Generation and evaluation of gross primary productivity using Landsat data through blending with MODIS data

International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

Contents lists available at ScienceDirect

International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag

Generation and evaluation of gross primary productivity using Landsat data through blending with MODIS data Devendra Singh Department of Science and Technology, Technology Bhawan, New Mehrauli Road, New Delhi 110016, India

a r t i c l e

i n f o

Article history: Received 16 March 2010 Accepted 10 June 2010 Keywords: Landsat Modis Blending Gross primary productivity Chlorophyll index Evapotranspiration

a b s t r a c t Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) has been used for the blending of Landsat and MODIS data. Specifically, the 30 m Landsat-7 ETM+ (Enhanced Thematic Mapper plus) surface reflectance was predicted for a period of 10 years (2000–2009) as the product of observed ETM+ and MODIS surface reflectance (MOD09A1) on the predicted and observed ETM+ dates. A pixel based analysis for six observed ETM+ dates covering winter and summer crops showed that the prediction method was more accurate for NIR band (mean r2 = 0.71, p ≤ 0.01) compared to green band (mean r2 = 0.53; p ≤ 0.01). A recently proposed chlorophyll index (CI), which involves NIR and green spectral bands, was used to retrieve gross primary productivity (GPP) as the product of CI and photosynthetic active radiation (PAR). The regression analysis of GPP derived from closet observed and synthetic ETM+ showed a good agreement (r2 = 0.85, p ≤ 0.01 and r2 = 0.86, p ≤ 0.01) for wheat and sugarcane crops, respectively. The difference between the GPP derived from synthetic and observed ETM+ (prediction residual) was compared with the difference in GPP values from observed ETM+ on the two dates (temporal residual). The prediction residuals (mean value of 1.97 g C/m2 in 8 days) was found to be significantly lower than the temporal residuals (mean value of 4.46 g C/m2 in 8 days) that correspondence to 12% and 27%, respectively, of GPP values (mean value of 16.53 g C/m2 in 8 days) from observed ETM+ data, implying that the prediction method was better than temporal pixel substitution. Investigating the trend in synthetic ETM+ GPP values over a growing season revealed that phenological patterns were well captured for wheat and sugarcane crops. A direct comparison between the GPP values derived from MODIS and synthetic ETM+ data showed a good consistency of the temporal dynamics but a systematic error that can be read as bias (MODIS GPP over estimation). Further, the regression analysis between observed evapotranspiration and synthetic ETM+ GPP showed good agreement (r2 = 0.66, p ≤ 0.01). © 2010 Elsevier B.V. All rights reserved.

1. Introduction There is a growing interest in monitoring the gross primary productivity (GPP) of crops due mostly to their carbon sequestration potential. Both within and between field variability are important components of crop GPP monitoring, particularly for the estimation of carbon budgets. However, the 16-day revisit cycle of Landsat has long limited its use especially over tropical areas because of prevailing cloudy conditions. In contrast, Moderate-resolution Imaging Spectroradiometer (MODIS) has a high temporal resolution, covering the earth up to multiple times per day. The MODIS instrument offers new possibilities for large area crop mapping by providing a near-daily global coverage of science-quality, intermediate resolution (250 m) data since February 2000 at no cost to the end user (Justice and Townshend, 2002). The moderate 250 m spatial reso-

E-mail address: [email protected]. 0303-2434/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jag.2010.06.007

lution of MODIS is appropriate for classifying cropping patterns in the U.S. Central Great Plains given the region’s relatively large field sizes, which are frequently 32.4 ha or larger and would spatially correspond to five or more 250 m pixels. Indian agriculture, in general, is characterized by the fragmentation of land and small land holdings. About 56% of the land holdings in India are less than 1 ha size and the size is going to diminish further, in the future. Of late, intensive cultivation by introduction of mixed cropping systems is a practice in several ecozones in India for maximizing the profits. As a result, agro ecosystems present a higher level of complexity due to small field sizes, diversified cropping pattern and field-tofield variability in crop phenology and management practices. Thus obtaining reliable information on the various crops grown in mixed cropping is of paramount importance in determination of the total area under crop; prediction of the yield per unit area and crop condition assessment. One solution is to combine the spatial resolution of Landsat with the temporal frequency of coarse-resolution sensors, such as MODIS. The Terra platform crosses the equator at about 10:30 A.M. local solar time, roughly 30 min later than Landsat-7

60

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

Table 1 Landsat ETM+ bands and the corresponding MODIS bands used in this study.

Bandwidth specifications Spatial resolution Radiometric resolution Data frequency

Terra- MODIS

Landsat-7 ETM+

Band 1: 545–565 nm Band 2: 841–876 nm 500 m 12 bits Daily

Band 3: 530–610 nm Band 4: 780–900 nm 30 m 8 bits 16 days

ETM+. Their orbital parameters are equal, and as such the viewing (near-nadir) and solar geometries are close to those of the corresponding Landsat acquisition. The Terra/Aqua MODIS provides frequent coarse-resolution observations, revisiting the globe once to twice per day. The MODIS observations include 250-m spatial resolution for red (band 1) and near-infrared (band 2) wavebands and 500-m spatial resolution for other five MODIS land bands (band 3–7). The MODIS land bands have corresponding bandwidths to the Landsat ETM+ sensor except their bandwidths are narrower than ETM+ (see Table 1). Several approaches describing existing fusion techniques are summarized in Table 2. An early example of a fusion model was illustrated by Carper et al. (1990) who combined 10 m spatial resolution SPOT panchromatic imagery with 20 m spatial resolution multispectral imagery by using an intensity–hue-saturation (IHS) transformation. The generated composite images have the spatial resolution of the panchromatic data and the spectral resolution of the original multispectral data. Other techniques to enhance the spatial resolution of multispectral bands include component substitution (Shettigara, 1992), and wavelet decomposition (Yocky, 1996). One of the first studies designed to increase the spatial resolution of MODIS using Landsat was introduced by Acerbi-Junior et al. (2006) using wavelet transformations. The algorithm yields classified land covers types and was used for mapping the Brazilian Savanna (Acerbi-Junior et al., 2006). Recently, Hansen et al. (2008) used regression trees to fuse Landsat and MODIS data based on the 500 m 16-day MODIS BRDF/Albedo land surface characterization product (Roy et al., 2008; Hansen et al., 2008) to monitor forest cover in the Congo Basin on a 16-day basis. While there are numerous examples existing in the current literature that fuse data from multiple sensors, only a few techniques yield calibrated outputs of spectral radiance or reflectance (Gao et al., 2006), a requirement to study vegetation dynamics or quantitative changes in reflectance over time. The Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) (Gao et al., 2006) is one such model and was designed to study vegetation dynamics at a 30 m spatial resolution. STARFM predicts changes in reflectance at Landsat’s spatial and spectral resolution using high temporal frequency observations from MODIS. STARFM predicts reflectance at up to daily time steps, depending on the availability of MODIS data.

