Evaluating vegetation indices for assessing productivity...

© 2012 Science From Israel / LPPLtd., Jerusalem

IsraelJournalofPlantSciences Vol.60 2012 pp.123–133DOI: 10.1560/IJPS.60.1-2.123

*Author to whom correspondence should be addressed.E-mail: [email protected]

Evaluating vegetation indices for assessing productivity along a tropical rain forest chronosequence in Western Amazonia

Andrés ViñA

CenterforSystemsIntegrationandSustainability,DepartmentofFisheriesandWildlife,1405S.HarrisonRoad,Suite115ManlyMilesBldg.,MichiganStateUniversity,EastLansing,MI48823-5243,USA

(ReceivedAugust1,2011;acceptedinrevisedformOctober6,2011)

HonoringAnatolyGitelsonontheoccasionofhis70thbirthday

AbstrAct

Tropical deforestation is leading not only to losses of biodiversity but also to re-gional losses of vegetation productivity. However, in many areas the deforestation process is usually accompanied by a fast forest regeneration that produces a mosaic of forest patches in different successional stages. These successional stages have dif-ferent productivities owing primarily to differences in species composition and soil nutrients. In order to assess the long-term consequences of deforestation and forest regeneration on atmospheric carbon dynamics, it is imperative to analyze the spatio-temporal patterns of forest productivity along different successional stages. This study evaluates the suitability of five different vegetation indices for assessing pro-ductivity along a tropical rain forest chronosequence located in Western Amazonia. Among the indices tested, the Red Edge Chlorophyll Index (CIRed Edge) and the Wide Dynamic Range Vegetation Index (WDRVI) proved to be the most suitable for this purpose. However, due to the current low availability of multi-spectral imagery acquired by spaceborne sensors with a band in the red edge region of the electro-magnetic spectrum, the WDRVI serves as the most practical index. The dynamics of the regional productivity in the study area during the 2000s were assessed using the WDRVI and proved to be consistent with observed land cover dynamics.

Keywords: Chlorophyll indices, EVI, Hyperion, Landsat, NDVI, WDRVI

IntroductIon

With the current trends of change in the global climate induced by human modifications in the amount of greenhouse gases in the atmosphere (IPCC, 2007), it is critically important to analyze the spatio-temporal pat-terns of vegetation productivity from local to regional and global scales. This analysis requires developing techniques to delineate biomass accumulation and dis-tribution at multiple scales (DeFries et al., 2007; Gibbs et al., 2007). Different methods have been developed for analyzing vegetation productivity, including field mea-surements derived from gas exchange towers (Suyker et al., 2004; Verma et al., 2005), simulation models (Ruimy et al., 1999), and analysis of greenness patterns

derived from vegetation indices (Gitelson et al., 2006b; Garbulsky et al., 2008; Gitelson et al., 2008; Nakaji et al., 2008). Such procedures have been successfully ap-plied in many vegetation types across the globe, includ-ing croplands, grasslands, and forests. However, the spatio-temporal dynamics of the productivity of tropical humid forests are little understood, partly attributed to their overall weak seasonality and high plant species diversity (Ewusie, 1992; Huete et al., 2008). This is un-fortunate since tropical humid forests constitute around 56% of the forests worldwide, and are home to about 50% of all plant species on earth (Mayaux et al., 2005).

IsraelJournalofPlantSciences 60 2012

124

Tropical deforestation may not only lead to drastic regional decreases in precipitation (Walker et al., 2009) and overall changes in surface climate (McGuffie et al., 1995), but also to reductions in the productivity of vegetation at regional scales (Gibbs et al., 2007). However, forest regeneration in abandoned cropland and pastureland increases the aboveground primary pro-ductivity (Feldpausch et al., 2004), which may lead to overall increases in the amount of carbon sequestered by vegetation in deforested areas. Therefore, the changes in productivity due to tropical deforestation and subse-quent forest regeneration are highly dynamic and need to be comprehensively and synoptically analyzed, in order to evaluate the overall effects of deforestation and forest regeneration on global carbon budgets.

The purpose of this study was to evaluate the suit-ability of vegetation indices developed by Professor Anatoly A. Gitelson and his colleagues (Gitelson et al., 2003a,c, 2005; Gitelson, 2004; Viña et al., 2004b; Viña and Gitelson, 2005) for assessing the changes in productivity along a chronosequence of successional stages of a tropical rain forest. These vegetation in-dices have been successfully used for evaluating the productivity of agroecosystems but their suitability for assessing changes in productivity in other ecosystems, particularly humid forests in tropical regions, has not been fully evaluated. Two widely used vegetation indi-ces developed by other researchers were also included in the analysis, for comparison purposes.

