Global Biogeochemical Cycles

Characteristics and drivers of global NDVI-based FPAR from 1982 to 2006

Authors

  • Dailiang Peng,

    1. Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, Beijing, China
    2. Key Laboratory of Digital Earth, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Bing Zhang,

    Corresponding author
    1. Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, Beijing, China
    2. Key Laboratory of Digital Earth, Chinese Academy of Sciences, Beijing, China
      Corresponding author: B. Zhang, Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, 9 Dengzhuang S. Rd., Haidian District, Beijing 100094, China. (zb@ceode.ac.cn)
    Search for more papers by this author
  • Liangyun Liu,

    1. Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, Beijing, China
    2. Key Laboratory of Digital Earth, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Hongliang Fang,

    1. Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Dongmei Chen,

    1. Department of Geography, Queen's University, Kingston, Ontario, Canada
    Search for more papers by this author
  • Yong Hu,

    1. Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Lingling Liu

    1. Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author

Corresponding author: B. Zhang, Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, 9 Dengzhuang S. Rd., Haidian District, Beijing 100094, China. (zb@ceode.ac.cn)

Abstract

[1] Fraction of Absorbed Photosynthetically Active Radiation (FPAR) is a state parameter in most ecosystem productivity models and is also the key terrestrial product. In this study, Normalized Difference Vegetation Index (NDVI) from Advanced Very High Resolution Radiometer (AVHRR) Global Inventory Modeling and Mapping Studies (GIMMS) was used to estimate FPAR from 1982 to 2006, using an intermediate model. Our research focused on the analysis of long-term global FPAR interannual trend patterns and driving forces involving climate and land cover changes. Results showed that interannual trend and spatial distribution patterns of global FPAR were independent of the changes in AVHRR instruments, and differed by season dynamics and vegetation types. Compared with other seasons, the period during JJA (June–August) exhibited more areas with decreasing FPAR and greater reduction range. For FPAR interannual trend, a wholly different correlation pattern was observed between temperature and precipitation, especially for arid and semi-arid regions. A significant influence of extreme droughts such as those associated with El Nino/Southern Oscillation (ENSO) on FPAR variability was found. The result also revealed the increasing and decreasing interannual trend of FPAR corresponding to the afforestation in the Three North Shelterbelts Program in China and deforestation in tropical forests in Southeast Asia. Driving factor analysis indicated that the climate and land cover changes had an interactive effect on the FPAR annual anomalous variation.

1. Introduction

[2] The Fraction of Absorbed Photosynthetically Active Radiation (FPAR) is generally defined as the fraction of Photosynthetically Active Radiation (PAR) absorbed by vegetation in the 0.4–0.7 μm spectrum. FPAR acts as one of the key state parameters in global models of climate, hydrology, biogeochemistry, and ecology [Sellers et al., 1996, 1997], and has been identified as one of the Essential Climate Variables (ECV) by the Global Climate Observing System [McCallum et al., 2010]. There are several global FPAR data sets, such as Moderate Resolution Imaging Spectroradiometer (MODIS) and Multi–angle Imaging Spectro Radiometer (MISR) FPAR, Carbon Cycle and Change in Land Observational Products from an Ensemble of Satellites (CYCLOPES) FPAR, Joint Research Center (JRC) FPAR, and GLOBCARBON FPAR, and the time series of above mentioned FPAR products are March 2000 to the present, 1999 to 2007, 1997 to 2006, and 1998 to 2006, respectively, they are not long enough to study the relationship between vegetation dynamics and global climate change or other anthropogenic activities. Several regional or short time period FPAR data studies have been reported [Fensholt et al., 2004; Fang et al., 2005; Hu et al., 2007; Weiss et al., 2007; McCallum et al., 2010; Gobron et al., 2006], but a globally comprehensive analysis of FPAR is lacking, especially for the global FPAR characteristics and drivers analysis. FPAR is linked closely to the maximum photosynthetic capacity of vegetation [Liu et al., 2006], and all factors related to vegetation cover, plant growth change, and plant photosynthesis will contribute to the anomalous variation of FPAR, including climatic and anthropogenic factors. Therefore, a long-term consistent FPAR product is necessary to study the spatial and temporal patterns of global vegetation dynamics associated with climatic changes and human activities.

[3] The purpose of this paper is to analyze the global FPAR interannual variation characteristics from 1982 to 2006, and the driving forces primarily involved in climate and land cover changes. Section 2 introduces the data and methods used to generate and analyze the global FPAR. In Section 3, we present the resultant FPAR data set and compare it to the MODIS FPAR and the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) FPAR over global FluxNet stations from 2000 to 2006, as well as the gross primary productivity (GPP) over North America Flux sites from 2000 to 2006. We focus on a long-term global FPAR interannual trend and spatial distribution patterns of global FPAR and the drivers of climatic factors and land cover changes. Finally, our concluding remarks are presented inSection 4.

2. Methodology

2.1. Data Description

2.1.1. AVHRR GIMMS NDVI

[4] The latest version of the Global Inventory Modeling and Mapping Studies (GIMMS) Normalized Difference Vegetation Index (NDVI) data sets from 1982 to 2006 was downloaded via http://glcf.umiacs.umd.edu/data/gimms/ [Tucker et al., 2005]. This data set is in geographic coordinates with an 8-km Albers Equal Area Conic projection, the Clarke 1866 ellipsoid, and WGS84 datum. The data is composited at a 15-day time step: the 15a composite is the maximum value composite from the first 15 days of the month, and the second (15b) is from day 16 to the end of the month. The latest GIMMS-NDVI data set that has been corrected for calibration, view geometry, volcanic aerosols, and other effects not related to actual vegetation change [Tucker et al., 2005; Brown et al., 2006]. The current GIMMS NDVI data sets had a satellite overpass time drift correction that largely eliminates the variation of NDVI due to changes in solar zenith angle [Tucker et al., 2005]. The on–site tests showed reasonable agreement between interannual variation in GIMMS NDVI and other measures of vegetation [Malmström et al., 1997; D'Arrigo et al., 2000; Tucker et al., 2005].