Crops are the most pervasive anthropogenic biome worldwide, playing an important role in the global cycles of carbon, water, and nutrients. Most production efficiency models (PEM) are based on the assumption of a close linear relationship between the fraction of photosynthetically active radiation (FAPAR) and the normalized difference vegetation index (NDVI), as well as on a constant, though biome-specific, LUE (Ruimy et al., 1999). Nevertheless, it has been shown that these assumptions do not hold in many circumstances. On the one hand, although LUE is a relatively conservative value among plant formations of the same metabolic type (Ruimy et al., 1999), its variability is species-specific rather than biome-specific (Ahl et al., 2004), it changes with phenological stage (Suyker et al., 2005; Gitelson et al., 2006), and it depends on environmental stress factors (Running et al., 2004). Many models for the remote estimation of GPP use lookup tables of maximum light use efficiency (LUE) for a given vegetation type and then adjust those values downward on the basis of environmental stress factors (Ruimy et al., 1999; Running et al., 2004), but they use coarse spatial resolution meteorological variables, which also result in significant GPP estimation errors (Turner et al., 2005; Zhao et al., 2005). A recently developed conceptual model with a solid physical background has been widely used for the nondestructive retrieval of various pigments in different media such as leaves and crop canopies (Gitelson et al., 2005, 2006). Chlorophyll index (CI) that employ bands in the nearinfrared and either green or red-edge spectral regions are a special case of this conceptual model, and have been linearly related with crop Chlorophyll content (Gitelson et al., 2005), as well as successfully used for maize and soybean midday GPP estimation (Gitelson et al., 2006). It was shown that the product of CI and incoming photosynthetically active radiation (PAR) accounted for more than 98% of GPP variation in both maize and soybeans in a wide range of PAR variation (Gitelson, 2004). This relationship (i.e., GPP versus CI x PAR) was nonspecies-specific, consistent and repeatable under rainfed and irrigated conditions (Gitelson, 2004). The objective of this study was to investigate the suitability of the prediction algorithm (STARFM) for generating synthetic (predicted) ETM+ surface reflectance for NIR and green bands and derivation of CI values using these synthetic surface reflectances. The CI technique is based entirely on remotely sensed data, the close and consistent relationship between GPP and product of PAR and crop chlorophyll content. Therefore, it constitutes an accurate surrogate measure for GPP estimation (Gitelson et al., 2008). The STARFM algorithm was used for agriculture land comprising of winter and summer crops over a tropical area unlike previous studies (Roy et al., 2008; Goa et al., 2006; Hilker et al., 2009) mainly for forest over extra tropical areas. The assessment of the quality of the synthetic ETM+ surface reflectance has been carried out, by comparing these predicted surface reflectance from real (observed) ETM+ images acquired through out two growing seasons over a period of 10 years (2000–2009). The synthetic ETM+

Table 2 Summary of data blending techniques found in the literature. Technique

Sensor 1

Scale 1 (m)

Sensor 2

Scale 2 (m)

Author

IHS transforms Component Substitution Multi-resolution wavelet decomposition Wavelet transforms Downscaling MODIS

SPOT Pan SPOT XS Landsat TM Landsat TM MODIS 1,2 MODIS 3–7

10 20 30 28.5–120 250 500

Spot XS SPOT Pan SPOT Pan SPOT Pan Landsat TM

20 10 10 10 30

Carper et al. (1990) Shettigara (1992)

Combining medium and coarse-resolution satellite data

MODIS 1,2 MODIS

250 250

MODIS 3–7 Landsat TM

500 30

Semi-physical data fusion approach using MODIS BRDF/Albedo STARFM

MODIS MODIS

500 500

Landsat TM Landsat TM

30 30

Pan = Panchromatic, Xs = multispectral.

Yocky (1996) Acerbi-Junior et al. (2006) Trishchenko et al. (2006) Busetto et al. (2008) Roy et al. (2008), Hansen et al. (2008) Gao et al. (2006)

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

61

Fig. 1. Map of the study area. The study site encompasses a Landsat scene (185 km × 185 km) of 15 September 2009 as R (red),G (NIR),B (green) image near Delhi, India. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

data was used to study the GPP over the 10 years time period for assessing seasonal changes (i.e. changes due to vegetation green-up and leaf senescence) in vegetation cover and vegetation status over the study site for which the potential of acquiring frequent higher spatial resolution data (and therefore the potential for mapping of vegetation dynamics) is otherwise low. 2. Methods 2.1. Study area The study area is Mawana subdivision of Meerut district of Uttar Pradesh state depicted by red polygon is shown in Fig. 1, covering about 1250 km2 area. The land cover is predominantly agriculture land with scattered trees and bushes. This area is around 75 km from national capital Delhi. The study area was chosen mainly because of the three important factors: (1) the identified location represents the agroclimatic conditions of large part of northern India, where sugarcane occupies around 2 million ha and wheat has about 10 million ha area. This area comes under Indo-Gangetic plains, where applications of organic matters in soil have gone down drastically resulting in decline of crop productivity. (2) The availability of continuous satellite data as the study site lies at the middle of tiles where Landsat-7 ETM+ has no missing lines due to SLC-off. (3) The availability of continuous observed meteorological

parameters like temperature, rainfall, evaporation and wind over a period of 15 years (1995–2009) over the study site. Fig. 2a shows the RGB with R (red), G (NIR), and B (green) bands of Landsat-7 ETM+ image of 15th September 2009 to have an idea about the land cover type of study site. There are basically two types of land cover found in the RGB (Fig. 2a) namely crop and non-crop land cover types. The pink color shows non-crop land, the Ganges river in eastern part of region and villages in the image while the green colour shows the crop land cover. The maximum temperature in the region is about 45 ◦ C in summer while minimum is about 5 ◦ C in winter. The region receives approximately 95 cm of rainfall during the year. Out of that about 90% rain occurs in the monsoon months i.e. July to middle of September (Fig. 2b). The average depth of water table is 6.0 meter in the study area. The soils characteristics vary in the study area from sandy loam in western part of the area to highly clay loam in the eastern part of the area in khadar of the river Ganges. 2.2. Satellite data Two pairs of contemporary images, acquired by the sensors Terra—MODIS and Landsat-7 ETM+, respectively, were used in the present study. The 8 days MODIS composites (MOD09A1) of surface reflectances (green and NIR) with a spatial resolution of 500 m for 10 years (2000–2009) time period were obtained

Fig. 2. (a) Shows RGB image as R (red), G (NIR), B (green) of 15 September 2009 for land cover types. (b) Shows the annual variation of rainfall over the study site from 2002 to 2009. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

62

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

Table 3 Regression analysis of the observed versus predicted ETM+ images, whose prediction date was closest to the observed scenes. Observed ETM+ scenes