MEthods

study area

The study was performed in an area comprising 3,197.9 km2 of the Amazon Basin (from 0°0′21″N to 0°27′25″N and 76°26′1″W to 76°59′56″W), with ca. 44% located in Colombia (Department of Putumayo) and ca. 56% in Ecuador (Province of Sucumbios). It includes parts of the San Miguel (international border), Putumayo, and Aguarico river basins (Fig. 1). Elevation ranges between 250 and 500 m. Mean annual rainfall and temperature are 3,528 mm and 25.3 °C, respectively (Puerto Asis weather station: 0°30′56″N, 76° 29′42″W, 260 m). While all months receive more than 200 mm of rainfall, two especially wet periods are distinguishable, one between April and June, and another in November. Ac-cording to these temperature and precipitation regimes, the vegetation of the study area is classified as Lowland Tropical Rain Forest in the Holdridge life zone system (Holdridge, 1967).

Since the study area is located within the upper Amazon region, it is considered to have a high diversity of tree species (Gentry, 1992). It also exhibits the high-

est diversity in Colombia of monkey species, some of which are endemic to the region (Hernández-Camacho et al., 1992). However, the region has been subjected to a drastic deforestation, with an overall annual rate of forest loss between 1985 and 1996 of 1.72% (Viña et al., 2004a). Nevertheless, forest regeneration is also conspicuous, particularly due to the practice of shifting cultivation (i.e., abandonment of some cropland areas after a few years of cropping in order to allow soil nutri-ent recovery through natural forest regeneration), which is prevalent throughout the study region (Viña et al., 2004a). Therefore, deforestation and shifting cultivation have transformed a once completely forested landscape into a mosaic of agricultural lands (e.g., cropland and pasture for cattle ranching), built-up/barren lands, and forests in different successional stages, ranging from early (2–3 m of canopy height) to mature (more than 20 m of canopy height).

Forest cover dynamics

Changes in forest cover of the study region over time were evaluated using land cover maps of 1973, 1985, and 1996 developed from Landsat MSS and TM im-agery in a previous study (Viña et al., 2004a). These maps were complemented with land cover maps of 2002 and 2008 developed in this study using Landsat ETM+ images acquired on October 14, 2002, September 12, 2008, and September 28, 2008 (obtained through the United States Geological Survey Global Visualization Viewer—http://glovis.usgs.gov). Several steps were performed to generate these maps. First, the level 1T digital number (DN) data of the reflective bands (1 to 5 and 7) of the Landsat ETM+ images were converted to at-sensor radiances and further to at-sensor reflectances following standard procedures (Chander et al., 2009). This procedure was also applied to the Landsat TM image of October 21, 1996 used in a previous study to generate a land cover map of 1996 (Viña et al., 2004a). Second, a linear regression technique was used to match the 2002 and 2008 ETM+ images (slaves) to the 1996 TM image (master). This regression technique was ap-plied separately for each individual band. Third, due to the malfunction of the scan-line corrector in the ETM+ sensor on May 31, 2003, the ETM+ images of 2008 exhibited gaps of missing data. The lacunae in the Sep-tember 12, 2008 image were filled using data from the September 28, 2008 image. Fourth, to classify the land cover in 2002 and 2008 without the need of ground-truth field data for calibration, the average spectral signatures of the different land cover types in the 1996 TM image were used to classify the 2002 and 2008 ETM+ images using a non-hierarchical supervised classification ap-proach based on maximum likelihood decision rules

Viña/Tropicalrainforestproductivity

125

Fig. 1. The study area covers 3197.9 km2, occupying portions of the Department of Putumayo, Colombia (44% of the study area) and the province of Sucumbios, Ecuador (56% of the study area).


126

(Jensen, 1996). Since no ground-truth field data were available for validating these maps, overall accuracy (Congalton, 1991) was assessed using ca. 50 data points randomly located within high spatial resolution images acquired on January 19, 2004 and October 27, 2006, and accessed through Google Earth (http://www.google.com/earth/).

