Effects of canopy photosynthesis saturation on the estimation of gross primary productivity from MODIS data in a tropical forest

Remote Sensing of Environment 121 (2012) 252–260

Contents lists available at SciVerse ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

Effects of canopy photosynthesis saturation on the estimation of gross primary productivity from MODIS data in a tropical forest P. Propastin a, c,⁎, A. Ibrom b, A. Knohl c, S. Erasmi a a b c

Department of Geography, Georg-August University Göttingen, Goldschmidtstr. 5, 37077, Göttingen, Germany Risø National Laboratory for Sustainable Energy, Technical University of Denmark (DTU), 4000 Roskilde, Denmark Bioclimatology, Büsgen Institute, Georg-August University Göttingen, Büsgenweg 2, 37077 Göttingen, Germany

a r t i c l e

i n f o

Article history: Received 25 January 2011 Received in revised form 3 February 2012 Accepted 4 February 2012 Available online 7 March 2012 Keywords: Gross primary production MODIS Light use efficiency FAPAR Tropical rainforest Sulawesi

a b s t r a c t The Moderate Resolution Imaging Spectroradiometer (MODIS) gross primary production (GPP) product (GPPMOD17A2) was evaluated against GPP from the eddy covariance flux measurements (GPPm) at a CO2 flux tower test site in a tropical rainforest in Sulawesi, Indonesia. The dynamics of 8-day GPPMOD17A2 averages generally showed similarities with observed values for the period 2004–2005 (r-value is 0.66, RMSE = 1.31 g C m − 2 d − 1). However, the results revealed some underestimation of GPP by the MOD17A2 product during phases of low photosynthetic production while it overestimated GPP during phases with clear sky conditions. Obviously, these seasonal differences are caused by too large seasonal amplitudes in GPPMOD17A2. The observed inconsistencies of the GPPMOD17A2with GPPm were traced to the inputs of the MODIS GPP algorithm, including fraction of absorbed photosynthetically active radiation (fAPAR) and light use efficiency (εg). This showed that underestimation of low values is caused by several uncertainties in the MODIS fAPAR input, whereas overestimation at high irradiance is caused by the MODIS light use efficiency approach which does not account for saturation of canopy photosynthesis under clear sky conditions. The performance of the MODIS GPP algorithm has been improved through the use of a site-validated fAPAR data set and a novel approach for εg adjustment which allows for saturation of gross photosynthesis at high irradiance. Our study revealed a weakness of a commonly used light use efficiency approach to estimate global GPP at the example of a moist tropical rain forest in Indonesia and demonstrated a potential need for MOD17 enhancement. © 2012 Elsevier Inc. All rights reserved.

1. Introduction Gross Primary Production (GPP) refers to an incoming carbon flux at the ecosystem level, which drives such ecosystem functions as respiration and growth. With respect to the contemporary global warming debate, the importance of GPP is recognized as one of the major processes controlling land-atmosphere carbon dioxide exchange, which can provide the capacity to compensate for anthropogenic emissions of carbon dioxide (Beer et al., 2010). In addition, terrestrial GPP is a very significant contributor to human welfare because it builds the basis for food, fiber, and wood production. Field estimates of carbon fluxes are based on CO2 flux measurements between ecosystems and the atmosphere commonly use the eddy covariance technique (Aubinet et al., 2000; Falge et al., 2002). In terms of spatial scale, the eddy covariance method provides representative GPP measurements over areas that range from a

⁎ Corresponding author at: Department of Geography, Georg-August University Göttingen, Goldschmidtstr. 5, 37077, Göttingen, Germany. Tel.: + 49 551 39 79 79. E-mail addresses: [email protected] (P. Propastin), [email protected] (A. Ibrom). 0034-4257/$ – see front matter © 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2012.02.005

few thousand square meters to several hectares, depending upon tower height, physical characteristics of canopy and atmospheric stratification. Scaling up data from flux tower measurements is an important challenge in understanding the terrestrial GPP across different spatial and temporal scales (Chiesi et al., 2005; Ruimy et al., 1999). One of the up-scaling approaches for regional analysis is integration of flux measurements with remotely sensed data (Running et al., 1999a, 2000; Hunt et al., 2004; Xiao et al., 2004a,b, 2005). Satellite-based models use the light use efficiency (LUE, ε) approach first described by Montheith (1977) to calculate GPP by linking the incoming radiation to GPP through an empirical biophysical conversion factor. In such models, a number of satellite-derived input parameters including leaf area index (LAI), fraction of absorbed photosynthetically active radiation (fAPAR), and information on land cover distribution are combined with meteorological data and derivates from the eddy covariance measurements. Since 2000, satellitebased GPP products have been produced continuously using data from the Moderate Resolution Imaging Spectroradiometer (MODIS) at a spatial resolution of 1 km. The MODIS GPP products provide global coverage at an 8-day temporal resolution and are intended for monitoring photosynthetic activity of vegetation (Heinsch et al., 2003; Running et al., 1999a,1999b). Global estimates of GPP from MODIS

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

are available for download at a NASA DAAC (Distributed Active Archive Center). Validation of MODIS GPP products is a fundamental step in the assessment of their effectiveness. The extensive validation of MODIS GPP products is in progress using ground truth data from different regions. However, validation is a challenging task because of scale differences between the spatial resolution of the MODIS products and plot-scale measurements on the ground (Turner et al., 2004; Running et al., 1999b; Heinsch et al., 2006). The assessment of MODIS GPP is generally based on the use of time series of GPP estimated from eddy covariance flux tower data (Turner et al., 2004; Turner et al., 2006a,b; Wu et al., 2009; Xiao et al., 2004a,b; Xiao et al., 2005). The reported results reveal a wide range of variability between the ground-based and MODIS-based GPP estimates. Since the MODIS algorithm incorporates a number of global biome-specific default parameters, which neglects the inter-regional and within-region variability of their values (Heinsch et al., 2006), specific causes of under- or over-prediction of GPP by the MODIS product can be traced to the MODIS algorithm inputs, including the climate input data, fAPAR, LAI and LUE. In case studies from various regions, the performance of the MODIS GPP algorithm could be improved by using modified input for meteorological data (Fensholt et al., 2006; Zhao et al., 2005, 2006), fAPAR (Xiao et al., 2004a,b; Xiao et al., 2005; Fensholt et al., 2006), or LUE (Turner et al., 2006a,b). With respect to the validity of MODIS GPP, MODIS fAPAR input to the algorithm has two main sources of uncertainty. The first source is the inconsistency of modeled fAPAR with the ground measurements. Rigorous validation of MODIS fAPAR is difficult and has been made in only a few cases, primarily in ecosystems with low fAPAR (Fensholt et al., 2006; Myneni et al., 2002; Weiss et al., 2007). In these validation studies, the most critical points were a poor correlation with the ground-based data and a problem with MODIS fAPAR outside the growing season. The second source of uncertainty is the contamination of MODIS fAPAR pixels by atmospheric noise leading to unreliable values. Although the temporal filling of unreliable fAPAR values greatly improves the accuracy of inputs to MODIS GPP algorithm, as discussed by Zhao et al. (2005) and Heinsch et al. (2003), the filled values are artificial and introduce uncertainties. Various issues associated with the effect of fAPAR on MODIS GPP have been addressed for test sites in hardwood forest (Xiao et al., 2004a,b; Turner et al., 2005), beech forest (Wang et al., 2005), grassland (Fensholt et al., 2006) and tundra (Stow et al., 2004). These studies found biome- and site-validated effects of MODIS fAPAR which resulted in under- or overestimation of the ground-based GPP by the MODIS algorithm. In the satellite-based LUE models, a potential value of LUE, εg, can be obtained from eddy covariance measurements (Hunt et al., 2004; Turner et al., 2003) and is usually considered as a biome-specific constant (Hill et al., 2004; Olofsson et al., 2007; Seaquest et al., 2003). Values of potential εg are summarized in several publications (Gower et al., 1999; Ruimy et al., 1999). The optimal value of εg is modified taking into account the constraints imposed by climatic limitations such as temperature, soil moisture or vapor pressure deficit. The MODIS GPP algorithm suggests simple linear ramp functions of climatic variables to calculate the scalars that attenuate the potential εg to produce the final εg used to predict GPP (Heinsch et al., 2003; Turner et al., 2006a,b; Xiao et al., 2004a,b). However, a number of studies have concluded that the light use efficiency rate is also dependent on incoming solar radiation and saturates on days with clear sky conditions and high amount of PAR (Turner et al., 2003; Lagergren et al., 2005; Ibrom et al., 2008). Tropical rainforests account for 34% of the global terrestrial GPP and have the highest GPP per unit area (Beer et al., 2010). Therefore, the ability to monitor GPP in tropical rainforests is of great interest in relation to understanding the current status of the global carbon cycle and climate forcing and to evaluate climate change mitigation measures. With respect to validation and the improvement of MODIS

