a monthly minimum mean of 21–22 ◦C occurring in December–January, and a maximum monthly mean of 29.5 ◦C occurring in April [16]. Forest cover during the study period was on average 68.5% of total province area. Deciduous forest accounts for 68.3% of the overall forest coverage (Figure1)[ 17]. Lampang is located on the area with the altitude ranging from 114 m to 1939 m above mean sea level. The main forest types in this study area are deciduous forests, mixed deciduous forest, and evergreen forest [18]. We selected Lampang as our study area for three reasons. Firstly, Lampang has a climate regime that is representative of northern Thailand and has experienced El Niño events during study period [19]. Secondly, there are long term ﬁeld observations of leaf area index (LAI) in the Mea Mo teak plantation (2001–2012). Thirdly, the study area consists of a complex mountainous topography and the forest diversity varies according to different elevation. For example, lower elevation (300 m to 800 m) is dominated by the 3 of 20 plantation (2001–2012). Thirdly, the study area consists of a complex mountainous topography and theForests forest2018, 9 diversity, 448 varies according to different elevation. For example, lower elevation3 of 20 (300 m to 800 m) is dominated by the dry dipterocarp forest and Tectona grandis L. f (teak), and the higherdry elevation dipterocarp (800 forest m andto 1000Tectona m) grandis is dominatedL. f (teak), andby the the highermixed elevation association (800 mof todeciduous 1000 m) is and evergreendominated hardwood, by the mixed and the association region above of deciduous 1000 m and is evergreendominated hardwood, by primary and theand region evergreen above with Pinus1000 kesiya m isRoy. dominated Ex Gord. by primary (Pinaceae, and evergreenpine) [20]. with ThePinus complex kesiya Roy. topography, Ex Gord. (Pinaceae, forest ecosystem pine) [20]. and functioningThe complex make topography, it ideal for forest evaluating ecosystem the and spatial functioning variation make itin ideal growing for evaluating season theof spatialdeciduous forest.variation in growing season of deciduous forest.

FigureFigure 1. Location 1. Location of of studystudy area area and and extent extent of Thailand of Thailand forest map forest (Thai Royalmap Forest(Thai Department Royal Forest 2007/2008); the red dot shows the location of climate station; the blue dot shows the location of site Department 2007/2008); the red dot shows the location of climate station; the blue dot shows the observation where the in situ leaf area index (LAI) was collected. location of site observation where the in situ leaf area index (LAI) was collected. 2.2. Data Set 2.2. Data Set 2.2.1. MODIS Fata 2.2.1. MODISThis studyFata used surface reﬂectance of MOD09Q1 from MODIS bands 1 (Red) and band 2 (NIR) at 250 m resolution, and MOD09A1 at 500 m resolution imagery captured in an 8-day period (2001–2016). This study used surface reflectance of MOD09Q1 from MODIS bands 1 (Red) and band 2 The data were downloaded from the EROS Data Center, US Geological Survey [21,22]. Cloud cover (NIR)was at present250 m inresolution, MOD09Q1 images,and MOD09A1 which limited at 500 the potentialm resolution of the imagesimagery for groundcaptured information in an 8-day periodextraction. (2001–2016) Removing. The anddata replacing were downloaded cloud-contaminated from pixelsthe EROS was then Data performed Center, following US Geological the Surveymethod [21,22]. of HoanCloud and cover Tateishi was [present23]. Cloud in removalMOD09Q1 processing images, for which band limited 1 and band the 2potential of MODIS of the imagesMOD09Q1 for ground was divided information into three extraction main steps:. Removing cloud mask and extraction, replacing interpolation cloud-contaminated to remove cloud pixels was thencoverage performed and median following smoothing. the Formethod cloud of mask, Hoan combination and Tateishi of internal [23]. Cloud cloud algorithm removal ﬂag processing (bit 10) and cloud shadow (bit 2) from 500 m State Flags layer of MOD09A1 (500 m) product was selected. for band 1 and band 2 of MODIS MOD09Q1 was divided into three main steps: cloud mask extraction, interpolation to remove cloud coverage and median smoothing. For cloud mask, combination of internal cloud algorithm flag (bit 10) and cloud shadow (bit 2) from 500 m State Flags layer of MOD09A1 (500 m) product was selected. The interpolation method for replacing cloud-contaminated areas was applied for each pixel (see detail in [23]). The NDVI was then calculated by using Red and NIR band [24].

Forests 2018, 9, 448 4 of 20

The interpolation method for replacing cloud-contaminated areas was applied for each pixel (see detail in [23]). The NDVI was then calculated by using Red and NIR band [24].

2.2.2. Landsat Imagery The dataset Operational Land Imager (OLI-8) images were downloaded from U.S. Geological Survey [25] by using the Global Visualization Viewer (GloVis). The images with less than 10% cloud cover were chosen to minimize the effect of cloud to interpreting forest areas. The red, green, blue, NIR, SWIR-1, and SWIR-2 bands were used in the forest classiﬁcation. List of LANDSAT imagery used in this study is shown in Table1.

Table 1. Speciﬁcations of Operational Land Imager-8 (OLI) imagery used in this study.

No Landsat Path/Row Date Acquired 1 Landsat OLI-8 130/47 16-01-2017 2 Landsat OLI-8 130/47 16-01-2017 3 Landsat OLI-8 131/47 08-02-2017 4 Landsat OLI-8 131/47 08-02-2017

2.2.3. Digital Elevation Models The Shuttle Radar Topography Mission (STRM) 30 m × 30 m resolution obtained from the U.S. Geological Survey (USGS) were used in this study for topographic correction [26]. STRM was resampled using a nearest neighborhood transformation [27,28].

2.2.4. Climate Variables Local climate variables including daily maximum air temperature and precipitation were obtained for the period of 2001–2016 from Thai Meteorological Department. There are total of 123 meteorological stations over Thailand. However, there is only one station located within the study area (18◦170 N; 99◦310 E), and this was used as a representative of climate in Lampang. It should be noted that this station was only used to evaluate climatic trends and variation of the study area. It is not intended to characterize the climate of the whole region. The internal consistency and temporal outliers check were performed for climate data set. The missing value and outlier values on data set was checked and removed [29,30]. The daily maximum temperatures and precipitation were averaged and aggregated. Annual anomaly was calculated as the difference between the 16-year average and the individual year [31,32]. A positive anomaly value indicates that the observed value was greater than the average, while a negative anomaly indicates that the observed value was less than the average for the period 2001–2016.

2.3. Methods This study consists of three main parts: forest classiﬁcation, evaluation of the difference between in situ and satellite based phenological metrics, and the inﬂuence of the El Niño–Southern Oscillation (ENSO) to phenological metrics variation. A detailed description of the data, methods and outputs used in the study is provided in Figure2. Forests 2018, 9, 448 5 of 20 5 of 20 Data Collection Data Processing Results

Figure 2. The flowchart ﬂowchart of methodology in this study.

