The use of MERIS Terrestrial Chlorophyll Index to study spatio-temporal variation in vegetation phenology over India

Remote Sensing of Environment 114 (2010) 1388–1402

Contents lists available at ScienceDirect

Remote Sensing of Environment j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / r s e

The use of MERIS Terrestrial Chlorophyll Index to study spatio-temporal variation in vegetation phenology over India J. Dash ⁎, C. Jeganathan, P.M. Atkinson School of Geography, University of Southampton, Southampton, SO17 1BJ, United Kingdom

a r t i c l e

i n f o

Article history: Received 18 May 2009 Received in revised form 27 January 2010 Accepted 30 January 2010 Keywords: Phenology Time-series Chlorophyll index India Growing season Satellite data

a b s t r a c t India has a diverse set of vegetation types ranging from tropical evergreen to dry deciduous. The phenology of these natural vegetation types is often controlled by climatic condition. Estimating phenological variables will help in understanding the response of tropical and subtropical vegetation to climate change. The study investigated the annual and inter-annual variation in vegetation phenology in India using satellite remote sensing. The study used time-series data of the only available satellite measured index of terrestrial chlorophyll content (MERIS Terrestrial Chlorophyll Index) from 2003 to 2007 at 4.6 km spatial resolution. A strong coincidence was observed with expected phenological pattern, in particular, in interannual and latitudinal variability of key phenological variables. For major natural vegetation type the onset of greenness had greater latitudinal variation compared to the end of senescence and there was a small or no leafless period. In the 2003–04 growing season a late start for the onset of greenness was detected at low-to-mid latitudes and it was attributed to the extreme cold weather during the early part of 2003. The length of growing season varied from east to west for the major cropping areas in the Indo-Gangetic plain, for both the first and second crops. For the first time, this study attempted to establish a broad regional phenological pattern for India using remotely sensed estimation of canopy chlorophyll content using five years of data. The overall patterns of phenological variables detected from this study broadly coincide with the pattern of natural vegetation phenology revealed in earlier community level studies. The results of this study suggest the need for an organised network combining ground and space observation which is at presently missing in India. © 2010 Elsevier Inc. All rights reserved.

1. Introduction Climate influences vegetation growth: for example, increased temperature and levels of atmospheric carbon dioxide increase vegetation productivity and carbon sequestration and modify ecosystem function (Badeck et al., 2004; Harmon et al., 1990; Taylor & Lloyd, 1992; White et al., 2005). Research has shown that an increase in mean global temperature between 1982 and 1999 resulted in an increase in global net vegetation productivity by 6% (3.4 Pg of carbon) (Nemani et al., 2003). The prediction, in space and time, of vegetation phenological variables such as time of onset of ‘greenness’, time of end of ‘greenness’, duration of the growing season, rate of ‘green up’ and rate of senescence can provide the information needed to increase understanding of the effects of climate change on vegetation. Such phenological variables can be measured from ground or extracted from remotely sensed data. Ground-measured phenological variables provide species-specific information with high temporal resolution, but lack a spatial component (Studer et al., 2007). In contrast,

⁎ Corresponding author. E-mail address: [email protected] (J. Dash). 0034-4257/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2010.01.021

temporally frequent remotely sensed data provide a unique opportunity to estimate phenological variables at a range of scales from the local to global. In addition, phenological variables extracted from remotely sensed data have the potential to characterise seasonal variation in the response (and function) of ecosystems to changes in climatic variables (Harmon et al., 1990; Malingreau, 1986; Reed et al., 1994). Such information, in turn, can be used as an input to global biogeochemical cycle modelling. In the last four decades many studies have demonstrated the potential of multi-temporal remote sensing data to extract variables reflecting the phenological development of natural vegetation related to change in climatic variables (Goward et al., 1985; Justice et al., 1985; Lloyd, 1990; Reed et al., 1994; White et al., 2005). During the mid-to-late 1980s several studies used phenological variables extracted from Advanced Very High Resolution Radiometer (AVHRR) to study the seasonal pattern of natural vegetation and crops at regional to global scales (Goward et al., 1985; Justice et al., 1985; Lloyd, 1990; Malingreau, 1986; Townshend et al., 1987; White et al., 2005). Later, these phenological variables were linked to changes in climatic variables. For example, using phenological variables extracted from AVHRR satellite sensor images, it was shown that terrestrial vegetation between 45° N and 70° N had “greened-up” from 1981 to 1991 (Myneni et al., 1997), a phenomenon that was tightly linked

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

with positive changes in land surface temperature in this period (Myneni et al., 1997). It was also estimated that there had been an advance in the beginning of the growing season of about 8±3 days and a lengthening of the growing season by 12±4 days in northern temperate regions (latitudes 45–70° N). Recently, data from the MODerate-resolution Imaging Spectroradiometer (MODIS) sensor were used to estimate the phenological transition dates for natural vegetation in the northern midto-high latitudes (Zhang et al., 2004). Most studies using satellite sensor extracted phenological variables used the normalized difference vegetation index (NDVI) to, first, extract phenological variables and then quantify ecosystem response to climate change over continents and decades (Myneni et al., 1997; Reed et al., 1994; White et al., 1997; Zhou et al., 2001). These studies have been made possible by the large correlation between NDVI and the amount of green vegetation biomass. However, most studies suffered from unexplained variation in a smooth growth curve, as a result of image mis-alignment, sensor mis-calibration (Vermote & Kaufman, 1995) and changing atmospheric conditions (Tanre et al., 1992), for example, temporal variation in the presence of cloud, water, snow or shadow (Goward et al., 1985; Huete et al., 2002). As a result, it has been difficult to extract phenological variables routinely and reliably from raw NDVI time-series data (Reed et al., 1994). The NDVI, varies with both the amount of green vegetation biomass and the concentration of chlorophyll (Gitelson & Merzlyak, 1998; Huete et al., 2002; Mutanga & Skidmore, 2004) and saturates at high levels of both. Satellite sensor systems such as the Earth Observation System (EOS) and Envisat and, in the future, Sentinel 3 could go some way to addressing this constraint. An operational European Space Agency Envisat product, the Medium Resolution Imaging Spectrometer (MERIS) Terrestrial Chlorophyll Index (MTCI), is related directly to canopy chlorophyll content (Dash & Curran, 2004), which is, in turn, a function of chlorophyll concentration and leaf area index (LAI). The MTCI is calculated as the ratio of the difference in reflectance (R) between band 10 and band 9 and the difference in reflectance between band 9 and band 8 of the MERIS standard band setting. MTCI =

RBand 10 −RBand 9 R −R708:75 = 753:75 RBand 9 −RBand 8 R708:75 −R681:25

ð1Þ

Where R753.75, R708.75, and R681.25 are reflectance in the centre wavelengths of band 8, 9 and 10 in the MERIS standard band setting. MTCI has limited sensitivity to atmospheric effects and also soil background and view angle (Dash et al., 2008) and with the availability of near real-time weekly and global MTCI composites (Curran et al., 2007) enables researchers to extract phenological variables accurately. This research evaluates the potential of MTCI to extract phenological variables and monitor change in phenological variables over the last five years in India. 2. Study area India exhibits much variation in climate and vegetation type. Climate in India can be classified into four seasons: (i) winter (December– February), (ii) summer (March–June), (iii) south-west monsoon season (June–September), and (iv) post monsoon season (October–November) (Prasad et al., 2007). Variation in vegetation type is mainly attributed to soil type, availability of rainfall and temperature. The major vegetation types broadly constitute tropical evergreen, semi-evergreen, moist and dry deciduous (Fig. 1). Depending on the climate and soil condition, evergreen, semi-evergreen and deciduous vegetation types are often found in close proximity. The phenology of these natural vegetation types is often controlled by climatic condition. In India, relatively few studies have investigated climate-driven change in phenology. Most earlier studies related change in climatic variables to vegetation response at higher latitudes of the globe and very few have investigated these effects in tropical and subtropical regions (Badeck et al., 2004; Jeyaseelan et al.,