The pixels under forest cover in the 2002 land cover map were sorted into a chronosequence of four different forest successional stages. These successional stages were assigned based on whether the pixels classified as forest cover in the 2002 image were also present in (and since) the 1973 land cover map (i.e., >30 years old), present in (and since) the 1985 land cover map (i.e., 18–30 years old), present in (and since) the 1996 land cover map (i.e., 7–17 years old), or only appeared in the 2002 land cover map (i.e., <7 years old).

comparison of the performance of vegetation indices along the forest chronosequence

Five different vegetation indices were used and com-pared for the assessment of productivity along the forest chronosequence (Table 1): the Normalized Difference Vegetation Index (NDVI), the Enhanced Vegetation In-dex (EVI), the Wide Dynamic Range Vegetation Index (WDRVI), and the Green and Red Edge Chlorophyll indices, CIGreen and CIRed Edge, respectively. The NDVI (Rouse et al., 1974, Tucker, 1979) has been by far the most widely used vegetation index for the analysis of vegetation dynamics. However, because the NDVI loses sensitivity under conditions of moderate to high green aboveground biomass (Myneni et al., 1997; Gitelson et al., 2003b,c; Gitelson, 2004), other vegetation indices

have been developed, and their use has increased over the last few years. The EVI is a soil- and atmosphere-resistant index specifically designed for the MODIS system (Huete et al., 1996, 1997), and has been suc-cessfully used for evaluating vegetation dynamics in high biomass systems such as the tropical humid forests (Xiao et al., 2006). The WDRVI constitutes a non-linear transformation of the NDVI specifically designed to increase the sensitivity to changes in vegetation when the NDVI saturates (Gitelson, 2004; Viña et al., 2004b). This index exhibits a linear relationship with the frac-tion of photosynthetically active radiation absorbed by vegetation (Viña and Gitelson, 2005) and has been used for analyzing the spatio-temporal heterogeneity of tropi-cal lowland forests (Aguilar-Amuchastegui and Hene-bry, 2006, 2008). The CIGreen and the CIRed Edge are semi-analytical surrogates of foliar (Gitelson et al., 2003a, 2006a) and canopy (Gitelson et al., 2005; Ciganda et al., 2008) chlorophyll content, and have been success-fully related with green leaf area index (Gitelson et al., 2003c; Viña et al., 2011) and gross primary productivity (Gitelson et al., 2006b, 2008).

The NDVI, EVI, WDRVI, and CIGreen were calculated using the 2002 Landsat ETM+ at-sensor reflectance data (Table 1). However, since the CIRed Edge requires a red edge band that is not available in the Landsat series, an image acquired on October 14, 2002 (i.e., same date as the 2002 Landsat ETM+ image acquisition) by the Hyperion system onboard the National Aeronautics and Space Administration’s EO-1 satellite was used for calculating the CIRed-edge. The Hyperion sensor collects data in 221 spectral bands (ca. 10 nm per band) within the range 356–2577 nm (Khurshid et al., 2006). The

Table 1Vegetation indices evaluated in the study

Index Formulation Reference

Normalized Difference Vegetation Index Rouse et al., 1974; Tucker, 1979

Enhanced Vegetation Index Huete et al., 1996, 1997

Wide-Dynamic Range Vegetation Index* Gitelson, 2004

Green Chlorophyll Index Gitelson et al., 2003a,c, 2005

Red Edge Chlorophyll Index Gitelson et al., 2003a,c, 2005

*The α coefficient used in this study was 0.25. This was determined using a heuristic procedure (Henebry et al., 2004).


127

swath path of the Hyperion is very narrow (ca. 7.5 km) compared to the width of a Landsat TM/ETM+ scene, but the spatial resolution (ca. 30 m/pixel) is similar to that of the Landsat TM/ETM+ sensor systems. The Hyperion image was also obtained through the United States Geological Survey Global Visualization Viewer (http://glovis.usgs.gov) and was received geometri-cally rectified and with pixel values already calibrated to radiance (level 1GST). Narrow bands 34 to 36 were averaged to obtain a broad red edge band (ca. 690–720 nm), while bands 40 to 55 were averaged to simulate the near-infrared band of the Landsat ETM+ system (ca. 750–900 nm) (Chen and Henebry, 2010).

Values from the five vegetation indices (i.e., NDVI, EVI, WDRVI, CIGreen, and CIRed Edge) were extracted from 3,000 pixels randomly distributed within the Hyperion image’s 7.5 km swath, which comprises about one-sixth of the study area. These values were used to assess the responses of different vegetation indices to forests in different successional stages, as well as to barren/built-up and agricultural areas. Pair-wise comparisons of vegetation index values among these different land cover types were performed using Bonferroni-corrected post-hoc Mann-Whitney U tests (SAS, 2004).

rEsults And dIscussIon

Forest cover dynamics

The land cover maps of 2002 and 2008 (Fig. 2) derived from the Landsat ETM+ imagery had an overall accu-racy of around 84%. Similar to the results obtained for the 1996 land cover map (Viña et al., 2004a) (Fig. 2), the agriculture class exhibited the lowest accuracies (ca. 72%), since it includes a wide array of land-cover categories, ranging from pastures to coffee plantations, with variable amounts of vegetative cover and differ-ent canopy architectures. In contrast, the forest cover class exhibited the highest classification accuracies (ca. 92%).