2.1.2. GlobCover Land Cover Product

[5] The GlobCover land cover product at 300 m resolution is derived by the Medium Resolution Imaging Spectrometer Instrument (MERIS) Full Resolution (FR) time series data. These products from December 2004 to June 2006 are available from the Pôle d'Observation des Surfaces Terrestres aux Echelles Larges (POSTEL) Service Centre (http://medias.obs- mip.fr/postel/). Twenty-two total land cover classes are defined with the United Nations Land Cover Classification System (LCCS). The GlobCover map was reprojected to AVHRR GIMMS NDVI coordinate with an 8-km spatial resolution. Corresponding to the new Simple Biosphere model (SiB2) vegetation classification [Sellers et al., 1994, 1996], the original 22 land cover classes were recombined to 9 vegetation types by the majority of class if the area of this class was larger than fifty percent (50%), or was taken as the mixed class if the area of all classes was smaller than 50%, and the recombined global land cover map was shown in Figure 1.

Figure 1.

Flux net station distribution and global land cover map.

2.1.3. Reference Data Set for FPAR Comparisons

[6] FluxNet is a global network of micrometeorological tower sites that uses the eddy covariance methods to measure the exchanges of carbon dioxide, water vapor, and energy between the terrestrial ecosystem and atmosphere [Running et al., 1999; Zhao et al., 2005]. More than 400 tower sites are in service on a long-term and continuous basis (Figure 1). The JRC FPAR products derived from SeaWiFS (denoted as SeaWiFS FPAR) over these sites were used to compare the FPAR estimation using the method in this study. The SeaWiFS sensor on the Orbview-2 spacecraft, in operation since autumn of 1997, collects measurements in 8 narrow spectral bands extending from the blue to the near–infrared region of the solar spectrum, ensuring a global coverage approximately every 2 days [Gobron et al., 2006]. The SeaWiFS FPAR values corresponding to the 10-day time composite products were spatially averaged over 3 × 3 pixels (6 × 6 km2), around the central pixel, and the comparisons with the ground-based data sets indicated that the accuracy range was within ±0.1 with quite good representation of seasonal cycles [Gobron et al., 2006].

[7] The version-5 MODIS/Terra standard FPAR product (MOD15A2) over Flux net stations from 2000 to 2006 was obtained by the Warehouse Inventory Search Tool (WIST), which has a spatial resolution of 1-km and an 8-day maximum value [Myneni et al., 2002]. The value is derived from either a main algorithm (i.e., Radiation Transfer process (RT)) when possible or a back-up algorithm (i.e., the empirical relationship between FPAR and NDVI) [Cohen et al., 2003]. According to the MODIS FPAR collection 5 changes introduction, the algorithm refinements are targeted to improve the quality of FPAR retrievals and consistency with field measurements over all biomes compared to an older version of the data. The accuracy of this data set has been estimated using independent measurements obtained from selected locations and time periods and ground-truth/field program efforts [Morisette et al., 2006].

[8] We also obtained the Level 4 GPP data from the AmeriFlux web site (http://public.ornl.gov/ameriflux) for 37 North America Flux sites (Figure 1) over the period 2000 to 2006. The Level 4 GPP products were gap-filled using the Marginal Distribution Sampling (MDS) method [Reichstein et al., 2005] and the Artificial Neural Network (ANN) method [Papale and Valentini, 2003]. We used the averages of MDS-GPP and ANN-GPP to compare with three FPAR products (SeaWiFS FPAR, MODIS FPAR, and estimated FPAR in this study).

2.1.4. Meteorological Data

[9] The monthly mean temperature and accumulated precipitation were collected from the Global Land Data Assimilation System Version 2 (GLDAS-2) data set, which have ingested satellite-and ground-based observational data products [Rodell et al., 2004]. It is simulated by the noah land surface model, and has a spatial resolution of 1-degree.

2.2. FPAR Estimation From AVHRR GIMMS NDVI

[10] Sellers [1985]investigated methods of integrating simple leaf-level models of light scattering, light absorption, photosynthesis, and stomata conductance over the depth of vegetation canopies. His analysis provided a strong mechanistic basis for the FPAR-Simple Ratio (SR) Vegetation Index correlation [Sellers et al., 1992]. In Sellers et al. [1996], relationships between FPAR and SR were derived by land cover class as follows:

display math

where SR = (1 + NDVI)/(1 − NDVI), FPARmax = 0.95, FPARmin = 0.001, SRmax and SRmin represent the SR values corresponding to 98% and 2% of the NDVI frequency distributions. Equation (1)is referred to as the SR-FPAR model.

[11] An alternative model, referred to as the NDVI-FPAR model [Choudhury, 1987; Goward and Huemmrich, 1992] is given by the following formula.

display math

where NDVImax and NDVImin mean the NDVI values corresponding to 98% and 2% of the NDVI frequency distributions. Equation (2)is referred to as the NDVI-FPAR model.Los et al. [2000]gave an intermediate model, calculating the average FPAR from the NDVI-FPAR and SR-FPAR models, as the following:

display math

[12] Los et al. [2000]compared the estimated FPAR using the above mentioned three models with ground-measured FPAR, and found that the intermediate model performed better than the other two models. In this study, we adopted this intermediate model to estimate global FPAR from 1982 to 2006. 98% and 2% NDVI and corresponding SR values for different vegetation types were calculated from GIMMS NDVI from 1982 to 2006, based on global land cover inTable 1. The estimated FPAR by the intermediate model is referred to as GIMMS NDVI SR FPAR.