1389

2007; Menzel et al., 2006). The phenological patterns of some tropical and subtropical species are quite different to those found at higher latitude. Therefore, estimating their phenological parameters will help in understanding the response of tropical and subtropical vegetation to climate change. Relatively few studies used remote sensing to extract phenological variables over India (Jeyaseelan et al., 2007; Sarkar & Kafatos, 2004; Prasad et al., 2007). Most of these studies were limited to specific vegetation types or have limited spatial coverage (Prasad et al., 2007). Therefore, it is anticipated that this research will extract, for the first time, the vegetation phenological pattern over India and later will be useful for annual monitoring of phenological events. 3. Data Two data sources were used for this research: (i) 8-day temporal composites of MERIS MTCI and (ii) a landcover map derived from the Global Landcover 2000 data set. The MTCI product effectively combines information on LAI and the chlorophyll concentration of leaves to produce an image of chlorophyll content. The MTCI is simple to calculate and yet it is sensitive to a wide range of values of chlorophyll content. Coupled with the virtues of the MERIS sensor (e.g., radiometrically it is the most accurate imaging spectrometer in space (Curran & Steele, 2005); fine spectral resolution; moderate spatial resolution (300 m and 1 km); three-day repeat cycle) this has lead to the adoption of the MTCI as an operational ESA Level 2 land product. MTCI has been validated for several different species using data from the laboratory (Boyd et al., 2007), field (Zhang et al., 2008) and even at the MERIS spatial resolution (Dash et al., in press). For each of these experiments there was a large positive correlation between MTCI and chlorophyll content. The MTCI has now been used in applications of varying scope (Berberoglu et al., 2007; Dash & Curran, 2006; EspanaBoquera et al., 2006) and its ready availability should lead to wider adoption of the index. Given that the MTCI is the only available chlorophyll index from a spaceborne sensor there is now a real opportunity for monitoring vegetation function and condition systemically and reliably. Eight day composites of MERIS MTCI data at 4.6 km spatial resolution from 2003 to 2007 were obtained from the NERC Earth Observation Data Centre (NEODC) website (http://www.neodc.rl.ac.uk). MTCI data were composited from standard ESA Level 2 (geophysical) products using an arithmetic mean. Because the arithmetic mean composite technique is less sensitive to temporal biases compared to the widely used maximum value compositing, with an optimised cloud mask it can produce images with greater spatial and spectral consistency than other techniques. The land cover data used in this study were derived from the Global Land Cover Map (Bartholome & Belward, 2005). The GLC2000 product was created using daily S1 data acquired by the VEGETATION sensor onboard SPOT-4 acquired between 1st November 1999 and 31st December 2000 (Bartholome & Belward, 2005). The GLC2000 Map was created using various classification methods (e.g., supervised, unsupervised and hybrid classification methods) chosen by regional experts on the basis of their local suitability. This approach had the advantage of making the product more accurate as local expert knowledge was taken into consideration during the development of the map. The clusters derived from the regionally tuned classification methods were analysed, grouped together and labelled into a global legend system with 22 classes. For this research, the existing regional landcover map for the Indian sub-continent derived from the GLC global database (http://www-gem.jrc.it/glc2000) was used to locate major vegetation types to be considered in later analysis. 4. Pre-processing The methodology used to extract phenological variables from the MTCI time-series data consists of four major procedures: (i) data cleaning and flagging (ii) data smoothing

1390

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

Fig. 1. Simplified vegetation cover map of India derived from GLC global database (http://www-gem.jrc.it/glc2000).

(iii) temporal base information extraction and (iv) phenological variable extraction.

The algorithm is as follows: F ′ ðt Þ

This methodology is independent of the study site and phenological pattern and can detect multiple annual cycles as well as cycles which spread across years using multi-year data sets.

= ½F ðt−1Þ + F ðt = ½F ðt−1Þ + F ðt = ½F ðt−2Þ + F ðt = ½F ðt−2Þ + F ðt = F ðt−1Þ = F ðt + 1Þ = F ðt Þ

; ; ; ; ; ; ;

If F ðt Þ = 0 If F ðt Þ = 0 & F ðt + 1Þ = 0 If F ðt Þ = 0 & F ðt−1Þ = 0 If F ðt Þ = 0 & F ðt−1Þ = 0 & F ðt + 1Þ = 0 If F ðt Þ = 0 & F ðt + 1Þ = 0 & F ðt + 2Þ = 0 If F ðt Þ = 0 & F ðt−1Þ = 0 & F ðt−2Þ = 0 If F ðt Þ≥1

ð2Þ

4.1. Data cleaning and flagging A data cleaning and flagging procedure was used to remove missing data values from the original data and create a flag depending upon the quality of the temporal information available in each pixel. Though the valid MTCI values range from 1 to 6, cloud, local climate fluctuations and sensor noise could lead to erroneous MTCI values or data dropout (MTCI = 0). Hence, the first step is to remove or reduce such obvious errors (drop outs) from the temporal data series followed by temporal smoothing (Section 4.2). A temporal moving average window function is utilised to correct errors at specific weeks. For each pixel, the MTCI value at time ‘t’ (F(t)) is checked for its correctness and if it is a dropout then an average (F (t)) of the immediate temporal neighbourhood of MTCI values at times ‘t − 1’ and ‘t + 1’ is calculated if they are valid. If the value of any temporal neighbour (say ‘F(t − 1)’ or ‘F(t + 1)’) is invalid, then the next immediate temporal neighbour value (i.e., ‘F(t − 2)’ or ‘F(t + 2)’ respectively) is utilised. The neighbourhood check is limited to two temporal neighbours in order to preserve the trend in MTCI values.

+ 1Þ = 2 + 2Þ = 2 + 1Þ = 2 + 2Þ = 2

Where the & symbol refers to the logical AND condition. A pixel is marked with an error flag if it has a dropout of data for three consecutive weeks or it has dropout at 10 or more weeks. Care must be taken while utilising error flagged pixels at the final stage. Though spatial averaging may help to minimize local anomalies in a homogeneous landscape, it would artificially alter the phenological continuity and could affect the detection of local scale changes in a heterogeneous landscape (Bradley et al., 2007). 4.2. Data smoothing The raw MTCI time-series data included errors that were minimized by applying a smoothing algorithm. Different techniques have been used for temporal interpolation of erroneous or missing data in a time-series. They include: best index slope extraction (BISE) (Viovy et al., 1992), median filters (e.g. Vandijk et al., 1987), splines and weighted least-squares (White et al., 2005), discrete Fourier