253

GPP product, tropical rainforests represent regions where these tasks are particularly challenging because of small seasonal variability, high cloudiness throughout the year, and saturation effects of radiation at high LAI values. Nonetheless, most of the studies on MODIS GPP validation have been carried out in non-tropical biomes. For this reason, each new related study from a tropical rainforest is of great interest and findings from such studies can be valuable for improvement of the MODIS algorithm (Zhao et al., 2005). The study presented here follows a previous study, where the phenomenon of canopy photosynthesis saturation in the observed tropical rain forest was shown and an improved light-use efficiency approach was suggested (Ibrom et al., 2008). Here we modify the LUE approach by means of incorporating canopy photosynthesis saturation into model's εg inputs. We combine the modified LUE model with satellite-derived fAPAR data (Propastin & Erasmi, 2010) to estimate GPP in a tropical rain forest in Sulawesi (Indonesia). The results from the modified LUE approach are compared with the currently used approach for the calculation of MODIS GPP products. Systematic comparison with GPP derived from measured CO2 flux data showed in which way that the quality of the MODIS GPP product could be improved by using more realistic input data, namely light use efficiency and fAPAR. 2. Material and methods 2.1. Site-validated carbon dioxide flux data The test measurements for this study were carried out at a flux tower site located in the southern part of the Lore-Lindu National Park (01°39′ southern latitude and 120°10′ eastern longitude, 1412 m above sea level) (Ibrom et al., 2007; Ibrom et al., 2008) (Fig. 1). The Lore-Lindu National Park, covering an area of 221,505.300 ha, represents one of the largest remaining areas of natural tropical rainforest in Sulawesi. The terrain around the flux tower site is generally horizontal with a very small slope (b3°). Average daytime temperature is about 19.5 °C, while average annual precipitation is about 1700 mm. Natural forests in the study region are generally classified into major tree types based on altitudinal distribution with lowland forest (b1000 m), pre-montane forest (1000–1400 m), and montane forest (>1400 m). According to this classification, the flux tower site is located approximately at the border between

Fig. 1. The area of the Lore-Lindu National Park on the map of Sulawesi with the location of the eddy covariance flux tower.

254

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

pre-montane and montane forests. The most important families dominating rainforests in the Lore-Lindu National Park are Fagaceae, Moraceae, Sapotaceae, Lauraceae and Meliaceae (Culmsee & Pitopang, 2009; Hertel et al., 2009). The forest canopy structure was inventoried at 22 sampling plots with a size of 150 m2 located around the flux tower (Grote, 2006). The total area covered by the inventory was 7.3 ha. Results of the forest inventory showed that the canopy structure at the flux tower site is relatively homogeneous with a mean value of tree height of 22.5 m, the mean tree density of 550 trees per ha, tree diameter at breast height of 28 cm, and crown diameter of about 8.2 m. The mean value of LAI is about 6.1 (Grote, 2006). Eddy-covariance flux measurements were carried out during the period October 2003–December 2008 using equipment installed on a 70-m high tower. Briefly, wind velocities were recorded by a 3-D ultrasonic anemometer–thermometer USA-1 (Metek, Elmshorn Germany) and CO2 mixing ratio by an open-path CO2 and H2O gas analyser LI-7500 (Li-Cor, Lincoln, NE) linked to a field computer. Halfhourly values of net ecosystem exchange and GPP were calculated from the covariance of the fluctuations in vertical wind speed and CO2 mixing ratio measured at 10 Hz. Ibrom et al. (2007) and Ibrom et al. (2008) described the measurements and post-processing of the data used in this study in more detail. For the purpose of this study, 8-day average flux values were generated from the originally half-hourly flux data in order to match the temporal resolution of the MODIS products. In our analyses, we used 8-day data from two subsequent years 2004–2005 comprising drier and wetter seasons, which are typical for the site. The period of two years is long enough to ensure that all relevant weather conditions are included in the analysis. 2.2. Meteorological data Daily meteorological data were derived from a climate station situated near the Bariri village (01°39.51′S and 120°10.94′E). This station is located less than 10 km away from the flux tower site. Because the MODIS product is usually calculated with data from long-term meteorological stations for model input, we used data from a nearby climate station rather than the meteorological measurement from the flux tower site. The data set contains daily temperatures, air humidity, precipitation, global radiation, and cloudiness for the years from 2002 to 2005. Vapor pressure saturation deficit (VPD) was calculated from air temperature and relative humidity using the equation given by Alisov et al. (1956). From global radiation, incident PAR, defined as the domain of incoming solar radiation exploited by green vegetation for photosynthesis, was calculated with empirical functions described by Ross (1981). The amount of the absorbed photosynthetically active radiation (APAR) was estimated through multiplication of PAR with fAPAR. 2.3. Site-validated LAI and fAPAR data set We used a 16-day MODIS LAI data set covering the area of the Lore-Lindu National Park for the derivation of a site-validated fAPAR data set. The LAI data set was generated from 250-m MODIS NDVI data and ground data on canopy structure measured at 165 sampling sites across the study area. The sampling sites for ground measurements of canopy structure variables were randomly distributed over the whole area of the Lore-Lindu National park. A nested sampling strategy organized by plot and sub-plot was employed to the ground measurements of canopy structure. All plots had a size of about 40 ∗ 40 m and were divided into 12 sub-plots. The single measurements at sub-plots were averaged to provide a mean value for each plot. In situ estimations of LAI were performed using a WinScanopy Image Acquisition instrument developed by REGENT INSTRUMENTS (http://www.regentinstruments.com). For processing of hemispherical

photographs and retrieval of vegetation structure variables we used the Can Eye software (INRA, France, http://www.avignon.inra.fr/can_eye/). Gap fraction was estimated at 5° zenith angle intervals and used for further calculations. LAI and other canopy structure variables were calculated using routine procedures included in the Can Eye software following the methods described by Jonckheere et al. (2004) and Weiss et al. (2004). The LAI is corrected for non-random distribution of foliage elements based on the clumping index, which is calculated using the logarithmic gap averaging technique given by Lang and Xiang (1986). The spatially enhanced MODIS LAI data set is based on a physical radiative transfer model that is described in Propastin and Erasmi (2010). This geometric-optical model was based on a relationship between LAI, fraction of vegetation cover, and given patterns of surface reflectance, view-illumination conditions, and optical properties of vegetation. The model used a number of input parameters related to optical and structural characteristics of canopy such as the extinction coefficient (k), the foliage projection coefficient (G) and the clumping index (Ω), which were derived from hemispherical photographs in situ and measurements of crown architecture obtained from forest inventories. View-illumination conditions were simulated by a view angle geometry model incorporating the solar zenith angle and the sensor viewing angle. For a detailed description of the model see Propastin and Erasmi (2010). The conversion of LAI to fAPAR uses a simple Beer's Law approach (Jarvis & Leverenz, 1983), which enables computation of fAPAR as a function of LAI and the canopy light extinction coefficient (k):
 −kLAI fPAR ¼ 0:95 1−e :