2.3.1.2.3.1. Forest Forest Classiﬁcation Classification in in Lampang Lampang

ForFor pre-processing pre-processing LANDSAT LANDSAT imagery, imagery, the top-of-atmosphere the top-of-atmosphere (TOA) reflectance (TOA) reflectance and atmospheric and . correctionatmospheric (dark correction pixel subtraction (dark pixel method) subtraction were applied method) [33 were]. The applied cloud, cloud[33] The shadow cloud, and cloud water pixelsshadow were and identified water pixels by applying were identified the Function by ofap Maskplying (Fmask) the Function algorithm of Mask [34,35 (Fmask)] and were algorithm excluded from[34,35] further and were analyses. excluded Additionally, from further Landsat analys sceneses. Additionally, contain jagged Landsat pixels scenes along contain scene jagged edges, whichpixels could alongresult scene in edges, incorrect which reflectance could result values in alongincorrect the edgereflectance of imagery. values Thesealong jaggedthe edge pixels of wereimagery. removed These using jagged a 450 pixels m buffer were appliedremoved from using the a 450 edge m of buffer the mask applied inward from [ 36the,37 edge]. The of Otsuthe . thresholdmask inward was then[36,37] performed The Otsu based threshold on NDVI was calculatedthen performed from Redbased and on NIRNDVI bandcalculated of Landsat from to . separateRed and the NIR forest band and of non-forestLandsat to areas separate [38]. the The forest topographic and non-forest correction areas with [38] StatisticalThe topographic Empirical Correctioncorrection method with Statistical was performed Empirical on each Correction stratified me forestthod areawas toperformed remove the on topographic each stratified relief forest effect . fromarea Landsat to remove imagery the topographic [28]. The forest relief object effect wasfrom then Landsat used imagery to classify [28] theThe forest forest type object [39]. was then usedThe to stratifiedclassify the random forest samplingtype [39]. approach was performed in this study to estimate the number of samplesThe pointsstratified for random each class sampling [28,40]. Theapproach selection was of aperformed random sample in this of study training to dataestimate across the all . landnumber use types of samples ensures points that samples for each haveclass class[28,40] proportionsThe selection representative of a random of sample each forest of training type. In data total, thereacross were all 458land sample use types points ensures selected that for samp all classesles have (Figure class 3proportions). These 458 representative points included of 50each pointsforest for . evergreentype. In foresttotal, (EF),there 133 were points 458 forsample mixed points deciduous selected forest for (MDF), all classes 50 points (Figure for 3) dryThese dipterocarp 458 points forest (DDF),included and 50 225 points points for for evergreen other vegetation forest (EF), types 133 and points non-forest for mixed class. deciduous A total of 40 forest of the (MDF), 458 points 50 werepoints validated for dry duringdipterocarp the field forest survey. (DDF), The and rest 225 were points selected for other manually vegetation using types a combination and non-forest of the class. A total of 40 of the 458 points were validated during the field survey. The rest were selected manually using a combination of the time series of MODIS NDVI during 2001–2016, the

Forests 2018, 9, 448 6 of 20 6 of 20 interpretationtime series of of MODIS high-resolution NDVI during imagery, 2001–2016, G theoogle interpretation Earth images, of high-resolution and aerial imagery, photographs Google in QuantumEarth images, GIS. andTo avoid aerial photographsthe spatial autocorrelation in Quantum GIS. Toa minimum avoid the spatial distance autocorrelation of 2000 m awas minimum required betweendistance selected of 2000 mpoints was required [28]. A betweenminimum selected of 50 points training [28]. samples A minimum were of 50used training for each samples forest were class [28].used for each forest class [28].

Figure 3. Location of training and validation points in Lampang. A total 458 points, 320 points (70%) Figure 3. Location of training and validation points in Lampang. A total 458 points, 320 points randomly selected for training data set and 138 points (30%) for validation data samples. (70%) randomly selected for training data set and 138 points (30%) for validation data samples.

TheThe Random Random ForestForest (RF) (RF) Classifier, Classifier, a widely a widely used non-parametric used non-parametric machine learning machine classifier, learning was then applied for forest classification. RF is based on a tree classifier and grows many classification classifier, was then applied for forest classification. RF is based on a tree classifier and grows trees [28,39]. From the trained samples, the random Forest and raster packages were used to map manyforest classificati types fromon the trees set of[28,39] 06 spectral. From predictor the trained variables samples, (n = 500 classificationthe random trees).Forest For and the RFraster packagesclassification, were 70%used of to sample map pointsforest (320 types points) from were the used set of to train06 spectral the model, predictor while the variables remaining (n 30% = 500 classificationof sample points trees). (138 For points) the RF were classification, used to validate 70% theof sample classifier points (Figure (3203). The points) overall were accuracy used to and train theKappa model, statistic while were the considered remaining for 30% assessing of sample the accuracy points of (138 RF classifier. points) Thewere two used major to of validate classified the classifierdeciduous (Figure forest 3). types The includingoverall accuracy MDF and and DDF Kappa were used statistic as mask were to considered analysis their for phenology assessing in the accuracyresponse of to RF climate classifier extreme. The event. two major of classified deciduous forest types including MDF and DDF were used as mask to analysis their phenology in response to climate extreme event.

Forests 2018, 9, 448 7 of 20

2.3.2. NDVI Time Series for Phenological Analysis To examine the temporal signatures of deciduous forest classes, we extracted Normalized Difference Vegetation Index (NDVI) from MODIS to produce the profiles of vegetation dynamics during each growing season. NDVI was calculated from Red and Near infrared as described by [24]. The remaining noise in NDVI time series data was removed by using the Savitzky-Golay built within the TIMESAT software. The smoothed NDVI series was used to extract phenological metrics by using the threshold-based method [41–44]. In this study, we examined mean value of NDVI time series during 2001–2016 of two different forest types: DDF and MDF. The average of NDVI pixel locations of DDF and MDF was confirmed by checking the coordination from field survey.

2.3.3. Determination of Forest Phenological Variables from Satellite TIMESAT software was used to extract the tropical deciduous forest phenology metrics for each year (2001–2015). An adaptive Savitsky-Golay ﬁltering function in TIMESAT software was used to ﬁt the curve of time series NDVI, which has been widely used for phenology monitoring [41,43,45,46]. Three phenological metrics were extracted as follows:

1. Start of growing season (SOS): This is deﬁned as the date of leaf unfolding (day of year, DOY) and this study considered SOS as a date when NDVI of the left edge has increases 20 percentage measured from the left minimum point. 2. End of the season (EOS): This is deﬁned as the dates of leaf discoloration (day of year, DOY) and leaf fall at the end of season. This study considered EOS as a date when NDVI of the right edge has decreases to 20 percentage of the right minimum level. 3. Length of the season (LOS): This is the duration (number of days) from the start to the end of the season.

The analysis of phenological metrics was performed for each pixel in this study. Note that one growing season cycle of tropical deciduous forest starts from current year and ends in the following year. Therefore, only 15 years of phenological metrics were extracted from NDVI time series of 2001–2016. To avoid long-term changes pixels classiﬁed as deciduous forest, only pixels for which the SOS ranged between 15 and 180 DOY for the whole study period were selected for further analysis.

2.3.4. Determination of in situ Derived Forest Phenological Metrics The daily in situ LAI was aggregated to 8-day temporal resolution by using the mean function, with the same temporal resolution to MODIS NDVI. The Savitsky-Golay ﬁltering was also performed with LAI data set. The phenological metrics were then extracted by the same method of satellite-based phenological metrics.

2.3.5. Validation of the Phenological Metrics The ground observation site was located in Mae Mo District, Lampang Province (18◦250 N, 99◦430 E), in teak plantations (Tectona grandis Linn. f.) which was a major forest type in Lampang. In situ observation of leaf area index (LAI) from January 2001 to December 2012 was used to validate the time series of NDVI derived from satellite imagery [47]. To avoid positional error in MODIS data, the 3 × 3 pixels NDVI time series was extracted corresponding to the center of in situ measurement location. The daily in situ LAI was aggregated to 8-day temporal resolution by using the mean function. The phenological metrics extracted from MODIS and LAI are henceforth referred to as SOSNDVI, LOSNDVI and SOSLAI, LOSLAI. To understand more relationship between LAI and NDVI in term of phenology, we extracted the phenological metrics by using same threshold-based of 20 percentage amplitude. Linear regression was used to examine the overall relationship between the phenology metrics from in situ measurements and satellite-derived estimates. 8 of 20