Table 1. The 98% and 2% NDVI and Corresponding SR Values for Different Vegetation Types
Vegetation TypeNDVI2%SRminNDVI98%SRmax
Evergreen broadleaf forest0.0151.0300.85012.333
Deciduous broadleaf forest0.0151.0300.84511.903
Needleleaf and broadleaf forest0.0151.0300.86513.815
Evergreen needleleaf forest0.0151.0300.8159.811
Deciduous needleleaf forest0.0151.0300.84011.500
Grassland0.0151.0300.7607.333
Shrubland0.0151.0300.7808.091
Barren or sparsely vegetated area0.0151.0300.6905.452
Cropland0.0151.0300.7958.756

2.3. The Analysis of the Characteristics and Drivers of Long-Term Global FPAR

[13] Based on the previous studies, MODIS FPAR values retrieved by the RT algorithm are more reliable than those retrieved by the back-up algorithm because the empirical relationship between FPAR and NDVI is employed mostly when cloud cover, strong atmospheric effects, or snow/ice are detected [Zhao et al., 2005; Yang et al., 2006a]. In this study, according to the quality flags accompanying the MODIS FPAR product, only the reliable MODIS FPAR without cloud, aerosols, or snow or snow disturbance retrieved by the RT algorithm were used to compare with the corresponding SeaWiFS FPAR, flux GPP, and GIMMS NDVI SR FPAR at the flux sites. The scatterplots of three FPAR products over global flux sites from 2000 to 2006 were used for spatial comparison for each different land cover types. We averaged the monthly FPAR values over North America flux sites from 2000 to 2006, as well as the flux GPP, for their seasonal variations comparisons. The absolute error (AE), mean absolute error (MAE), root mean square error (RMSE) between GIMMS NDVI SR FPAR and SeaWiFS and MODIS FPAR were used as the measurements of comparisons.

display math
display math
display math

where n is the total number of FPAR retrieved by RT algorithm and without cloud, aerosols, or snow disturbance, and i denotes each selected FPAR.

[14] A globally comprehensive analysis of interannual trend of FPAR were provided, including the interannual trend of FPAR from 1982 to 2006 at the global, northern and southern hemisphere scales; and GIMMS NDVI SR FPAR averaged for the periods of March–May (MAM), June–August (JJA), September–November (SON), December–February (DJF), and the whole year (Annual), respectively, across the different vegetation types. The monthly mean temperature and accumulated precipitation from GLDAS-2 data set were used to examine the climatic impact on the FPAR interannual trend from 1982 to 2006. The Amazon, Three North Shelterbelts in China, and tropical forests in Southeast Asia were selected as the typical areas to explore the influence of drought, afforestation, and deforestation on FPAR variability.

3. Results and Discussion

3.1. Intercomparisons With SeaWiFS, MODIS FPAR, and Flux GPP

[15] Comparisons of GIMMS NDVI SR FPAR with SeaWiFS FPAR and MODIS FPAR for different land cover types over global Fluxnet sites from 2000 to 2006 are shown in Figure 2 and Table 2. RMSE and MAE values of GIMMS NDVI SR FPAR were lower compared with MODIS FPAR than with SeaWiFS FPAR for all vegetation types. Most RMSE values were approximately 0.204 for SeaWiFS FPAR, and 0.121 for MODIS FPAR, MAE of GIMMS NDVI SR FPAR compared to 0.174 for SeaWiFS FPAR and 0.094 for MODIS FPAR (Table 2). Most points plotted in Figure 2a fell above the y = x line for all land cover types, while the points in Figure 2b had a concentrated distribution around y = x, especially for grassland, shrubland, cropland, and sparsely vegetated area (Figure 2b). However, most points fell below the y = x line for all forest cover area over global flux sites used in the study. The distribution of scatterplots in Figure 2 indicated that in most class GIMMS NDVI SR FPAR were smaller than MODIS FPAR, and larger than SeaWiFS FPAR. Compared to SeaWiFS FPAR, GIMMS NDVI SR FPAR was closer to MODIS FPAR. Although there were biases among three FPAR products, both SeaWiFS and MODIS FPAR had significant correlations (P < 0.001) with GIMMS NDVI SR FPAR across all land cover types (Figure 2).

Figure 2.

Comparisons of GIMMS NDVI SR FPAR with (a) SeaWiFS FPAR and (b) MODIS FPAR for different land cover types over Fluxnet sites from 2000 to 2006. Only the reliable MODIS FPAR extracted by the RT algorithm and without cloud, aerosols, or snow disturbance was used for further comparisons with corresponding SeaWiFS FPAR, flux GPP, and GIMMS NDVI SR FPAR at flux sites.

Table 2. The Mean Absolute Error (MAE), Root-Mean-Square Error (RMSE), and Coefficients of Determination (R2) Between GIMMS NDVI SR FPRA and SeaWiFS and MODIS FPAR
Vegetation TypeSeaWiFS FPARMODIS FPAR
MAERMSER2MAERMSER2
Evergreen broadleaf forest0.2390.2260.410.0850.1130.656
Deciduous broadleaf forest0.1450.1750.4060.1160.1370.345
Needleleaf and broadleaf forest0.1830.2090.2290.1090.1420.283
Evergreen needleleaf forest0.2010.230.1640.0970.1180.476
Deciduous needleleaf forest0.1960.2260.2850.1230.1450.518
Grassland0.1610.2010.5630.0780.1130.68
Shrubland0.1470.170.6260.0650.0910.72
Barren or sparsely vegetated area0.1520.1870.4390.0930.1250.55
Cropland0.1420.1760.3950.0790.1040.607