ð1Þ

The value of the extinction coefficient k was set to 0.5 as estimated from PAR measurements in the study area by Propastin and Erasmi (2010). Analysis of hemispherical photographs from the forest sampling plots taken in the Lore Lindu National Park provided a leaf inclination value (the relationship between the horizontal semi-axis length and the vertical semi-axis length) of about 1.0 (Propastin & Erasmi, 2010) assuming spherical leaf angle distribution (Norman & Campbell, 1989). PAR albedo was assumed to have a value of 5% as estimated for an Amazonian tropical rainforest by Senna et al. (2005). Taking into account the uncertainty regarded to our assumption, we assessed the sensitivity of the GPP estimate (see Eqs. 2 and 3) on the value of the extinction coefficient. Corresponding to our calculations, a difference of 10% from the assigned value of k would change the final GPP result by 4.2% for a LAI value of 4.0, and by 3.2% for a LAI value of 6.1. Bearing in mind that the LAI at the test site is about 6.1, these errors can be tolerated in our further calculations. Considering this sensitivity of the final result to the value of k and reports from several other studies, assigning the value of 0.5 to the extinction coefficient for our test site was appropriate. The resulting 250-m fAPAR product was resampled to 1-km spatial resolution using the nearest-neighbor algorithm to facilitate comparison with MODIS 1-km products. In addition, 8-day fAPAR values were calculated from the original 16-day fAPAR using a simple linear interpolation procedure. 2.4. MODIS products The MODIS GPP (MOD17A2 V.4.5) product, GPPMOD17A2, is a global product on an 8-day basis (Heinsch et al., 2003). The GPPMOD17A2 is modeled using the light use efficiency (LUE) approach based on the assumption that rates of primary productivity are proportional to rates of solar radiation absorbed by vegetation (Eq. 2, Ruimy et al., 1994): GPP ¼ ε g APAR

ð2Þ

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

where APAR represents the amount of incident solar radiation (MJ m − 2) that is absorbed by vegetation per unit time (e.g. day, month or year), and εg is the LUE value for GPP. The value of εg is modified taking into account the constraints imposed by climatic limitations such as temperature, soil moisture and water vapor deficit. APAR is calculated as: APAR ¼ fAPAR⋅PAR

ð3Þ

where PAR is the incident photosynthetically active radiation (MJ m − 2) per unit time, and fAPAR is the fraction of PAR absorbed by the vegetation canopy. The MODIS GPP algorithm uses input data from MODIS fAPAR, MODIS land cover product (the UMD scheme in MOD12) and global meteorological simulations of incoming PAR, daily maximum and minimum temperature and daily minimum and maximum vapor pressure deficit provided by the NASA Global Modeling and Assimilation Office (GMAO). The GMAO meteorological data are based on a general circulation model that is continuously assimilating observations from space and ground stations. The spatial resolution of the GMAO data is 1 by 1.25°. The representativeness of the GMAO data for various regions and validation sites was examined in numerous studies. On the one hand, several studies reported that despite of its coarse resolution, the GMAO data set agrees well with ground-based measurements of climate data (among them Turner et al., 2005, 2006a; Zhao et al., 2006). On the other hand, meteorology data are the main contributor to the error in MODIS GPP estimates (Heinsch et al., 2006). With respect to representativeness of this data for our study area, we assume that the GMAO data should have comparable validity as in other regions and have comparable effects on accuracy of MOD17A2 GPP estimates as shown in previous studies (Heinsch et al., 2006; Zhao et al., 2006; Turner et al., 2006a). The MODIS landcover product in general agrees well with other data sources for the study area. However, it slightly overestimates forests compared to Landsat based maps and field surveys (Erasmi et al., 2007). The Bariri flux tower site is located within the class “evergreen broadleaf forest” of the MOD12 product. The original parameterisation of the optimum εg for representative vegetation in each biome type in MOD17 was based on an analysis of modeled global terrestrial GPP with a complex ecosystem model, BIOME-BGC (Running et al., 2000). Biome-specific values of εg were subject to sporadic adjustments as the potential LUE was measured at an increasing number of flux tower sites. For generating the MODIS GPP, 8-day estimates of fAPAR from the MOD15 product are used. The 1-km resolution MODIS fAPAR and LAI products (MOD15A1 and MOD15A2 respectively) are derived from a global-scale process model. The MOD15 algorithm is described by Knyazikhin et al. (1999) and Myneni et al. (2002). The MOD15 algorithm solves the inverse problem of retrieving LAI and fAPAR from MODIS channels 1–7 involving a number of biome-specific constants such as leaf angle distribution, canopy heterogeneity, and soil and wood optical properties (Knyazikhin et al., 1999). 2.5. MODIS post-processing For our analysis, 8-day time series of the 1-km GPPMOD17A2 and MOD15A1 fAPAR products covering the period from 2004 to 2005 including rainy and drier seasons in order to consider typical weather conditions at this site were used. Since the study area belongs to regions with high frequency of cloud cover, the original GPPMOD17A2 and MOD15A1 fAPAR time series show many contaminated pixels due to clouds and other noise. MODIS Quality Control (QC) fields were used to distinguish between pure and contaminated pixels. The Savitzky–Golay weighted-average filter (Chen et al., 2004; Erasmi et al., 2006) was applied to the MOD17A2 GPP time series in order to fill missing values and replace contaminated pixels. Note