Forests2.3.6.2018 Assessing, 9, 448 the ENSO-Related Patterns in Annual Phenological Metrics 8 of 20 To examine the influence of ENSO events on tropical deciduous forest vegetation dynamics, 2.3.6.the difference Assessing of the phenological ENSO-Related metrics Patterns during inAnnual El Niño Phenological and La Niña Metrics years were evaluated for the study area. The extreme climate and neutral years were determined using the following criteria: To examine the influence of ENSO events on tropical deciduous forest vegetation dynamics, (1) any year with five consecutive Oceanic Niño Index (ONI) periods in excess of +0.5 °C (−0.5 °C) the difference of phenological metrics during El Niño and La Niña years were evaluated for the study area.is an TheEl Niño extreme (La Niña) climate year and [48], neutral (2) comparing years were against determined the annual using anomaly the following climate criteria: at Lampang (1) any year(see withdetail five in consecutiveSection 3.1). OceanicTo evaluate Niño the Index influence (ONI) of periods ENSO in events excess on of canopy+0.5 ◦C( phenology,−0.5 ◦C) is the an El Niñodifference (La Niña) between year El [48 Niño], (2) comparingand La Niña against impact the patterns annual were anomaly evaluated climate for at the Lampang study area (see detailusing in Sectiona simple 3.1 difference). To evaluate between the influence average El of Niño ENSO or events La Niña on canopyyear and phenology, neutral year the SOS difference and LOS. between To Elinvestigate Niño and potential La Niña impactdifferences patterns in ENSO were evaluatedimpacts on for different the study forest area types, using these a simple differences difference betweenwere grouped average by El forest Niño ortype La an Niñad the year significance and neutral of yeardifference SOS and was LOS. compared To investigate based on potential one differencesstandard deviation in ENSO (±1SD) impacts and on analysis different of forest variance types, (ANOVA) these differences at the confidence were grouped level byof 95% forest (p type ≤ and0.05). the significance of difference was compared based on one standard deviation (±1SD) and analysis of variance (ANOVA) at the confidence level of 95% (p ≤ 0.05). 3. Results and Discussion 3. Results and Discussion 3.1. Variations in Temperature and Precipitation during 2001–2016 3.1. Variations in Temperature and Precipitation during 2001–2016 A clear seasonality of temperatures and precipitation is a main feature of climate characteristicA clear seasonality in Northern of temperatures Thailand, including and precipitation in Lampang is a (Figure main feature 4). Three of climate clear seasons characteristic are as in Northernfollows: (1) Thailand, rainy or including southwest in Lampang monsoon (Figure season4). (mid-May Three clear to seasons mid-October) are as follows: with August (1) rainy to or southwestSeptember monsoon as wettest season period; (mid-May (2) tocool mid-October) or northeast with monsoon August to Septemberseason (mid-October as wettest period; to (2)mid-February), cool or northeast is the monsoon mild temper seasonature (mid-October period of tothe mid-February), year with the iscoolest the mild period temperature in December period ofand the January; year with and the (3) coolest summer period or inpre-monsoon December and season January; (mid-February and (3) summer to mid-May). or pre-monsoon This is seasonthe (mid-Februarytransitional period to mid-May). from the This northeast is the transitional to southwest period monsoons. from the northeastMarch to to May southwest is the monsoons.hottest Marchperiod to of May the isyear the with hottest maximum period of temperatures the year with often maximum near or temperatures exceeding 40 often °C. The near subsequent or exceeding 40onset◦C. Theof rainy subsequent season onsetsignificantly of rainy reduces season significantlythe temperatures reduces from the temperaturesmid-May, which from leads mid-May, to whichintensive leads rainfall to intensive from mid-May rainfall from until mid-May early October. until early October.

Precipitation Maximum Temperature Minimum Temperature 200 42 180

35 160 140

C 28

0 120 100 21 80 14 60 Temperature, 40 mm Precipitation, 7 20 0 0 1 3 88 67 46 25 90 69 51 30 173 345 152 324 131 303 110 282 260 239 218 197 175 347 154 326 133 305 112 284 262 241 223 202 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

FigureFigure 4. 4. DailyDaily temperature temperature and and precipitation precipitation records records at at Lampang Lampang station station during during 2001 2001–2016.–2016.

BasedBased on on the the criterion criterion that that any any year year with with five fi consecutiveve consecutive the Oceanicthe Oceanic Niño Niño Index Index (ONI) (ONI) periods, beginningperiods, beginning with the with preceding the preced Augusting throughAugust through October October (ASO) period (ASO) andperiod ending and ending with February with throughFebruary April through (FMA) April of the (FMA) currentof the year, current El Niño year, or LaEl Niño Niña yearsor La areNiña defined years are as anydefined year as excess any of + . . +0.5year◦ C(excess−0.5 of◦C)0 [485 °C]. All(−0 other5 °C) years[48]. All are other defined years as neutralare define years.d as Following neutral years. this criteria,Following the this years 2001, 2006, 2008, 2011, 2012 are considered as La Niña, the years 2003, 2005, 2007, 2010, 2015, 2016 are El Niño year (with 2010 and 2016 defined as strong El Niño years).

9 of 20

Forestscriteria,2018, 9the, 448 years 2001, 2006, 2008, 2011, 2012 are considered as La Niña, the years 2003, 2005,9 of 20 2007, 2010, 2015, 2016 are El Niño year (with 2010 and 2016 defined as strong El Niño years). Climate data analysis at Lampang station shows that the maximum temperature anomaly (+1.04Climate °C) was data observed analysis during at Lampang the El Nino station year shows (2010), that which the also maximum had a pronounced temperature negative anomaly ◦ (+1.04precipitationC) was anomaly observed (− during133.9 mm). the El Precipitation Nino year (2010), anomalies which during also hadthe aperiod pronounced of study negative were precipitationmostly negative, anomaly particularly (−133.9 mm).in El Niño Precipitation events, whereas anomalies the during precipitation the period anomalies of study values were mostlywere negative,highest in particularly 2006 and 2011 in El (which Niño events, indicates whereas the wettest the precipitation conditions) anomaliesduring the valuesstudy period were highest (Figure in 20065). andBased 2011 on (whichthese results indicates we the identified wettest conditions)2010 and 2011 during as strong the study El Nino period and (Figure La Nina5). Based years, on theserespectively, results we for identiﬁed use as extreme 2010 and events 2011 as in strong the impact El Nino analysis. and La The Nina remaining years, respectively, years (2001–2009 for use as extremeand 2012–2015, events in theincluding impact analysis.the weak The El remainingNiño and yearsLa Niña) (2001–2009 with andless 2012–2015,variation of including anomaly the weaktemperature El Niño andandLa anomaly Niña) with precipitation less variation were of considered anomaly temperature as neutral years and anomaly in this study. precipitation Two werephenological considered metrics as neutral during years neutral in this and study. extreme Two phenologicalevent years were metrics extracted during for neutral each andpixel, extreme then eventthe difference years were between extracted them for eachwas investigated. pixel, then the difference between them was investigated.

1.5 800

C) b) 0 a) 600 1

400 0.5 200 0 0 -0.5 -200

Maximum temperature anomaly ( -1 Yearly precipitation Yearly precipitation (mm) anomaly -400

-1.5 -600 2001 2003 2005 2007 2009 2011 2013 2015 2001 2003 2005 2007 2009 2011 2013 2015 Year Year

FigureFigure 5. 5.Annual Annual maximum maximum temperaturetemperature anomalyanomaly ( a) and and precipitation precipitation anomaly anomaly (b (b) )at at Lampang Lampang stationstation during during 2001–2016. 2001–2016.

3.2.3.2. Forest Forest Classification Classification RFRF classification classification forfor various different different forest forest types types from from OLI-8 OLI-8 is presented is presented in Figure in Figure 6. The6 . Theclassification classification of of land land use use in LampangLampang 20172017 was was conducted conducted with with five five land land use anduse landand coverland cover classes: EF,classes: MDF, DDF,EF, MDF, other DDF, vegetation other vegetation and non-forest. and non The-forest. total areasThe total of classified areas of deciduousclassified deciduous forest were 878,831forest hawere (70.4% 878,831 of total ha (70.4% areas), whereof total MDF area ands), where DDF occupiedMDF and 45.4% DDF andoccupied 24.9%, 45.4% respectively. and 24.9%, Areas ofrespectively. other land use Areas types of including other land evergreen use types forest, including other vegetation evergreen andforest, non-forest other vegetation were 370,428 and ha (29.7%non-forest of total were areas). 370,428 The machineha (29.7% learning of total RF areas). classifier The canmachine be implemented learning RF effectively classifier forcan forest be classificationimplemented in thiseffectively study site.for Theforest overall classification accuracy in of this the classifiedstudy site. map The was overall 87.8%. accuracy of the classified map was 87.8%.

Forests 2018, 9, 448 10 of 20 10 of 20

Figure 6. Forest classiﬁcationclassification map in Lampang (2017).(2017).