[16] SeaWiFS FPAR algorithms involve the input data from the SeaWiFS sensor, and a set of ancillary data mathematical functions [Govaerts et al., 1999]. The algorithm includes two steps. First, the spectral Bi-directional Reflectance Factors (BRFs) measured in the red and near infrared bands are rectified to ensure their optimal decontamination from atmospheric and angular effects. Second, two bands are combined together to estimate the instantaneous FPAR value [Gobron et al., 2006]. Many assumptions and uncertainties, such as the horizontally homogeneous plant canopies, ignored orographic and adjacency effects, would result in errors in the estimation of SeaWiFS FPAR, especially for the regions of hilly or mountainous where the retrieval of vegetation characteristics may not be reliable [Gobron et al., 2002]. Gobron et al. [2006]found that the SeaWiFS FPAR over FluxNet sites appeared biased low with respect to the ground-based measurements, especially under conditions where the structural effects become significant and/or the contribution of the woody elements of the canopy to the interception process was not negligible.McCallum et al. [2010] compared the SeaWiFS FPAR and MODIS FPAR over Northern Eurasia for the year 2000, their results showed that the SeaWiFS FPAR was smaller than MODIS FPAR over all the land cove types, which was consistent with the results of comparisons in this study. In addition, several previous studies have confirmed the tendency of the MODIS FPAR product to overestimate actual FPAR or other satellite FPAR products, especially for mixed stands and dense canopies [Cohen et al., 2003; Fang et al., 2005; Bacour et al., 2006; Weiss et al., 2007; McCallum et al., 2010]. Weiss et al. [2007] validated the CYCLOPES FPAR using MODIS FPAR, and found CYCLOPES FPAR was smaller than MODIS FPAR for forest regions, and was closer to MODIS FPAR for low canopy vegetation. We drew similar conclusions from Figure 2b, although the intermediate model (equations (1)(3)) used in this study was different from the CYCLOPES FPAR algorithms based on the radiative transfer model inversion. Los et al. [2000] found that the estimating FPAR by the intermediate model in this study had a good consistence with the measured data, they did not show a lack of fit at the 1% significance level, and the RMSE was 0.08.

[17] The NDVI-based FPAR algorithms (equations (1)(3)) in this study are a function only of NDVI. For the NDVI–FPAR model (equation (2)), NDVImax, NDVImin, FPARmax, and FPARminare constants; therefore the NDVI-FPAR model is a simple linear function of NDVI, and FPAR increases linearly but rapidly. The SR-FPAR model (equation (1)) is a nonlinear function of NDVI because SR, FPAR increases slowly and then rapidly at higher NDVI values (about NDVI > = 0.8). While the intermediate model (equation (3)), which was used in Los et al. [2000]and in this study, provides a better fit than the NDVI-FPAR and SR-FPAR models, the FPAR increases dramatically after NDVI > 0.9. Previous studies have also confirmed that AVHRR NDVI exhibited saturated signals for high biomass conditions [Huete, 1988]. Those may account for the difference among the three FPAR products.

[18] The seasonal variations of GIMMS NDVI SR FPAR, SeaWiFS FPAR, and MODIS FPAR were shown in Figure 3. There were significant seasonal variations, with the maximum FPAR occurring in summer season, and minimum FPAR in winter season. GIMMS NDVI SR FPAR always had closer values and more consistent seasonal patterns with MODIS FPAR than with SeaWiFS FPAR. SeaWiFS FPAR seasonal dynamics showed a later onset of greenness than GIMMS NDVI SR and MODIS FPAR, especially in forests (as listed in Table 1) and cropland. With the exception of a small peak in July, evergreen needleleaf forest kept at FPAR values of around 0.5 for GIMMS NDVI SR and MODIS FPAR, and 0.38 for SeaWiFS FPAR. For shrubland, peak values of FPAR were observed in April and August. This may have been caused by grazing between the months April and June followed by a regrowth period during good moisture conditions [C. Wang et al., 2010].

Figure 3.

Comparisons of the seasonal variations of GIMMS NDVI SR FPAR with SeaWiFS FPAR, MODIS FPAR, and flux GPP for different land cover types over North America Flux sites from 2000 to 2006. Only the reliable MODIS FPAR extracted by the RT algorithm and without cloud, aerosols, or snow disturbance was used for further comparisons with corresponding SeaWiFS FPAR, flux GPP, and GIMMS NDVI SR FPAR at flux sites.

[19] GPP is the amount of carbon fixed by vegetation through photosynthesis and is a key component of ecosystem carbon fluxes and the carbon balance between the biosphere and the atmosphere [Mäkelä et al., 2008]. GPP is directly related to FPAR through light use efficiency [Zhao et al., 2005]. Without the field measured FPAR data, GPP from eddy covariance flux towers over North America from 2000 to 2006 were used to analyze the similarities and differences among three FPAR products (Figure 3). Because the deciduous dominated canopy and cropland was bare and low air temperature and frozen soil inhibit photosynthetic activities of conifer trees [Xiao et al., 2004], GPP values, in the region of deciduous broadleaf forest, deciduous needleleaf forest, and cropland, were near zero during the winter season. Due to the data selection procedure according to the MODIS FPAR by the RT algorithm and without cloud, aerosols, or snow disturbance, and the missing periods of GPP data because of weather conditions or inactive flux towers, we didn't get the flux GPP for the comparisons with three FPAR products over evergreen broadleaf forest and sparsely vegetated area. GIMMS NDVI SR and MODIS FPAR had consistent seasonal dynamics with flux GPP for evergreen needleleaf forest, deciduous needleleaf forest, shrubland, and cropland (Figure 3). The consistent seasonal dynamics were found between SeaWiFS FPAR and flux GPP for needleleaf and broadleaf forest (Figure 3). For deciduous broadleaf forest and grassland, flux GPP increased continuously before reaching its maximum values in summer season (Figure 3). In addition to the estimating errors of three FPAR products in this study, the estimation error of GPP at the daily time scale should also be considered when analyzing for biases among FPAR product and GPP. GPP was calculated from field-measured net ecosystem exchange (NEE) and ecosystem respiration (including day and nighttime ecosystem respiration). A large uncertainty was found in estimating daytime ecosystem respiration. Therefore, daytime ecosystem respiration can result in an error in the estimation of GPP. Although the Level 4 GPP products used in this study were gap-filled using MDS and ANN methods [Papale and Valentini, 2003; Reichstein et al., 2005; Xiao et al., 2010], these methods required subjective decisions and were currently the subject of a great deal of discussion [Falge et al., 2001; Xiao et al., 2004].