255

that the Savitsky–Golay filtering procedure was also applied to improve unreliable pixel values in the site-validated fAPAR data set described above (Section 2.2). The MODIS GPPs were transformed from 8-day sums to 8-day means, and were converted from kg C m− 2 to g C m − 2. For our analyses, we used data from an area of 3 ∗ 3 1-km² MODIS (MOD17A2) cells centered on the flux tower at Bariri. Extracting a 3 ∗ 3 pixel area for comparison with site data allows avoiding potential errors in georectification of satellite data. For a detailed discussion of the issue of georectification when comparing site data with satellite products see Heinsch et al. (2006) and Cohen et al. (2003). The extracted values of GPPMOD17A2 and fAPAR were used in further analyses. 2.6. Evaluation of the MODIS algorithm and its input parameters In order to directly compare GPP measured from the eddy covariance flux at the flux tower site (GPPm) and GPPMOD17A2, GPPm were averaged to 8-day periods corresponding with the GPPMOD17A2 data. Similarly, fAPAR values (MOD15A1) were compared with fAPAR values derived from the site-validated LAI data set. Common accuracy statistics were used to quantify the consistency between the data sets: a Pearson correlation coefficient (r), Root Mean Square Error (RMSE), relative RMSE (RMSE(%)) and F-test. Before calculating a Pearsons r, the data were also tested for normality. To examine effects of fAPAR and εg inputs on MOD17 GPP algorithm outputs, we employed the MODIS GPP algorithm with different types of input data. To examine the effect of the different fAPAR input data on the GPP estimate, we compared simulations using either MOD15A1 fAPAR or the site-validated fAPAR data set (Propastin & Erasmi, 2010). The effect of the light use efficiency approach was tested by using the common MODIS approach using linear ramp functions (Heinsch et al., 2003), and the εg adjustment taking the saturation of canopy photosynthesis during periods with clear sky conditions into account (Ibrom et al., 2008). The tested models and the used input data are summarized in Table 1. GPP outputs from the 4 models (GPPA, GPPB, GPPC, and GPPD) were compared with respect to GPPm data. The outputs from the 4 models were also compared with the MOD17A2 GPP data. 2.7. Simulating regulation of LUE by climatic factors The εgmax value for evergreen broadleaved forest, which is used for computation of the current MODIS GPP products (Collection 4.5) is 1.159 g C MJ − 1 and is reduced according to unfavorable climate conditions indicated by extremes of minimum/maximum temperature and VPD. The εg adjusted using linear scaling functions of temperature and VPD represents a value of LUE that would be expected under clear sky conditions. However, observations at flux towers suggest that canopy photosynthesis saturates on clear sky days even

Table 1 Overview of models for GPP estimation that have been applied in this study. Input parameters Meteorological data

fAPAR

GPPMOD17A2

GMAO data

MOD15A1 fAPAR

GPPA

Bariri climate station

MOD15A1 fAPAR

GPPB

Bariri climate station

Site-validated fAPAR

GPPC

Bariri climate station

MOD15A1 fAPAR

GPPD

Bariri climate station

Site-validated fAPAR

εg adjusting function Linear ramp (MODIS algorithm) Linear ramp (MODIS algorithm) Linear ramp (MODIS algorithm) Saturating function (Eqs. 4–5) Saturating function (Eqs. 4–5)

256

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

at the daily time-scale (Ibrom et al., 2008; Turner et al., 2003). The saturation of canopy photosynthesis occurs when the amount of incident APAR achieves a certain value (600 μmol m − 2 s − 1 for the flux tower site in Bariri, see Ibrom et al., 2008). However, the saturated GPP has a substantial sensitivity to climatic conditions. The analysis by Ibrom et al. (2008) showed that the saturated GPP value was mostly affected by VPD. To simulate attenuation of LUE by climatic constraints taking into account the light saturation, we used the hyperbolic function described by Ibrom et al. (2008): ε gsat ¼

8 < εg max f ðVPDÞ

APAR ≤ APARsat

APARsat : εg max f ðVPDÞ APAR

APAR > APARsat

GPP estimation

Mean g C m− 2 d − 1

Min g C m− 2 d − 1

Max g C m− 2 d − 1

SD g C m− 2 d − 1

GPPm GPPMOD17A2 GPPA GPPB GPPC GPPD

7.56 7.43 7.09 7.61 7.21 7.51

5.25 4.67 4.95 5.56 5.01 5.36

8.94 9.51 9.36 9.05 9.21 9.00

0.85 1.36 1.19 0.87 1.08 0.92

ð4Þ

where εgmax is the maximum light use efficiency, f(VPD) is the function simulating the effect of air moisture on plant photosynthesis, and APARsat is the amount of APAR at which the photosynthesis rate saturates. The effect of moisture was modeled using a function of vapor pressure deficit of which definition and parameters are given by Ibrom et al. (2008):
 a− b f ðVPDÞ ¼ 1−e VPD2

Table 2 Descriptive statistics of tower estimates of GPP (GPPm) and GPP values computed from the different model definitions in Table 1.

ð5Þ

where a and b are empirical parameters. The value for the daily APARsat (5.66 MJ d − 1) was taken from Ibrom et al. (2008) as reported for the flux tower site in Bariri. 3. Results 3.1. Evaluation of GPPMOD17A2 The observed GPP time-series at the tower site shows a slightlypronounced seasonal variation and a more pronounced withinseason variation (Fig. 2, A). The GPPm ranges between a maximum

of 8.94 g C m − 2 d − 1 and a minimum of 5.25 g C m − 2 d − 1 around the mean value of 7.56 g C m − 2 d − 1 (see Table 2). For both years, the GPPm was higher during the dry season (May–October) and lower during the wet season (November–April) (compare Fig. 2, B). The temporal dynamics of GPPMOD17A2 agreed reasonably well with that from the eddy covariance flux-based measurements (Fig. 2, A). The GPPMOD17A2 showed larger GPP in the dry season and smaller GPP in the wet season, which is generally consistent with eddy-covariance measurements for the Bariri flux tower site. Nonetheless, some differences remain between the two estimates. The variability of GPPMOD17A2 is larger than that of GPPm; the standard deviation for GPPMOD17A2 was 1.36 g C m − 2 d − 1 compared to a SD of 0.85 g C m − 2 d − 1 for GPPm (Table 2). Similarly, the total range of GPPm is lower than that of GPPMOD17A2. The range of the GPPMOD17A2 was more than 62% of the average 8-day GPP (minimum GPP = 4.67 g C m − 2 d − 1, maximum GPP = 9.51 g C m− 2 d− 1, and the average GPP = 7.43 g C m − 2 d − 1), whereas the range of the GPPm was about 50%. GPPMOD17A2 estimates were generally higher in the wet season, but equal or lower in the dry season than the corresponding GPPm values. The evaluation statistics for the GPPMOD17A2 versus GPPm are provided in Table 3. The relationship between the GPPMOD17A2 and GPPm was found to be significant at the level p b 0.001 (RMSE = 1.31 g C m − 2 d − 1, RMSE(%) = 17.66), even though the correlation was not very high ( r = 0.663). Fig. 3 (A) shows a considerable scattering of the data points around the 1:1 line. The slope of GPPMOD17A2 versus GPPm was significantly larger than the 1:1 line suggesting that the MODIS algorithm somewhat underestimates low GPP values and overestimates high GPP values. However, the over-/underestimation of extreme values by the GPPMOD17A2 had only a small impact on the predicted mean value. The 8-day average GPP is predicted accurately: 7.43 g C m − 2 d − 1 (GPPMOD17A2) versus 7.56 g C m − 2 d − 1 for the GPPm. The inaccurate prediction of the range GPP values by GPPMOD17A2 is also reflected in the histograms (Fig. 3, B). The histogram produced from GPPm shows a bell-shaped distribution, whereas the GPPMOD17A2 distribution is more flat, i.e. uniform. 3.2. Analysis of variables determining MODIS GPP 3.2.1. FAPAR inputs to the MODIS GPP algorithm The MOD15A1 fAPAR product was evaluated against the sitevalidated fAPAR data set (Fig. 4, Table 3). The MOD15A2 fAPAR was generally very close to the site-validated fAPAR values (r = 0.86, RMSE = 0.14). However, the comparison also exposed that the Table 3 Statistics of GPPMOD17A2 and MOD15A1 fAPAR evaluation. Model

Fig. 2. (A) Dynamics of observed 8-day GPPm and the GPPMOD17A2 during 2004–2005 at the eddy flux tower site of the Lore-Lindu National Park, Central Sulawesi, Indonesia. (B) Dynamics of vapor pressure saturation deficit (VPD) and global radiation at the tower site during 2004–2005.