3.3. Variations in NDVI during 2001–2016 3.3. Variations in NDVI during 2001–2016

3.3.1.3.3.1. Relationship Relationship between between Satellite-Based Satellite-Based NDVI NDVI and and Observed Observed LAI LAI of of Teak Teak Forest Forest Plantation Plantation InIn this this study, study, satellite-based satellite-based NDVI NDVI waswas comparedcompared against the observed observed LAI LAI in in a alocal local teak teak forestforest plantation. plantation. NDVI NDVI showed showed a a strong strong exponential exponential relationship relationship with with the the observed observed LAI LAI (Figure (Figure7). During7). During the dry the season,dry season, LAI decreasedLAI decreased to the to minimum the minimum value value of about of abou zero,t zero, while while minimum minimum NDVI remainedNDVI remained around 0.3–0.4.around However,0.3–0.4. However, NDVI seemed NDVI toseemed be more to be sensitive more sensitive than LAI than during LAI theduring onset ofthe growing onset of season, growing as NDVIseason, increased as NDVI more increas rapidlyed more and rapidly reached and the reached maximum the valuemaximum earlier value than LAIearlier (Figure than7a). LAI The (Figure relationship 7a). The between relationship LAI andbetween NDVI LAI can and be expressedNDVI can asbe LAIexpressed= 0.013 ase 6.53NDVILAI , . R20.013e= 0.80. For this, R2 deciduous= 0.80. For teak this forest, deciduous the NDVI teak becomesforest, the saturated NDVI becomes at LAI around saturated≥ 2 at m LAI2/m 2 (Figurearound7b). ≥ 2 Potithep m2/m2 (Figure et al. [7b).49] Potithep also found et al. an [49] exponential also found relationship an exponential between relationship NDVI and between in situ LAINDVI in a and deciduous in situ LAI forest in anda deciduous NDVI was forest saturated and NDVI at values was saturated over 0.8. at Thus, values it is over conﬁrmed 0.8. Thus, that it theis pre-processingconfirmed that NDVI the datapre-processing set has a signiﬁcant NDVI data relationship set has a tosignificant LAI and itsrelationship capability forto LAI capturing and its the patterncapability of seasonality for capturing tropical the pattern deciduous of seasonality forest in this tropical study. deciduous forest in this study.

Forests 2018, 9, 448 11 of 20 11 of 20 11 of 20

3.5 1 3.5 LAI NDVI 1 5 LAI NDVI 5 0.9 3 0.9 4.5 3 4.5 y = 0.013e6.5301x 0.8 y = 0.013e6.5301x

) 0.8 4 R² = 0.80 2

) 4 R² = 0.80

2 2.5

/m 0.7 2.5 2

/m 0.7 3.5 2 3.5 ) 2

0.6 ) 2 3

2 0.6 /m 3 2

2 /m 2 0.5 2.5 0.5 2.5 1.5 NDVI value LAI (m 1.5 0.4NDVI value LAI (m 2 0.4 2 0.3 1.5 1 0.3 1 m (LAI, Leaf area index 1.5 Leaf area index m (LAI, Leaf area index 0.2 1 0.2 1 0.5 0.5 0.1 0.5 0.1 0.5 0 0 0 0 0 0 10/1/2000 6/28/2003 3/24/2006 12/18/2008 9/14/2011 0 0.2 0.4 0.6 0.8 1 10/1/2000 6/28/2003 3/24/2006 12/18/2008 9/14/2011 0 0.2 0.4 0.6 0.8 1 Time (mm/dd/yy) NDVI value Time (mm/dd/yy) NDVI value (a) (a) (b) (b) Figure 7. (a) Time series of in situ LAI against NDVI based on satellite data period 2001–2012; (b) the Figure 7. (a) Time series of in situ LAI against NDVI based on satellite data period 2001–2012; (b) the Figurescatter 7. (a )plot Time of series in situ of in LAI situ LAIand againstNDVI NDVIbased basedon satellite on satellite. The datasolid period line 2001–2012;shows the (bexponential) the scatter plot of in situ LAI and NDVI based on satellite. The solid line shows the exponential scatterrelationship plot of in situ (R2 LAI = 0. and80). NDVI based on satellite. The solid line shows the exponential relationship relationship (R2 = 0.80). (R2 = 0.80).

Figure 8 shows the relationship between and at same thresholds-based Figure 8 shows the relationship between and at same thresholds-based (20%).(20%).Figure Overall, 8Overall, shows the thetheSOS relationshipSOS derived derived from between from MODIS MODISSOS NDVILAI NDVIand agreedSOS agreedNDVI well wellat with same with that thresholds-based that derived derived from from LAI (20%). LAI Overall,observation the SOS data. derived However, from MODIS relationship NDVI agreedbetween well with thatand derived fromwas LAIrather observation weak. On observation data. However, relationship between and was rather weak. On data.average, However, relationship is 18 betweendays earlier LOS LAIthanand LOSNDVI andwas rather is weak. 30 days On average,longer than SOS NDVIis , average, is 18 days earlier than and is 30 days longer than , 18 daysrespectively. earlier than It SOSis notedLAI and thatLOS NDVIis and 30 days longer were than delayedLOSLAI ,about respectively. 4 days It and is noted 22 days, that respectively. It is noted that and were delayed about 4 days and 22 days, SOSrespectivelyLAIrespectivelyand SOS in ElNDVI in Niño Elwere Niño year delayed year (2010). (2010). about 4 days and 22 days, respectively in El Niño year (2010).

340 340 170 170 320 320 150 150 300 300 130 130 (day)

(day) 280 280 LAI (DOY) LAI (DOY) LAI

110 LOS

LAI y = 1.07x + 10.31 110 y = 1.07x + 10.31 LOS 260 SOS R² = 0.67 260 y = 0.68x + 65.19 SOS R² = 0.67 y = 0.68x + 65.19 R² = 0.28 90 240 R² = 0.28 90 240

70 220 70 220 80 100 120 275 295 315 335 80 100 120 275 295 315 335 SOSNDVI (DOY) LOS (day) SOSNDVI (DOY) LOS (day)NDVI NDVI (a) (a) (b) (b) Figure 8. The variation in temporal of phenological metrics derived from LAI and NDVI FigureFigure 8. 8.The The variation variation in temporalin temporal of phenological of phenolog metricsical metrics derived derived from LAI from and NDVILAI and observations NDVI observations period 2001–2011; (a) relationship between and and (b) periodobservations 2001–2011; period (a) relationship 2001–2011; between (a) relationshipSOSLAI and SOSbetweenNDVI and (b) relationship and between and LOS (b)LAI . relationship between and . andrelationship LOSNDVI between. and 3.3.2. Temporal Variations of NDVI in Lampang Province during 2001–2016 3.3.2.3.3.2. TemporalTemporal Variations Variations of of NDVI NDVI in in Lampang Lampang Province Province during during 2001–2016 2001–2016 In Lampang the rainy season starts around March to April, the same time as the deciduous InIn Lampang Lampang the the rainy rainy season season starts starts around around March Marc toh to April, April, the the same same time time as theas the deciduous deciduous forest forest starts to bud. A gradual increase in NDVI is also observed during this period (65–100 DOY). startsforest to starts bud. to A bud. gradual A gradual increase increase in NDVI in NDVI is also is observed also observed during during this period this period (65–100 (65– DOY).100 DOY). Rainy seasonRainyRainy season ends season around ends ends around October around October to October November, to November, to November, corresponding corresponding corresponding to the to decrease the to thedecrease decrease of NDVI of NDVI of values NDVI values during values thisduring periodduring this (Figurethis period period9 ).(Figure The (Figure timing 9). The9). of The timing the timing minimum of the of theminimum NDVI minimum value NDVI variedNDVI value amongvalue varied varied different among among forest different different types. / Theforest minimumforest types. types. The NDVI The minimum valueminimum of NDVI DDF NDVI occurredvalue value of DDF inofMarch/April DDF occurred occurred in and March in itMarch is/April lowerApril and than and it theis itlower minimumis lower than than NDVIthe the minimumminimum NDVI NDVI value value of MDF of MDF in general. in general. On Onthe theother other hand, hand, the themaximum maximum NDVI NDVI value value of DDF of DDF