3.2. Interannual Variation Patterns of Global FPAR

3.2.1. Interannual Variations

[20] In general, there was a slight increase in annual FPAR from 1982 to 2006, with an average annual growth of 0.009, 0.01, and 0.008 for global, Northern Hemisphere, and Southern Hemisphere, respectively (Figure 4). FPAR declined during the periods of 1991–1994, 1997–2000 and 2002–2005, with the annual trend being more obvious in the Southern Hemisphere than that in the Northern Hemisphere. The FPAR interannual change patterns were independent of AVHRR instruments changes (Figure 4). FPAR interannual change patterns were varied by season and vegetation types. In MAM, SON and DJF, FPAR increased at an average rate of 0.014, with the exception of evergreen broadleaf forest in MAM and deciduous needleleaf forest in DJF. For all forest types (as listed in Table 1), FPAR values decreased at an average rate of 0.016 from 1982 to 2006 in JJA periods, which showed more significantly annual fluctuations than those of non-forest (as listed inTable 1). Non-forest FPAR showed the smallest interannual growth increment across the four seasons with an interannual average rate of 0.006. The average FPAR interannual changes across the different vegetation types increased with an average rate of 0.009, with the exception of evergreen needleleaf forest.

Figure 4.

Interannual trend of FPAR at the global, northern, and southern hemisphere scales from 1982 to 2006, with the AVHRR instruments changes.

3.2.2. Spatial Patterns of FPAR Interannual Trends

[21] The interannual change of FPAR from 1982 to 2006 was developed, and the global spatial patterns in four seasons were mapped in Figure 5. Both increasing and decreasing trends in annual change in FPAR were observed except in the Australian desert area, which had a year-round decreasing trend. In addition, there were more areas with decreasing FPAR in JJA than in other periods (Figure 5). Most interannual changes of FPAR ranged around −0.1–0.1. However, significant changes (larger than 0.1) were observed in the following regions: in the tropical rain forest of Southern America during the period of MAM, southern Alaska, some tropical rain forests in Africa and Southern America, tropical semi-deciduous forest in Asia, non-desert areas of Australia, middle latitude deciduous forest during the period of JJA, tropical semi-deciduous forest in the central part of Southern America, Africa and Asia during the period of SON, and Tanzania during the period of DJF. A few regions with a change of 0.2 were found in Zimbabwe and Botswana, Northern China and western Australia during the period of MAM, the tropical rain forests in Africa and the Amazon, northwestern North America, Southeast Asia, and non-desert areas of Australia during the period of JJA, forest area in Southern America during the period of SON, and northern India and African tropical rain forest in the period of DJF. Because of low (or even no) solar radiation [Upchurch et al., 1998], the very low air temperature and frozen soil (or covered by the snow/ice) inhibited photosynthetic activities of plant [Xiao et al., 2004] causing FPAR values in the high northern latitudes area to be zero during the period of DJF (Figure 5).

Figure 5.

Global FPAR interannual changes in the different seasons (MAM, March–May; JJA, June–August; SON, September–November; DJF, December–February) and annual average (Annual) from 1982 to 2006. The background and zero-value of FPAR areas were colored in white. The black, green, and purple label area shows Three North Shelterbelts in China, tropical rain forest in Amazon basin, and tropical rain forest in Southeast Asia, respectively. These places were used as the typical areas for the driver analysis of the FPAR interannual changes.

[22] The global FPAR interannual change averaged over 25 years (1982–2006) revealed that the change values were within 0.1. Furthermore, there were more regions with the increasing FPAR than decreasing FPAR for interannual change (Figure 5, Annual). An increase in FPAR from 1982 to 2006 was observed in areas of coniferous forest in Northern Russia and America, most areas in Europe, northern China, tropical rain forest in Africa, and western and southern Australia. FPAR decreased in wetland areas in North America, the central part of Africa and Australia, and Southeast Asia. Most evergreen needleleaf forest located in North America and Eurasia, had a significant decrease from 1982 to 2006. A reduction of larger than 0.2 was found in the tropical rain forest in Africa, northwestern North America, Southeast Asia, and non-desert area in Australia during the period of JJA (Figure 5, JJA). In conclusion, the average FPAR interannual decreased overall, even though it increased during the periods of MAM, SON and DJF. The annual change in the typical regions in Figure 5—such as Three North Shelterbelts in China, tropical forests in Southeast Asia, and the Amazon—will be further discussed in the following sections.