Independent

Dependent

GPPm Site-validated fAPAR

GPPMOD17A2 MOD15A1 fAPAR

Pearsons correlation, r

RMSE g C m− 2 d − 1

RMSE (%)

p-value

0.663 0.860

1.31 0.14

17.66 16.82

7.374E-06 1.102E-10

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

257

Table 4 Statistics of the performance of the 8-day GPP LUE models with different input data as compared to the observed GPP from the eddy-covariance measurements (see Table 1 for combinations of input data). Model

GPPA GPPB GPPC GPPD

Fig. 3. Comparison between 8-days observed GPP and GPPMOD17A2 during 2004: A) scatterplot, and B) histogram.

MOD15A1 product considerably underestimates fAPAR in the low value range (Fig. 4B). To investigate the sensitivity of GPPMOD17A2 to the fAPAR input, we ran the MOD17 algorithm using both MOD15A1 fAPAR and the site-validated fAPAR. The correlations between GPP estimates using models A and B with GPPM were highly significant and estimated the GPPm well (Tables 2 and 4), when considering

Fig. 4. Comparison between site-validated fAPAR and MODIS fAPAR at the flux tower site in Bariri: A) time series, and B) scatterplot of the 8-day values.

Accuracy statistics Pearsons correlation, r

RMSE g C m− 2 d − 1

RMSE (%)

p-value

0.649 0.688 0.671 0.716

1.25 0.94 1.04 0.76

16.48 12.31 13.21 10.08

0.0008 0.0001 7.62E-05 2.15E-07

values of RMSE and the amount of the explained variance. Using the site-validated fAPAR data set improved the GPP estimation considerably more than using MODIS fAPAR. As shown in scatter plots (Fig. 5), the highest discrepancies between GPPA and GPPB were apparent for lower GPPm values (b7.5 g C m − 2 d − 1). In this range, GPPA showed a general tendency to underestimate GPPm. Distributions of the GPPA and GPPB were much closer to that of GPPm than the distribution from the GPPMOD17A2 (Figs. 3 B and 5B). GPPB showed a bell-shaped distribution around a mean value of 7.61 g C m − 2 d− 1 and a standard deviation of 0.87 g C m − 2 d − 1. These statistics were almost identical with the statistics of GPPm (mean GPP = 7.54 g C m− 2 d − 1, and st.dev. GPP = 0.85 g C m− 2 d − 1). The distribution of GPPA revealed higher frequency in the lower classes at the expense of the higher classes. Because of this shift, the mean value of the GPPA was lower in comparison to GPPm (7.09 g C m − 2 d − 1), whereas the standard deviation was essentially higher, 1.19 g C m − 2 d − 1. 3.2.2. LUE inputs to GPPMOD17A2 Time series of the 8-day εg- and εgsat-values calculated using the common MOD17 approach and the alternative procedure are presented in Fig. 6 (A). The value of LUE is not a constant entity but showed considerable variation caused by variability of climate parameters. However, the effects of VPD and temperature, which are the two adjusting parameters in the simple linear ramp function in

Fig. 5. (A) Scatterplots between 8-day GPP modeled by Model A (quadrates) and B (circles) and the flux tower GPP. (B) Histograms produced from Model A and Model B.

258

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

Fig. 6. (A) Two-year trajectory of 8-day GPPm, εg, and εgsat at the Bariri flux tower site calculated using the sample linear ramp scaling procedure (dots) and the alternative non-linear procedure adjusted for overcast conditions. (B) Trajectory of 8-day APAR and cloudiness for similar period.

Fig. 7. Two-year trajectories of GPP estimated by Models A, B, C, and D in comparison with measured GPPm.

4. Discussion MOD17 algorithm, are relatively small compared with the saturating effect of incoming PAR (compare Fig. 6 B). The value of εgsat ranged from 0.70 g C MJ − 1 during dry seasons to 1.18 g C MJ − 1 during wet seasons with a mean value of 1.04 g C MJ − 1 and standard deviation of 0.12 g C MJ − 1. The variability of εg was considerably lower ranging from 0.85 g C MJ − 1 to 1.18 g C MJ − 1 with a standard deviation of 0.07 g C MJ − 1. Both εg and εgsat were sensitive to incoming solar radiation showing highest values during overcast periods and lower values during periods with clear sky conditions (compare the profile of εgsat in Fig. 6A with profiles of APAR and cloudiness in Fig. 6B). Nonetheless, the effect of high APAR on εg was not as large as for εgsat. The decrease of εg during periods with clear sky conditions was only driven by Tmax and VPD, masking the effect of saturation through high amounts of APAR. The impact of LUE saturation on GPP modeling was examined by calculation of MOD17 with εgsat (GPPC and GPPD). The results suggest that the outputs from MOD17 algorithm are sensitive to GPP saturation. The accuracy of MOD17 GPP prediction was improved through the incorporation of εgsat into the algorithm. Using models GPPC and GPPD increased values of r and decreased RMSE in comparison with their corresponding variants, GPPA and GPPB, respectively (Table 4). Values of RMSE decreased by about 1 g C m − 2 d − 1 for the models with MOD15 fAPAR (GPPA and GPPC), and by more than 1.5 g C m − 2 d − 1 for the models with the site-validated fAPAR (GPPB and GPPD). Also the descriptive statistics of GPPB and GPPD were much closer to those of the tower GPP as shown in Table 2. The F-test revealed statistically significant (p b 0.05) differences for the outputs from both GPPA and GPPC, and GPPB and GPPD, respectively. The impact of LUE on MODIS GPP can also be seen in time series of 8-day GPP (Fig. 7). GPPD, which has LUE regulated by APAR, captured the ground-based GPP during the periods with clear sky conditions much better than GPPB. The improvement of GPPD can be seen clearly during the periods from 193 to 321 DOY, from 17 to 81 DOY, and from 209 to 273 DOY (Fig. 7). Values of GPPB are mostly above the measured GPP, whereas GPPD estimates are much closer to GPPm.

The seasonal dynamics of in situ 8-day GPP and fAPAR were generally captured by MOD17A2 GPP, however, the accuracy was relatively not high (r-value of 0.66, RMSE(%) = 17.66). Similar studies from seasonally tropical evergreen forest sites (Xiao et al., 2005), temperate deciduous broadleaf forest (Xiao et al., 2004a), and temperate mixed forest (Wu et al., 2009) reported much closer associations between flux tower GPP and MOD17A2 GPP. However, recent studies showed that forest areas with high biomass and LAI such as humid tropical rainforests are generally characterized by a weaker sensitivity of the canopy signal in remote sensing data to LAI (Foody et al., 2001, 2003; Huete et al., 2008). Taking into account the challenges of MODIS GPP validation in tropical rainforest such as small seasonal and inter-seasonal variability of GPP, high biomass amount causing rapid saturation of the canopy radiation signal, and relatively frequent cloudiness, the standard MODIS 8-day GPP is suitable to some extent. However, taking into account the contribution of tropical evergreen forests to global primary productivity it might lead to considerable mismatch in global environmental modeling studies. The differences in the MOD17A2 GPP estimates and the flux tower GPP are more pronounced during the wet season. Here, the 8-day values of the MOD17A2 GPP product were too low for the test site located in the humid forest biome. These results correspond well with the conclusions made by Turner et al. (2006a,b) and Heinsch et al. (2006) who found a tendency of the MODIS products to underestimate GPP at high productivity sites. Our results also showed that the MOD17 algorithm tends to overestimate GPP values under clear sky conditions during dry seasons. As a consequence, the MOD17A2 GPP product is characterized in part by larger seasonal amplitude of GPP compared to the eddy covariance-flux data. The study examined two reasons for the mismatch between GPPMOD17A2 and GPPm. For this purpose, we examined the sensitivity of the MOD17 algorithm to changes in input data with respect to their effects on the accuracy of the MODIS GPP prediction. The effect of fAPAR in the MODIS algorithm was tested by using a site-validated fAPAR data set (in model B) as an alternative to the MOD15A1

