Next Article in Journal
Detection of Coal Fire Dynamics and Propagation Direction from Multi-Temporal Nighttime Landsat SWIR and TIR Data: A Case Study on the Rujigou Coalfield, Northwest (NW) China
Previous Article in Journal
Remotely Sensed Monitoring of Small Reservoir Dynamics: A Bayesian Approach

2014, 6(2), 1211-1233;

The Generalized Difference Vegetation Index (GDVI) for Dryland Characterization
ICARDA (International Center for Agricultural Research in the Dry Areas), CGIAR, P.O. Box 950764, Amman 11195, Jordan
This paper is an extension version of the paper presented in the 8th International Soil Sciences Congress (ISSC), Izmir, Turkey, 15–17 May 2012.
Received: 3 November 2013; in revised form: 24 December 2013 / Accepted: 2 January 2014 / Published: 29 January 2014


A large number of vegetation indices have been developed and widely applied in terrestrial ecosystem research in the recent decades. However, a certain limitation was observed while applying these indices in research in dry areas due to their low sensitivity to low vegetation cover. In this context, the objectives of this study are to develop a new vegetation index, namely, the Generalized Difference Vegetation Index (GDVI), and to examine its applicability to the assessment of dryland environment. Based on the field investigation and crop Leaf Area Index (LAI) measurement, five spring and summer Landsat TM and ETM+ images in the frame with Path/Row number of 174/35, and MODIS (Moderate Resolution Imaging Spectroradiometer) LAI and vegetation indices (VIs) data (MOD15A2 and MOD13Q1), of the same acquisition dates as the Landsat images, were acquired and employed in this study. The results reveal that, despite the same level of correlation with the fractional vegetation cover (FVC) as other VIs, GDVI shows a better correlation with LAI and has higher sensitivity and dynamic range in the low vegetal land cover than other vegetation indices, e.g., the range of GDVI is higher than Normalized Difference Vegetation Index (NDVI),Soil-Adjusted Vegetation Index (SAVI), Enhanced Vegetation Index (EVI), Wide Dynamic Range Vegetation Index (WDRVI), and Soil-Adjusted and Atmospherically Resistant Vegetation Index (SARVI), by 164%–326% in woodland, 185%–720% in olive plantation, and 190%–867% in rangeland. It is, hence, concluded that GDVI is relevant for, and has great potential in, land characterization, as well as land degradation/desertification assessment in dryland environment.
GDVI; LAI; vegetation indices; sensitivity; dynamic range; land characterization; dryland environment

1. Introduction

Spectral vegetation indices (VIs) are mathematical combinations of different spectral bands, mostly in the visible and near infrared regions of the electromagnetic spectrum [1,2], and are optical measures of vegetation canopy greenness, a composite property of leaf chlorophyll, leaf area, canopy cover, and canopy architecture [3]. Since the development of the Simple Ratio Index (SR) [46] and the Normalized Difference Vegetation Index (NDVI) [1,7], a number of vegetation indices, such as the Transformed Vegetation Index (TVI) [8], Perpendicular Vegetation Index (PVI) [9], Weighted Difference Vegetation Index (WDVI) [10,11], Non-Linear Vegetation Index (NLI) [12]), Green Red Vegetation Index (GRVI) [13,14], and Wide Dynamic Range Vegetation Index (WDRVI) [15], have been proposed. In addition, a certain number of indices taking the atmospheric effects and/or soil influence into account, namely, the Global Environmental Monitoring Index (GEMI) [16], Soil-Adjusted Vegetation Index (SAVI) [17], Transformed SAVI (TSAVI) [18], Modified SAVI (MSAVI) [19], Optimized SAVI (OSAVI) [20], Generalized SAVI (GeSAVI) [21], Green SAVI (GSAVI) [22], Modified Non-Linear Vegetation Index (MNLI) [23], Atmospherically Resistant Vegetation Index (ARVI), and Soil-Adjusted and Atmospherically Resistant Vegetation Index (SARVI) [24], SARVI2 [25], and later renamed as the Enhanced Vegetation Index (EVI) [26], and its two-band version (EVI2) [3], Green Atmospherically Resistant Index (GARI) [27], Visible Atmospherically Resistant Index (VARI) [13], etc., have been developed; and those frequently applied are listed in Table 1.
As well as these two-band and three-band indices, Kauth and Thomas [28], and Crist and Cicone [29,30], have developed the Tasseled Cap (TC) Transformation respectively for Landsat MSS (Multispectral Scanner System) and TM (Thematic Mapper) data. Recently, Huang et al. [31] have produced the TC coefficients for the ETM+ (Enhanced Thematic Mapper Plus) data. TC Transformation can provide different thematic information of land surface, namely, the Brightness (TCB), Greenness (TCG), and Wetness (TCW). The TCG is, in fact, an indicator of vegetation vigor or greenness, and was termed the Green Vegetation Index, a multiple linear spectral combination, e.g., four bands for MSS and six bands for TM and ETM+ data.
In terms of their development rationale, all of these indices, either ratio-based (SR, NDVI, SAVI, OSAVI, MSAVI, ARVI, GRVI, EVI, EVI2, etc.) or in the form of linear combination (PVI, WDVI, TVI, and TCG), were formulated on the basis of the strong contrast in reflectance between the near infrared (NIR) and red (R) bands as vegetation has strong reflection of the incident radiation in NIR band while a strong absorption in R band [1,15]. Considering the limitation or shortcoming of the most frequently applied NDVI, e.g., its saturation in densely vegetated area and its sensitivity to atmospheric effects and soil influence [13,15,17,2426,32], a number of improved variants, such as SAVI, OSAVI, ARVI, SARVI, EVI, EVI2, and MNLI, have been derived to avoid the mentioned limitation by introducing either an adjustment factor (e.g., L in SAVI, SARVI, EVI, MNLI, EVI2), based on the soil line information or the blue band to correct the aerosol effect (e.g., ARVI, SARVI, and EVI) or both the blue band and L (e.g., SARVI and EVI) to reduce both soil influence and atmospheric effects. All these vegetation indices have been widely applied and contributing to the ecosystem research and characterization by remote sensing, e.g., land cover classification, biomass and net primary production (NPP) quantification, crop yield estimation, land degradation monitoring, soil mapping, vegetation-climate interaction assessment, etc.
The main biophysical character of the dryland environment is its low vegetation cover (despite that it may be locally dense, e.g., in low-lying lands, riparian plains) due to aridity. According to the United Nations (UN) [33], the aridity index, the ratio of precipitation to potential evapotranspiration (denoted as P/PET), of drylands, is between 0.05 and 0.65 (<550–600 mm of annual rainfall). Hence, such aridity (or water scarcity) affects food security, and, at the same time, makes land susceptible to degradation [33]. For this reason, a number of authors have conducted studies on monitoring of the vegetation condition and its change trend of and the impacts of anthropogenic activity on the vulnerable dryland environment either by using albedo [34,35] or by VIs, such as NDVI in combination with rainfall [3640].
However, it was noted that even though some of these VIs have taken the soil and atmospheric influences into consideration, they are mostly sensitive to moderately and densely vegetated land cover, such as forests and croplands rather than to low vegetal drylands. Taking SAVI as an instance, Qi et al. [19] noted that, after introduction of the adjustment factor, L, the dynamic range of SAVI is reduced, which is not favorable for study in sparsely vegetated areas. In fact, not only SAVI, but also other indices, such as NDVI, EVI, and ARVI have low dynamic range in these areas. While conducting an assessment on the effectiveness of sand-control and land degradation in the Ordos Desert rangelands in China, using Landsat imagery, it came to the attention of the authors [40] that it was almost impossible to discern the subtle increase in greenness in the newly controlled areas (e.g., <4–5 years) based on the available vegetation indices due to their low dynamic range and insensitivity. Such insensitivity constitutes the limitation of these indices when applied to dryland research. Therefore, there is a need to develop another new vegetation index relevant for such low vegetation dryland environment.
The main objectives of this paper are to propose a new vegetation index and investigate its relevance for, sensitivity to, and applicability in low vegetation land cover in arid and semi-arid environment. A test site, located in the northwest of Syria, was selected to demonstrate the development and evaluation of the new index.