12 of 20 occurred in June/July and it is also lower than the maximum NDVI of MDF. The seasonal pattern for growing season each year cycle can be described as follows: NDVI was generally at its lowest (0.3–0.4) during the hottest period of the year (February to ForestsApril),2018 which, 9, 448 is comparable to the field observations of LAI. NDVI then increased rapidly for 4–512 of 20 weeks, overlapping with the period of decreasing in temperature and increasing in precipitation resulting in leaf expansion. NDVI reached their maximum peak (0.8–0.9) in July–August,12 of 20 value of MDF in general. On the other hand, the maximum NDVI value of DDF occurred in June/July corresponding to maximum and saturation LAI value. NDVI then gradually decreased to below and it is also lower than the maximum NDVI of MDF. The seasonal pattern for growing season each 0.5occurred and LAI in Junebelow/July 2.0 and in November, it is also lower overlapping than the maximumwith the period NDVI with of MDF. lower The rainfall seasonal and pattern cooler year cycle can be described as follows: temperature.for growing season each year cycle can be described as follows: NDVI was generally at its lowest (0.3–0.4) during the hottest period of the year (February to April), which is comparable Timeto the series field NDVI observatio of DDF ns of LAI.Time NDVI series then NDVI increased of MixedDF rapidly for 4–5 0.9 weeks, overlapping with the period of decreasing in temperature and increasing in precipitation resulting0.8 in leaf expansion. NDVI reached their maximum peak (0.8–0.9) in July–August, corresponding to maximum and saturation LAI value. NDVI then gradually decreased to below 0.5 and0.7 LAI below 2.0 in November, overlapping with the period with lower rainfall and cooler temperature. 0.6 Time series NDVI of DDF Time series NDVI of MixedDF NDVI value 0.9 0.5

0.8 0.4

0.7 0.3 10/01/00 06/28/03 03/24/06 12/18/08 09/14/11 06/10/14 03/06/17 0.6 Time (mm/dd/yy) NDVI value Figure0.5Figure 9.9. Example of of seasonal seasonal different different in in NDVI NDVI timing timing values values of of DDF DDF (green (green)) and and MDF MDF (red) (red) periodperiod 2001–2016.2001–2016. 0.4 3.4. VariationsNDVI was of generally Phenological at its Metrics lowest and (0.3–0.4) the Effects during of ENSO the hottest period of the year (February to April), 0.3 which is comparable to the ﬁeld observations of LAI. NDVI then increased rapidly for 4–5 weeks, 3.4.1.10/01/00 Temporal Variations 06/28/03 03/24/06 12/18/08 09/14/11 06/10/14 03/06/17 overlapping with the period of decreasing inTime temperature (mm/dd/yy) and increasing in precipitation resulting in leafThe expansion. temporal NDVIvariations reached in tropical their maximum deciduous peak forest (0.8–0.9) phenology in July metrics–August, were corresponding examined for to Figure 9. Example of seasonal different in NDVI timing values of DDF (green) and MDF (red) maximumboth the dry and dipterocarp saturation LAI(DDF) value. and NDVI the mixed then graduallydeciduous decreased forests (MDF) to below pixels 0.5 andwhich LAI were below period 2001–2016. 2.0confirmed in November, from field overlapping survey locations. with the periodThe results with are lower shown rainfall in Figure and cooler 10. temperature.

3.4.3.4. Variations Variations of of Phenological Phenological Metrics Metrics and and the the Effects Effects of of ENSO ENSO 140 SOS_DDF SOS_MDF LOS_DDF LOS_MDF 370 3.4.1.130 Temporal Variations 3.4.1. Temporal Variations 350 120 110TheThe temporal temporal variations variations in in tropical tropical deciduous deciduou forests forest330 phenology phenology metrics metrics were were examined examined for for both both the dry dipterocarp (DDF) and the mixed deciduous forests (MDF) pixels which were the dry100 dipterocarp (DDF) and the mixed deciduous forests310 (MDF) pixels which were conﬁrmed from confirmed from field survey locations. The results are shown in Figure 10. ﬁeld survey90 locations. The results are shown in Figure 10. 290 Start of season (DOY) Start of 80 140 SOS_DDF SOS_MDF 270 Length of Season (day) Season Length of LOS_DDF LOS_MDF 70 370 130 250 60 350 1202001 2003 2005 2007 2009 2011 2013 2015 2001 2003 2005 2007 2009 2011 2013 2015 Year Year 110 330 (a) (b) 100 310 90 Figure 10. Inter-annual variations of tropical deciduous290 forest phenology metrics in different Start of season (DOY) Start of forest80 types period 2000–2015; (a) Start of growing season (SOS); (b) length of growing season 270 (LOS).70 (day) Season Length of 60 250 2001 2003 2005 2007 2009 2011 2013 2015 2001 2003 2005 2007 2009 2011 2013 2015 The average start of growing season for dry dipterocarp (SOS) and start growing season Year Year for mixed deciduous forests (SOS ) during 2001–2015 were at DOY 113 ± 16, and 113 ± 15, (a) (b)

Figure 10. Inter-annual variations of tropical deciduous forest phenology metrics in different Figure 10. Inter-annual variations of tropical deciduous forest phenology metrics in different forest forest types period 2000–2015; (a) Start of growing season (SOS); (b) length of growing season types period 2000–2015; (a) Start of growing season (SOS); (b) length of growing season (LOS). (LOS).

The average start of growing season for dry dipterocarp (SOS) and start growing season for mixed deciduous forests (SOS) during 2001–2015 were at DOY 113 ± 16, and 113 ± 15,

Forests 2018, 9, 448 13 of 20 13 of 20

The average start of growing season for dry dipterocarp (SOS ) and start growing season for respectively. On the other hand, length growing season of dry DDFdipterocarp (LOS) was about mixed deciduous forests (SOS ) during 2001–2015 were at DOY 113 ± 16, and 113 ± 15, respectively. 302 ± 18 days, slightly shorterMDF than the LOS of 309 ± 17 days. During the El Niño (Year 2010), On the other hand, length growing season of dry dipterocarp (LOS ) was about 302 ± 18 days, SOS and SOS was delayed about 26 days and 23 days, DDFrespectively and LOS and slightly shorter than the LOS of 309 ± 17 days. During the El Niño (Year 2010), SOS and LOS was 19 days and 30MDF days shorter than the neutral year, respectively. On the otherDDF hand, SOS was delayed about 26 days and 23 days, respectively and LOS and LOS was 19 days duringMDF La Niña (Year 2011) SOS and SOS was 23 and 18DDF days advanced,MDF whereas and 30 days shorter than the neutral year, respectively. On the other hand, during La Niña (Year 2011) LOS and LOS was 24 and 13 days, respectively, longer than in the neutral years. An SOSexampleDDF and ofSOS theMDF differencewas 23 andin phenological 18 days advanced, metrics whereas betweenLOS El NiñoDDF and (YearLOS 2010)MDF andwas La 24 andNiña 13 (Year days, respectively,2011) to average longer of than phenology in the neutral metrics years. during An 2001–20 example15 of is theillustrated difference in Figure in phenological 11. It is obvious metrics betweenthat ENSO El Niño significantly (Year 2010) affects and La the Niña start (Year of 2011) the season to average while of phenologythe effects metricsof ENSO during on the 2001–2015 end of is illustratedseason was in not Figure clear 11 as. Itthat is obvious of the start that of ENSO the season. signiﬁcantly affects the start of the season while the effects of ENSO on the end of season was not clear as that of the start of the season.

Neutral years El Nino (2010) La Nina (2011) 0.9

0.8

0.7

SOSMDF in La Nina year, 92DOY 0.6 NDVI value SOSMDF in neutral years, 107DOY 0.5

SOSMDF in El Nino year, 126DOY 0.4

0.3 11-Dec 30-Jan 21-Mar 10-May 29-Jun 18-Aug 7-Oct 26-Nov 15-Jan 6-Mar Time (dd/mm)

FigureFigure 11. 11An. An example example of of difference difference in in phenological phenological metricsmetrics amongamong El Niño (Year (Year 2010), 2010), La La Niña Niña (Year(Year 2011) 2011) and and Neutral Neutral years years at group at group pixels pixels of MDF of MDF derived derived from from MODIS MODIS NDVI. NDVI.