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

fAPAR (in model A). The difference between models A and B reflects the influence of fAPAR on the final GPP result. The prediction accuracy by model B increased significantly (A: R 2 = 0.47, RMSE = 12.3%; B: R 2 = 0.42, RMSE = 16.5%). A comparison of the MODIS and the site-validated fAPAR time series enabled us to trace a general difference between these data sets. There was a tendency in the MOD15A1 fAPAR product towards underestimation of the sitevalidated fAPAR for the wet season when fAPAR has lower values. This difference could not be explained by external factors such as cloud contamination because both data sets had been processed using the similar procedure for temporal filling of unreliable fAPAR. Estimates of fAPAR for the tropical moist forest are very rare in literature. This is the reason why the validity of MODIS fAPAR product in this biome is assessed only at a very small number of sites in the Amazonian region. Some MODIS fAPAR validation for Amazon tropical forest is given by Senna et al. (2005). Senna et al. (2005) indicated a general reliability of the MODIS fAPAR product for the tropical rain forest but they also detected a slight underestimation of groundbased fAPAR by the MODIS algorithm. To our current knowledge, there is no study on validation of MOD15A1 fAPAR in tropical moist forest from South-East Asia to be compared with our findings. However, the results of this study agree with the findings by Propastin and Erasmi (2010) who reported that MODIS underestimates LAI for the seasonal minimum values while the LAI estimations for the annual peaks demonstrated good agreement with the ground-based data. With respect to the impact of fAPAR input to the MODIS algorithm, taking into account that the MOD15A1 fAPAR seasonal peaks were close to the site-validated values, the only critical point of information derived from MODIS fAPAR could be its effect on low values of GPP which occur in wet seasons. This was demonstrated here by the better fit between measured and simulated GPP using a model based on site validated fAPAR (Model B) as compared to one that uses MOD15A1 fAPAR (Model A). The results clearly showed that the radiation absorption in the canopy strongly depends on the amount of diffuse radiation in the total incoming solar radiation (Fig. 6B). The value of APAR was much lower during the overcast periods that are characterized by higher amount of diffuse radiation and vice versa. These observations correspond to previous reports coming from the Amazon basin rainforest studies (Myneni et al., 2007). The seasonal increase in solar radiation of moist tropical rainforests during the dry season serves as a proximate signal for increasing leaf production (Myneni et al., 2007). The increased leaf production enforces the absorption of radiation by vegetation and causes an increase of FAPAR, which is reflected in higher productivity. The general decrease of the APAR in the wet season is attributed to (i) the much smaller amount of incident photosynthetically active radiation (radiation limitation for APAR) and (ii) the effect on phenology, i.e. reduced LAI and consequently FAPAR, in tropical rainforest (Goulden et al., 2004). The effect of saturation in GPP under high incident PAR as suggested by Ibrom et al. (2008) can be seen by comparing models GPPA versus GPPC or models GPPB versus GPPD (Fig. 7). The results demonstrate that the MOD17 algorithm does not account for saturation of gross primary photosynthesis at the 8-day time-step. Both models using the adjusted LUE approach (GPPC and GPPD) provided more accurate prediction of GPP than the models using the common linear approach (GPPA and GPPB). The analyses revealed a general tendency of MOD17 to overestimate GPP values for 8-day periods with clear sky conditions due to disregarding saturation of εg. Previous studies reported that saturation of canopy photosynthesis might occur on clear sky days at the half-hourly and daily timescale (Turner et al., 2003; Turner et al., 2006a; Lagergren et al., 2005; Ibrom et al., 2008). At the short-time scale, neglecting saturation can result in overestimation of GPP at high values of APAR and to underestimation at low APAR, if εg was fitted to a flux data set

259

comprising weather conditions with both high and low APAR. To capture saturation of canopy photosynthesis on clear sky days in the satellite-based GPP modeling, Turner et al. (2006b) developed an index of cloudiness that adjusted the light use efficiency depending on the degree of cloudiness. The approach for simulation of εg saturation used in this study seems to be simpler and more direct than the approach presented by Turner et al. (2006b) because it incorporates PAR as input variable into εgsat calculations directly, whereas the approach by Turner et al. (2006b) needs several parameters and uses cloudiness as the main indicator for εg saturation. Vapor pressure deficit is another key limiting factor for GPP at clear days, when canopy photosynthesis saturates (Ibrom et al., 2008). Being directly proportional to incident radiation (Fig. 2B), VPD does not constrain the canopy photosynthesis during the periods with overcast conditions, when global radiation is low. Synchronization of VPD and radiation dynamics has strict consequences for the modeling and leads to overestimation of GPP by the MODIS algorithm, since the MODIS algorithm accounts only for VPD and neglects the photosynthesis saturation at clear sky conditions. Due to the strong coincidence between low cloudiness, high VPD and high beam fraction the empirical data do not reveal the cause of the reduced light use efficiency under clear sky conditions. However they include the possibly combined effects of radiative regime, i.e. diffuse / direct radiation, and leaf physiological responses to VPD, i.e. stomatal limitation of photosynthesis. The overall effect of these processes on light use efficiency is subsumed in the model under VPD effect. In order to further improve estimates one would need to employ biophysical canopy models (e.g. Ibrom et al., 2006). Finally, the combined effect of both, site adapted fAPAR and saturating GPP could be seen in the difference between models GPPD and GPPA. The fAPAR- and εgsat-improved model (GPPD) predicted much better than the original version (GPPA): RMSE = 0.76 g C m − 2 d − 1 versus 1.25 g C m − 2 d − 1. This difference demonstrates a potential for improvement of the MOD17 GPP algorithm. Even though Collection 4.5 of MOD17 had already been improved and considered to be mature for a variety of applications (Zhao et al., 2005), GPP validation studies around the world provide a very important information for the further improvement of MOD17 GPP and lead to refinements of global GPP modeling. The results of the study might be used for potential methodological improvements of the global MODIS GPP modeling in this biome. Acknowledgments This study is part of the German-Indonesian collaborative research project STORMA (‘Stability of Rain Forest Margins in Indonesia’, subprojects B1 and D6) and ELUC-PAK that has been funded by the German Research Foundation (DFG). The financial support is gratefully acknowledged. We thank three anonymous reviewers for their thorough reviews and well conceived recommendations. References Alisov, B. P., Drosdow, O. A., & Rubinstein, E. S. (1956). Lehrbuch der Klimatologie. Berlin: VEB Deutscher Verlag der Wissenschaften 536 pp. Aubinet, M., Grelle, A., Ibrom, A., Rannik, U., Moncrieff, J., Foken, T., et al. (2000). Estimates of the annual carbon and water exchange of Eropean forests: The EUROFLUX methodology. Advances in Ecological Research, 30, 113–175. Beer, C., Reichstein, M., Tomelleri, E., Ciais, P., Jung, M., Carvalhais, N., et al. (2010). Terrestrial gross carbon dioxide uptake: Global distribution and covariation with climate. Science, 329, 834–837. 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, 332–344. Chiesi, M., Maselli, F., Bindi, M., Fibbi, L., Cherubini, P., Arlotta, E., et al. (2005). Modelling carbon budget of Mediterranean forests using round and remote sensing measurements. Agricultural and Forest Meteorology, 135, 22–34. Cohen, W. B., Maiersperger, T. K., Yang, Z., Gower, S. T., Turner, D. T., Ritts, W. D., et al. (2003). Comparisons of land cover and LAI estimates derived from ETM +