2. Materials and Methods

2.1. Rationale

While other authors use soil line information in the NIR–R space to derive vegetation indices [3,911,17,18,20,21,25,32], this paper attempts to generate a new index from the already popularized and frequently applied NDVI and SR. As Pinty and Verstraete [16] and Jackson and Huete [41] revealed, NDVI and SR can be converted from each other, that is,
NDVI = ( SR 1 ) / ( SR + 1 )
where SR = ρNIR/ρR, ρNIR and ρR are respectively reflectance of the NIR and R bands. Clearly, Equation (1) can be considered as a function of NDVI with respect to the variable SR, that is, any change in SR will lead to change in NDVI. However, for a given land cover, SR is fixed to certain extent and so is NDVI. If one wants to augment the value, i.e., the dynamic range of NDVI of the given land cover type, the only possibility is to increase SR through certain operation, e.g., power operation, in the form of SRn (n is an integer >0). A simple experiment using Landsat images reveals that through power operation (when n = 2, 3, and 4, for example), SRn is to a varying degree amplified depending on land cover types and power number n (see Table 2). This implies that power operation on SR can lead to an amplification of the dynamic range of the new derivatives of NDVI (Table 2, when n > 1). To facilitate the description, we term the new derivatives of NDVI the Generalized Difference Vegetation Index (GDVI), and mathematically, Equation (1) can be, thus, generalized as:
GDVI n = ( SR n 1 ) / ( SR n + 1 ) = ( ρ NIR n ρ R n ) / ( ρ NIR n + ρ R n )
where n is power, an integer of the values of 1, 2, 3, 4... n. The global dynamic range of GDVI is the same as NDVI, from −1 to 1; and, when n = 1, GDVI = NDVI. For facilitating the description, GDVI is denoted respectively GDVI^2, GDVI^3, and GDVI^4 when n = 2, 3, and 4.
In order to understand the characteristics of the new index, GDVI was projected against its origin, NDVI, taking the same images from which Table 2 was derived, as an example. The projection is shown in Figure 1. Two features of GDVI can be easily observed: (1) amplification of the dynamic range in the low and moderate vegetation cover with respect to NDVI, e.g., for a typical low vegetal rangeland/desert, NDVI is 0.2 but GDVI^2, GDVI^3, and GDVI^4 are respectively about 0.38, 0.54, and 0.67, a clear increase in the dynamic range (h in Figure 1a, 1b and 1c); and (2) saturation of the high power GDVI in the densely vegetated areas, e.g., GDVI^3 and GDVI^4 get saturated, respectively, when NDVI is >0.65 and >0.5 (the S point in Figure 1b,c).
As Goel and Qin [12], Pinty and Verstraete [16], and Gong et al. [23] have explained, the relationships between different surface biophysical features often tend to be non-linear. Thus, the objective of the non-linear operation is to linearize the relationships between the biophysical parameters [23]. However, the non-linear power operation in GDVI is not for such a linearization but for an enlargement of the contrast between the reflection in NIR (vegetation signal) and absorption in R bands. This is the reason that GDVI has a wider dynamic range in the low vegetation biome than NDVI and other ratio-based indices.