The annual rates of deforestation for the entire study area during the 1996–2002 period correspond to around 2.14% and reached a value of around 4.11% during the 2002–2008 period. Thus, deforestation rates during the 2000s were more than double those observed during the 1990s (Viña et al., 2004a). Consequently, the study region continues to exhibit a fast and dramatic loss of forest cover. However, forest regeneration is also conspicuous. Among the total forest area left in 2002 (ca. 1590 km2), ca. 13%, 7.5%, and 3% correspond to forests <7 years old, 7–17 years old, and 18–30 years old, respectively (Fig. 3). The rest corresponds to ma-ture forests (>30 years old) that have been present since

at least 1973 (Fig. 3) (Viña et al., 2004a). The larger amount of young secondary forest (<7 years old) pres-ent, compared to the older secondary stages (i.e., 7–17 and 18–30 years old) suggests that shifting cultivation is conspicuous and widespread across the entire study area, but secondary forests are seldom being allowed to reach maturity.

Analysis of productivity across the forest chronosequence

Relationships among the five different vegetation indi-ces studied were obtained using the 3,000 random pixels selected along the 7.5 km swath of the Hyperion image (Fig. 3). The NDVI, the EVI, and the WDRVI exhibited a non-linear relationship with the CIRed Edge, while the CIGreen exhibited an almost linear relationship (Fig. 4). These relationships show the saturation effect which in-dices like NDVI, EVI, and WDRVI experience at mod-erate-to-high green aboveground biomass values, as reported in several studies (Gitelson et al., 2003c, 2005, 2006b, 2008; Viña and Gitelson, 2005). However, the NDVI exhibited saturation at lower CIRed Edge values, fol-lowed by the EVI and the WDRVI (Fig. 4). In contrast, the almost linear relationship between the CIGreen and the CIRed Edge suggests that both of these indices are equally sensitive to changes in green biomass. Nevertheless, the CIGreen has been reported to be species-specific, while the CIRed Edge tends to be insensitive to different species (Gi-telson et al., 2005, 2006b; Viña et al., 2011). Therefore, the CIRed Edge seems more appropriate for analyzing the changes in productivity among different successional stages of a forest chronosequence.

With the exception of the EVI, all vegetation indi-ces evaluated exhibited a monotonic increase in their average values along the forest chronosequence (Figs. 5, 6). However, only the average values between the young (<7 years) and mature (>30 years) stages were significantly (p < 0.01) different. This pattern suggests that, with the exception of young vs. mature stages, forests in different successional stages have a wide vari-ability in their productivities, deeming them difficult to isolate using vegetation indices alone. It has been reported that differences in productivity among forests of different ages and/or in different successional stages are primarily driven by differences in soil nutrients (Feldpausch et al., 2004). In addition, all indices exhib-ited no significant differences between agricultural land and young secondary forests (<7 years old), with the only exception being the EVI in which the agricultural land exhibited significantly higher values (Figs. 5, 6). Therefore, the productivity of agricultural land seems to be comparable to that of young secondary forests, and


128

Fig. 2. Panels on the left (A, C, E) correspond to false color composites of the satellite images used in the study. Band com-binations (in the red, green, and blue color planes) correspond to bands 7, 4, 3 of the Landsat TM/ETM+ sensor systems, and bands 50 (around 850 nm), 33 (around 680 nm), and 20 (around 550 nm) of the Hyperion sensor system (i.e., image comprising a 7.5 km swath in panel C). Panels on the right (B, D, F) correspond to land cover classifications obtained from the Landsat TM/ETM+ images.


129

Fig. 3. Forest cover of the study area in 2002 sorted into four successional stages, from young (<7 years old) to mature (>30 years old). This forest chronosequence was obtained from land cover classifications of a Landsat MSS image acquired in 1973 (Viña et al., 2004a), Landsat TM imagery acquired in 1985 and 1996 (Viña et al., 2004a), and a Landsat ETM+ image acquired in 2002 (this study). The black dots along a 7.5 km swath correspond to 3,000 pixels randomly selected for analyzing vegetation index data obtained from the Landsat ETM+ and Hyperion images of 2002 (see Fig. 2C).