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

transformation (DFT) (Geerken et al., 2005; Jakubauskas et al., 2001), locally adjusted cubic-splines (Chen et al., 2006), and the asymmetric Gaussian (Jönsson & Eklundh, 2002), Savitzky–Golay (Chen et al., 2004) and double logistic functions (Zhang et al., 2004). Recently, Hird and McDermid (2009) assessed several smoothing techniques using a model-based empirical comparison and found that both the double logistic and asymmetric Gaussian function-fitting methods performed comparatively more accurately. Most of these techniques require an iterative approach to adjust the model parameters such as noise-threshold and size of temporal neighbourhood to achieve reliable smoothing (Atkinson et al., 2009) and most of the time these adjustments are not static due to phenological variation within and across landcover classes. Fourier transform-based approaches have the advantage of minimal user input (only need to decide the number of harmonics to reconstruct the time-series) and have been applied successfully to many regional-to-global AVHRR time-series datasets (e.g., FourierAdjusted, Sensor and Solar zenith angle corrected, Interpolated, Reconstructed (FASIR) data set (Los et al., 2000), Atmospheric, Bidirectional, and Contamination Corrections of CCRS (ABC3) data set (Cihlar et al., 1997), temporal Fourier analysis (TFA) dataset (Hay et al., 2006)). The DFT decomposes any complex waveform into a series of sinusoids of different frequency. Individual sinusoids and their frequencies can be amalgamated into a complex waveform for which noise has been removed. The DFT is given by: FðuÞ =

1 N−1 −2πux = T ∑ f ðxÞ Te N x=0

ð3Þ

Where f(x) is the xth value in the time-series, u is the number of Fourier components, x is the dekad number, T is the length of time period cover (number of dekad), and here T is equal to N. Eq. (3) consists of two parts: cosine (real) and sine (imaginary) parts, where the cosine part is: FC ðuÞ =


ux 1 N−1
∑ f ðxÞ T cos 2π N x=0 T

ð4Þ

And the sine part is FSðuÞ =


ux 1 N−1
∑ f ðxÞ T sin 2π N x=0 T

ð5Þ

Using Eqs. (4) and (5), the Fourier magnitude (Fm) can be calculated as FmðuÞ =

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi FC2ðuÞ + FS2ðuÞ

ð6Þ

And the phase (Fp) can be calculated as FpðuÞ = a tan 2

FC ðuÞ FSðuÞ

! ð7Þ

The first two harmonics of the Fourier transform usually account for 50–90% of the variability in a data set; in this case, variability in the vegetation index time-series (Jakubauskas et al., 2001). A complete reconstruction of the phenological signals from the Fourier transform needs to take into account the appropriate number of harmonics. If the first two harmonics only are used, representing only annual and semi-annual cycles, then it may be difficult to represent a naturally varying phenological cycle. For example, changes at the beginning and end of the growing season, bimodal patterns in phenology and double or triple cropping agricultural systems, cannot be represented with only two harmonics. Jakubauskas et al. (2001) demonstrated that the first four harmonics could adequately represent uni-modal vegetation

1391

growth patterns and the same has been adopted in the current study. For uni-modal growth patterns it was found that the first four harmonics avoid capturing local oscillations during the growing season which were detected using higher harmonics. Since the focus of this study was to extract the “onset” or “end” of season, the first four harmonics provided a reliable means to capture these variables. As a double cropping pattern is common in India, we utilised six harmonics to extract the phenology in agricultural areas. Using the DFT, the MTCI time-series data were decomposed into a series of sinusoids of different frequency. The first four/six harmonics were used in the inverse DFT to generate a smoothed time-series. 4.3. Temporal base information extraction Since a vast area (i.e., 3,280,000 km2 of India) needs to be processed, the whole computing process was divided into smaller subroutines to make the processing memory efficient. In this way, the required temporal base information (the first derivative, annual minimum and maximum MTCI value and corresponding week, all visible peaks (i.e., local trend) and annual dominant peak) were extracted for each pixel from the smoothed MTCI data. The first derivative refers to the change in MTCI value from its immediate previous neighbour. For each pixel, the minimum and maximum MTCI values, and their corresponding weeks were identified. The peak in the annual cycle was identified by applying a logical condition, which checks for continuity over the annual MTCI data. The rule checks the continuity in the variation of MTCI values before and after each temporal composite data band. If a particular temporal composite (i.e., F (t)) is a peak then it must have increasing trend at the four successive preceding composites, and decreasing trend amongst the four successive following composites. If there is more than one peak in a uni-modal phenological pattern then the peak with the maximum amplitude is identified and retained as a dominant peak (Fig. 2). Peak information is needed to trace the start and end of the growth cycle. 5. Methodology Several quantitative methods have been used to extract variables related to vegetation phenology. These methods can be grouped into three broad categories: threshold-based methods, trend derivative methods and inflection point methods (Reed et al., 1994). Thresholdbased methods use either a pre-defined or relative reference value to define phenology transition dates (Fisher & Mustard, 2007; Lloyd, 1990). The trend or curve derivative phenology method attempts to identify points of departure between the original vegetation temporal signal data and a derivative curve. The inflection point phenology method is based on detecting points where maximum curvature occurs in a plotted time-series of vegetation indices. The inflection point-based method was used in this study as it has the advantage of being easy to implement and also permits discrimination of multiple growing seasons for land cover types with multiple growth seasons such as crops (Reed et al., 1994). This was combined with simple logical and continuity functions to derive key phenological variables such as: onset of greenness, end of senescence. Onset of greenness was defined as a valley point at the beginning of a growing cycle, and end of senescence was defined as a valley point occurring at the decaying end of a phenology cycle (Figs. 2 and 3). It was relatively easy to extract the phenological variables given a single phenology cycle following a smooth sinusoidal pattern. In reality, the ecosystem may exhibit a complex pattern and local climatic fluctuation may further alter normal behaviour. Hence, the phenology curves are complex and do not follow any pre-defined shape and size over several years. Therefore, the main objective was to extract the variables from any given MTCI time-series irrespective of shape and size. In a diverse and vast country, like India, many of the ecosystem's phenology cycles span across

1392

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

Fig. 2. Schematic representation of the phenological variables extracted in this study.

the years and hence complete information about onset of greenness and end of senescence would not be obtained using single year data. Irrespective of the available data the algorithm is designed in such a way that it can extract the desired information, whether “onset” or “end” of season, from a given annual time-series. If two year data are available then the approach expands its temporal search domain and looks for missing variable information in the next year. For every phenology cycle two variables (onset of greenness and end of senescence) were extracted (Fig. 3). The algorithm starts from the dominant peaks and searches both forwards and backwards in time checking the derivative information. While moving backwards, a change in derivative value from positive to negative may indicate a valley point. Such a valley point may occur even at larger values of MTCI due to local

fluctuations or stabilisation, which can be removed by checking the difference between the MTCI value at the peak and valley points. If the difference is greater than one fifth of the maximum MTCI value then the valley point is accepted as a phenological variable, otherwise the process is continued until a definite valley point is detected. If a valley is detected in the backward search, then it is assigned as the onset of greenness and if a valley is detected in the forward search then it is assigned as the end of senescence. To study the phenological variation within and across major vegetation types, first, the GLC2000 map of the Indian sub-continent was resampled using the majority method to match the spatial resolution of the MTCI data. Then, the median values for onset of greenness and end of senescence were estimated for each pixel for the five years of the study

Fig. 3. Flow chart of phenology extraction process.