2.2. Test Site

To test and demonstrate the applicability and sensitivity of GDVI, an area located in northwestern Syria including a part of the governorates of Aleppo, Idleb, and Lattakia, covered by the Landsat frame/scene with Path/Row number 174/35 was selected (Figure 2). The landform in the test site is coastal slope and mountains with peak altitude of 1,500–1,710 m in the west, depression/valley in the middle (about 160–170 m in altitude), and plateau/plain with interspersed hills with an altitude range of 310–550 m (locally reaching 870 m) in the east. The area is dominated by Mediterranean climate with mean annual rainfall from 600 mm in the northwest to 200 mm in the southeast. The precipitation is mostly concentrated in winter and spring. These landscape and climatic characteristics constrain largely the diversity of land cover types and land use in agriculture, e.g., forest (mainly coniferous) and maquis (Mediterranean shrubland) in the mountainous chain, and on the slope, in the west, irrigated and rainfed croplands (wheat, barley, and vegetables) in spring, and irrigated cropland (e.g., cotton, maize, sesame, sunflower, watermelon) in summer in the valley and plateau, and rangeland/bare land in the low rainfall areas in the east. Permanent tree crops (mainly perennial olive tree crops, citrus/orchard including lemon, orange, fig, cherry, peach, and other fruit trees) are widely grown in rainfed areas crossing from the coastal slope to the depression/valley and then to the plateau in the east. The test site is one of the most important agricultural production areas in Syria, especially, known for its products of olive and wheat.

2.3. Data

Five spring and summer Landsat images, and MODIS LAI and EVI/NDVI including the reflectance of the B (blue), R, NIR, and MIR (middle infrared) bands corresponding to the acquisition dates of Landsat images were obtained and are listed in Table 3.
MODIS LAI was produced based on vegetation index, such as NDVI, directional spectral reflectance and radiative transfer models proposed by Myneni et al. [42], Knyazikhin et al. [43], and Myneni et al. [44]. The LAI products of collections 3 and 4 (C003 and C004) were generally found to have an overestimate compared to in situ observations [45,46], but the C005 product seems much more realistic, especially, for coniferous forest [47]. Land use/cover data were obtained, based on four field investigations, in the spring of 2007, 2009, 2010, and 2011, in the test site.
A limited number of LAI measurements, using AccuPAR Ceptometer (LP-80, Decagon Devices, Inc.) for wheat under supplemental irrigation and rainfed barley, were conducted in the campus of ICARDA (International Center for Agricultural Research in the Dry Areas), near Aleppo in Syria (see Figure 2 for location), during the peak growing period of wheat and barley from 10 to 15 April 2010, and from 21 to 31 March 2011.
As experiment fields in the campus, cultivation of wheat, barley, and other crops, such as lentil, faba bean, oat, and coriander were interleaved from each other and often different from year to year for the same parcel of land due to rotation practices. The size of different cropland patches varies from 0.1 to 3 ha depending on the slope and location.
The AccuPAR readings were made in 12 plots (1 × 1 m2 in size) distributed, respectively, in two parcels of wheat land where crop performance was visually distinguished good (two plots), intermediate (two plots), and relatively poor (two plots) in each parcel (<2 ha in size). Another 12 plots were measured in two tracts of rainfed barley (for harvesting), based on the same sampling scheme as wheat (six plots for each tract). Some sporadic measurements (five plots in total) were also made in irrigated barley, which is not common in Syria. All these measurements were undertaken during noon-time, from 11:00 am to 1:30 pm, while it was sunny (cloud-free) in the spring 2010, and repeated in wheat and rainfed barley lands in 2011, but in different locations due to crop rotation.
It was noted that the maximum MODIS LAI of the ICARDA Campus dated 1 May 2007 was only 1.07–1.1, much lower than the measured values, 2.6 for irrigated wheat, 5.4 and 1.9, respectively, for irrigated and rainfed barley. It was, hence, considered better to use the average values of the measured LAI of the irrigated wheat (2.01) and rainfed barley for harvesting (1.46) to represent these two croplands in spring images for this study. However, there were no field measurement for other biomes, such as forest, woodland, and rangeland. To use MODIS LAI for other biomes in the successive calibration is the only choice despite of its coarse resolution.

2.4. Evaluation Procedure

As Gitelson [15], Gong et al. [23] and Richardson et al. [48], have demonstrated, to test the reliability and the efficiency of a vegetation index, a common approach is to calibrate the index with LAI crossing different biomes to ascertain their relationship.
Carlson and Ripley [49] reported that the scaled NDVI, i.e., (NDVI-NDVI0)/(NDVIS-NDVI0), where NDVI0 and NDVIS are, respectively, the NDVI values of bare soil and full vegetation cover (100%), is out of the influence of atmospheric effects. They recommended using its square to represent the fractional vegetation cover (FVC) for land surface studies. Gutman and Ignatov [50] proposed to use directly the scaled NDVI as FVC; and this was followed by Jiménez-Muñoz et al. [51], Zhang et al. [52], and Kallel et al. [53]. One of the advantages of FVC lies in its simplicity to be derived from satellite images, e.g., from NDVI [49,50,52], or from Spectral Mixing Analysis (SMA) [51], or from isoline parameterization in the NIR-R space [53].
Based on the above understanding, both LAI and FVC were used for evaluation of the relevance of GDVI. Moreover, Sirikul [54] found that the relation between the field measured LAI and MODIS NDVI/EVI is poor in semi-arid areas and broadleaf forest biomes. It will be interesting for us to examine the relationships between MODIS GDVI/NDVI/EVI with MODIS LAI for the observed biomes in the test area. The procedure to assess the relevance, sensitivity and applicability of GDVI is demonstrated as follows:

2.4.1. Atmospheric Correction

This correction was conducted for Landsat images using FLAASH (Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes) Model [55,56], which was developed based on the MODTRAN (MODerate resolution atmospheric TRANsmission) theory by the Air Force Research Laboratory, USA. This model can correct both additive and multiplicative atmospheric effects. The FLAASH reflectance of each band was rescaled to normal reflectance range from 0 to 1.