Fig. 4. Relationships among the vegetation indices evaluated in the study. Colored symbols correspond to the vegetation index values in the 3,000 pixels randomly selected (see Fig. 3). Black lines correspond to best-fit regression lines.


130

Fig. 5. Average values of the (A) Normalized Difference Vegetation Index, NDVI, (B) Enhanced Vegetation Index, EVI, and (C) Wide Dynamic Range Vegetation Index, WDRVI, in different land cover types, derived from 3,000 randomly selected pixels (see Fig. 3) of the Landsat ETM+ image of 2002. Land cover types with different letters ex-hibit significantly (p < 0.01) different average vegetation index values, as determined by Bonferroni-corrected post-hoc Mann–Whitney U tests. Error bars correspond to 2 standard errors from the mean.

Fig. 6. Average values of the (A) Green Chlorophyll Index, CIGreen, and (B) Red Edge Chlorophyll Index, CIRed Edge, in dif-ferent land cover types, derived from 3,000 randomly selected pixels (see Fig. 3) of the Landsat ETM+ image of 2002 and the Hyperion image of 2002, respectively. Land cover types with different letters exhibit significantly (p < 0.01) different aver-age vegetation index values, as determined by Bonferroni-cor-rected post-hoc Mann–Whitney U tests. Error bars correspond to 2 standard errors from the mean.

thus difficult to be isolated using the vegetation indices evaluated. However, the WDRVI and the CIRed Edge were the only indices capable of distinguishing between agri-cultural land and intermediate successional stages of the

forest (7–17 and 18–30 years old forests), as intermedi-ate forests exhibited significantly higher values than agricultural areas (Figs. 5, 6). Thus, these two indices (among the five evaluated) seem to perform better for analyzing the changes in productivity along the tropical forest chronosequence. The WDRVI was also identified to be suitable for analyzing the productivity of crops us-ing Landsat ETM+ data (Gitelson et al., 2008).

Due to its proven accuracy for assessing the spatio-temporal changes in vegetation productivity, as well as for being insensitive to different leaf structures and canopy architectures (characteristic of different spe-cies), the CIRed Edge constitutes the index of choice for as-sessing vegetation productivity (Gitelson et al., 2006b). Preference for this index is also supported in this study,


131

Fig. 7. Frequency distribution of WDRVI values (derived from the Landsat TM image of 1996 and the Landsat ETM+ images of 2002 and 2008) among the pixels (30 m/pixel) comprising the entire study area (see Fig. 1). Frequency is expressed as total area. The secondary X-axis represents the transformation of the WDRVI scale into a Gross Primary Productivity (GPP) scale (i.e., grams of carbon, per square meter, per day), using the coefficients reported by (Gitelson et al., 2008).

Fig. 8. Pixel-based dynamics (i.e., losses or gains) of the WDRVI in the study area derived from the Landsat TM image of 1996 and the Landsat ETM+ images of 2002 and 2008. (A) Difference between WDRVI values of 2002 and 1996. (B) Difference between WDRVI values of 2008 and 2002. Black lines in (A) and (B) correspond to the international border between Colombia (North) and Ecuador (South).

in which this index was suitable for assessing productiv-ity across a tropical forest chronosequence. However, the CIRed Edge requires a red edge band that is not available in any current operational satellite sensor system, with the exception of the European Space Agency’s Medium Resolution Imaging Spectroradiometer (MERIS) and the EO-1 Hyperion. The MERIS system acquires data with a high temporal resolution and with an almost global coverage. However, the spatial resolution of MERIS data is relatively coarse (ca. 300 m/pixel over land). The Hyperion acquires data with a higher spatial resolution (ca. 30 m/pixel), but data acquisition is “on-demand”, thus has a very low temporal resolution and a low spatial coverage. Therefore, the WDRVI is the pre-ferred index, since it can be obtained using all current operational satellite sensor systems, ranging from those with high spatial resolution but low temporal resolution and limited areal coverage (e.g., Ikonos and Quickbird), to those with high temporal resolution and global or near-global coverage but with coarse spatial resolutions (e.g., MODIS, MERIS, AVHRR, VIIRS).