3 February 2009 24 April 2009 26 May 2009 17 October 2009 18 November 2009 20 December 2009

NIR

Green

r2

a

B

r2

a

b

0.72 0.73 0.68 0.81 0.67 0.64

0.10 0.09 0.12 0.08 0.13 0.14

0.93 0.97 0.92 1.07 0.89 0.90

0.58 0.55 0.53 0.53 0.51 0.49

0.11 0.10 0.12 0.09 0.11 0.13

0.94 0.92 0.92 0.91 0.92 0.89

from the EOS data gateway of NASA’s Goddard Space Flight Center (http://redhook.gsfc.nasa.gov). Landsat-7 ETM+ orthorectified images (path 146, row 040) were acquired through the USGS GLOVIS portal (http://glovis.usgs.gov/) for the same period. Following STARFM algorithm input requirements, the MODIS data were reprojected to the geographical projection using the MODIS reprojection tool (Kalvelage and Williams, 2005), clipped to the extent of the available Landsat imagery, and resampled to a 30 m spatial resolution with nearest-neighbor resampling to maintain the MODIS pixel values and 30 m output pixel dimensions to reduce nearest-neighbor resampling pixel shifts (i.e., position errors) (Roy and Dikshit, 1994). These datasets were forced to assume the same size in order to make the results comparable. The basic features of the two sets of data are given in Table 1. The radiometric and atmospherically corrections were applied to Landsat-7 ETM+ scenes. The equations and parameters to convert calibrated Digital Numbers (DNs) to physical units, such as at-sensor radiance or TOA reflectance were utilized in this study are from the previous studies (Chander and Markham, 2003; Chander et al., 2007; Markham et al., 2004). Images were atmospherically corrected using the QUAC module (Bernstein et al., 2005) of environment for visualizing images (ENVI) processing package. 3. Evaluation of synthetic ETM+ imagery The Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) developed by Gao et al. (2006) predicts pixel values based upon a spatially weighted difference computed between the ETM+ and the MODIS scenes acquired at date T1, and the ETM+ T1-scene and one or more MODIS scenes of prediction day (T2), respectively. The input pairing (T1) criteria for the prediction of synthetic ETM+ (T2) was based on least amount of cloud cover (almost 0%) and minimal temporal difference in order to reduce the likelihood for changes in land cover resulting from harvesting or phenological changes. The input pairs (T1) were selected as March and September months. Since, the study area comprises winter and summer crops. Generally, early March is the peak time (middle of season) for winter and September is peak time for summer crops. The validation of synthetic ETM+ with observed ETM+ surface reflectance was carried out for the year 2009. The algorithm yielded 46 (8 days composites) 30 m spatial resolution, synthetic ETM+ images for the growing seasons (December to April for winter crop and May to December for summer crop) using a ETM+ and a MODIS scene acquired on 7th March 2009 and 15th September 2009 as the T1 images, and 8-day MODIS composites between January to December 2009 as the T2 images for prediction. No images were synthesized for March and September 2009 as these imageries were used as T1 input. Also, the validation was not done for June, July and August months because of non-availability of cloud-free ETM+ images. The statistical analysis between synthetic and observed ETM+ surface reflectance for NIR and green bands for winter and summer crops has been carried out. The results are provided in Table 3.

The first column in each sub-table is showing the coefficient of determination, the second column is showing the intercept, normalized to percent total reflectance of the observed image (for instance a value of 0.1 means the predicted image overestimated the reflectance by 10%), the third column is showing the slope of the relationship between observed and predicted. A two-sided t-test was used to determine whether there is a statistically significant difference between observed and predicted surface reflectance values (i.e., whether the mean difference between observed and predicted data varied significantly from zero). The high correlations between observed and predicted pixel values were found for the NIR band (0.64 ≤ r2 ≤ 0.81, p ≤ 0.01), while the green band yielded slightly weaker relationships (0.49 ≤ r2 ≤ 0.58, p ≤ 0.01) (Table 3). In all cases the intercepts of the relationship between observed and predicted images were greater than zero (Table 3) which can be interpreted as a noise signal likely due to atmospheric and BRDF effects. These findings are also confirmed by previous studies (Gao et al., 2006). Figs. 3a–f and 4a–f show a per-pixel comparison between observed and predicted ETM+ NIR and green surface reflectance for the winter and summer crops. The first row shows the scatter plots for NIR band and second row for green band. The relationship between observed and predicted pixel values closely followed the 1-to-1 line (Figs. 3a–f and 4a–f) thereby showing that ETM+ surface reflectances were accurately predicted by STARFM. Some deviations from this 1-to-1 line, however, were found for the NIR towards the end of the summer crops period (Fig. 4b and c). The harvesting of summer crops (sugarcane) and simultaneously sowing of winter crops (wheat) during the month of December creates a complex mixture of land cover type, as a result the scatter plots (Fig. 4b and c) are bit noisy and the coefficient of determinations for NIR are also relatively low (r2 = 0.67 and r2 = 0.64) (Table 3). The similar situation was also observed for winter crops (Fig. 3b and c), but less noisy compared to that of summer crops (Fig. 4b and c). The ability of 30 m resolution synthetic ETM+ images for the predictability of changes depends upon the capacity of MODIS to detect these changes, particularly when they occur in vegetation structure or stand composition or at sub-pixel ranges (Gao et al., 2006). Complex mixtures of land cover type are a challenge for all methods of data fusion. Therefore, it will be difficult to identify or spatially define individual change events as it is not possible to depict changes occurring in the sub-MODIS pixel range. As a result, the algorithm in its current form seems less suited for the prediction of changes in vegetation structure (such as originating from clear cut harvesting or thinning) or changes in land cover. Changes will also not be detected by STARFM when two contradicting changes occur within a coarse-resolution pixel simultaneously and compensate for each other (Gao et al., 2006).

4. Results One of the objectives of this paper was to test the suitability of canopy chlorophyll content related vegetation indices using synthetic ETM+ data for GPP estimation. Many current models of ecosystem carbon exchange based on remote sensing, such as the MODIS product termed MOD17, still require considerable input from ground based meteorological measurements and look up tables based on vegetation type. Since these data are often not available at the same spatial scale as the remote sensing imagery, they can introduce substantial errors (either in the original estimate of LUE for a particular vegetation type or in the assignment of vegetation type to a pixel into the carbon exchange estimates (Yang et al., 2007; Sims et al., 2008). Therefore, it is worthwhile to explore alternative methods for the estimation of GPP that may not require as many input parameters. Gitelson et al. (2006) have success-

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

63

Fig. 3. Per-pixel comparison between observed and predicted ETM+ surface reflectance for three different observed ETM+ dates for winter crops. The first row represents the surface reflectance for the NIR and second row for green band, respectively.