3.4.2.3.4.2. Spatial Spatial Variations Variations InIn order order to investigateto investigate more more details details of phenological of phenological metrics metrics for two for differenttwo different deciduous deciduous forest typesforest (DDF types and (DDF MDF), and we MDF), created we the created mask forthe differentmask for forest different types, forest then types, phenological then phenological metrics for eachmetrics forest for types each were forest extracted types were and extracted analyzed and separately analyzed (Figures separately 12 and (Figures 13). We12 and can 13). see We that can the earliestsee thatSOS theDDF earliestwas in SOS March was (DOY in 80–90)March (DOY which 80–90) mainly which occurred mainly in southernoccurred partin southern of Lampang. part Theof latest Lampang.SOSDDF Themainly latest SOS appeared mainly in May appeared (DOY 130–140).in May (DOY This 130–140). mainly occurredThis mainly in theoccurred northern in partthe of northern Lampang part which of mayLampang be caused which by may differences be caused in precipitation by differences and/or in precipitation altitude (Figure and/or 12).

Onaltitude the other (Figure hand, 12). the On average the otherSOSMDF hand,ranged the average between SOS DOY 80ranged and DOY between 140. DOY Similar 80 toandSOS DOYDDF , SOS140.MDF Similaroccurred to earlierSOS in, SOS the loweroccurred altitude earlier of the in southern the lower part altitude compared of tothe northern southern parts part of Lampangcompared (Figure to northern 12). parts of Lampang (Figure 12).

ForFor LOS, LOS,LOS LOSDDF mainlymainly ranged ranged from from 280 280 toto 330330 daysdays whilewhile LOSLOSMDF fromfrom 290 290 to to 340 340 days. days. ThereThere is nois obviousno obvious difference difference in LOS in DDFLOSand andLOS MDFLOS betweenbetween northern northern and and southern southern parts parts as those as foundthose in found SOS (Figurein SOS (Figure13). There 13). isThere a slight is a differenceslight difference of phenology of phenology (SOS (SOS and LOS) and LOS) between between DDF andDDF MDF, and and MDF, their and responses their responses to ENSO year to ENSO is analyzed year below.is analyzed The averagebelow. SOSTheDDF averagein neutral SOS yearsin duringneutral 2001–2015 years during occurred 2001–2015 at DOY occurred 106.4 ± at7, DOY similarily 106.4 to± 7,SOS similarilyMDF at DOY to SOS 106 ±at7. DOY On the 106 other ± 7. hand,On the averageother hand, of LOS theDDF averagein neutral of LOS years in was neutral about years 307 ± was14 days,about slightly307 ± 14 shorter days, thanslightly the LOSshorterMDF with than 319 the± LOS13 days with (Table 3192). ± 13 days (Table 2).

Forests 2018, 9, 448 14 of 20 14 of 20

(a) (b)

Figure 12. The spatial difference of SOS and SOS between El Niño (a,c) and La Niña (b,d) Figure 12. The spatial difference of SOSDDF and SOSMDF between El Niño (a,c) and La Niña (b,d) respectively, compared to that of the neutral year. The significant difference is based on 1SD. respectively, compared to that of the neutral year. The signiﬁcant difference is based on 1SD.

Forests 2018, 9, 448 15 of 20 15 of 20

(a) (b)

Figure 13. The spatial different of LOS and LOS between neutral year to El Niño (a,c), La Figure 13. The spatial different of LOSDDF and LOSMDF between neutral year to El Niño (a,c), La Niña Niña (b,d) respectively. The significant difference is based on 1SD. (b,d) respectively. The signiﬁcant difference is based on 1SD.

Forests 2018, 9, 448 16 of 20

Table 2. Results of phenological metrics difference during extreme climate events across two forest types in Lampang, Northern of Thailand. The percentage showing the pixels which signiﬁcant different in temporal between average neutral years and ENSO year.

The Area Size The Areasize Neutral Year El Niño Year La Niña Year Delayed by El Niño Advanced by La Niña (Mean ± 1SD) (2010) (2011) (2010) (±Ha, 1SD) (2011) (±Ha, 1SD)

SOSDDF (DOY) 106.4 ± 7.1 131.8 86.5 168,337.5 (75.3%) 147,718.8 (66.0%) SOSMDF (DOY) 106.9 ± 6.9 126.9 92.4 210,662.5 (65.8%) 153,762.5 (48.0%) LOSDDF (Day) 306.9 ± 13.9 279.0 321.9 214,806.3 (96.0%) 221,818.8 (99.2%) LOSMDF (Day) 319.0 ± 12.5 292.7 327.3 307,262.5 (96.0%) 307,931.3 (96.2%)

Overall, the growing season (SOSDDF) started on the 106 ± 7 DOY during the neutral years, delayed to 132 DOY during El Niño (Year 2010) and advanced to 87 DOY in La Niña year (Year 2011). It was delayed about 19 to 33 days in El Niño and was earlier about 13 to 27 days in La Niña (1SD). SOSMDF started almost same time as of SOSDDF on the 107 ± 7 DOY during the neutral years but was delayed to 127 DOY and advanced to 92 DOY during El Niño and La Niña years, respectively. The SOSMDF was delayed by about 12 to 28 days in El Niño and was earlier by about 8 to 22 days in La Niña (1SD). The LOS of these tropical deciduous forests, as indicated by the presence of live canopy, was significantly shorter in that El Niño year compared to neutral years in this study. The results from ANOVA test confirmed that the difference in SOS or LOS between ENSO and neutral years is statistically significant (p ≤ 0.05). During the El Niño (Year 2010), SOSDDF and SOSMDF was delayed about 35 days and 29 days, respectively and LOSDDF and LOSMDF was 31 days and 29 days shorter than the neutral years, respectively. On the other hand, during La Niña (Year 2011) SOSDDF and SOSMDF was 24 and 23 days advanced, whereas LOSDDF and LOSMDF was 15 and 9 days, respectively, longer than in the neutral years (p ≤ 0.05). However, we acknowledge that the large sample sizes may result in significant test even if the difference is small and biologically insignificant. The analysis presented here showed a significant impact of extreme climate events (El Niño and La Niña years) on the timing of leaf flush and the length of growing season in two tropical deciduous forest types in northern Thailand. Moreover, the degree and nature of impact was different between the forest types. In this study, during the El Niño year, the anomaly of maximum temperature significantly increased and summer monsoon rainfall significantly decreased (Figure5). Suepa et al. [43] also indicated that at there is significant reduction of vegetation growth during EL Niño year. The mechanisms controlling leaf phenology in dry tropical forest are related to water limitation as opposed to light limitation in tropical rainforests [9]. A previous study at the same location [47] found that the interannual variation of the timing of leaf flush responded to increasing soil moisture during March to May. Recent studies of climate impacts on leaf phenology in deciduous species within Asian monsoon and neotropical climate regions observed that differences in rehydration processes following the dry season associated with differences in micro-sites water availability were key in determining the primary factors controlling leaf flush and its response to rainfall [50,51]. In our study, higher maximum temperature and lower precipitation during summer (March, April) likely result in higher water stress during the dry season, which leads to a significant delay in start of growing season during El Niño. In contrast, the higher precipitation and lower maximum temperature in dry season leads to advanced SOS during La Niña year. The response of tropical deciduous forest was also found in the other El Niño years (e.g., in 2004, Figure5) where anomaly temperature was higher (+0.37 ◦C) and precipitation was much lower (−373.3 mm) than neutral years. The average of SOSDDF and SOSMDF in 2004 were 122 ± 12 DOY and 120 ± 13 DOY, a delay of 16 and 13 days, respectively. The average of LOSDDF and LOSMDF were 277 ± 18 days and 291 ± 18 days, 30 and 28 days respectively shorter compared to the neutral years. This result confirms the effect of El Niño to shifting of tropical deciduous forest phenology metrics. From the analysis we did not find a significant difference in SOS and LOS between DDF and MDF in Forests 2018, 9, 448 17 of 20

El Niño year. However, LOS of DDF was signiﬁcantly longer (15 days) than that LOS of MDF (8 days) in La Niña years (Table2). In other years, La Niña (e.g., 2006) where anomaly temperature was close to zero and precipitation was much higher than the average, the variations of tropical deciduous forest phenology metrics were small. It could be explained by very small variation of maximum temperature anomaly and precipitation anomaly in March and April of year 2006. Differences in phenological response to ENSO events between the two forest types were likely related to variations in soil moisture and elevation. The deciduous dipterocarp forests and the teak plantation area are generally located in lower elevation areas ranging from 300 m to 800 m, while mixed deciduous forest extends mostly from 800 m to 1000 m [20]. At the lower elevation, deciduous dipterocarp forest canopy is more open and soils tend to be drier with thinner organic layers and coarser texture relative to the mixed deciduous forest [20]. This is consistent with lower NDVI values during the dry season in the DDF forest type relative to the MDF (Figure9). It is noted that this study does not consider other related factors that may affect phenology as solar radiation, species diversity, soil layer and soil moisture, which should be addressed in follow-up studies. Further investigation could reveal how the micro-climate, forest species, and soil moisture differ along elevation gradients. Long-term ﬁeld measurements at different topographic locations would help to enhance our understanding of micro-climatic impact on this forest ecosystem.