Application of the WdrVI to analyze regional changes in productivity

As shown in Fig. 2, the study area exhibited a drastic conversion of forest cover to other land cover types dur-ing the 2000s. Thus, it was hypothesized that these loss-es of forest cover induced drastic changes in the regional productivity of the study area. The WDRVI was used to analyze these changes in productivity during the 2000s

by using the Landsat TM image of 1996 and the Landsat ETM+ images of 2002 and 2008. A severe reduction in the WDRVI was observed between the 1996 and the 2002, which translates into drastic reductions in the re-gional productivity (Fig. 7). Nevertheless, the changes in WDRVI were not as drastic between 2002 and 2008, suggesting that the study area might be stabilizing in terms of vegetation productivity, despite continuous and drastic forest cover losses. Comparison of the dynam-ics of WDRVI values between 1996, 2002, and 2008 at pixel level reveals that, during the 1996–2002 period, the Colombian side of the study area exhibited wide-


132

spread reductions in WDRVI values (Fig. 8A). These reductions were mostly associated with the conversion of forest cover to barren lands (Fig. 2). In contrast, the Ecuadorian side exhibited overall increases in WDRVI values (Fig. 8A). However, during the 2002–2008 an opposite trend occurred in which the Ecuadorian side experienced mostly reductions in WDRVI values, while the Colombian side exhibited mostly gains (Fig. 8B). This inversion in the spatial dynamics of WDRVI is mainly associated with the conversion of barren land into agricultural land observed in the Colombian side, while the Ecuadorian side drastically increased its areas under built-up/barren (Fig. 2). Therefore, the dynamics of the regional productivity of the study area during the 2000s, assessed with the WDRVI, were consistent with observed land cover dynamics.

concludIng rEMArks

This study evaluated the use of different vegetation indices for the analysis of the spatio-temporal changes in productivity of tropical rain forests exposed to dras-tic human-induced dynamics. Results showed that the WDRVI and the CIRed Edge, two vegetation indices previ-ously identified as suitable for analyzing crop productiv-ity, seem to also constitute the best indices for analyzing the productivity of tropical humid forests. However, due to issues of data availability, the WDRVI constitutes the most feasible to be used in an operational way, as it can be applied to a wide range of sensor systems, including the Landsat series.

While the study reports significant results, it is neces-sary to underline that it was based entirely on remotely sensed data. Therefore, further assessments are required to fully support the conclusions obtained, particularly using ground-level vegetation productivity measures.

AcknoWlEdgMEnts

I am extremely grateful to Professor Anatoly A. Gi-telson for a decade-long friendship and professional relationship that has helped me grow both as a profes-sional and as a person. I also thank Geoffrey Henebry and an anonymous reviewer for their useful comments to the manuscript. The study was produced using data obtained through the Global Visualization Viewer (http://glovis.usgs.gov), developed and maintained by the United States Geological Survey (USGS).

rEFErEncEs

Aguilar-Amuchastegui, N., Henebry, G.M. 2006. Monitoring sustainability in tropical forests: how changes in canopy

spatial pattern can indicate forest stands for biodiversity surveys. IEEE Geosci. Remote Sens. Lett. 3: 329–333.

Aguilar-Amuchastegui, N., Henebry, G.M. 2008. Character-izing tropical forest spatio-temporal heterogeneity using the Wide Dynamic Range Vegetation Index (WDRVI). Int. J. Remote Sens. 29: 7285–7291.

Chander, G., Markham, B.L., Helder, D.L. 2009. Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sens. Environ. 113: 893–903.

Chen, W.R., Henebry, G.M. 2010. Spatio-spectral heteroge-neity analysis using EO-1 Hyperion imagery. Comput. Geosci. 36: 167–170.

Ciganda, V., Gitelson, A., Schepers, J. 2008. Vertical profile and temporal variation of chlorophyll in maize canopy: quantitative “crop vigor” indicator by means of reflec-tance-based techniques. Agron. J. 100: 1409–1417.

Congalton, R.G. 1991. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens. Environ. 37: 35–46.

DeFries, R., Achard, F., Brown, S., Herold, M., Murdiyarso, D., Schlamadinger, B., de Souza, C. 2007. Earth observa-tions for estimating greenhouse gas emissions from defor-estation in developing countries. Environ. Sci. Policy 10: 385–394.

Ewusie, J.Y. 1992. Phenology in tropical ecology. Accra: Ghana Universities Press.

Feldpausch, T.R., Rondon, M.A., Fernandes, E.C.M., Riha, S.J., Wandelli, E. 2004. Carbon and nutrient accumulation in secondary forests regenerating on pastures in central Amazonia. Ecol. Appl. 14: S164–S176.