fully estimated GPP with chlorophyll indices and clearly indicated that the product of total crop chlorophyll content and PAR could be good indicator of GPP. The GPP used in the present study was calculated as a product of PAR and crop chlorophyll content. PAR was calculated from European Center for Medium Range Weather Forecasting (ECMWF) analysis data. 4.1. Spatiotemporal variation of GPP The spatiotemporal variation of the GPP is illustrated in Fig. 5 over the study area. The absolute difference of GPP values (temporal residuals) from MODIS data on 14th September 2009 (Fig. 5a) and 16th October 2009 (Fig. 5b) and observed ETM+ data on 15th September 2009 (Fig. 5d) and 17th October 2009 (Fig. 5e) were computed and shown in Fig. 5c and f, respectively. Similarly the prediction residuals were computed from the absolute difference of GPP values from synthetic ETM+ data on 16th October 2009 and observed ETM+ data on 17th October 2009 (Fig. 5i). The study site

is predominately agricultural land containing many small fields of approximately less than a hectare. Sugarcane is the dominate crop over this area, planted during May and being harvested from late September to December month. Close examination of the GPP derived from predicted ETM+ reveals a faint blocky pattern that corresponds spatially to the locations of the resampled 500 m MODIS pixel dimensions. This pattern is most evident across some of the fields (Fig. 5a and b) across some of the fields (Fig. 5h). This is due to the fact that, the dynamics of surface reflectance are different between one field and its neighbor (e.g., due to harvesting in one field and not in an adjacent field), underlying the fact that method are less likely to be valid where the Landsat reflectance heterogeneity at the sub-MODIS pixel scale changes temporally. The GPP derived from observed ETM+ data for 15th September (Fig. 5d) is higher than the October 2009 (Fig. 5e) with many fields harvested by the later acquisition date. Despite the evident land cover change complexity, the GPP derived from predicted ETM+ data on 16th October 2009 captures many of the temporal changes

Fig. 4. Per-pixel comparison between observed and predicted ETM+ surface reflectance for three different observed ETM+ dates for summer crops. The first row represents the surface reflectance values for the NIR and second row for green band, respectively.

64

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

Fig. 5. Illustrating spatiotemporal variation of GPP (g C m2 in 8 days); (a) and (b) GPP from MODIS (MOD09A1) on 14th September and 16th October 2009, (c) the temporal residual (absolute value of (a) and (b)), (d) and (e) GPP from observed ETM+ on 15th September and 17th October 2009, (f) the temporal residual (absolute value of (d) and (e)), (g) and (h) GPP from observed ETM+ on 17th October 2009 and predicted ETM+ on 16th October 2009, (i) the prediction residual (absolute value of g and h). (a), (b), (d), (e), (g) and (h) Are shown with the same contrast stretch. Residual values in c, f and i are colored as: 0 ≤ purple ≤ 0.5, 0.5 ≤ blue ≤ 1.5, 1.0 ≤ green ≤ 1.5, 1.5 ≤ yellow ≤ 2.0, 2.0 ≤ orange ≤ 2.5, red ≤ 2.5. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

(Fig. 5h). In general, the prediction residuals (spatial mean value of GPP 1.97 g C/m2 in 8 days) was considerably lower than the temporal residuals (spatial mean value of GPP was 4.46 g C/m2 in 8 days for ETM+ data) that correspond to 12%, 27% of the spatial mean GPP 16.53 g C/m2 in 8 days) from ETM+ data on 16th October 2009 and 17th October 2009, respectively, implying that the prediction method is on an average better than temporal pixel substitution. Temporal residuals of MODIS derived GPP were significantly higher compared to ETM+ data sets (Fig. 5c and f). However, prediction residuals were also found to be greater than the temporal residuals at few locations along contrast boundaries such as river flowing along the eastern part of the study region. The higher predictions

residuals may be due to regions of textural change in addition to misregistration and resampling impacts that are greatest in regions of high spatial variation (Roy, 2000). 4.2. Comparisons of GPP The majority of ETM+ images used for the prediction of NIR and green surface reflectances were close to 100% cloud-free. However, the GPP calculated using synthetic ETM+ data still include some contaminated pixels. MODIS and ETM+ data sets are generally welldocumented, quality-controlled data sources that have been preprocessed to reduce many of these problems (Smith et al., 1997;

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

65

Fig. 6. (a) Scatter plot of observed ETM+ GPP (x-axis) versus sensor synthetic ETM+ GPP (y-axis) from the nearest maximum value composite to Landsat acquisition date for wheat crop. (b) Similar to (a) but for sugarcane. (c) The seasonal dynamics of GPP data derived from observed and synthetic ETM+ scenes. The line plot show the 8 days mean GPP values over study site derived from synthetic ETM+ GPP. (d) Same as (c) for MODIS derived GPP. (e) and (f) Show monthly temperature and rainfall anomalies derived from 15 years (1995–2009) observed data over study site.

Tucker and Pinzon, 2005; James and Kalluri, 1994; Gutman, 1999). However, some noise is still present in the downloadable data sets and, therefore, CI time series need to be smoothed before being used. Therefore, Savitzky–Golay filtering technique (Savitzky and Golay, 1964) was used in order to remove the noise from CI time series. This technique was applied to smooth every pixel’s time series profile for 10-year (2000–2009) period. The scatter plots of GPP derived from synthetic and observed ETM+ data for winter and summer crops from the nearest composite period are shown in Fig. 6a and b, respectively. The GPP values derived from observed ETM+ were close to 100% cloudfree thus provide a reference value for the comparison of GPP derived from synthetic ETM+ data. The total number observed ETM+ scenes used for winter and summer seasons were 20 and

44, respectively. A strong linear relations (winter crops, r2 = 0.85, p ≤ 0.01 and summer crops, r2 = 0.86, p ≤ 0.01) exist between GPP values from observed and synthetic ETM+ data. The values of coefficient of determination (winter crops, r2 = 0.85, p ≤ 0.01 and summer crops, r2 = 0.86, p ≤ 0.01) were found to be almost equal for two canopies. It means that chlorophyll content is main driver of GPP and that structure plays a small role in this relation. This result is also consistent with the study of Gitelson et al. (2005) which shows small differences between GPP estimation for soybeans and maize that have very different canopy structure. Higher chlorophyll content is directly correlated with increased photosynthetic capacity through increased light use efficiency (Field and Mooney, 1986). These conclusions imply that GPP estimation from chlorophyllrelated indices and PAR is of particular significance for remote