3.3. Climatic Influence on Global FPAR Variations

3.3.1. Relationship Between Interannual Trend of FPAR and Temperature, Precipitation

[23] We mapped the significance level of the correlations between interannual trend of FPAR and temperature, precipitation during the periods of MAM, JJA, SON, and DJF from 1982 to 2006, as shown in Figures 6 and 7, and calculated the area percentage of different significance levels in Table 3. In the global area where interannual trends of FPAR were observed, more than 70% showed no significant correlation with climate factors (precipitation and temperature), except for the correlation of FPAR interannual trend with temperature during MAM (Table 3). A wholly different correlation pattern was observed between temperature and precipitation, especially for arid and semi–arid regions, such as central and northwestern Australia and southern Africa, where temperature and precipitation showed significant negative and positive correlation (P < 0.01) with FPAR interannual trend, respectively (Figures 6 and 7).

Figure 6.

Significant correlations between the interannual trend of FPAR and temperature in the different seasons (MAM, March–May; JJA, June–August; SON, September–November; DJF, December–February) from 1982 to 2006. SNCL is short for significant negative correlation level; SPCL is short for significant positive correlation level. The background and zero-value of FPAR areas were colored in white. The black, green, and purple label area shows Three North Shelterbelts in China, tropical rain forest in Amazon basin, and tropical rain forest in Southeast Asia, respectively. These places were used for the typical areas for the drivers' analysis of the FPAR interannual changes.

Figure 7.

Significant correlations between the interannual trend of FPAR and precipitation in the different seasons (MAM, March–May; JJA, June–August; SON, September–November; DJF, December–February) from 1982 to 2006. SNCL is short for significant negative correlation level; SPCL is short for significant positive correlation level. The background and zero-value of FPAR areas were colored in white. The black, green, and purple label area shows Three North Shelterbelts in China, tropical rain forest in Amazon basin, and tropical rain forest in Southeast Asia, respectively. These places were used for the typical areas for the drivers' analysis of the FPAR interannual changes.

Table 3. The Percentage of the Area of the Different Significant Correlation Between the Interannual Trend of FPAR and Precipitation and Temperature in the Different Seasons from 1982 to 2006 at the Global Scalea
Climate FactorCorrelation LevelMAMJJASONDJFAnnual
  • a

    MAM, March–May; JJA, June–August; SON, September–November; DJF, December–February.

  • b

    SNCL is short for Significant Negative Correlation level.

  • c

    SPCL is short for Significant Positive Correlation level.

PrecipitationSNCb < 0.011.0540.8900.8400.7580.740
0.01 < SNCL < 0.052.3452.4512.4102.6562.048
0.05 < SNCL < 0.12.2682.4932.3023.0732.138
No significant79.39377.04678.53887.41674.735
0.05 < SPCc < 0.14.8594.8204.5732.8745.816
0.01 < SPCL < 0.055.8246.6476.3152.3946.926
SPCL < 0.014.2575.6525.0220.8307.596
       
TemperatureSNCL < 0.012.8534.4882.8520.8734.012
0.01 < SNCL < 0.052.6046.1874.9452.1343.501
0.05 < SNCL < 0.12.4934.9494.4112.1473.313
No significant64.19377.55580.53677.25773.086
0.05 < SPCL < 0.16.7532.4953.1266.7025.294
0.01 < SPCL < 0.0510.6112.6632.9527.5746.459
SPCL < 0.0110.4931.6641.1783.3134.334

[24] In many regions around the world, climate changes (temperature and precipitation in this study) had the different effects on FPAR interannual trends during the period of MAM, JJA, SON, and DJF (Figures 6 and 7). For example, there were more areas where interannual trend of FPAR had a significant positive correlation with temperature during MAM and DJF, but not in any other seasons (Table 3). About 20% of the global areas showed a significant positive correlation at the 99% or 95% level (P < 0.01 or P < 0.05) with temperature during MAM (Table 3), and was mainly observed in North America and Eurasia, and in the Northern Russia and central North America during DJF. However, there were significant negative correlations during JJA, and no significant correlations during SON for North America and Eurasia (Figure 6). In southern Sahara, southwest Russia, and areas around Central Asia, interannual trend of FPAR showed significant positive correlations with precipitation at the level of 99% or 95% (P < 0.01 or P < 0.05) only during the period of JJA and SON, and no significant correlation was observed in other seasons (Figure 7).

[25] According to Figures 6 and 7(Annual), we found that FPAR interannual trend almost had a negative correlation with temperature (P < 0.01 or P < 0.05) in the semi-arid and arid regions in the Southern Hemisphere (throughout the tropics). Most regions in the Northern Hemisphere showed a positive or no correlation between FPAR and temperature, except for areas in the southern part of North America and Sahara, northwestern India, southwest Russia, and around of the Central Asia, where a significant negative correlation was observed. The relationship of precipitation and FPAR interannual trend had an inverse spatial distribution pattern compared to that of temperature.

3.3.2. The Influence of Drought on Interannual Variability of FPAR in the Amazon

[26] According to Figure 5, during MAM and JJA from 1982 to 2006, FPAR in most areas in the northern Amazon decreased, and some place in the southern and eastern Amazon increased, especially for the Amazon River region. In the period of SON, a significant change of 0.2 in FPAR was observed in the northern Amazon. However, for most regions in the Amazon, the change of interannual FPAR was smaller than 0.1 during DJF, as well as for the average FPAR in the whole year. The significant negative and positive correlation (P < 0.01 or P < 0.05) of interannual FPAR trend was observed with temperature in the Amazon River region during MAM and SON, and a significant negative correlation (P < 0.01) was found in SON with precipitation (Figures 6 and 7). However, for most areas in the Amazon, there was no significant correlation between interannual trend of FPAR and climate factors (temperature and precipitation). The spatial distribution of correlation between interannual trend of FPAR and precipitation was approximately uniform with the correlation between LAI and precipitation in Myneni et al. [2007]'s study, which found that these forests in the southern Amazon could be water-limited [Myneni et al., 2007].