260

P. Propastin et al. / Remote Sensing of Environment 121 (2012) 252–260

and MODIS for four sites in North America: A quality assessment of 2000/2001 provisional MODIS products. Remote Sensing of Environment, 88, 233–255. Culmsee, H., & Pitopang, R. (2009). Tree diversity in sub-montane and lower montane primary rain forest in Central Sulawesi. Blumea, 54, 21–35. Erasmi, S., Bothe, M., & Petta, R. A. (2006). Enhanced filtering of MODIS time series data for the analysis of desertification processes in northern Brazil. The international Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 34, Part XXX. Erasmi, S., Kappas, M., Twele, A., & Ardiansyah, M. (2007). From global to regional scale: Remote Sensing based concepts and methods for mapping land-cover and land-cover change in tropical regions. In T. Tscharntke (Ed.), Stability of tropical rainforest margins: Linking ecological, economic and social constraints (pp. 437–462). Heidelberg: Springer. Fensholt, R., Sandholt, I., Rasmussen, M. S., Stisen, S., & Diouf, A. (2006). Evaluation of satellite based primary production modelling in the semi-arid Sahel. Remote Sensing of Environment, 105, 173–188. Falge, E., Tenhunen, J., Baldocchi, D., Aubinet, M., Bakwin, P., Berbigier, P., et al. (2002). Phase and amplitude of ecosystem carbon release and uptake potentials as derived from FLUXNET measurements. Agricultural and Forest Meteorology, 113, 75–95. Foody, G. M., Boyd, D. S., & Cutler, M. E. (2003). Predictive relations of tropical forest biomass from Landsat TM data and their transferability between regions. Remote Sensing of Environment, 85, 463–474. Foody, G. M., Cutler, M. E., Mcmorrow, J., Pelz, D., Tangki, H., Boyd, D. S., et al. (2001). Mapping the biomass of Bornean tropical rain forest from remotely sensed data. Global Ecology and Biogeography, 10, 379–387. Gower, S. T., Kucharik, C. J., & Norman, J. M. (1999). Direct and indirect estimation of leaf area index, FAPAR, and net primary production of terrestrial ecosystems. Remote Sensing of Environment, 70, 29–51. Grote, S. (2006). Struktur eines submontanen tropischen natürlichen Regenwaldes bei Bariri auf der indonesischen Insel Sulawesi. Masterarbeit, angefertigt im wissenschaftlichen Studiengang Forstwissenschaft an der Georg-August-Universität Göttingen. Goulden, M. L., Miller, S. D., da Rocha, H. R., Menton, M., de Freitas, H. C., Silva Figueira, A. M. E., et al. (2004). Diel and seasonal patterns of tropical forest CO2-exchange. Ecological Applications, 14, 542–554. Heinsch, F. A., Reeves, M., & Bowker, C. F. (2003). User's quide, GPP and NPP (MOD 17A2/A3) products, NASA MODIS Land Algorithm. http://www.forestry.umt.edu/ntsg/ Heinsch, F. A., Zhao, M., Running, S. W., Kimball, J. S., Nemani, R. R., Davis, K. J., et al. (2006). Evaluation of remote sensing based terrestrial productivity from MODIS using regional tower eddy flux network observations. IEEE Transactions on Geoscience and Remote Sensing, 44, 1908–1925. Hertel, D., Moser, G., Culmsee, H., Erasmi, S., Horna, V., Schuldt, B., et al. (2009). Belowand aboveground biomass and net primary production in a paleotropical natural rainforest (Sulawesi, Indonesia) as compared to neotropical forests. Forest Ecology and Management, 258, 1904–1912. Hill, M. J., Donald, G. E., Hyder, M. W., & Smith, R. C. G. (2004). Estimation of pasture growth rate in the south west of Western Australia from AVHRR NDVI and climate data. Remote Sensing of Environment, 93, 528–545. Huete, A. R., Restrepo-Coupe, N., Ratana, P., Didan, K., Saleska, S. R., Ichii, K., et al. (2008). Multiple site tower flux and remote sensing comparisons of tropical forest dynamics in Monsoon Asia. Agricultural and Forest Meteorology, 148, 748–760. Hunt, E. R., Kelly, R. D., Smith, W. K., Fahnestock, J. T., Welker, J. M., & Reiners, W. A. (2004). Estimation of carbon sequestration by combining remote sensing and net ecosystem exchange data for northern mixed-grass prairie and sagebrush-steppe ecosystems. Environmental Management, 33, 432–441. Ibrom, A., Jarvis, P. G., Clement, R. B., Morgenstern, K., Oltchev, A., Medlyn, B., et al. (2006). A comparative analysis of simulated and observed photosynthetic CO2 uptake in two coniferous forest canopies. Tree Physiology, 26(7), 845–864. Ibrom, A., Oltchev, A., June, T., Ross, T., Kreilein, H., Falk, U., et al. (2007). Effects of land use change on matter and energy exchange between a tropical rain forest margin and the atmosphere. In T. Tscharntke, C. Leuschner, M. Zeller, E. Guhardja, & A. Bidin (Eds.), The stability of tropical rainforest margins: Linking ecological, economic and social constraints (pp. 463–492). Springer. Ibrom, A., Oltchev, A., June, T., Kreilein, H., Rakkibu, G., Ross, T., et al. (2008). Variation in photosynthetic light-use efficiency in a mountainous tropical rainforest in Indonesia. Tree Physiology, 28, 499–508. Jarvis, P. G., & Leverenz, J. W. (1983). Productivity of temperate deciduous and evergreen forests. In O. L. Lange, P. S. Nobel, C. B. Osmond, & H. Zigler (Eds.), Ecosystem Processes: Mineral Cycling, Productivity and Man's Influence. Physiological Plant Ecology, New Series, vol. 12D. (pp. 233–280) New York: Springer-Verlag. Jonckheere, I., Fleck, S., Nackaerts, K., Muysa, B., Coppin, P., Weiss, M., et al. (2004). Review of methods for in situ leaf area index determination. Part I. Theories, sensors and hemispherical photography. Agricultural and Forest Meteorology, 121, 19–35. Knyazikhin, Y., Glassy, J., Privette, J. L., Tian, P., Lotsch, A., & Zhang, Y. (1999). MODIS Leaf Area Index (LAI) and Fraction of Photosynthetically Active Radiation Absorbed by Vegetation (FAPAR) Product (MOD15) Algorithm. Theoretical basis document. Available on-line at: http://eospso.gsfs.nasa.gov/atbd/modistables.html Lagergren, F., Eklundh, L., Grelle, A., Lundblad, M., Mölder, M., Lankreijer, H., et al. (2005). Net primary production and light use efficiency in a mixed coniferous forest in Sweden. Plant, Cell & Environment, 28, 412–423. Lang, A. R., & Xiang, Y. (1986). Estimation of leaf area index from transmission of direct sunlight in discontinuous canopies. Agricultural and Forest Meteorology, 35, 229–243. Montheith, J. L. (1977). Climate and the efficiency of crop production in Britain. Philosophical Transactions of the Royal Society of London, B, 281, 277–294.