2.4.2. Calculation of Different Vegetation Indices and FVC

A number of vegetation indices namely NDVI, EVI, SAVI, OSAVI, SARVI, VARI, WDRVI, NLI, MNLI, and GDVI (n = 2, 3 and 4) were transformed from the multispectral bands. As Wu et al. [57] noted, most of the pixel values of VARI and WDRVI (e.g., a = 0.1 and 0.2) were negative in woodland and rangeland. Similarly, it was found that not only VARI and WDRVI, but also NLI and MNLI (L = 0.5 for intermediate vegetation cover in semi-arid areas) have negative values in almost all biomes except for wheat cropland. For this reason, only the Landsat image dated 1 May 2007, was used to derive these indices for comparison.
FVC [5053] was also derived from the NDVI of Landsat image, dated 1 May 2007, after fixing the values of NDVI0 (=0.11) and NDVIS (=0.96).
For MODIS LAI (eight-day interval), taking that of 1 May 2007 as an instance, three dates of LAI, DOY (Day of Year) 113 (24 April), 121 (1 May) and 129 (9 May), were combined to compose a multi-band image. Then, a multi-band maximization algorithm [57] was applied to get the maximum value of each pixel from the three LAI images in order to have a cloud-free or cloud influence minimized LAI image.
For MODIS NDVI/EVI (16-day interval), only two dates (113 and 129) of data were available for 1 May. After combination, the same procedure as for MODIS LAI was followed to derive the cloud-free NDVI, EVI data for 1 May (121).
Apart from these two indices, the cloud-free GDVI^2 was also produced from the NIR and R reflectance bands of the same dataset as NDVI and EVI (MOD13Q1).

2.4.3. Selection of Sampling Areas

Based on the field investigation, the identified land cover types are forest, maquis (Mediterranean shrubland), woodland, spring croplands (irrigated wheat, rainfed barley for harvesting, rainfed barley for grazing), summer irrigated croplands (maize, cotton, and water melon), residue of spring crop after harvesting, permanent tree croplands (citrus, orchard, and olive plantation), rangelands, stone pits/mining sites, fallow, and bareland. Sampling areas in the form of polygons were first randomly defined in each land cover class in the color composites of Landsat images in making reference to the ground observation and Google Earth. They were defined as many as possible for each biome in order to have enough spatial representativeness, e.g., >3%–5% of the whole land cover class. Then, these polygons were adjusted and reshaped against MODIS LAI to avoid mixed and bad pixels (underestimation or no data) in MODIS LAI data. The adjusted polygons were considered as the sampling areas for the Landsat-MODIS image pair. The location of the polygons of the same biome may be slightly different from image pair to image pair due to the difference in acquisition dates, land use change (between spring and summer, and among different years), and due to mixed or bad pixels in MODIS LAI data. More concretely, each Landsat-MODIS pair has its own sampling areas for each land cover class.
The sampling areas take up about 3.6% on average in each biome class. The location and pixel number of the sampling areas before resizing (see next section) are shown in Figure 3 and Table 4 taking the image pair of 1 May 2007, as an example.

2.4.4. Calculation of the Mean Values and the Random Point Values of Different VIs in the Sampling Areas

In order to have more comparability between Landsat and MODIS data, a pixel resizing from 30 m to 250 m for Landsat VIs and from 1,000 m to 250 m for MODIS LAI was conducted using nearest neighbor resampling. After importation of these images into ArcGIS, 890 random points were generated in the defined polygons in which samples were assigned to each land cover type, based on its importance in the test site (see Table 4); and the point values and biome mean values of all VIs including GDVI derived from different Landsat images and cloud-free LAI, NDVI, GDVI, and EVI from MODIS data were calculated for each land cover class, namely forest, maquis, woodland, rangeland, rainfed barley for grazing, and bare soil for the spring image pairs. In the summer image pairs, irrigated crops (cotton, watermelon, and sunflower) and spring crop residue left after harvesting were included, instead of wheat, rainfed barley for harvesting and grazing.
The mean FVC of 1 May 2007 was calculated for all biomes without pixel resizing as it will be directly calibrated with the VIs from the same Landsat images.

2.4.5. Calibration

Landsat VIs vs. MODIS LAI: The studies conducted by Myneni et al. [42], Richardson et al. [48], Price and Bausch [58], and Gitelson et al. [59] revealed allometric relations between LAI and other vegetation indices, such as NDVI and WDRVI, which can be calibrated by either logarithmic or exponential or polynomial function. To compare the results, this calibration was conducted in two ways: (1) Random samples of LAI (from MODIS LAI without integration of field measured LAI) were linked with their corresponding VIs across all different land cover types; (2) the mean LAI values of different land cover were coupled with the mean VIs (using the measured mean LAI for wheat and rainfed barley). Both calibrations were conducted by simple linear regression model (least-square) at the confidence level of 95% after the conversion of LAI into different variants such as ln(LAI) and exp(LAI).
Landsat VIs vs. Landsat FVC: The mean VIs and GDVI of Landsat images dated 1 May 2007 were also coupled with the mean FVC and its exponential and logarithmic variants across all land cover types to check whether GDVI is better correlated with FVC than other indices.
MODIS VIs vs. MODIS LAI: The mean MODIS NDVI, EVI, and GDVI^2 were linked to the mean LAI from MOD15A2 across different land cover to check whether GDVI has better correlation with LAI than that of NDVI and EVI.