4. Conclusions It is important to gain a better understanding of the phenological responses of tropical deciduous forests to extreme climate events as it will help inform the development of adaptation strategies to reduce the risk declining forest health associated with future climate change. In this study, various data types including meteorological observation, remote sensing, forest inventory maps, digital elevation models, and in situ LAI observation were applied to evaluate the response of tropical deciduous forest to extreme climate anomalies including ENSO events. The results indicate that phenological metrics of tropical deciduous forest (including dry dipterocarp forest and mixed deciduous forest) varied in response to precipitation and temperature anomalies associated with ENSO events. During the El Niño, significant delay 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, significant advancement 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. Since frequency and intensity of extreme climate events including ENSO are predicted to increase in the future, tropical deciduous forests may become increasingly vulnerable. Furthermore, climatic impacts on the distribution and productivity of this large region could have significant feedback on global climate systems.

Author Contributions: P.K.D., A.C., U.P. and P.V. provided the overall idea for this research and designed the methodology and wrote the manuscript. P.K.D., U.P., A.S., S.P., and K.L. planned and carried out the field survey for topographic correction, forest classification, accuracy assessment. P.K.D. implemented the analysis on the effect of ENSO to forest phenology. K.T. provided and suggested the data for validation of phenology. M.Y.L. provided comments and edited on the paper. Funding: This research was funded by the 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). This study is part of a project entitled “Analysis of historic forest carbon changes in Myanmar and Thailand and the contribution of climate variability and extreme weather events”. Acknowledgments: We gratefully thank M. Alber for supports to this project. The research was carried out in collaboration with the Royal Forest Department, Thailand. We thank N. Yoshifuji of Forestry and Forest Products Research Institute, who provided LAI data at Mae Mo teak plantation and R. Kaewthongrach King Mongkut’s University of Technology Thonburi for supports of specific site information. Conflicts of Interest: The authors declare no conflict of interest. Forests 2018, 9, 448 18 of 20

References

1. U.S. DOE. Research Priorities for Tropical Ecosystems Under Climate Change Workshop Report; DOE/SC-0153; U.S. Department of Energy Ofﬁce of Science: Washington, DC, USA, 2012. 2. European Climate Foundation. The UN Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5): Implications for Business; European Climate Foundation: Cambridge, UK, 2013. 3. Nakagawa, M.; Tanaka, K.; Nakashizuka, T.; Ohkubo, T.; Kato, T.; Maeda, T.; Sato, K.; Miguchi, H.; Nagamasu, H.; Ogino, K.; et al. Impact of Severe Drought Associated with the 1997–1998 El Niño in a Tropical Forest in Sarawak. J. Trop. Ecol. 2000, 16, 355–367. [CrossRef] 4. Asner, G.P.; Townsend, A. Satellite Observation of El Nino Effects Effects on Amazon Forest Phenology and Productivity. Geophys. Res. Lett. 2000, 27, 981–984. [CrossRef] 5. FAO. Forest Resources Assessment 1990—Non-Tropical Developing Countries Mediterranean Region; FAO: Rome, Italy, 1990. 6. Cavaleri, M.A.; Coble, A.P.; Ryan, M.G.; Bauerle, W.L.; Loescher, H.W.; Oberbauer, S.F. Tropical Rainforest Carbon Sink Declines during El Niño as a Result of Reduced Photosynthesis and Increased Respiration Rates. New Phytol. 2017, 216, 136–149. [CrossRef][PubMed] 7. Rolim, S.G.; Jesus, R.M.; Nascimento, H.E.M.; Do Couto, H.T.Z.; Chambers, J.Q. Biomass Change in an Atlantic Tropical Moist Forest: The ENSO Effect in Permanent Sample Plots over a 22-Year Period. Oecologia 2005, 142, 238–246. [CrossRef][PubMed] 8. Holmgren, M.; Scheffer, M.; Ezcurra, E.; Gutiérrez, J.R.; Mohren, G.M.J. El Niño Effects on the Dynamics of Terrestrial Ecosystems. Trends Ecol. Evol. 2001, 16, 89–94. [CrossRef] 9. Pau, S.; Okin, G.S.; Gillespie, T.W. Asynchronous Response of Tropical Forest Leaf Phenology to Seasonal and El Niño-Driven Drought. PLoS ONE 2010, 5, e11325. [CrossRef][PubMed] 10. Slik, J.W.F. El Nino Droughts and Their Effects on Tree Species Composition and Diversity in Tropical Rain Forests. Oecologia 2004, 141, 114–120. [CrossRef][PubMed] 11. Patra, P.K.; Ishizawa, M.; Maksyutov, S.; Nakazawa, T.; Inoue, G. Role of Biomass Burning and Climate

Anomalies for Land-Atmosphere Carbon Fluxes Based on Inverse Modeling of Atmospheric CO2. Global Biogeochem. Cycles 2005, 19, 1–10. [CrossRef] 12. Li, J.; Fan, K.; Xu, Z. Asymmetric Response in Northeast Asia of Summer NDVI to the Preceding ENSO Cycle. Clim. Dyn. 2016, 47, 2765–2783. [CrossRef] 13. Richardson, A.D.; Keenan, T.F.; Migliavacca, M.; Ryu, Y.; Sonnentag, O.; Toomey, M. Climate Change, Phenology, and Phenological Control of Vegetation Feedbacks to the Climate System. Agric. For. Meteorol. 2013, 169, 156–173. [CrossRef] 14. Saigusa, N.; Yamamoto, S.; Hirata, R.; Ohtani, Y.; Ide, R.; Asanuma, J.; Gamo, M.; Hirano, T.; Kondo, H.;

Kosugi, Y.; et al. Temporal and Spatial Variations in the Seasonal Patterns of CO2 Flux in Boreal, Temperate, and Tropical Forests in East Asia. Agric. For. Meteorol. 2008, 148, 700–713. [CrossRef] 15. Nanda, A.; Prakash, H.M.; Suresh, H.S.; Murthy, Y.L.K. Canopy, Understorey Leaf Phenology and Seasonality in Tropical Dry Forest, Southern India. Adv. For. Lett. 2016, 5.[CrossRef] 16. Igarashi, Y.; Kumagai, T.; Yoshifuji, N.; Sato, T.; Tanaka, N.; Tanaka, K.; Suzuki, M.; Tantasirin, C. Environmental Control of Canopy Stomatal Conductance in a Tropical Deciduous Forest in Northern Thailand. Agric. For. Meteorol. 2015, 202, 1–10. [CrossRef] 17. Royal Forest Department of Thailand. Forestry in Thailand; Royal Forest Department of Thailand: Bangkok, Thailand, 2009. 18. Vaidhayakarn, C.; Maxwell, J.F. Ecological Status of the Lowland Deciduous Forest in Chang Kian. J. Sci. Technol. 2010, 4, 268–317. 19. Kirtphaiboon, S.; Wongwises, P.; Limsakul, A.; Sooktawee, S.; Humphries, U. Rainfall Variability over Thailand Related to the El Nino-Southern Oscillation (ENSO). J. Sustain. Energy Environ. 2014, 5, 37–42. 20. Maxwell, J.F.; Elliot, S.; Anusarnsunthorn, V. The Vegetation of Jae Sawn National Park, Lampang Province, Thailand. Nat. Hist. Bull. Siam Soc. 1997, 45, 71–97. 21. Land Processes Distributed Active Archive Center - LP DAAC, USGS. Available online: https://lpdaac.usgs. gov/products/modis_products_table/mod09q1 (accessed on 01 July 2017). 22. Land Processes Distributed Active Archive Center - LP DAAC, USGS. Available online: https://lpdaac.usgs. gov/products/modis_products_table/mod09a1 (accessed on 01 July 2017). Forests 2018, 9, 448 19 of 20