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

period. Finally, the majority of onset of greenness and end of senescence was estimated for 30 minute increments of latitude within the country masked by land cover classes obtained from the resampled GLC2000 map. While analysing the result it is important to realise that these values within the 30 minute range of latitude are subject to variability resulting from varying soil, illumination and percentage coverage of land cover type within the region. 6. Results 6.1. Smoothing The effect of data smoothing on the time-series for an evergreen pixel is illustrated in Fig. 4. The DFT was able to capture the broad phenological pattern and major variation during a phenological cycle. For the evergreen vegetation type the MTCI value was larger than 2 through the period and the seasonality was related to leaf flushing and leaf fall. It can be seen in the top-most plot that there were a large number of missing data due to cloud cover. These were removed by the data cleaning operation. The final smoothed curve was constructed from the first four harmonics of the DFT. 6.2. Annual MTCI variation 6.2.1. Spatial variation Fig. 5 represents the spatial variation in the 8-day mean MTCI across India through 2006. Time periods in Fig. 3 represent the 8-day compositing period. For example, composite 1 consists of data from

1393

Julian days 1 to 8, and composite 2 consists of data from Julian days 9 to 16 and so on. The figure suggests that there are two broad patterns across India: one related to agricultural crops and one related to natural vegetation. Most agricultural regions in the north and northwest part of the country exhibited large MTCI values from January to March (composites 1 to 9). These were mainly attributed to the winter crop which is planted in December and is harvested by April. Natural vegetation started greening up from composite 17 (beginning of May) with a peak value of MTCI representing the peak of the growing season during composite 29 (middle of July). Growth in natural vegetation starts to decline from composite 37 (middle of October). Natural vegetation growth in India broadly depends on the monsoon rain (Corlett & Lafrankie, 1998; Joshi et al., 2006) and the pattern described by the 8-day MTCI composites broadly coincide with the arrival and withdrawal of the south-west monsoon. 6.2.2. Vegetation type variation Fig. 6 describes the broad phenological pattern for five years for four major vegetation types (evergreen, semi-evergreen, dry deciduous and moist deciduous) as well as agricultural crops. For the Evergreen vegetation type, the MTCI value remains large throughout the year. However, there was a distinct increase in MTCI during May (onset of greenness), a peak during October and a decline which reaches a minimum by April next year (end of leaf fall). The increase in MTCI can be attributed to the phenomenon of leaf flushing. As reported in earlier studies (Elliott et al., 2006; Kikim & Yadava, 2001) this process ensures that young photosynthetically competent leaves are in

Fig. 4. Effect of smoothing on the MTCI time-series data for an Evergreen pixel.

1394

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

Fig. 5. Spatial variation in MTCI across India during one year (2006) for various compositing periods.

place when the monsoon rains begin. From Fig. 6, May is a transition period where the MTCI values decrease up to a certain limit and then start increasing. This pattern was attributed to the phenomenon of leaf exchange, where shedding of old leaves during the early dry season is accompanied or immediately followed by bud break and expansion of new leaves. The semi-evergreen vegetation type (overstory consists of deciduous vegetation and the understory consists of evergreen vegetation) shows a similar phenological pattern as the Evergreen vegetation with large values of MTCI throughout the year. However, during the peak of the growing season the curve remained flat from July to October. The onset of greenness is attributed due to the first leaf flush during March and April (Kikim & Yadava, 2001) and the decline in greenness is attributed to leaf shedding from November. The flatness during the peak of the growing season is caused by a second leaf flushing during the mid wet season (August–September) in both overstory and understory vegetation. The semi-evergreen vegetation exhibits a similar leaf exchange period to that of Evergreen vegetation. The moist deciduous vegetation exhibits a similar pattern to both evergreen and semi-evergreen vegetation types, with a sharp increase

in MTCI values from March, a peak during October and a decline which reaches a minimum by March next year. In moist deciduous forest in India there is much understory vegetation due to the availability of water and leaf fall and leaf flushing processes slightly overlapping (Bhat, 1992; Mishra et al., 2006; Newton, 1988). The phenology of dry deciduous forest was different compared to the other species described above. The leaf flush (onset of greenness) was late in the year starting around June and the leaf fall ended earlier (February) with approximately 3 months without leaf (March to May). These species are generally found in the rain shadow regions and their phenology is limited by the availability of rainfall. By shedding leaves in the dry season they limit evapotranspiration (Singh & Kushwaha, 2005). The crop phenology, shown in Fig. 6, was from a double cropping region (winter and summer crop). The summer crop which starts in June and ends in November has larger maximum MTCI values compared to the winter crop which starts in November and ends in May next year. Variation in MTCI values for both crops is reflected by the crop type: the summer crop is dominated by major grain crops (e.g., rice, wheat), thus, producing large MTCI values, whereas the winter crop is dominated by lentil crops, thus, producing small MTCI values.

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

1395

Fig. 6. Inter- and intra-annual variation in MTCI values for major natural vegetation types and double cropping agricultural land. (a) Double cropping agriculture; (b) dry deciduous; (c) moist deciduous; (d) semi-evergreen and (e) evergreen.

6.3. Spatial variation of phenological variables Fig. 7 shows the spatial variation in the onset of greenness and end of senescence over India for the growing seasons which started in 2006. Three pairs of images were required to represent the onset of greenness and end of senescence mainly due to the triple cropping pattern in some areas. Fig. 7a and b represent the first growing season; Fig. 7c and d represent the second growing season and Fig. 7e and f represent the third growing season. Depending on water availability,

cropping activity continues all year in India (http://agricoop.nic.in/). In northern India, there are two distinct seasons, kharif (July to October), and rabi (October to March). The land may be occupied by one crop during one season (mono-cropping), or by two crops (double cropping) which may be grown in a year in sequence. Recent reports suggest that in certain parts of the country there is a trend for growing more than two crops in a year (http://agricoop.nic.in/). Fig. 7 suggests geographical and agro-ecological patterns for the phenological variables that are consistent with the reported

1396

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

Fig. 7. Spatial distribution of (a, c, e) onset of greenness and (b, d, f) end of senescence for the (a and b) first growing season, (c and d) second growing season and (e and f) third growing season.

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

phenological dates for the country. Earlier onset of greenness (between 1 to 8 compositing period: January to middle of March) was detected in the southern part of the country. This area has a combination of landcover types including barren land, shrubs and grassland. The corresponding end of senescence in these areas was towards the end of the year (Compositing period 44 to 47: December). This indicates that this area has no clear phenological pattern or that the phenological variables are hard to detect. The majority of the central, eastern and north-eastern parts of the country exhibited a start of greenness between compositing periods 9 and 12 (March to April). The landcover in these areas is dominated by deciduous forest and agricultural land. However, the end of senescence in these regions fell into two broad compositing periods: one between 44 and 47 (November–December) and the other between periods 52 to 59 (February–April next year), with the earlier one representing mono crop agricultural land and the later one representing deciduous forest. The majority of the northern and part of central and western India (dominated by agricultural land), some of north-eastern India and the western-Ghat region (dominated by evergreen and semi-evergreen forest) exhibited a start of greenness between compositing periods 13 and 16 (May to June) (Sundarapandian et al., 2005). However, the end of senescence in these regions varied between compositing periods 39 to 43 (October–November) for agricultural land and between 57 and 60 (March–April next year) for evergreen and semi-evergreen forest. In some parts of the agricultural regions in northern India the end of senescence was detected earlier (compositing periods 27 to 32: July–August). This is consistent with the fact that farmers in this region produce a crop grown between March and June known as zaid (http://agricoop.nic.in/). For some parts of northwestern India the onset of greenness (between compositing periods 17 and 20 (May–June)) and end of senescence (between compositing periods 39 and 43 (October–November)) are spatially homogeneous. This pattern was attributed to the major summer crops in these regions. Finally, there are a few agricultural areas where the start of greenness was detected late as compared to other agricultural crops and these may be the result of different crop types grown in these areas. The second set of phenological variables (Fig. 7b and c) capture the double cropping pattern mostly concentrated in northern India. Within these regions, the start of greenness for the second crop was detected quite late in the year (compositing periods: 40 to 46 (November– December)) except towards the eastern part of the region where the start of greenness for the second crop was detected earlier (compositing periods: 26 to 28 (July–August). For this region, except the area where there was an earlier onset of greenness, crop growth extended to the next year and end of senescence was detected between compositing periods 57 to 68 (March to May next year). It was also observed that, when moving from east to west along this region the end of senescence was delayed. Finally, for the area where there was an early onset of greenness, the end of senescence was also detected earlier (compositing periods 44 to 45 (December). The final set of phenological variables (Fig. 7e and f) seems to capture the areas with triple crops within a year and these figures show the start of greenness and end of senescence for the third crop in these regions. They cover a relatively small area in the country and they represent the regions where there was an earlier onset of greenness in the second growing season. The onset of greenness for this region was detected between compositing period 44 to 46 (December) and the crop growth continues to next year with an end of senescence detected during compositing period 59 to 60 (March next year).