2.4.6. Sensitivity Analysis

Gitelson [15] has evaluated the sensitivity of WDRVI vs. NDVI. By following his approach, the sensitivity of the new vegetation index, GDVI, against other referenced VI can be quantitatively analyzed using the following expression:
S r ( GDVI ) = [ d ( GDVI ) / d ( VI ) ] × [ Δ ( GDVI ) / Δ ( VI ) ] 1
where Sr(GDVI) is the relative sensitivity of GDVI vs. the referenced VI (such as NDVI, EVI, SAVI and SARVI); d(GDVI)/d(VI) are the first derivative of GDVI with respect to VI or the infinitesimal change in GDVI corresponding to that of VI; and Δ(GDVI) = GDVImaxGDVImin and Δ(VI) = VImaxVImin are the ranges of GDVI and VI, i.e., the difference between the maximum and minimum GDVI and VI values of the same land cover observed in the image. In operation, d(GDVI) and d(VI), the tiny change in GDVI and VI can be obtained by measuring the GDVI and VI in the adjacent pixels, which have very tiny differences in values in the same land cover type.
According to Gitelson [15], if Sr > 1, the observed index, in our case, GDVI, is more sensitive than the referenced VI; if Sr = 1, they have the same sensitivity; and if Sr < 1, GDVI is less sensitive than the referenced VI.

3. Results and Discussions

Following the above procedures, the most relevant relationships either in logarithmic or exponential function between vegetation indices and LAI of the observed five pairs of images were established. It was noted that logarithmic functions, i.e., VI–ln (LAI), have much higher correlation coefficients than those of exponential ones; so the former are listed in Tables 5 and 6. Those of the image pairs of 12 July 2010 and 1 May 2007 are presented in Figures 4 and 5; and the values of vegetation indices of some typical dryland biomes are presented in Table 7.

3.1. GDVI vs. LAI

3.1.1. Advantages

Better correlation with LAI than other indices: It is evident that both random point-based (Table 5) and biome mean-based (Table 6, Figures 4 and 5) calibrations illustrate that GDVI^2 is better correlated with LAI than any other index, such as NDVI, EVI, SAVI, OSAVI, SARVI, NLI, MNLI, and WDRVI across all land cover types. GDVI^3 and GDVI^4 are more or less at the same level as NDVI and SARVI, depending on images. Moreover, the correlation with LAI resulted from biome mean-based calibration (1 May 2007, in Table 6) is higher than that from random point-based calibration (Table 5), probably resulting from the sample mixing among different biomes in the latter while modeling was conducted.
As mentioned earlier, despite of the consideration of the “non-linear” factor in WDRVI (i.e., a × ρNIR), NLI and MNLI (i.e., square of ρNIR), their values in most dryland biomes are negative (Figure 5c and Table 7). These indices are more applicable to study of densely vegetated areas, for example, forests and croplands, rather than dryland areas, as one may not be used to seeing all negative values of a vegetation index crossing all land cover types.
Amplification of the dynamic range: As already discerned in Section 2.1, this advantage of GDVI, i.e., amplification of the dynamic range in the low vegetated areas is obvious. As revealed in Table 7, the values of GDVI are magnified respectively by 164%–326%, 185%–720%, 190%–867%, and 196%–2,088% for the biomes namely woodland, olive plantation, rangeland, and bareland, in comparison with NDVI, SAVI, EVI, SARVI, NLI, MNLI, and WDRVI. This implies that GDVI gains more dynamic range for low vegetal biomes than other vegetation indices.

3.1.2. Sensitivity

Based on the measurements, such as tiny change, maximum, and minimum values of VIs and GDVI in the image dated 1 May 2007, the relative sensitivity of GDVI vs. other VIs across different biomes were calculated and are presented in Figure 6.
It is easily seen in Figure 6 that, for the biomes wheat and forest, the relative sensitivity, Sr(GDVI), with respect to other vegetation indices is <1, implying that all power GDVI is less sensitive than other vegetation indices in the densely vegetated biomes. For rainfed barley, GDVI^2 has more or less similar sensitivity to others (Figure 6a). However, for the biomes rangeland, bareland, and olive groves, Sr(GDVI) >1, indicating that GDVI has higher sensitivity than other vegetation indices in the sparsely vegetated areas.

3.2. GDVI vs. FVC

Together with other vegetation indices, GDVI was checked against FVC across all land cover types, and the results are listed in Table 8.
It is seen that, among the VI-FVC relationships, GDVI^2 is at the same level of correlation as NDVI, SARVI, EVI, WDRVI, among which, OSAVI is the best. As for VI-ln(FVC), GDVI^2 is the same as NDVI but GDVI^4 is the best. Among the VI-exp(FVC) functions, WDRVI, show better correlation than GDVI.


The evaluation of MODIS GDVI, EVI and NDVI vs. MODIS LAI (without integration of the field measurement), taking again the cloud-free MODIS data of 1 May 2007 as an example, reveals that all three vegetation indices are well correlated with the logarithmic variant of MODIS LAI if all observed land cover types are taken into account; and NDVI is the best one followed by GDVI^2 (Figure 7). MODIS NDVI and EVI are poorly correlated with the field measured LAI [53], but MODIS GDVI, NDVI, and EVI are highly correlated with MODIS LAI, especially NDVI and GDVI.