66

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

sensing applications provided one can obtain accurate information of chlorophyll as canopy structure difference exerts a relatively small effect. The GPP time series from synthetic ETM+ and MODIS (MOD09A1) are presented in Fig. 6c and d, respectively. The GPP time series from synthetic ETM+ was generated at 30 m spatial resolution by taking the mean of GPP (Fig. 6c) over study area for 46 weeks (8-day composites) for 10 years period (2000–2009). This time series was resampled to 500 m spatial resolution in order to compare with corresponding MODIS derived GPP time series as shown in Fig. 6d. The first maxima correspondences to winter crops (wheat) while the second one for summer crops (sugarcane) in each year. All the curves of GPP for the year 2000–2009 show distinctive seasonal dynamics. It may be observed that vegetation green-up and leaf-down at the beginning and at the end of the vegetation period were well captured by the GPP derived from synthetic ETM+ (Fig. 6c) corresponding to MODIS derived GPP (Fig. 6d). The GPP time series by and large represent crops, since only a small area of scattered trees and bushes are present over the study site. The GPP values are maximum around week 8 (winter crops) followed by second maxima around 38 week (summer crops). The long-term spatial mean GPP values (Fig. 6c and d, black line) were found to be higher in the summer season than in the winter season, for instance, first maxima (15 g C/m2 in 8 days) around February and second maxima (18 g C/m2 in 8 days) in September. The relatively low GPP in the winter season can be attributed to the length of crop season and as a result more accumulation of chlorophyll contents for the crops of longer time period. The winter crops are mainly wheat crop of four month time period and sugarcane is main summer crop of eight month time period over the study site. This may suggests, that synthetic ETM+ derived GPP may potentially be used to quantify seasonal changes in vegetation, the rate of increase and decrease of the GPP, the dates of the beginning, end and peak(s) of the growing season, the length of the growing season, the timing of the annual maximum GPP and the GPP value at a fixed date at fine spatial scales (Gao et al., 2006). Similar patterns are observed in case of MODIS derived GPP (Fig. 6d). However, there were two major differences observed between these two data sets. First, the interannual variations were found to be more distinct in case of synthetic ETM+ GPP compared to MODIS. Second, MODIS GPP was overestimated. A seasonal pattern of synthetic ETM+ GPP (Fig. 6c) shows clear effects by two severe drought events during the year 2002 and 2004, when one can find considerable overestimation of the MODIS GPP (Fig. 6d). The recent studies also support these findings (Susmitha et al., 2009). The mean spatial GPP (mean of 10 years) over study site is shown as black line (Fig. 6c and d), which clearly separates the years of healthy and poor vegetation conditions during the summer season. The maximum value of synthetic ETM+ mean spatial GPP values (Fig. 6c observed during the year 2002 and 2004 were significantly lower (12.5 and 13.5 g C/m2 in 8 days, respectively) compared to mean spatial GPP value (18 g C/m2 in 8 days)) and also with other years. The lower value of GPP for the year 2002 attributed to no rainfall event during month of July 2002 (Susmitha et al., 2009). The year 2004 also suffered with four short or long monsoon breaks, but these were intermittent and of less duration compared to the July month of year 2002 (Susmitha et al., 2009). The extent of negative deviation of observed rainfall and positive deviation of observed temperature from its long-term mean for a pixel, district or region, and the duration of continuous negative and positive deviations are powerful indicators of drought magnitude and persistence. The monthly temperature and rainfall anomalies were compute as a difference of value minus long-term mean (1995–2009) as shown in Fig. 6e and f, respectively. Fig. 6e and f clearly shows that, overall, the study region was dry in 2002 and 2004 during the monsoon period (June to September) com-

pared to that of other years. The precipitation dynamics is a key link between climate fluctuation and vegetation dynamics in space and time. The less precipitation reduces plants water availability in the dry season. Generally, the sowing time of summer crops over this area starts from middle of June and therefore, July rainfall is very critical for the growth of vegetation. The drought years 2002 and 2004 are very well brought by synthetic ETM+ derived GPP at 30 m spatial resolution. Similar results were also observed for MODIS GPP (Fig. 6d), but with overestimation. This finding was also found to be in confirmation with previous study (Hwang et al., 2008). 4.3. The relationship between ETM+ GPP and evapotranspiration The crop chlorophyll content was shown to be sensitive to changes in vegetation conditions, since it is directly influenced by the chlorophyll’s absorption of the sun’s radiation (Gitelson et al., 2008). Because the chlorophyll status integrates the effects of numerous environmental factors, the GPP derived as a product of crop chlorophyll content and PAR could be useful parameter for crop monitoring. However, the validation of ETM+ derived GPP at 30 m resolution is problematic because there are very limited available field data. Ideally, the testing sites should cover as many as biomes types and climate regimes as possible. Eddy flux towers offer invaluable opportunities to validate process-based ecosystem models and satellite data because they measure carbon, water and energy exchange on a long-term and continuous basis (Baldocchi et al., 2001; Running et al., 1999). Unfortunately, there are no eddy flux towers over this region. In addition to CO2 flux data, H2 O flux data from the eddy flux tower sites should be used for evaluating the process-based and satellite-based models, particularly in those flux sites where there were large numbers of missing CO2 flux data in the wet season (Saleska et al., 2003). As photosynthesis is closely coupled with H2 O flux (evapotranspiration), therefore observed evapotranspiration was used to study the relationship between synthetic ETM+ GPP. Atmospheric and radiation variables are continuously measured at the weather station of the study site. The evaporation measurements are made using the class A pan evaporimeter by taking into account not only of evaporation loss but also gains due to rainfall. A rain gauge situated nearby is used to assess the depth of rain falling in the pan. The evaporation measurements are done twice a day (8:30 A.M. and 5:30 P.M.) on the days without rain. Reference evapotranspiration by pan method was estimated from measurements of daily Class A evaporation pan, based on the relationship Eo = Ep × kp , where Eo is the daily reference evapotranspiration (mm), kp is the pan coefficient (Doorenbos and Pruitt, 1977) and Ep is the pan evaporation (mm). The coefficient of determination was obtained from the linear regression analysis in order to evaluate the performance of the relationship between synthetic ETM+ GPP derived in this research and observed evapotranspiration data. The regression analysis has been carried out on entire data set without bifurcating into winter and summer seasons. Fig. 7 shows the results of the statistical comparison of the daily reference evapotranspiration values for intervals of 1–8 days, from the year 2000 to 2009.These two data sets are found to be in good agreement (r2 = 0.665, p ≤ 0.01). Even though this study site by and large represents agriculture land, still there are some human land uses such as road, building and some waste land. Therefore, when applying the spatial pattern of vegetation in the regression analysis, these patches of different land cover types could also be a potential source of error. These data sets also include two drought years 2002 and 2004 (Fig. 6c and d). The impact of drought on GPP has been shown to vary with the intensity of the drought (Reichstein et al., 2002; Barr et al., 2004; Kljun et al., 2006; Krishnan et al., 2006). These climatic perturbations could also have impact on regression analysis between observed evapotranspiration and GPP from synthetic ETM+ data.

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

Fig. 7. The scatter plot of observed evapotranspiration (ET) and GPP from predicted ETM+.