6.4. Relationship with latitude and longitude Four major vegetation types (evergreen, semi-evergreen, dry deciduous and moist deciduous) were examined to quantify the effect of latitude, and the extensive agricultural area in northern India was examined to quantify the effect of longitude on the phenological variables (Figs. 8 and 10).

1397

6.4.1. Major vegetation type The greatest latitudinal variation was found for the onset of greenness variables (Table 1). Among major vegetation types moist and dry deciduous vegetation had a larger positive correlation with latitude (r 2 = 0.63 for moist deciduous and r2 = 0.68 for dry deciduous). Evergreen and semi-evergreen vegetation types had a relatively small positive correlation with latitude (r2 = 0.42 for Evergreen and r2 = 0.35 for semi-evergreen). This suggests a delay in the onset of greenness at higher latitudes, but the magnitude of this delay depends upon the vegetation type. At lower latitudes the onset of greenness showed a lot of fluctuation irrespective of vegetation type. This may be due to: (i) a relatively small number of pixels for that specific class within the 30 minute increment and (ii) misclassification in the landcover map. For the end of senescence, the latitudinal variation was less pronounced than that for the onset of greenness. This relates to the fact that senescence is a much more complex phenomenon compared to onset of greenness as reported in earlier studies (Schaber & Badeck, 2003). However, for the evergreen and semi-evergreen vegetation types there was a small positive correlation with latitude (r2 = 0.39 for evergreen and r2 = 0.24 for semi-evergreen). No relationship was found between latitude and end of senescence for moist and dry deciduous vegetation (r2 ≈ 0). There were a few outliers at very low latitudes and these were mainly from the Andaman and Nicobar Islands which have a different climatic condition than the mainland. For the season length variable, there was a small negative correlation with latitude for moist and dry deciduous vegetation (r2 = 0.35 for moist deciduous and r2 = 0.15 for dry deciduous). This suggests a shortening of growing season at higher latitudes. However, for evergreen and semi-evergreen vegetation, season length had a very small positive correlation with latitude (r2 = 0.11 for evergreen and r2 = 0.07 for semi-evergreen). 6.4.2. Agricultural crops Agricultural crops have a different phenological pattern compared to natural vegetation. As demonstrated above, multiple growing seasons (mostly double) within a year are quite common in the agricultural areas of India. The major agricultural region in India is situated in the IndoGangetic Plain. It extends from Punjab in the north-west to west Bengal in the east. Therefore, this region has greater longitudinal (74° to 86° East) than latitudinal (25° to 32° North) variation. Phenological data in this region were analysed against longitude. Fig. 9 shows the majority of the phenological variables for the first growing season within a 30 minute increment of latitude and longitude in this region. At low latitude and high longitude (eastern part of the country) the onset of greenness for the first growing season crop was earlier (≈14 compositing period) than the high latitude and low longitude regions (north-western part of the country) (≈18 compositing period). Onset of greenness varies near-linearly between these extremes. The difference between the start of season for these regions was approximately 4 compositing periods (equivalent to a month). However, in the low latitude and high longitude region there are some areas where the start of season was late in the year and starts around compositing period 20 (Fig. 9a). These are mainly in the regions where there are triple cropping patterns. However, there was no clear pattern for the end of senescence in this region and the variation depended on crop type and irrigation (http://agricoop.nic.in/). The delay in onset of greenness for the first crop is mainly controlled by arrival of the southeast Monsoon. The date of arrival in the eastern part of this region was earlier compared to the western part. As a result, the onset of greenness in the western area was delayed compared to the eastern area. For the second crop growing season, the onset of greenness was late (≈44 compositing period) and end of senescence was early (≈60 compositing period) in the low latitude and high longitude (eastern part of the country) compared to the high latitude and low longitude regions (north-western part of the country) (onset of greenness ≈ 40 compositing period; end of senescence ≈ 64 compositing period)

1398

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

Fig. 8. Latitudinal variation in (a) onset of greenness, (b) end of senescence and (c) length of growing season for four major vegetation types.

(Fig. 9b). However, for the whole region the end of senescence for the second crop was detected in the following calendar year. The duration of growing season was shorter in the eastern part of the region than the north-western part of the region. This reflects the cropping type and management practice in these regions and matches with the official data (http://agricoop.nic.in/). The crops in the second growing season in the eastern area are dominated by non-cereal crops such as oilseeds and lentils which have a shorter growing season, whereas the second growing season crops in the western part of the country are dominated by cereal crops such as rice, wheat and maize, which have a longer growing season. 6.5. Phenological variation between years The overall phenological variation for all natural vegetation in India over the five years of study including four complete growing seasons is given in Fig. 10. The figure represents the three major phenological variables plotted against a 30 minute increment of latitude for four complete growing seasons for all major natural vegetation types, and excludes agricultural crops. Each phenological variable (OG, ES and length of season) for a growing season in each 30 minute increment is represented as a discrete unit. The dimension

of this unit is proportional to the value of the phenological variables in the compositing period (Fig. 10). At lower latitudes for all four years, the onset of greenness was earlier than for the higher latitudes. Exceptions to this trend were observed at very low (less than 8°) and very high (greater than 32°) latitudes. At mid latitudes between 13° and 25°, the onset of greenness in 2003 was later than for other years. This can be explained by the extreme cold wave during the early part of 2003 experienced in this region (http://www.imd.gov.in/). Thus, it took longer for this region to reach the optimum temperature to start the greening up process. The end of senescence was detected around compositing period 50 between 7° to 30° latitude and above this latitude the end of senescence was detected at least 6 compositing periods earlier. There was little variation in end of senescence between different years. In general, the length of season at higher latitudes was shorter than at lower latitudes. The longest length of season was detected between 9° and 13° latitude. The length of season in the mid latitudes between 13° to 25° was shorter in 2003 compared to the other years. This resulted from a late onset of greenness in 2003 as explained earlier. The Kruskal–Wallis test was used to determine if there was a significant difference between phenological variables detected using MTCI composites for the four years. It was observed that for onset of

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

1399

Table 1 Coefficient of determination, slope and intercept of the linear regression line for latitude and vegetation type (OG—Onset of greenness; ES—End of senescence, SL— season length). Type