[27] The El Nino/Southern Oscillation (ENSO) extreme phases are usually related to major episodes of floods and droughts [Barlow et al., 2001]. To further identify the relative roles of ENSO on the long-term Amazon FPAR variability,Figure 8a showed the interannual variations FPAR in the Amazon area (Figure 5) from 1982 to 2006, the corresponding variations of Southern Oscillation Index (SOI) and observed rainfall can be found in Zeng et al. [2008]. We found a coherent FPAR reduction with SOI in Zeng et al. [2008], especially during 1982–1983, 1986–1987, 1991, 1997, and 2002–2003 (Figure 8a). The El Niño events (SOI < 0) caused large negative rainfall anomalies and drought in the Amazon, and often immediately followed by La Niña events (SOI > 0) which resulted in anomalously wet conditions [Zeng et al., 2008]. Saleska et al. [2007]showed a green-up during the drought of 2005 in Amazon, which seems to be identical with FPAR increase from 2004 to 2005 inFigure 8a. However, Samanta et al. [2010, 2011a, 2011b, 2012]found no evidence of large-scale greening of intact Amazon forests during the 2005 drought. To elucidate the influence of drought in 2005 on variability of FPAR in Amazon, the seasonal variation of FPAR in 2005 over the Amazon area (Figure 5) was shown in Figure 8b; it was found that FPAR significantly decreased during the dry season (July–August–September) of 2005 [Samanta et al., 2010]. The increasing FPAR from 2004 to 2005 over the whole Amazon area may be attributed to the drought of 2005 was especially severe during the dry season (July–August–September) in southwestern Amazon, but did not impact the central and eastern regions [Marengo et al., 2008; Samanta et al., 2010; Xu et al., 2011].

Figure 8.

(a) Interannual variation of FPAR from 1982 to 2006 and (b) the seasonal variation of FPAR in 2005 over Amazon basin.

[28] The long duration of drought caused the FPAR drop, and which mechanism may be explained that water-limited vegetation responds promptly to initial drought by reducing transpiration and photosynthesis [Saleska et al., 2007], and the green-up of vegetation may be prevented if deep roots are insufficient to overcome dry-season rainfall deficits during the strong drought over El Niño period [Huete et al., 2006]. In addition, higher rates of tree mortality and increased forest flammability are known to be happened, when the plant available soil moisture stays below a critical threshold level for a prolonged period of extreme droughts such as those associated with the ENSO [Nepstad et al., 2004, 2007; Samanta et al., 2010]. However, interannual trend of FPAR was observed to have no significant correlation with rainfall (Figure 7). Besides the climate, the human drivers, such as deforestation and land use pressures may influence the balance of moisture and sunlight controls on Amazon phenology with important consequences to sustainable land use in Amazonia [Huete et al., 2006].

3.4. The Influence of Afforestation, Deforestation on FPAR Variability

[29] According to the interannual change of FPAR in Three North Shelterbelts from Figure 5 and Figure 9a, it was found that FPAR showed an increasing trend for most regions from 1982 to 2006 with an average annual growth of 0.01, which is mostly attributed to the afforestation program. In 1978, a tremendous afforestation program was executed in Three North Shelterbelts (northeast China, north China, and northwest China). The objective of this program is to improve forest coverage from 5% to 15% in arid and semiarid China, and thus to combat desertification and to control sand storms [Wang and Zhou, 2003; X. M. Wang et al., 2010]. Government bulletins, forestry statistical yearbooks, forestry yearbooks, and previous studies have reported great afforestation achievements under this program [Zhu et al., 2004; Yu et al., 2006].

Figure 9.

(a) Interannual variation of FPAR in Three North Shelterbelts in China and (b) tropical rain forest of Southeast Asia with the characteristics of afforestation and deforestation, respectively.

[30] We found that the annual change of FPAR from 1982 to 2006 exhibited an evident decreasing trend in the periods of MAM and JJA in the northern Amazon (Figure 5) and nearly all of Southeast Asia (Figures 5 and 9b). This trend was also observed in year-round averaged result shown inFigure 5 (Annual) and the annual mean decrease in FPAR was about 0.02 (Figure 9b). The deforestation is widely recognized as one of the world's leading environmental problems, which leads to the loss of original forests, reduction of the size of forest fragments [Armenteras et al., 2006; Ewers, 2006]. Global estimates found that 9 million km2 of tropical forests were lost in less than 50 years, and Brazilian Amazonia and Southeast Asia have the world's highest absolute rates of forest deforestation [Bhattarai and Hammig, 2001; Armenteras et al., 2006]. This contributed to the annual decreasing tendency of FPAR in the Amazonian and Southeast Asian rain forest regions. In addition, the primary forests were being replaced by secondary forests, and this new land cover and afforestation might have contributed to the FPAR recovery dynamic in Figure 9b. Those land cover change caused by anthropogenic activities may contribute to the non–significant correlation between climate changes (temperature and precipitation) and FPAR interannual variation in the Amazon and Southeast Asia (Figures 6 and 7).

3.5. Interactive Effects of Climate and Land Cover Changes on FPAR Anomalous Variation

[31] Theoretically, temperature and precipitation influence time series variation of FPAR by heat energy and plant water stress, and then directly affect plant phenology and growth [Huete et al., 2006]. For the interannual FPAR variability, a primary driving force in addition to global climate change is land cover change - deforestation, afforestation, fire disaster, urban expansion, etc., also cause FPAR annual variability and affects the correlation of FPAR annual trend and climate change.Zeng et al. [2008] found that a dramatic indicator of the 2002–2005 droughts could be correlated to fire occurrences over the southern Amazon, where the fire events in 2005 were more than twice as frequent as the average of the previous seven years over the whole basin. Large coverage areas of fire events hint that the land cover change, floods, and weather disturbances caused by ENSO events caused the anomalous FPAR variation. On the other hand, tremendous deforestation and afforestation in turn exacerbate the climate change. Deforestation releases the carbon stored in plants and soil and alters the physical properties of the surface [Bala et al., 2007]. Tropical rain forest ecosystems are the most productive ecosystems, and their changes are likely to have the greatest impact on climate change. In contrast, afforestation, especially in the tropics, could help mitigate global warming, since they are the most effective carbon sinks in the short term [Bala et al., 2007; Malhi et al., 2002]. In turn, climate change also influences vegetation growth. Therefore, climate and land cover changes have interactive effects on FPAR's anomalous variation.