Garbulsky, M.F., Penuelas, J., Papale, D., Filella, I. 2008. Remote estimation of carbon dioxide uptake by a Mediter-ranean forest. Glob. Change Biol. 14: 2860–2867.

Gentry, A.H. 1992. Tropical forest biodiversity: distributional patterns and their conservational significance. Oikos 63: 19–28.

Gibbs, H.K., Brown, S., Niles, J.O., Foley, J.A. 2007. Moni-toring and estimating tropical forest carbon stocks: making REDD a reality. Environ. Res. Lett. 2:DOI:10.1088/1748-9326/1082/1084/045023.

Gitelson, A.A. 2004. Wide dynamic range vegetation index for remote quantification of biophysical characteristics of vegetation. J. Plant Physiol. 161: 165–173.

Gitelson, A.A., Gritz, Y., Merzlyak, M.N. 2003a. Relation-ships between leaf chlorophyll content and spectral re-flectance and algorithms for non-destructive chlorophyll assessment in higher plant leaves. J. Plant Physiol. 160: 271–282.

Gitelson, A.A., Verma, S.B., Viña, A., Rundquist, D.C., Key-dan, G., Leavitt, B., Arkebauer, T.J., Burba, G.G., Suyker, A.E. 2003b. Novel technique for remote estimation of CO2 flux in maize. Geophys. Res. Lett. 30:486, doi:410.1029/2002GL016543.

Gitelson, A.A., Viña, A., Arkebauer, T.J., Rundquist, D.C., Keydan, G., Leavitt, B. 2003c. Remote estimation of leaf area index and green leaf biomass in maize canopies. Geo-phys. Res. Lett. 30:1248, doi:1210.1029/2002GL016450.


133

Gitelson, A.A., Viña, A., Ciganda, V., Rundquist, D.C., Arkebauer, T.J. 2005. Remote estimation of canopy chlo-rophyll content in crops. Geophys. Res. Lett. 32:L08403, doi:08410.01029/02005GL022688.

Gitelson, A.A., Keydan, G.P., Merzlyak, M.N. 2006a. Three-band model for noninvasive estimation of chlorophyll, ca-rotenoids, and anthocyanin contents in higher plant leaves. Geophys. Res. Lett. 33:DOI: 10.1029/2006GL026457.

Gitelson, A.A., Viña, A., Verma, S.B., Rundquist, D.C., Arkebauer, T.J., Keydan, G., Leavitt, B., Ciganda, V., Burba, G.G., Suyker, A.E. 2006b. Relationship between gross primary production and chlorophyll content in crops: implications for the synoptic monitoring of vegeta-tion productivity. J. Geophys. Res.-Atmos. 111:D08S11, doi:10.1029/2005JD006017.

Gitelson, A.A., Viña, A., Masek, J.G., Verma, S.B., Suyker, A.E. 2008. Synoptic monitoring of gross primary produc-tivity of maize using Landsat data. IEEE Geosci. Remote Sens. Lett. 5: 133–137.

Henebry, G.M., Viña, A., Gitelson, A.A. 2004. The Wide Dy-namic Range Vegetation Index and its potential utility for Gap Analysis. Gap Analysis Program Bulletin 12: 50–56.

Hernández-Camacho, J.A., Hurtado-Guerra, A., Ortiz-Qui-jano, R., Walschburger, T. 1992. Unidades biogeográficas de Colombia. In: Halffter, G., ed. La diversidad biológica de Iberoamérica. Instituto de Ecologia, Mexico D.F., Mexico.

Holdridge, L.R. 1967. Life zone ecology. Tropical Science Center, San José, Costa Rica.

Huete, A.R., Justice, C., van Leeuwen, W. 1996. MODIS vegetation index (MOD13). Algorithm Theoretical Basis Document. Version 2. NASA Goddard Space Flight Cen-ter, Greenbelt, Maryland 20771. USA.

Huete, A.R., Liu, H.Q., Batchily, K., vanLeeuwen, W. 1997. A comparison of vegetation indices global set of TM images for EOS-MODIS. Remote Sens. Environ. 59: 440–451.

Huete, A.R., Restrepo-Coupe, N., Ratana, P., Didan, K., Sales-ka, S.R., Ichii, K., Panuthai, S., Gamo, M. 2008. Multiple site tower flux and remote sensing comparisons of tropical forest dynamics in Monsoon Asia. Agr. Forest Meteorol. 148: 748–760.