Evergreen Semi-evergreen Moist deciduous Dry deciduous

r2

Slope

Y-intercept

OG

ES

SL

OG

ES

SL

OG

ES

SL

0.42 0.35 0.63 0.68

0.39 0.24 0.00 0.01

0.11 0.07 0.35 0.15

0.36 0.30 0.46 0.63

0.72 0.57 − 0.01 0.13

0.36 0.27 − 0.44 − 0.51

3.87 4.9 2.48 0.13

39.54 43.42 54.73 47.78

35.68 38.51 51.63 47.64

greenness (p = 0.016) and length of season (p = 0.018) the difference among years was marginally significant and for end of season (p = 0.23) the difference between years was insignificant (confidence level 99%). Since a late onset of greenness was detected in 2003, the Mann–Whitney U test was used to find if there was a significant difference between the onset of greenness between years (Table 2). It was observed that the early onset of greenness detected in 2003 was marginally significant (p b 0.05) as compared to other years. However, for other years there was no significant difference between the onset of greenness values. 7. Discussion Phenological patterns in the tropical and subtropical regions are the most diverse and least understood (Bhat, 1992; Corlett & Lafrankie, 1998; Prasad et al., 2007). In India, several studies were undertaken to characterise vegetation phenology (Bhat, 1992; Kikim & Yadava, 2001; Mishra et al., 2006; Newton, 1988; Ralhan et al., 1985; Sundarapandian et al., 2005). These studies highlighted (i) the large spatial variation in key phenological events and (ii) the response of vegetation to the south-east Monsoon. However, most of these studies focused on local scale (community level) phenological variation, for a short time period and provided phenological information at a coarse temporal resolution (mainly monthly). Though detailed community level phenological studies help to understand the response of vegetation to local climatic fluctuations, they do not provide enough information for input into bio-geochemical models at regional to global scales. In addition, these studies are time consuming and are prone to biotic interference and natural disturbances such as forest fires, disease and environmental degradation. Therefore, repeated observations of phenology for monitoring and managing natural vegetation over large areas can only be possible though satellite sensor observation. The extracted phenological variables were evaluated with respect to ground-based observations from previous research papers (Bhat, 1992; Kikim & Yadava, 2001; Mishra et al., 2006; Newton, 1988; Ralhan et al., 1985; Sundarapandian et al., 2005). Ground-based phenological observations are usually limited in that they are: (i) at the individual species level, (ii) too dispersed spatially, (iii) at too coarse a temporal resolution, and (iv) at too fine a spatial resolution. However, some studies (e.g., Fisher & Mustard, 2007; Studer et al., 2007) attempted to validate satellite-derived phenological information with that from direct ground data or indirectly by a combination of ground and high spatial resolution remote sensing data. At present, there is no ground observation plot in India that can provide phenological information for the study period (2003–2007), which could have been used for direct/indirect validation of the results from this study. Therefore, it was not possible to establish a one-to-one relation between the extracted phenological variables and those reported from ground-based observations. However, the overall patterns of phenological variables detected from this study broadly coincide with the pattern of natural vegetation phenology revealed by earlier studies. For the first time, this study attempted to establish a broad regional phenological pattern for India using remotely sensed estimation of canopy chlorophyll content for five years of data. Though this is a short time

Fig. 9. Latitudinal and longitudinal variation in onset of greenness and end of senescence for (a) first crop and (b) second crop in the Indo-Gangetic plain.

period, the automated algorithm developed in the study can be implemented in the Envisat processing chain to produce an annual phenological pattern for the whole country. This could be evaluated through simultaneous ground measurements and in the longer term this could provide a valuable data set to study the effect of climate change on the phenological pattern of natural vegetation in India. The phenological patterns detected from this study for key natural vegetation broadly match with ground observations as reported in earlier studies (Bhat, 1992; Elliott et al., 2006; Kikim & Yadava, 2001; Mishra et al., 2006; Newton, 1988; Ralhan et al., 1985; Singh & Kushwaha, 2006; Sundarapandian et al., 2005). Spatial mixing of species and vegetation type is quite common in Indian forests. For example, (Newton, 1988) reported three major vegetation types (moist deciduous, dry deciduous and grassland) with 29 families and 63 species of tree within a 74.5 ha area in Kanha National Park, Madhya Pradesh. Jayakumar et al. (2002) reported five vegetation types (semi-evergreen, deciduous, southern thorn, euphorbia scrub and riverine forest) with 35 families, 39 genera and 67 species within a 2806 ha area in Puliyanjolai reserve forest, Tamil Nadu. Therefore, it is important to note that the spatial resolution used in this study was quite coarse (4.6 km). For natural vegetation, in most cases, the response from the pixel was, thus, attributable to a mix of species or vegetation types. Unfortunately, there are no long term phenological monitoring stations in India (Kushwaha & Singh, 2008) and, therefore, validation was based on data reported from community level studies (Bhat, 1992; Elliott et al., 2006; Kikim and Yadava, 2001; Mishra et al., 2006;

1400

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

Fig. 10. Phenological variation with latitude (y-axis) for all major natural vegetation types from 2003 to 2007. The number represents the compositing period.

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402 Table 2 p values from the Mann–Whitney U test for the onset of greenness between different years.

2003–04 2004–05 2005–06 2006–07

2004–05

2005–06

2006–7

0.009

0.035 0.596

0.004 0.79 0.491

Newton, 1988; Ralhan et al., 1985; Singh & Kushwaha, 2006; Sundarapandian et al., 2005). In addition, phenology within a certain latitude range can also vary with altitude. For example, Shukla & Ramakrishnan (1984) reported a delay of 4 weeks for start of leaf flushing for an evergreen species at higher compared to lower altitude. An indirect validation of these results could have been done by comparing the phenological variables exacted from MTCI with those from the MODIS phenological product (MOD12) (Zhang et al., 2003). However, MOD12 data for India were not available to the public during this study. Moreover, it would be challenging to compare these products which use different inputs, geo-computation techniques for smoothing and have different spatial resolutions and, thus, such a comparison was beyond the scope of this study. Therefore, considering all the above points it was difficult to validate the phenological matrix derived from this research. At the same time, the study poses a challenge to researchers working on community level phenology to develop methods for validating these results using the sparse ground data. A strong coincidence was observed with expected phenological pattern for major vegetation types. In particular, the expected trend in inter-annual and latitudinal variability of key phenological variables was observed. For most vegetation types there was a small or no leafless period. This was in contrast to similar landcover types in higher latitude regions of the globe which have a large leafless period (Elliott et al., 2006). This was mainly due to a direct leaf transition stage from mature to young leaves. In general, the length of growing season was higher at lower latitudes. This was characterised by two monsoon seasons in these regions supporting secondary growth during the latter part of the year (Corlett & Lafrankie, 1998; Sundarapandian et al., 2005). For tropical plants, leaf flush is attributed to onset of rain after a period of dry conditions, water stress, day light duration and temperature. In many tropical sites the onset of greenness was early in the dry season of the year when the temperature reached a maximum (Singh & Kushwaha, 2005; Sukumar et al., 1995). The onset of greenness for most natural vegetation followed a latitudinal pattern with earlier onset of greenness at lower latitudes (Table 1 and Fig. 8c). However, for dry deciduous forest this correlation was largest. On the other hand, the end of season was earlier at high latitudes. In the 2003–04 growing season a late start for the onset of greenness was detected at low-to-mid latitudes and it was attributed to the extreme cold weather during the early part of 2003. The phenological variables detected for other years showed a consistent pattern with variation within two compositing weeks. This research was able to identify the cropping patterns across India and in some parts of the country three growing seasons were detected (Fig. 6e and f). The length of growing season varied from east to west for the major cropping areas in the Indo-Gangetic plain, for both the first and second crops. For the first crop, the length of season was greater in the eastern part of this region. However, for the second crop, the length of season was greater in the western part. 8. Conclusion This paper maps for the first time a suite of phenological variables for the whole of India using multi-temporal composites of a remote sensing-based chlorophyll index. Each pixel with a spatial resolution