3.4. Discussion on Limitation and Problems Encountered

Despite of the advantages, the limitation of GDVI is also clear. Apart from the saturation, as raised in Section 2.1 and Table 7, GDVI has low sensitivity in the densely vegetated biomes, e.g., forest, maquis, irrigated croplands (e.g., wheat), as shown in Figure 6. The applicability of GDVI of different power is summarized in Table 9.
Problems were encountered in this study. Due to limited field measurements, MODIS LAI data have to be used for the non-measured biomes. As concerned by some authors [4547,53], the quality of MODIS LAI eight-day product relies heavily on the weather condition in subtropical areas; especially, it is difficult to have completely cloud-free condition during the crop-growing period in spring. For this reason, underestimation may arise. Another factor influences the quality of MODIS LAI data lies in, perhaps, its biome mapping [44]. If a higher-resolution biome map is available, we believe that more relevant MODIS LAI product will be made available. In this study, the multi-date LAI maximization procedure can minimize, more or less, the problem related to cloud cover or bad pixels but cannot remove that resulted from biome mapping. Another shortcoming of MODIS LAI for such calibration is its coarse resolution. Clearly, it is not ideal to match VIs of 30 m resolution to LAI of 1,000 m resolution. Both degradation of 30 m VIs to 1000 m resolution and upgrading of 1000 m LAI to 30 m resolution will lead to either a loss or an unrealistic improvement in quality of the data. A feasible compromise is to degrade 30 m data to 250 m, and to upgrade 1,000 m LAI to 250 m as done in this study. It is worthy of mentioning that, for such study involving multi-resolution and multi-source data, selection of a reasonable resampling approach is also important. It is known that both bilinear and cubic convolution resampling will change the original value of pixel due to an averaging of the observed pixel with its adjacent pixels. Thus, the best choice is to use the nearest neighbor resampling, in which the original values of pixels can be kept after pixel resizing.
Moreover, limited field measurements do not allow us to assess whether field LAI is better than MODIS LAI for such calibration. The equations in Table 5 were obtained from 890 random samples but their correlation is lower than that of biome mean-based equations presented in Table 6, which are a result of the mixed LAI (field measured ones were used for wheat and rainfed barley while MODIS LAI for other land cover types). With the increase of the field measurement, field LAI-based calibration will be made possible. This can be a follow-up topic of research in the future.
FVC is a choice for calibration. As it is derived directly from NDVI and has a good linear correlation with most of the vegetation indices; and GDVI shows a better correlation with it only in logarithmic function.

4. Conclusions

This paper presented the development of the new vegetation index, GDVI, and explored its relevance for and applicability in dryland ecosystem research. Although more additional work, e.g., calibration and verification with more field measurements in different areas, is still needed, based on the conducted random sample-based and biome mean-based calibrations, we can already see that the proposed index has its uniqueness different from and advantages over other vegetation indices, i.e., higher sensitivity and dynamic range in the low vegetal biomes in spite of its disadvantage and limitation (e.g., low sensitivity and saturation) in densely vegetated areas. Additionally, GDVI^2 seems able to catch better LAI than other VIs across different land cover types. Broadly speaking, GDVI^2 is mainly operational for land characterization in dryland biomes but, if applied to forest, shrubland, and cropland studies, care has to be taken due to its low sensitivity to these biomes. The higher power GDVI (n = 3, 4) is only applicable in sparsely vegetated areas, e.g., for monitoring land degradation and desertification in dryland ecosystems. After an extensive validation, GDVI can serve as a complement of the other vegetation indices for research in the dry areas. In the future, it is also possible to integrate GDVI into MODIS products and serve for dryland system research.
In this study, atmospheric effects have been considered in image preprocessing; however, soil influence has not been yet. As well as calibration in different dry areas, future work should take this correction into account as either done by Wang et al. [60] by applying the FFT (Fast Fourier Transform) to remove the background noise, or by Huete [17], Rondeaux et al. [20], and Gong et al. [23], by using a self-correction factor in different magnitude adapted to the power of GDVI.


Four Landsat images dated 27 March 2003, 21 August 2007, 12 July 2010, 29 August 2010, and all MODIS data (MOD15A2, MOD13Q1) were freely obtained from the USGS data server (

Conflicts of Interest

The author declares no conflict of interest.