5. Discussions As currently only the SLC-off Landsat ETM+ and the aging Landsat-5 TM (Chander et al., 2007; Helder and Ruggles, 2004) systems are acquiring data, and only a single successor Landsat Data Continuity Mission (LDCM) sensor is scheduled for launch early in the next decade (Irons and Masek, 2006), a potential solution to provide more frequent high resolution surface observations is to fuse Landsat observations with data from other remote sensing systems. This study investigated the capability of STARFM (Gao et al., 2006) over a tropical region (India) to predict seasonal changes in winter and summer crops for a higher level of complexity due to small field sizes, diversified cropping pattern and field-to-field variability in crop phenology and management practices, which is different than its use in previous studies (Gao et al., 2006; Hilker et al., 2009). The prediction residuals (Fig. 5i) were found to be considerably lower than the temporal residuals (Fig. 5c and f), suggesting that the prediction method is on an average better than temporal pixel substitution. The higher correlations between observed and predicted pixel values for the NIR and green band (Table 3) may suggest their use for the derivation of the product such as green chlorophyll at 30 m spatial resolution. The validation of synthetic 8 day ETM+ NIR reflectance with single day observed ETM+ NIR showed a very good agreement (Figs. 3 and 4), implying the capability of prediction method. In the present study, MODIS 8 day composite images are used rather than daily MODIS reflectance products, because they yielded largely cloud-free images and can therefore help to predict changes in reflectance of vegetation over tropical area of the earth where cloud cover prevents frequent cloud-free observations. The use of MODIS composites rather than single observations may, however, impact the average reflectance brightness for a given image region, depending on the MODIS scenes used in the MOD09A1 product and is therefore at the same time also a limitation to the applied technique as the composition of data originating from multiple viewing angles and the variation of vegetation within the 8-day production period which differs from the Landsat acquisition date, also provides a possible source of error. GPP is often used as a monitoring tool for the productivity, vegetation health and dynamics, enabling easy temporal and spatial comparisons (Myneni et al., 1997). Benefits of this work include the successful data fusion of ETM+ and MODIS data sets, enabling the use of the longer Landsat TM/ETM+ GPP time series for winter and summer seasons over a tropical area. Direct comparisons of synthetic ETM+ GPP with MODIS GPP revealed several important distinctions and similarities. One obvious difference was associated with image/map resolution. Synthetic ETM+ captured much of the spatial complexity of land cover at the study site (Fig. 5f). In con-

67

trast, the relatively coarse-resolution of MODIS did not allow for that level of spatial detail (Fig. 5c). The greatest difference was an over estimation by MODIS GPP (Fig. 6d). This is because MODIS data have significantly larger grain size (500 m), which can be expected to lower overall spatial variance (Woodcock and Strahler, 1987; Cohen et al., 1990). Furthermore, at larger grain sizes, there is less likelihood that class-specific relationships are practical. However, due to difference in spectral bands, temporal compositing, spatial resolution and other sensor characteristics, there has never been the expectation that GPP values from these two sensors (synthetic ETM+ and MODIS) here will match perfectly. Empirical techniques can be used to force agreement, yet there remains a question of how well the resulting time series matches reality. Comparison provides evidence that GPP time series does represent what is actually happening on the ground. The regression analysis between synthetic ETM+ GPP and observed evapotranspiration over the study site is shown in Fig. 7. The observed ET and GPP are found to be in good agreement (r2 = 0.66, p ≤ 0.01). High correlation was observed between GPP and ET, since vegetation grows well for high precipitation and therefore, increases in resulting GPP (Korner, 1994; McMurtrie et al., 1992). Plant transpiration is controlled by canopy conductance, which further represents the average status of leaf level stomatal conductance. Therefore, a fluctuation in stomatal conductance usually leads to a commensurately large fluctuation in transpiration, and hence, ET. In semiarid or arid systems, soil evaporation is a major component of ET, yet little is known quantitatively about it over large spatial scales. Soil evaporation is reported to range from a few percent to more than 80% of the measured or estimated total ET depending on the vegetation cover (Bethenod et al., 2000; Hsiao and Xu, 2005; Villalobos and Fereres, 1990; Wilson et al., 2000). Few studies have provided in situ seasonal measurements of leaf optical properties over plant growing seasons in the tropical area (Roberts et al., 1998). The future field work should focus on seasonal measurements of leaf water content, chlorophyll, dry matter, and leaf phenology (leaf fall and emergence of new leaves) over seasons, in support of temporal analyses of GPP. Algorithms like the one used in this study are important components of current research efforts seeking to map high spatial resolution changes in vegetation cover and status with high temporal density, over larger areas. Data blending approaches, such as STARFM can help in minimizing the technical limitations and tradeoffs associated with information needs that require data with both high spatial and high temporal resolutions. Applications such as monitoring seasonal changes in vegetation biophysical and structural attributes over tropical areas can benefit from the synergies of multiple data sources such as MODIS and Landsat ETM+. Advances in data blending can also influence the design of new sensors, where the advantages of different spatial and temporal resolutions may be fully realized in the creation of different sensors on different platforms, with the complementary nature of these systems in a data blending approach, considered from the outset of the design process. Tactical decision making on land management can benefit from immediate access to the synthetic data especially over the heterogeneous areas of very small crop field. At this stage, however, the synthetic ETM+ data should be considered only a general solution. Until more detailed models are ready that begin with synthetic ETM+ and add computations of species, and even of cultivar-specific developmental sequences, it will lack the accuracy desired for specific crops and areas.

6. Conclusions The STARFM algorithm has been successfully used in this study for complex mixture of agriculture land over a tropical area to