1401

of 4.6 km was analysed temporally for the years 2003 to 2007 and phenological variables were extracted using the inflection point method. The first four harmonics of the DFT were used to smooth the signal while maintaining the overall phenological pattern. The major landcover types in India were obtained from the GLC2000 landcover map. Spatial variation with latitude was observed for the phenological variables for natural vegetation. Among all phenological variables analysed the onset of greenness was highly related with latitude irrespective of vegetation type. Among all natural vegetation types considered the onset of greenness for the Dry deciduous vegetation type was the most clearly demarcated and most readily extracted. There was marginal variation in the phenological variables between different years except 2003, where there was a statistically significant (Mann–Whitney U test p b 0.005) earlier onset of greenness. The cropping patterns in one of the major agricultural regions (IndoGangetic plain) were analysed. It was found that the length of the growing season decreases from east to west with early onset of greenness in the east for the first crop, but late onset of greenness in the east for the second crop. The extracted phenological variables from this study can be used as an important input to investigation of the impact of climate change on natural vegetation, for example, through bio-geochemical modelling. In addition, spatial information on phenological variables is important for planning, management and conservation strategies for terrestrial ecosystems. Acknowledgement We are grateful to the NERC Earth Observation Data Centre for providing MTCI data, the School of Geography for funding and two anonymous referees for their helpful comments in improving the manuscript. References Atkinson, P. M., Jeganathan, C., & Dash, J. (2009, November–December). Analysing the effect of different geocomputational techniques on estimating phenology in India. In B. G. Lees & S.W. Laffan (Eds.), 10th International conference on GeoComputation. Sydney: UNSW. Badeck, F. W., Bondeau, A., Bottcher, K., Doktor, D., Lucht, W., Schaber, J., et al. (2004). Responses of spring phenology to climate change. New Phytologist, 162, 295−309. Bartholome, E., & Belward, A. S. (2005). GLC2000: A new approach to global land cover mapping from Earth observation data. International Journal of Remote Sensing, 26, 1959−1977. Berberoglu, S., Evrendilek, F., Ozkan, C., & Donmez, C. (2007). Modeling forest productivity using envisat MERIS data. Sensors, 7, 2115−2127. Bhat, D. M. (1992). Phenology of tree species of tropical moist forest of Uttara–Kannada district, Karnataka, India. Journal of Biosciences, 17, 325−352. Boyd, D., Almond, S., Dash, J., & Curran, P. J. (2007). Investigating the factors affecting the relationship between the Envisat MERIS Terrestrial Chlorophyll Index and chlorophyll content: preliminary findings. Proceedings of the RSPSoc, annual conference-07, Newcastle (CD Rom). Bradley, B. A., Jacob, R. W., Hermance, J. F., & Mustard, J. F. (2007). A curve fitting procedure to derive inter-annual phenologies from time series of noisy satellite NDVI data. Remote Sensing of Environment, 106, 137−145. Chen, J. M., Deng, F., & Chen, M. Z. (2006). Locally adjusted cubic-spline capping for reconstructing seasonal trajectories of a satellite-derived surface parameter. IEEE Transactions on Geoscience and Remote Sensing, 44, 2230−2238. Chen, J., Jönsson, P., Tamura, M., Gu, Z., Matsushita, B., & Eklundh, L. (2004). A simple method for reconstructing a high-quality NDVI time-series data set based on the Savitzky–Golay filter. Remote Sensing of Environment, 91, 332−334. Cihlar, J., Ly, H., Li, Z. Q., Chen, J., Pokrant, H., & Huang, F. T. (1997). Multitemporal, multichannel AVHRR data sets for land biosphere studies — Artifacts and corrections. Remote Sensing of Environment, 60, 35−57. Corlett, R., & Lafrankie, J. V. (1998). Potential impacts of climate change on tropical Asian forests through an influence on phenology. Climate Change, 39, 439−453. Curran, P. J., Dash, J., Lankester, T., & Hubbard, S. (2007). Global composites of the MERIS terrestrial chlorophyll index. International Journal of Remote Sensing, 28, 3757−3758. Curran, P. J., & Steele, C. M. (2005). MERIS: The re-branding of an ocean sensor. International Journal of Remote Sensing, 26, 1781−1798. Dash, J., & Curran, P. J. (2004). The MERIS terrestrial chlorophyll index. International Journal of Remote Sensing, 25, 5403−5413. Dash, J., & Curran, P. J. (2006). Relationship between herbicide concentration during the 1960s and 1970s and the contemporary MERIS Terrestrial Chlorophyll Index

1402

J. Dash et al. / Remote Sensing of Environment 114 (2010) 1388–1402

(MTCI) for southern Vietnam. International Journal of Geographical Information Science, 20, 929−939. Dash, J., Curran, P. J., & Foody, G. M. (2008). Remote sensing of terrestrial chlorophyll content. In A. P. Cracknell, V. F. Krapivin, & C. A. Varotsos (Eds.), Problems of global climatology and ecodynamics: Anthropogenic effects on the state of Planet Earth. Germany: Springer. Dash, J., Curran, P. J., Tallis, M. J., Llewellyn, G. M., Taylor, G., & Snoeij, P. (in press). Validating the MERIS Terrestrial Chlorophyll Index (MTCI) with ground chlorophyll content data at MERIS spatial resolution. International Journal of Remote Sensing. Elliott, S., Baker, P. J., & Borchert, R. (2006). Leaf flushing during the dry season: The paradox of Asian monsoon forests. Global Ecology and Biogeography, 15, 248−257. Espana-Boquera, M. L., Cardenas-Navarro, R., Lopez-Perez, L., Castellanos-Morales, V., & Lobit, P. (2006). Estimating the nitrogen concentration of strawberry plants from its spectral response. Communications in Soil Science and Plant Analysis, 37, 2447−2459. Fisher, J. I., & Mustard, J. F. (2007). Cross-scalar satellite phenology from ground, Landsat, and MODIS data. Remote Sensing of Environment, 109, 261−273. Geerken, R., Zaitchik, B., & Evans, J. P. (2005). Classifying rangeland vegetation type and coverage from NDVI time series using Fourier Filtered Cycle Similarity. International Journal of Remote Sensing, 26, 5535−5554. Gitelson, A. A., & Merzlyak, M. N. (1998). Remote sensing of chlorophyll concentration in higher plant leaves. Advances in Space Research, 22, 689−692. Goward, S. N., Tucker, C. J., & Dye, D. G. (1985). North-American vegetation patterns observed with the NOAA-7 Advanced Very High-Resolution Radiometer. Vegetatio, 64, 3−14. Harmon, M. E., Ferrell, W. K., & Franklin, J. F. (1990). Effects on carbon storage of conversion of old-growth forests to young forests. Science, 247, 699−702. Hay, S. I., Tatem, A. J., Graham, A. J., Goetz, S. J., & Rogers, D. J. (2006). Global environmental data for mapping infectious disease distribution. Advances in Parasitology, 62, 37−77. Hird, J., & McDermid, G. J. (2009). Noise reduction of NDVI time series: An empirical comparison of selected techniques. Remote Sensing of Environment, 113, 248−258. Huete, A., Didan, K., Miura, T., Rodriguez, E. P., Gao, X., & Ferreira, L. G. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83, 195−213. Jakubauskas, M. E., Legates, D. R., & Kastens, J. H. (2001). Harmonic analysis of timeseries AVHRR NDVI data. Photogrammetric Engineering and Remote Sensing, 67, 461−470. Jayakumar, S., Arockiasami, D. I., & Britto, S. J. (2002). Forest type mapping and vegetation analysis in part of Kolli hills, eastern ghats of Tamil Nadu. Tropical Ecology, 43, 345−349. Jeyaseelan, A. T., Roy, P. S., & Young, S. S. (2007). Persistent changes in NDVI between 1982 and 2003 over India using AVHRR GIMMS (global inventory Modeling and mapping studies) data. International Journal of Remote Sensing, 28, 4927−4946. Jönsson, P., & Eklundh, L. (2002). Seasonality extraction by function-fitting to time series of satellite sensor data. IEEE Transactions on Geoscience and Remote Sensing, 40, 1824−1832. Joshi, P. K., Roy, P. S., Singh, S., Agrawal, S., & Yadav, D. (2006). Vegetation cover mapping in India using multi-temporal IRS Wide Field Sensor (WiFS) data. Remote Sensing of Environment, 103, 190−202. Justice, C. O., Townshend, J. R. G., Holben, B. N., & Tucker, C. J. (1985). Analysis of the phenology of global vegetation using meteorological satellite data. International Journal of Remote Sensing, 6, 1271−1318. Kikim, A., & Yadava, P. (2001). Phenology of tree species in subtropical forests of Manipur in north eastern India. Tropical Ecology, 42, 269−276. Kushwaha, C. P., & Singh, K. P. (2008). India needs phenological stations network. Current Science, 95, 832−834. Lloyd, D. (1990). A phenological classification of terrestrial vegetation cover using shortwave vegetation index imagery. International Journal of Remote Sensing, 11, 2269−2279. Los, S. O., Collatz, G. J., Sellers, P. J., Malmstrom, C. M., Pollack, N. H., DeFries, R. S., et al. (2000). A global 9-yr biophysical land surface dataset from NOAA AVHRR data. Journal of Hydrometeorology, 1, 183−199. Malingreau, J. P. (1986). Global vegetation dynamics — Satellite-observations over Asia. International Journal of Remote Sensing, 7, 1121−1146. Menzel, A., Sparks, T. H., Estrella, N., Koch, E., Aasa, A., Ahas, R., et al. (2006). European phenological response to climate change matches the warming pattern. Global Change Biology, 12, 1969−1976. Mishra, R. K., Upadhyay, V. P., Bal, S., Mahapatra, P. K., & Mohanty, R. C. (2006). Phenology of species of moist deciduous forest sites of Similipal biosphere reserve. Lyonia, 11, 5−17. Mutanga, O., & Skidmore, A. K. (2004). Narrow band vegetation indices overcome the saturation problem in biomass estimation. International Journal of Remote Sensing, 25, 3999−4014. Myneni, R. B., Keeling, C. D., Tucker, C. J., Asrar, G., & Nemani, R. R. (1997). Increased plant growth in the northern high latitudes from 1981 to 1991. Nature, 386, 698−702.