  1. Tucker, C.J. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 1979, 8, 127–150. [Google Scholar]
  2. Vina, A.; Gitelson, A.A.; Nguy-Robertson, A.L.; Peng, Y. Comparison of different vegetation indices for the remote assessment of green leaf area index of crops. Remote Sens. Environ. 2011, 115, 3468–3478. [Google Scholar]
  3. Jiang, Z.; Huete, A.R.; Didan, K.; Miura, T. Development of a two-band enhanced vegetation index without a blue band. Remote Sens. Environ. 2008, 112, 3833–3845. [Google Scholar]
  4. Birth, G.S.; McVey, G.R. Measuring the color of growing turf with a reflectance spectrophotometer. Agron. J. 1968, 60, 640–643. [Google Scholar]
  5. Jordan, C.F. Derivation of leaf-area index from quality of light on the forest floor. Ecology 1969, 50, 663–666. [Google Scholar]
  6. Knipling, E.B. Physical and physiological bases for the reference of visible and near infrared radiation from vegetation. Remote Sens. Environ. 1970, 1, 155–159. [Google Scholar]
  7. Rouse, J.W.; Haas, R.H.; Schell, J.A.; Deering, D.W. Monitoring Vegetation Systems in the Great plains with ERTS. Proceedings of the 3rd ERTS-1 Symposium NASA SP-351, Washington, DC, USA, 10–14 December 1973; 1, pp. 309–317.
  8. Deering, D.W.; Rouse, J.W.; Haas, R.H.; Schel, J.A. Measuring Forage Production of Grazing Units from Landsat MSS Data. Proceedings of the 10th International Symposium of Remote Sensing of Environment, Ann Arbor, MI, USA, 6–10 October 1975; II, pp. 1169–1178.
  9. Richardson, A.J.; Wiegand, C.L. Distinguishing vegetation from soil background information. Photogramm. Eng. Remote Sens. 1977, 43, 1541–1552. [Google Scholar]
  10. Clevers, J.G.P.W. The derivation of a simplified reflectance model for the estimation of leaf area index. Remote Sens. Environ. 1988, 35, 53–70. [Google Scholar]
  11. Clevers, J.G.P.W. Application of the WDVI in estimating LAI at the generative stage of barley. ISPRS J. Photogramm. Remote Sens. 1991, 46, 37–47. [Google Scholar]
  12. Goel, N.S.; Qin, W. Influences of canopy architecture on relationships between various vegetation indices and LAI and FPAR: A computer simulation. Remote Sens. Rev. 1994, 10, 309–347. [Google Scholar]
  13. Gitelson, A.A.; Kaufman, Y.J.; Stark, R.; Rundquist, D. Novel algorithms for remote estimation of vegetation fraction. Remote Sens. Environ. 2002, 80, 76–87. [Google Scholar]
  14. Motohka, T.; Nasahara, K.N.; Oguma, H.; Tsuchida, S. Applicability of green-red vegetation index for remote sensing of vegetation phenology. Remote Sens. 2010, 2, 2369–2387. [Google Scholar]
  15. Gitelson, A.A. Wide dynamic range vegetation index for remote quantification of crop biophysical characteristics. J. Plant Physiol. 2004, 161, 165–173. [Google Scholar]
  16. Pinty, B.; Verstraete, M.M. GEMI: A non-linear index to monitor global vegetation from satellites. Plant Ecol. 1992, 101, 15–20. [Google Scholar]
  17. Huete, A.R. A soil adjusted vegetation index (SAVI). Remote Sens. Environ. 1988, 25, 295–309. [Google Scholar]
  18. Baret, F.; Guyot, G.; Major, D. TSAVI: A Vegetation Index which Minimizes Soil Brightness Effects on LAI and APAR Estimation. Proceedings of the 12th International Canadian Symposium Geoscience and Remote Sensing (IGARSS’89), Vancouver, BC, Canada, 10–14 July 1989; pp. 1355–1358.
  19. Qi, J.; Chehbouni, A.; Huete, A.R.; Kerr, Y.H.; Sorooshian, S. A modified soil adjusted vegetation index. Remote Sens. Environ. 1994, 48, 119–126. [Google Scholar]
  20. Rondeaux, G.; Steven, M.; Baret, F. Optimization of soil-adjusted vegetation index. Remote Sens. Environ. 1996, 55, 95–107. [Google Scholar]
  21. Gilabert, M.A.; Gonzalez-Piqueras, J.; Garcia-Haro, F.J.; Melia, J. A generalized soil-adjusted vegetation index. Remote Sens. Environ. 2002, 82, 303–310. [Google Scholar]
  22. Sripada, R.P.; Heiniger, R.W.; White, J.G.; Meijer, A.D. Aerial color infrared photography for determining early in-season nitrogen requirements in corn. Agron. J. 2006, 98, 968–977. [Google Scholar]
  23. Gong, P.; Pu, R.; Biging, G.S.; Larrieu, M.R. Estimation of forest leaf area index using vegetation indices derived from Hyperion hyperspectral data. IEEE Trans. Geosci. Remote Sens. 2003, 41, 1355–1362. [Google Scholar]
  24. Kaufman, Y.J.; Tanré, D. Atmospherically resistant vegetation index (ARVI) for EOS-MODIS. IEEE Trans. Geosci. Remote Sens. 1992, 30, 261–270. [Google Scholar]
  25. Huete, A.R.; Liu, H.Q.; Batchily, K.; van Leeuwen, W. A comparison of vegetation indices global set of TM images for EOS-MODIS. Remote Sens. Environ. 1997, 59, 440–451. [Google Scholar]
  26. Huete, A.; Didan, K.; Miura, T.; Rodriguez, E.P.; Gao, X.; Ferreira, L.G. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 2002, 83, 195–213. [Google Scholar]
  27. Gitelson, A.; Kaufman, Y.; Merzlyak, M. Use of a green channel in remote sensing of global vegetation from EOS-MODIS. Remote Sens. Environ. 1996, 58, 289–298. [Google Scholar]
  28. Kauth, R.J.; Thomas, G.S. The Tasseled Cap—A Graphical Description of the Spectral-Temporal Development of Agricultural Crops as Seen by Landsat. Proceedings of the Symposium on Machine Processing of Remotely Sensed Data, Purdue University, West Lafayette, IN, USA, 29 June–1 July 1976; pp. 4B41–4B51.
  29. Crist, E.P.; Cicone, R.C. Application of the tasseled cap concept to simulated thematic mapper data. Photogramm. Eng. Remote Sens. 1984, 50, 343–352. [Google Scholar]
  30. Crist, E.P.; Cicone, R.C. A Physically-based transformation of thematic mapper data—The TM tasseled cap. IEEE Trans. Geosci. Remote Sens. 1984, 22, 256–263. [Google Scholar]
  31. Huang, C.; Wylie, B.; Yang, L.; Homer, C.; Zylstra, G. Derivation of a tasseled cap transformation based on Landsat 7 at-satellite reflectance. Int. J. Remote Sens. 2002, 23, 1741–1748. [Google Scholar]