68

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69

predict reflectance for NIR and green bands over a period of 10 years (2000–2009). The accuracy was found to be better for NIR (mean r2 = 0.71, p ≤ 0.01) compared to green band (mean r2 = 0.53; p ≤ 0.01) for winter and summer crops. The prediction method maintained a high spatial level of detail in the predicted scenes, as the prediction residuals were significantly lower than temporal residuals, it seems however, less well suited to predict sudden changes in land cover, such as induced by harvesting or sowing of crops. Gitelson et al. (2006) have successfully estimated GPP with chlorophyll indices and clearly indicated that the product of total crop chlorophyll content and PAR could be good indicator of GPP. Thus, this study has also demonstrated the utility of canopy chlorophyll content derived from synthetic ETM+ data in the estimation of GPP that will provide new avenue for future remote sensing of GPP. The synthetic ETM+ GPP demonstrated the capability of mapping the seasonal and interannaul variations in vegetation at ETM+ spatial resolution and 8-day time intervals. The two drought years 2002 and 2004 were clearly brought out by synthetic ETM+ GPP. The study of the relationship of synthetic ETM+ GPP with evapotranspiration yielded good agreement (r2 = 0.66; p ≤ 0.01)). This study suggests that MODIS composites can be a useful alternative to daily observations, since the prevailing cloud conditions prevent frequent clear sky observations during monsoon period. Composites may however reduce the quality of STARFM predictions due to changes in pixel brightness resulting from remaining directional or atmospheric impacts in the different MODIS images. Acknowledgements I am very much thankful to Dr. T. Ramasami, Secretary, Department of science and Technology, for his kind support and encouragement during the course of this study. Author would like to thank Dr. Feng Gao, Biospheric Sciences Branch, NASA Goddard Space Flight Center, USA for providing the STARFM algorithm and Dr. Thomas Hilker, Department of Forest Resource Management, University of British Columbia, Vancouver, Canada, for useful discussions. I thank the subject editor and two anonymous reviewers for comments on the manuscript. Author is also thankful to NASA for providing Landsat-7 ETM+ and MODIS data and ECMWF for analysis data. References Acerbi-Junior, F.W., Clevers, J., Schaepman, M.E., 2006. the assessment of multisensor image fusion using wavelet transforms for mapping the Brazilian savanna. International Journal of Applied Earth Observation and Geoinformation 8, 278–288. Ahl, D.E., Gower, S.T., Mackay, D.S., Burrows, S.N., Norman, J.M., Diak, G.R., 2004. Heterogeneity of light use efficiency in a northern Wisconsin forest: implications for modeling net primary production with remote sensing. Remote Sensing of Environment 93 (1/2), 168–178. Baldocchi, D., Falge, E., Gu, L., Olson, R., Hollinger, D., Running, S.W., 2001. FLUXNET: a new tool to study the temporal and spatial variability of ecosystem-scale carbon dioxide, water vapor, and energy flux densities. Bulletin of the American Meteorological Society 82, 2415–2434. Barr, A.G., Black, T.A., Hogg, E.H., Kljun, N., Morgenstern, K., Nesic, Z., 2004. Interannual variability in the leaf area index of a boreal aspen-hazelnut forest in relation to net ecosystem production. Agriculture and Forest Meteorology 126, 237–255. Bernstein, L.S., Adler-Golden, S.M., Sundberg, R.L., Levine, R.Y., Perkins, T.C., Berk, A., Ratkowski, A.J., Felde, G., Hoke, M.L., 2005. A new method for atmospheric correction and aerosol optical property retrieval for VIS-SWIR multi- and hyperspectral imaging sensors: QUAC (QUick atmospheric correction). In: Proceedings of IGARSS, pp. 3549–3552. Bethenod, O., Katerji, N., Goujet, R., Bertolini, J.M., Rana, G., 2000. Determination and validation of corn crop transpiration by sap flow measurement under field conditions. Theoretical and Applied Climatology 67, 153–160. Busetto, L., Michele, M., Roberto, C., 2008. Combining medium and coarse spatial resolution satellite data to improve the estimation of sub-pixel NDVI time series. Remote Sensing of Environment 112 (1), 118–131. Chander, G., Markham, B.L., 2003. Revised Landsat-5 TM radiometric calibration procedures, and post-calibration dynamic ranges. IEEE Transactions on Geoscience and Remote Sensing 41, 2674–2677.

Chander, G., Markham, B.L., Barsi, J.A., 2007. Revised Landsat 5 Thematic Mapper radiometric calibration. IEEE Transactions on Geoscience and Remote Sensing 44, 490–494. Carper, W.J., Lillesand, T.M., Kiefer, R.W., 1990. The use of intensity–hue-saturation transformations for merging SPOT panchromatic and multispectral image data. Photogrammetric Engineering and Remote Sensing 56, 459–467. Cohen, W., Spies, T., Bradshaw, G., 1990. Semivariograms of digital imagery for analysis of conifer canopy structure. Remote Sensing of Environment 34, 167–178. Doorenbos, J., Pruitt, W.O., 1977, Guidelines for Predicting Crop Water Requirements. FAO Irrigation and Drainage Paper no. 24, Rome, Italy. Field, C., Mooney, H.A., 1986. The photosynthesis–nitrogen relationship in wild plants. In: Givnish, T.J. (Ed.), On the Economy of Plant Form and Function. Cambridge University Press, Cambridge, pp. 25–55. Gao, F., Masek, J., Hall, H., Schwaller, M., 2006. On the blending of the Landsat and MODIS surface reflectance: predicting daily Landsat surface reflectance. IEEE Transactions on Geosciences and Remote Sensing 44, 2207–2218. ˜ A., Verma, S.B., Rundquist, R.D., Arkebauer, T.J., Keydan, g., LeavGitelson, A.A., Vina, itt, B., Ciganda, V., Burba, G.G., Suyker, A.E., 2006. Relationship between gross primary production and chlorophyll content in crops: implications for the synoptic monitoring of vegetation productivity. Journal of Geophysical. Research 111, D08 S11. ˜ A., Rundquist, D.C., Ciganda, V., Arkebauer, T.J., 2005. Remote Gitelson, A.A., Vina, estimation of canopy chlorophyll content in crops. Geophysical Research Letter 32, L08 403. Gitelson, A.A., 2004. Wide dynamic range vegetation index for remote quantification of biophysical characteristics of vegetation. Journal of Plant Physiology 161 (2), 165–173. ˜ A., Masek, J.G., VermaF S.B., Suyker, A.E., 2008. Synoptic monitorGitelson, A.A., Vina, ing of gross primary productivity of maize using Landsat data. IEEE Geoscience and Remote Sensing Letters 5 (2). Gutman, G.G., 1999. On the use of long-term global data of land reflectances and vegetation indices derived from the AVHRR. Journal of Geophysical Research 104, 6241–6255. Hansen, M.C., Roy, D.P., Lindquist, E., Adusei, B., Justice, C.O., Altstatt, A., 2008. A method for integrating MODIS and Landsat data for systematic monitoring of forest covers and changes in the Congo Basin. Remote Sensing of Environment 112, 2495–2513. Helder, D., Ruggles, T., 2004. Landsat thematic mapper reflective-band radiometer artifacts. IEEE Transactions on Geoscience and Remote Sensing 42 (12), 2704–2716. Hilker, T., Wulder, M.A., Coops, N.C., Seitz, N., White, J.C., Gao, F., Masek, J.G., Stenhouse, G., 2009. Generation of dense time series synthetic Landsat data through data blending with MODIS using a spatial and temporal adaptive reflectance fusion model. Remote Sensing of Environment 113, 1988–1999. Hsiao, T., Xu, L., 2005. Evapotranspiration and relative contribution by the soil and the plant. California Water Plan Update. UC Davis. Hwang, T., Kang, S., Kim, J., Kim, Y., Lee, D., Band, L., 2008. Evaluating drought effect on MODIS Gross Primary Production (GPP) with an eco-hydrological model in the mountainous forest, East Asia. Global Change Biology 14, 1–20. Irons, J., Masek, J., 2006. Requirements for a Landsat data continuity mission. Photogrammetic Engineering and Remote Sensing 72 (10), 1102–1108. James, M.E., Kalluri, S.N.V., 1994. The pathfinder AVHRR land data set: an improved coarse resolution data set for terrestrial monitoring. International Journal of Remote Sensing 15, 3347–3363. Justice, C.O., Townshend, J., 2002. Special issue on the moderate resolution spectroradiometer (MODIS); a new generation of land surface monitoring. Remote Sensing of Environment 83 (1), 1–2. Kalvelage, T., Williams, J., 2005. Supporting users through integrated retrieval processing and distribution systems at the land processes distributed active archive center. Acta Astronaut 56 (7), 681–687. Kljun, N., Black, T.A., Griffis, T.J., Barr, A.G., Gaumont-Guay, D., Morgenstern, K., McCaughey, J.H., Nesic, Z., 2006. Response of net ecosystem productivity of three boreal forest stands to drought. Ecosystems 9, 1128–1144. Krishnan, P., Black, T.A., Grant, N.J., Barr, A.G., Hogg, E.T.H., Jassal, R.S., Morgenstern, K., 2006. Impact of changing soil moisture distribution on net ecosystem productivity of a boreal aspen forest during and following drought. Agriculture and Forest Meteorology 139, 208–223. Korner, Ch., 1994. Leaf diffusive conductances in the major vegetation types of the globe. In: Schulze, E.-D., Caldwell, M.M. (Eds.), Ecophysiology of Photosynthesis. Springer, New York, pp. 463–490. Markham, B.L., Thome, K., Barsi, J., Kaita, E., Helder, D., Barker, J., Scaramuzza, P., 2004. Landsat-7 ETM+ on orbit reflective band radiometric stability and absolute calibration. IEEE Transactions and Geoscience and Remote Sensing 42, 2810–2820. McMurtrie, R.E., Leuning, R., Thompson, W.A., Wheeler, A.M., 1992. A model of canopy photosynthesis and water use incorporating a mechanistic formulation of leaf CO2 exchange. Forest Ecology and Management 52, 261–278. Myneni, R.B., Keeling, C.D., Tucker, C.J., Asrar, G., Nemani, R.R., 1997. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386, 698–702. Reichstein, M., Tenhunen, J.D., Roupsard, O., 2002. Severe drought effects on ecosystem CO2 and H2 O fluxes at three Mediterranean evergreen sites: revision of current hypotheses? Global Change Biology 8, 999–1017. Roberts, D.A., Nelson, B.W., Adams, J.B., Palmer, F., 1998. Spectral changes with leaf aging in Amazon caatinga. Trees-Structure and Function 12, 315–325. 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 (1), 56–64.