Nemani, R. R., Keeling, C. D., Hashimoto, H., Jolly, W. M., Piper, S. C., Tucker, C. J., et al. (2003). Climate-driven increases in global terrestrial net primary production from 1982 to 1999. Science, 300, 1560−1563. Newton, P. N. (1988). The structure and phenology of s moist deciduous forest in the central Indian highlands. Vegetatio, 75, 3−16. Prasad, V. K., Badarinath, K. V. S., & Eaturu, A. (2007). Spatial patterns of vegetation phenology metrics and related climatic controls of eight contrasting forest types in India — Analysis from remote sensing datasets. Theoretical and Applied Climatology, 89, 95−107. Ralhan, P. K., Khanna, R. K., Singh, S. P., & Singh, J. S. (1985). Phenological characteristics of the tree layer of Kumaun Himalayan forests. Vegetatio, 60, 91−101. Reed, B. C., Brown, J. F., Vanderzee, D., Loveland, T. R., Merchant, J. W., & Ohlen, D. O. (1994). Measuring phenological variability from satellite imagery. Journal of Vegetation Science, 5, 703−714. Sarkar, S., & Kafatos, M. (2004). Interannual variability of vegetation over the Indian sub-continent and its relation to the different meteorological parameters. Remote Sensing of Environment, 90, 268−280. Schaber, J., & Badeck, F. -W. (2003). Physiology based phenology models for forest tree species in Germany. International Journal of Biometeorology, 47, 193−201. Shukla, R. P., & Ramakrishnan, P. S. (1984). Leaf dynamics of tropical trees related to successional status. New Phytologist, 97, 697−706. Singh, K. P., & Kushwaha, C. P. (2005). Paradox of leaf phenology: Shorea robusta is a semi-evergreen species in tropical dry deciduous forests in India. Current Science, 88, 1820−1824. Singh, K. P., & Kushwaha, C. P. (2006). Diversity of flowering and fruiting phenology of trees in a tropical deciduous forest in India. Annals of Botany, 97, 265−276. Studer, S., Stockli, R., Appenzeller, C., & Vidale, P. L. (2007). A comparative study of satellite and ground-based phenology. International Journal of Biometeorology, 51, 405−414. Sukumar, R., Suresh, H. S., & Ramesh, R. (1995). Climate change and its impact on tropical montane ecosystems in southern India. Journal of Biogeography, 22, 533−536. Sundarapandian, S. M., Chandrasekaran, S., & Swamy, P. S. (2005). Phenological behaviour of selected tree species in tropical forests at Kodayar in the Western Ghats, Tamil Nadu, India. Current Science, 88, 805−810. Tanre, D., Holben, B. N., & Kaufman, Y. J. (1992). Atmospheric correction algorithm for NOAA-AVHRR products — Theory and application. IEEE Transactions on Geoscience and Remote Sensing, 30, 231−248. Taylor, J. A., & Lloyd, J. (1992). Sources and sinks of atmospheric CO2. Australian Journal of Botany, 40, 407−418. Townshend, J. R. G., Justice, C. O., & Kalb, V. (1987). Characterization and classification of south-American land cover types using satellite data. International Journal of Remote Sensing, 8, 1189−1207. Vandijk, A., Callis, S. L., Sakamoto, C. M., & Decker, W. L. (1987). Smoothing vegetation index profiles — An alternative method for reducing radiometric disturbance in NOAA/AVHRR Data. Photogrammetric Engineering and Remote Sensing, 53, 1059−1067. Vermote, E., & Kaufman, Y. J. (1995). Absolute calibration of AVHRR visible and nearinfrared channels using ocean and cloud views. International Journal of Remote Sensing, 16, 2317−2340. Viovy, N., Arino, O., & Belward, A. S. (1992). The best index slope extraction (BISE) — A method for reducing noise in NDVI time-series. International Journal of Remote Sensing, 13, 1585−1590. White, M. A., Hoffman, F., Hargrove, W. W., & Nemani, R. R. (2005). A global framework for monitoring phenological responses to climate change. Geophysical Research Letters, 32, L04705. doi:10.1029/2004GL021961 White, M. A., Thornton, P. E., & Running, S. W. (1997). A continental phenology model for monitoring vegetation responses to interannual climatic variability. Global Biogeochemical Cycles, 11, 217−234. Zhang, Y. Q., Chen, J. M., Miller, J. R., & Noland, T. L. (2008). Leaf chlorophyll content retrieval from airborne hyperspectral remote sensing imagery. Remote Sensing of Environment, 112, 3234−3247. Zhang, X. Y., Friedl, M. A., Schaaf, C. B., & Strahler, A. H. (2004). Climate controls on vegetation phenological patterns in northern mid- and high latitudes inferred from MODIS data. Global Change Biology, 10, 1133−1145. Zhang, X., Friedl, M. A., Schaaf, C. B., Strahler, A. H., Hodges, J. C. F., Gao, F., et al. (2003). Monitoring vegetation phenology using MODIS. Remote Sensing of Environment, 84, 471−475. Zhou, L. M., Tucker, C. J., Kaufmann, R. K., Slayback, D., Shabanov, N. V., & Myneni, R. B. (2001). Variations in northern vegetation activity inferred from satellite data of vegetation index during 1981 to 1999. Journal of Geophysical Research-Atmospheres, 106, 20069−20083.