IPCC. 2007. Climate Change 2007: The Physical Science Ba-sis. Cambridge University Press, New York, NY, USA.

Jensen, J.R. 1996. Introductory digital image processing, a remote sensing perspective. Second edition. Prentice Hall., New Jersey: .

Khurshid, K.S., Staenz, K., Sun, L.X., Neville, R., White, H.P., Bannari, A., Champagne, C.M., Hitchcock, R. 2006. Preprocessing of EO-1 Hyperion data. Can. J. Remote Sens. 32: 84–97.

Mayaux, P., Holmgren, P., Achard, F., Eva, H., Stibig, H., Branthomme, A. 2005. Tropical forest cover change in the 1990s and options for future monitoring. Philos. T. Roy.Soc. B. 360: 373–384.

McGuffie, K., Henderson-Sellers, A., Zhang, H., Durbridge, T.B., Pitman, A.J. 1995. Global climate sensitivity to tropi-cal deforestation. Global and Planetary Change 10: 97–128.

Myneni, R.B., Nemani, R.R., Running, S.W. 1997. Estima-

tion of global leaf area index and absorbed PAR using radiative transfer models. IEEE Trans. Geosci. Electron. 35: 1380–1393.

Nakaji, T., Ide, R., Takagi, K., Kosugi, Y., Ohkubo, S., Na-sahara, K.N., Saigusa, N., Oguma, H. 2008. Utility of spectral vegetation indices for estimation of light conver-sion efficiency in coniferous forests in Japan. Agr. Forest Meteorol. 148: 776–787.

Rouse, J.W., Haas Jr., R.H., Schell, J.A., Deering, D.W. 1974. Monitoring vegetation systems in the Great Plains with ERTS. pp 309–317 In: Third ERTS-1 Symposium. NASA, Washington, DC.

Ruimy, A., Kergoat, L., Bondeau, A., Intercomparison, P.P. N.M. 1999. Comparing global models of terrestrial net primary productivity (NPP): analysis of differences in light absorption and light-use efficiency. Glob. Change Biol. 5: 56–64.

SAS. 2004. SAS/STAT User’s Guide Version 9.1. SAS Insti-tute Inc., Cary, NC, USA.

Suyker, A.E., Verma, S.B., Burba, G.G., Arkebauer, T.J., Wal-ters, D.T., Hubbard, K.G. 2004. Growing season carbon dioxide exchange in irrigated and rainfed maize. Agr. For-est Meteorol. 124: 1–13.

Tucker, C.J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8: 127–150.

Verma, S.B., Dobermann, A., Cassman, K.G., Walters, D.T., Knops, J.M., Arkebauer, T.J., Suyker, A.E., Burba, G.G., Amos, B., Yang, H.S., Ginting, D., Hubbard, K.G., Gitel-son, A.A., Walter-Shea, E.A. 2005. Annual carbon dioxide exchange in irrigated and rainfed maize-based agroecosys-tems. Agri. Forest Meteorol. 131: 77–96.

Viña, A., Gitelson, A.A. 2005. New developments in the remote estimation of the fraction of absorbed photosyn-thetically active radiation in crops. Geophys. Res. Lett. 32:L17403.

Viña, A., Echavarria, F.R., Rundquist, D.C. 2004a. Satel-lite change detection analysis of deforestation rates and patterns along the Colombia-Ecuador border. Ambio 33: 118–125.

Viña, A., Henebry, G.M., Gitelson, A.A. 2004b. Satellite mon-itoring of vegetation dynamics: Sensitivity enhancement by the wide dynamic range vegetation index. Geophys. Res. Lett. 31:L04503.

Viña, A., Gitelson, A.A., Nguy-Robertson, A.L., Peng, Y. 2011. Comparison of different vegetation indices for the remote assessment of green leaf area index of crops. Re-mote Sens. Environ.: DOI:10.1016/j.rse.2011.1008.1010.

Walker, R., Moore, N.J., Arima, E., Perz, S., Simmons, C., Caldas, M., Vergara, D., Bohrer, C. 2009. Protecting the Amazon with protected areas. Proc. Natl. Acad. Sci. U.S.A. 106: 10582–10586.

Xiao, X.M., Hagen, S., Zhang, Q.Y., Keller, M., Moore, B. 2006. Detecting leaf phenology of seasonally moist tropi-cal forests in South America with multi-temporal MODIS images. Remote Sens. Environ. 103: 465–473.

Evaluating vegetation indices for assessing productivity...

Documents

Transcript of Evaluating vegetation indices for assessing productivity...