D. Singh / International Journal of Applied Earth Observation and Geoinformation 13 (2011) 59–69 Roy, D.P., Dikshit, O., 1994. Investigation of image resampling effects upon the textural information content of a high spatial resolution remotely sensed image. International Journal of Remote Sensing 15, 1123–1130. Roy, D.P., 2000. The impact of misregistration upon composited wide field of view satellite data and implications for change detection. IEEE Transactions on Geoscience and Remote Sensing 38, 2017–2032. Roy, P., Junchang, J., Lewis, P., Schaaf, C., Gao, F., Hansen, M., Lindquist, E., 2008. Multitemporal MODIS–Landsat data fusion for relative radiometric normalization, gap filling, and prediction of Landsat data. Remote Sensing of Environment 112, 3112–3130. Running, S.W., Nemani, R., Glassy, J. M., Thornton, P., 1999, MODIS daily photosynthesis (PSN) and annual net primary production (NPP) product (MOD17). Algorithm theoretical basis document, version 3.0. http://modis.gsfc.nasa.gov. Running, S.W., Nemani, R.R., Heinsch, F.A., Zhao, M.S., Reeves, M., Hashimoto, H., 2004. A continuous satellite-derived measure of global terrestrial primary production. Bioscience 54 (6), 547–560. Saleska, S.R., Miller, S.D., Matross, D.M., Goulden, M.L., Wofsy, S.C., da Rocha, H.R., de Camargo, P.B., Crill, P., Daube, B.C., deFreitas, H.C., Hutyra, L., Keller, M., Kirchhoff, V., Menton, M., Munger, J.W., Pyle, E.H., Rice, A.H., Silva, H., 2003. Carbon in Amazon forests: unexpected seasonal fluxes and disturbance induced losses. Science 302, 1554–1557. Savitzky, A., Golay, M.J.E., 1964. Analytical Chemistry 36, 1627–1639. Shettigara, V.K., 1992. A generalized component substitution technique for spatial enhancement of multispectral images using a higher resolution data set. Photogrammetric Engineering and Remote Sensing 58, 561– 567. Smith, P.M., Kalluri, S.N.V., Prince, S.D., Defries, R., 1997. The NOAA/NASA pathfinder AVHRR 8-km land data set. Photogrammetric Engineering and Remote Sensing 63, 12–31. Sims, D.A., Rahman, A.F., Cordova, V.D., El-Masri, B.Z., Baldocchi, D.D., Bolstad, P.V., Flanagan, L.B., Goldstein, A.H., Hollinger, D.Y., Misson, L., Monson, R.K., Oechel, W.C., Schmid, H.P., Wofsy, S.C., Liukang Xu, 2008. A new model of gross primary productivity for North American ecosystems based solely on the enhanced vegetation index and land surface temperature from MODIS. Remote Sensing of Environment 112, 1633–1646.

69

Susmitha, J., Sahai, A.K., Goswami, B.N., 2009. Eastward propagating MJO during boreal summer and Indian monsoon droughts. Climate Dynamics 32 (7–9), 1139–1153. Suyker, A.E., Verma, S.B., Burba, G.G., Arkebauer, T.J., 2005. Gross primary production and ecosystem respiration of irrigated maize and irrigated soybean during a growing season. Agriculture and Forest Meteorology 131 (3/4), 180–190. Trishchenko, A.P., Luo, Y., Khlopenkov, K.V., 2006. A method for downscaling MODIS land channels to 250 spatial resolution using adaptive regression and normalization. SPIE, paper No:6366-07, 1–8. Tucker, C.J., Pinzon, J.E., 2005. An extended AVHRR 8-km NDVI dataset compatible with MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing 26, 4485–4498. Turner, D.P., Ritts, W.D., Cohen, W.B., Maeirsperger, T.K., Gower, S.T., Kirschbaum, A.A., Running, S.W., Zhao, M.S., Wofsy, S.C., Dunn, A.L., Law, B.E., Campbell, J.L., Oechel, W.C., Kwon, H.J., Meyers, T.P., Small, E.E., Kurc, S.A., Gamon, J.A., 2005. Site-level evaluation of satellite-based global terrestrial gross primary production and net primary production monitoring. Global Change Biology 11 (4), 666–684. Villalobos, F.J., Fereres, E., 1990. Evaporation measurements beneath corn, cotton, and sunflower canopies. Agronomy Journal 82, 1153–1159. Wilson, K.B., Hansonb, P.J., Baldocchi, D.D., 2000. Factors controlling evaporation and energy partitioning beneath a deciduous forest over an annual cycle. Agricultural and Forest Meteorology 102, 83–103. Woodcock, C.E., Strahler, A.H., 1987. The factor of scale in remote sensing. Remote Sensing of Environment 21, 311–332. Yang, F., Ichii, K., White, M.A., Hashimoto, H., Michaelis, A.R., Votava, P., Zhu, A.X., Huete, A., Running, S., Nemani, R.R., 2007. Developing a continental-scale measure of gross primary production by combining MODIS and Ameriflux data through support vector machine. Remote Sensing of Environment 110, 109–122. Yocky, D.A., 1996. Multiresolution wavelet decomposition image merger of Landsat Thematic Mapper and SPOT panchromatic data. Photogrammetric Engineering and Remote Sensing 62, 1067–1074. 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 (2), 164–176.