23. Hoan, N.T.; Tateishi, R. Global MODIS 250 M Dataset for 10 Years (2003–2012) User’s Manual; Chiba University: Chiba, Japan, 2013. 24. Tucker, C.J. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 1979, 8, 127–150. [CrossRef] 25. The Land Processes Distributed Active Archive Center. Available online: https://lpdaac.usgs.gov/data_ Access/glovis (accessed on 10 April 2017). 26. The Shuttle Radar Topographic Mission (SRTM). Available online: https://lta.cr.usgs.gov/SRTM (accessed on 22 October 2017). 27. Wu, Q.; Jin, Y.; Fan, H. Evaluating and Comparing Performances of Topographic Correction Methods Based on Multi-Source DEMs and Landsat-8 OLI Data. Int. J. Remote Sens. 2016, 37, 4712–4730. [CrossRef] 28. Pimple, U.; Sitthi, A.; Simonetti, D.; Pungkul, S.; Leadprathom, K.; Chidthaisong, A. Topographic Correction of Landsat TM-5 and Landsat OLI-8 Imagery to Improve the Performance of Forest Classiﬁcation in the Mountainous Terrain of Northeast Thailand. Sustainability 2017, 9, 258. [CrossRef] 29. Feng, S.; Hu, Q.; Qian, W. Quality Control of Daily Meteorological Data in China, 1951–2000: A New Dataset. Int. J. Climatol. 2004, 24, 853–870. [CrossRef] 30. Peterson, T.C.; Russell, V.; Richard, S.; Vyachevslav, R. Global Historical Climatology Network (GHCN) Quality Control of Monthly Temperature Data. Int. J. Climatol. 1998, 1179, 1169–1179. [CrossRef] 31. Friedl, M.A.; Gray, J.M.; Melaas, E.K.; Richardson, A.D.; Hufkens, K.; Keenan, T.F.; Bailey, A.; O’Keefe, J.A. Tale of Two Springs: Using Recent Climate Anomalies to Characterize the Sensitivity of Temperate Forest Phenology to Climate Change. Environ. Res. Lett. 2014, 9, 054006. [CrossRef] 32. Wolf, S.; Baldocchi, D.; Wolf, S.; Keenan, T.F.; Fisher, J.B.; Baldocchi, D.D.; Desai, A.R. Warm Spring Reduced Carbon Cycle Impact of the 2012 US Summer Drought Warm Spring Reduced Carbon Cycle Impact of the 2012 US. Summer Drought. 2016, 113, 5880–5885. 33. Bruce, C.M.; Hilbert, D.W. Pre-Processing Methodology for Application to Landsat TM/ETM + Imagery of the Wet Tropics; Cooperative Research Centre for Tropical Rainforest Ecology and Management, Rainforest CRC: Cairns, Australia, 2004. 34. Foga, S.; Scaramuzza, P.L.; Guo, S.; Zhu, Z.; Dilley, R.D.; Beckmann, T.; Schmidt, G.L.; Dwyer, J.L.; Joseph Hughes, M.; Laue, B. Cloud Detection Algorithm Comparison and Validation for Operational Landsat Data Products. Remote Sens. Environ. 2017, 194, 379–390. [CrossRef] 35. Zhu, Z.; Woodcock, C.E. Object-Based Cloud and Cloud Shadow Detection in Landsat Imagery. Remote Sens. Environ. 2012, 118, 83–94. [CrossRef] 36. Robinson, N.P.; Allred, B.W.; Jones, M.O.; Moreno, A.; Kimball, J.S.; Naugle, D.E.; Erickson, T.A.; Richardson, A.D. A Dynamic Landsat Derived Normalized Difference Vegetation Index (NDVI) Product for the Conterminous United States. Remote Sens. 2017, 9, 863. [CrossRef] 37. Pimple, U.; Simonetti, D.; Sitthi, A.; Pungkul, S.; Leadprathom, K.; Skupek, H.; Som-ard, J.; Gond, V.; Towprayoon, S. Google Earth Engine Based Three Decadal Landsat Imagery Analysis for Mapping of Mangrove Forests and Its Surroundings in the Trat Province of Thailand. J. Comput. Commun. 2018, 6, 247–264. [CrossRef] 38. Otsu, N.A. Threshold Selection Method from Gray-Level Histograms. IEEE Trans. Syst. Man. Cybern. 1979, 9, 62–66. [CrossRef] 39. Millard, K.; Richardson, M. On the Importance of Training Data Sample Selection in Random Forest Image Classiﬁcation: A Case Study in Peatland Ecosystem Mapping. Remote Sens. 2015, 7, 8489–8515. [CrossRef] 40. Olofsson, P.; Foody, G.M.; Herold, M.; Stehman, S.V.; Woodcock, C.E.; Wulder, M.A. Good Practices for Estimating Area and Assessing Accuracy of Land Change. Remote Sens. Environ. 2014, 148, 42–57. [CrossRef] 41. Jönsson, P.; Eklundh, L. TIMESAT—A Program for Analyzing Time-Series of Satellite Sensor Data. Comput. Geosci. 2004, 30, 833–845. [CrossRef] 42. Xu, L.; Li, B.; Yuan, Y.; Gao, X.; Zhang, T. A Temporal-Spatial Iteration Method to Reconstruct NDVI Time Series Datasets. Remote Sens. 2015, 7, 8906–8924. [CrossRef] 43. Suepa, T.; Qi, J.; Lawawirojwong, S.; Messina, J.P. Understanding Spatio-Temporal Variation of Vegetation Phenology and Rainfall Seasonality in the Monsoon Southeast Asia. Environ. Res. 2016, 147, 621–629. [CrossRef][PubMed] 44. He, Y.; Fan, G.F.; Zhang, X.W.; Li, Z.Q.; Gao, D.W. Vegetation Phenological Variation and Its Response to Climate Changes in Zhejiang Province. J. Nat. Resour. 2013, 2, 220–233. Forests 2018, 9, 448 20 of 20

45. Chen, J.; Chen, J.; Jo, P. A Simple Method for Reconstructing a High- Quality NDVI Time-Series Data Set Based on the Savitzky-Golay Filter A Simple Method for Reconstructing a High-Quality NDVI Time-Series Data Set Based on the Savitzky—Golay Filter. Remote Sens. Environ. 2004, 91, 332–344. [CrossRef] 46. Palacios-Orueta, A.; Huesca, M.; Whiting, M.L.; Litago, J.; Khanna, S.; Garcia, M.; Ustin, S.L. Derivation of Phenological Metrics by Function Fitting to Time-Series of Spectral Shape Indexes AS1 and AS2: Mapping Cotton Phenological Stages Using MODIS Time Series. Remote Sens. Environ. 2012, 126, 148–159. [CrossRef] 47. Yoshifuji, N.; Igarashi, Y.; Tanaka, N.; Tanaka, K.; Sato, T.; Tantasirin, C.; Suzuki, M. Inter-Annual Variation in the Response of Leaf-out Onset to Soil Moisture Increase in a Teak Plantation in Northern Thailand. Int. J. Biometeorol. 2014, 58, 2025–2029. [CrossRef][PubMed] 48. Dannenberg, M.P.; Song, C.; Hwang, T.; Wise, E.K. Empirical Evidence of El Nino-Southern Oscillation Inﬂuence on Land Surface Phenology and Productivity in the Western United States. Remote Sens. Environ. 2015, 159, 167–180. [CrossRef] 49. Potithep, S.; Nagai, S.; Nasahara, K.N.; Muraoka, H.; Suzuki, R. Two Separate Periods of the LAI-VIs Relationships Using in Situ Measurements in a Deciduous Broadleaf Forest. Agric. For. Meteorol. 2013, 169, 148–155. [CrossRef] 50. Borchert, R. Water Status and Development of Tropical Trees During Seasonal Drought. Trees. 1994, 8, 115–124. [CrossRef] 51. Elliott, S.; Baker, P.J.; Borchert, R. Leaf Flushing during the Dry Season: The Paradox of Asian Monsoon Forests. Glob. Ecol. Biogeogr. 2006, 15, 248–257. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).