3.6. Study Limitation

[32] A major limitation of this study is the impact of mixing pixels and spatial heterogeneity. The 300-m spatial resolution GlobCover land cover product were re-projected to an AVHRR GIMMS NDVI coordinate with 8-km spatial resolution, and the main vegetation type with the highest percentage in the area of 8 × 8 km2was taken as the land cover type for the resampled pixel. In addition, we used 1-degree GLDAS-2 climate data set to examine the climate driving effect of FPAR interannual trend from 1982 to 2006. Although we found the general characteristics and drivers of FPAR from 1982 to 2006 at the global scale and some typical region, the potential impact of the mixing pixels and heterogeneity of landscape should be studied further in the future. Land surface models all require a high-quality time series of global FPAR as part of the set of land-surface input parameters, the estimation of which is feasible only through remote sensing. More research on algorithm refinement and product validation are needed in the future, including theoretical improvements to better account for spatial heterogeneity, adjustments of parameters for a join AVHRR/MODIS and future Visible Infrared Imaging Radiometer Suite (VIIRS) time series data. A compilation of an extensive database of field measurements is also needed to further assess the accuracy of the products [Yang et al., 2006b]. We are planning more detailed FPAR product validation for several current available global FPAR products including MODIS, MISR, CYCLOPES, JRC, GLOBCARBON FPAR, and GIMMS NDVI SR FPAR in the near future.

4. Conclusions

[33] In this study, we presented a global FPAR product estimated from AVHRR GIMMS NDVI from 1982 to 2006, with the focus on the analysis of the long - term global FPAR interannual variation patterns and the driving forces primarily involving climate and land cover changes. Our main contributions are can be summarized in three aspects:

[34] First, the comparisons of GIMMS NDVI SR FPAR with SeaWiFS FPAR, MODIS FPAR for different land cover types indicated that GIMMS NDVI SR FPAR were smaller than but closer to MODIS FPAR, and larger than SeaWiFS FPAR. In addition, SeaWiFS FPAR seasonal dynamics showed a later onset of greenness than GIMMS NDVI SR and MODIS FPAR for forest land covers (as listed in Table 1) and cropland, which was consistent with the seasonal dynamics of flux GPP derived from North AmeriFlux web site over the period 2000–2006.

[35] Second, we found that the spatial patterns of global FPAR varied by seasons and vegetation types as it is expected, and they were independent of changes in AVHRR instruments. We observed both increasing and decreasing tendencies of annual variation in FPAR, except in the Australian desert, which had a year-round decreasing trend. There were more areas with decreasing FPAR in JJA than in other periods, and the interannual FPAR changes were mainly within 0.1. In MAM, SON, and DJF, almost all vegetation FPAR increased, at an average rate of 0.014. Forest FPAR values, especially evergreen needleleaf forest, had a decreasing tendency, with an average rate of 0.016 from 1982 to 2006 in JJA periods.

[36] Third, we examined the climate-driving effect on FPAR interannual change from 1982 to 2006 with the global temperature and precipitation from the GLDAS-2 data set, as well as the impact of drought associated with El Niño occurrences. We found that in the whole of the Southern Hemisphere, FPAR interannual trend almost had a significant negative correlation (P < 0.01 or P < 0.05) with temperature, and the relationship of FPAR interannual trend and precipitation had an inverse spatial distribution pattern compared to its relationship to temperature, especially for arid and semi-arid regions. In addition, a significant influence of extreme droughts such as those associated with the ENSO on FPAR variability was found, especially during the dry season (July–August–September) of 2005 in southwestern Amazon. Afforestation in Three North Shelterbelts in China and deforestation in the Amazon and Southeast Asia rain forest were used as an example to analyze the impact of land cover changes caused by anthropogenic activities, and their influences on the interannual variability of FPAR. We found that FPAR in Three North Shelterbelts showed an increasing tendency from 1982 to 2006 with an average annual growth of 0.01. In contrary, a significant reduction was observed in Southeast Asia, and large area of deforestation may contribute to the non-significant correlation between climate changes (temperature and precipitation) and FPAR interannual variation in the Amazon and Southeast Asia.

Acknowledgments

[37] This research was primarily funded by the National Basic Research Program of China (contract 2009CB723902), National High-tech R&D Program of China (contract 2012AA12A301), and Key Project of Digital Earth Science Platform CEODE (contract Y01002101A). The authors acknowledge the following data support: the AVHRR GIMMS NDVI from the Global Land Cover Facility (GLCF); the GlobCover products from the European Space Agency (ESA) and the ESA GlobCover Project led by the Medium Resolution Imaging Spectrometer Instrument (MERIS) France; Temperature and precipitation data acquired as part of the mission of NASA's Earth Science Division and archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC); WIST provides access to a complete data record of all MODIS FPAR products available from the LP DAAC; The SeaWiFS FPAR product from the SeaWiFS Project (code 970.2) and the Distributed Active Archive Center (code 902) at the Goddard Space Flight Center, Greenbelt, MD 20771, for the production and distribution of these data, respectively.