Myneni, R., Hoffman, R., Knyazikhin, Y., Privette, J., Glassy, J., & Tian, H. (2002). Global products of vegetation leaf area and fraction absorbed PAR from one year of MODIS data. Remote Sensing of Environment, 83, 214–231. Myneni, R. B., Yang, W., Nemani, R. R., Heute, A. R., Dickinson, R. E., Knyazikhin, Y., et al. (2007). Large seasonal swings in leaf area of Amazon rainforests. PNAS, 104, 4820–4823. Norman, J. M., & Campbell, S. G. (1989). Canopy structure. In R. Pearcy, J. R. Ehleringer, H. A. Mooney, & P. W. Rundel (Eds.), Plant physiological ecology (pp. 301–325). London, Chapman and Hall: Field methods and instrumentation. Olofsson, P., Eklundh, L., Lagergren, F., Jönsson, P., & Lindbroth, A. (2007). Estimating net primary production for Scandinavian forests using data from Terra/MODIS. Advances in Space Research, 39, 125–130. Propastin, P., & Erasmi, S. (2010). A physically based approach to model LAI from MODIS 250 m data in a tropical region. International Journal of Applied Earth Observation and Geoinformation, 12, 47–59. Ross, J. (1981). The radiation regime and architecture of plant stands. The Hague: Dr. W. Junk Publishers 391 p. Ruimy, A., Dedieu, G., & Saugier, B. (1994). Methodology for estimation of terrestrial net primary production from remotely sensed data. Journal of Geophysical Research, 99, 5263–5283. Ruimy, A., Kergoat, L., & Bondeau, A. (1999). Comparing global models of terrestrial net primary productivity (NPP): Analysis of differences in light absorption and light use efficiency. Global Change Biology, 5, 56–64. Running, S. W., Baldocchi, D. D., Turner, D. P., Gower, S. T., Bakwin, P. S., & Hibbard, K. A. (1999). A global terrestrial monitoring network integrating tower fluxes, flask sampling, ecosystem modelling and EOS satellite data. Remote Sensing of Environment, 70, 108–127. Running, S. W., Nemani, J. M., Glassy, J. M., & Thornton, P. E. (1999b). MODIS daily photosynthesis and annual net primary production (NPP) product (MOD17). Algorithm theoretical basis, document V.3.0. Running, S. W., Thornton, P. E., & Nemani, R. (2000). Global terrestrial gross and net primary productivity from the earth observing system. In O. E. Sala, R. B. Jackson, H. A. Mooney, & R. W. Howarth (Eds.), Methods in ecosystem science (pp. 44–57). New York: Springer-Verlag. Seaquest, J. W., Olsson, L., & Ardo, J. (2003). A remote sensing-based primary production model for grassland biomes. Ecological Modelling, 169, 131–155. Senna, M. C. A., Costa, M. H., & Shimabukuro, Y. E. (2005). Fraction of photosynthetically active radiation absorbed by Amazon tropical forest: A comparison of field measurements, modelling, and remote sensing. Journal of Geophysical Research, 110, G01008, doi:10.1029/2004JG000005. Stow, D. A., Hope, A., McGuire, D., Verbyla, D., Gamon, J. A., & Huemmrich, R. (2004). Remote sensing of vegetation and land cover change in arctic tundra ecosystems. Remote Sensing of Environment, 89, 281–306. Turner, D. P., Urbanski, S., Bremer, D., Wofsy, S. C., Meyers, T., Gower, S., et al. (2003). A cross-biome comparison of daily light use efficiency for gross primary production. Global Change Biology, 9, 383–395. Turner, D. P., Ollinger, S., Smith, M. L., Krankina, O., & Gregory, M. (2004). Scaling net primary production to a MODIS footprint in support of Earth Observing System Product Validation. International Journal of Remote Sensing, 25, 1961–1979. Turner, D. P., Ritts, W. D., Cohen, W. B., Maeirsperger, T., & Gower, S. T. (2005). Sitelevel evaluating of satellite-based global terrestrial gross primary production and net primary production monitoring. Global Change Biology, 11, 666–684. Turner, D. P., Ritts, W. D., Cohen, W. B., Gower, S. T., Running, S. W., Zhao, M., et al. (2006a). Evaluation of MODIS GPP and NPP products across multiple biomes. Remote Sensing of Environment, 102, 282–292. Turner, D. P., Ritts, W. D., Styles, J. M., Yang, Z., Cohen, W. B., Law, B. E., et al. (2006b). A diagnostic carbon flux model to monitor the effects of disturbance and interannual variation in climate on regional NEP. Tellus, 58 B, 476–490. Wang, Q., Tenhunen, J., Dinh, N. Q., Reichstein, M., Otieno, D., & Granier, A. (2005). Evaluation of seasonal variation of MODIS derived leaf area index in two European deciduous broadleaf forest sites. Remote Sensing of Environment, 96, 475–484. Weiss, M., Baret, F., Smith, G. J., Jonckheere, I., & Coppin, P. (2004). Review of methods for in situ leaf area index determination. Part II. Estimation of LAI, errors and sampling. Agricultural and Forest Meteorology, 121, 37–53. Weiss, M., Baret, F., Garriques, S., & Lacaze, R. (2007). LAI and fAPAR CYCLOPES global products derived from VEGETATION. Part 2: Validation and comparison with MODIS collection 4 products. Remote Sensing of Environment, 110, 317–331. Wu, J. B., Xiao, X. M., Guan, D. X., Shi, T. T., Jin, C. J., & Han, S. J. (2009). Estimation of the gross primary production of an old-growth temperate mixed forest using eddy covariance and remote sensing. International Journal of Remote Sensing, 30, 463–479. Xiao, X., Hollinger, D., Aber, J., Goltz, M., Davidson, E. A., Zhang, Q., et al. (2004a). Satellitebased modelling of gross primary production in an evergreen needleaf forest. Remote Sensing of Environment, 89, 519–534. Xiao, X., Zhang, Q., Braswell, B., Urbanski, S., Boles, S., Wolfsy, S., et al. (2004b). Modelling gross primary production of temperate deciduous broadleaf forest using satellite images and climate data. Remote Sensing of Environment, 91, 256–270. Xiao, X., Zhang, Q., Saleska, S., Hutyra, L., De Camargo, P., Wolfsy, S., et al. (2005). Satellite-based modelling of gross primary production in a seasonally moist tropical evergreen forest. Remote Sensing of Environment, 94, 105–122. Zhao, M., Heinsch, F. A., Nemani, R. R., & Running, S. W. (2005). Improvements of the MODIS terrestrial gross and net primary production global data set. Remote Sensing of Environment, 95, 164–176. Zhao, M., Running, S. W., & Nemani, R. R. (2006). Sensitivity of Moderate Resolution Spectroradiometer (MODIS) terrestrial primary production to the accuracy of meteorological reanalyses. Journal of Geophysical Research, 111, doi:10.1029/2004JG000004.