  32. Baret, F.; Jacquemoud, S.; Hanocq, J.F. The soil line concept in remote sensing. Remote Sens. Rev. 1993, 7, 65–82. [Google Scholar]
  33. United Nations (UN). Global Drylands: A UN System-Wide Response, 2011. Available online: (accessed on 8 February 2012).
  34. Courel, M.-F.; Kandel, R.S.; Rasool, S.I. Surface albedo and the Sahel drought. Nature 1984, 307, 528–531. [Google Scholar]
  35. Otterman, J.; Tucker, C.J. Satellite measurements of surface albedo and temperatures in semi-desert. J. Appl. Meteorol. 1985, 24, 228–234. [Google Scholar]
  36. Malo, A.R.; Nicholson, S.E. A study of rainfall and vegetation dynamics in the African Sahel using Normalized Difference Vegetation Index. J. Arid Environ. 1990, 19, 1–24. [Google Scholar]
  37. Tucker, C.J.; Dregne, H.E.; Newcomb, W.W. Expansion and contraction of Sahara Desert from 1980 to 1990. Science 1991, 253, 299–301. [Google Scholar]
  38. Evans, J.; Geerken, R. Discrimination between climate and human-induced dryland degradation. J. Arid Environ. 2004, 57, 535–554. [Google Scholar]
  39. Herrmann, S.M.; Anyamba, A.; Tucker, C.J. Recent trends in vegetation dynamics in the African Sahel and their relationship to climate. Glob. Environ. Chang. 2005, 15, 394–404. [Google Scholar]
  40. Wu, W.; de Pauw, E.; Zucca, C. Use remote sensing to assess impacts of land management policies in the Ordos rangelands in China. Int. J. Digit. Earth 2013, 6, 81–102. [Google Scholar]
  41. Jackson, R.D.; Huete, A.R. Interpreting vegetation indices. Prev. Vet. Med 1991, 11, 185–200. [Google Scholar]
  42. Myneni, R.B.; Nemani, R.R.; Running, S.W. Estimation of global leaf area index and absorbed par using radiative transfer models. IEEE Trans. Geosci. Remote Sens 1997, 35, 1380–1893. [Google Scholar]
  43. Knyazikhin, Y.; Martonchik, J.V.; Myneni, R.B.; Diner, D.J.; Running, S.W. Synergistic algorithm for estimating vegetation canopy leaf area index and fraction of absorbed photosynthetically active radiation from MODIS and MISR data. J. Geophys. Res. 1998, 103, 32257–32276. [Google Scholar]
  44. Myneni, R.B.; Hoffman, S.; Knyazikhin, Y.; Privette, J.L.; Glassy, J.; Tian, Y.; Wang, Y.; Song, X.; Zhang, Y.; Smith, G.R.; et al. Global products of vegetation leaf area and fraction absorbed PAR from year one of MODIS data. Remote Sens. Environ. 2002, 83, 214–231. [Google Scholar]
  45. Fang, H.; Liang, S. A hybrid inversion method for mapping leaf area index from MODIS data: Experiments and application to broadleaf and needle leaf canopies. Remote Sens. Environ. 2005, 94, 405–424. [Google Scholar]
  46. Leuning, R.; Cleugh, H.A.; Zeglin, S.J.; Hughes, D. Carbon and water fluxes over a temperate Eucalyptus forest and tropical wet/dry savanna in Australia: Measurements and comparison with MODIS remote sensing estimates. Agric. For. Meteorol. 2005, 129, 151–173. [Google Scholar]
  47. De Kauwe, M.G.; Disney, M.I.; Quaife, T.; Lewis, P.; Williams, M. An assessment of the MODIS collection 5 leaf area index product for a region of mixed coniferous forest. Remote Sens. Environ. 2011, 115, 767–780. [Google Scholar]
  48. Richardson, A.J.; Wiegand, C.L.; Wanjura, D.F.; Dusek, D.; Steiner, J.L. Multisite analysis of spectral-biophysical data for sorghum. Remote Sens. Environ. 1992, 47, 71–82. [Google Scholar]
  49. Carlson, T.N.; Ripley, D.A. On the relation between NDVI, fractional vegetation cover, and leaf area index. Remote Sens. Environ. 1997, 62, 241–251. [Google Scholar]
  50. Gutman, G.; Ignatov, A. The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. Int. J. Remote Sens 1998, 19, 1533–1543. [Google Scholar]
  51. Jiménez-Muñoz, J.C.; Sobrino, J.A.; Plaza, A.; Guanter, L.; Moreno, J.; Martínez, P. Comparison between fractional vegetation cover retrievals from vegetation indices and spectral mixture analysis: Case study of PROBA/CHRIS data over an agricultural area. Sensors 2009, 9, 768–793. [Google Scholar]
  52. Zhang, X.; Liao, C.; Li, J.; Sun, Q. Fractional vegetation cover estimation in arid and semi-arid environments using HJ-1 satellite hyperspectral data. Int. J. Appl. Earth Obs. Geoinf. 2012. [Google Scholar] [CrossRef]
  53. Kallel, A.; le Hégarat-Mascle, S.; Ottlé, C.; Hubert-Moy, L. Determination of vegetation cover fraction by inversion of a four-parameter model based on isoline parametrization. Remote Sens. Environ. 2007, 111, 553–566. [Google Scholar]
  54. Sirikul, N. Comparison of MODIS Vegetation Index Products with Biophysical and Flux Tower Measurements. Ph.D. Dissertation. University of Arizona: Tucson, AZ, USA, 2006; 197. [Google Scholar]
  55. Matthew, M.W.; Adler-Golden, S.M.; Berk, A.; Richtsmeier, S.C.; Levine, R.Y.; Bernstein, L.S.; Acharya, P.K.; Anderson, G.P.; Felde, G.W.; Hoke, M.P.; et al. Status of atmospheric correction using a MODTRAN4-based algorithm. Proc. SPIE 2000, 4049, 199–207. [Google Scholar]
  56. Perkins, T.; Adler-Golden, S.; Matthew, M.; Berk, A.; Anderson, G.; Gardner, J.; Felde, G. Retrieval of Atmospheric Properties from Hyper and Multispectral Imagery with the FLAASH Atmospheric Correction Algorithm. Proc. SPIE 2005, 5979. [Google Scholar] [CrossRef]
  57. Wu, W.; de Pauw, E.; Hellden, U. Assessing woody biomass in African tropical savannas by multiscale remote sensing. Int. J. Remote Sens. 2013, 34, 4525–4549. [Google Scholar]
  58. Price, J.C.; Bausch, W.C. Leaf area index estimation from visible and near-infrared reflectance data. Remote Sens. Environ. 1995, 52, 55–65. [Google Scholar]
  59. Gitelson, A.A.; Wardlow, B.D.; Keydan, G.P.; Leavitt, B. An evaluation of MODIS 250-m data for green LAI estimation in crops. Geophys. Res. Lett. 2007, 34, L20403. [Google Scholar]
  60. Wang, Q.; Adiku, S.; Tenhunen, J.; Granier, A. On the relationship of NDVI with leaf area index in a deciduous forest site. Remote Sens. Environ. 2005, 94, 244–255. [Google Scholar]
Figure 1. Characteristics of GDVI vs. NDVI ((a) GDVI^2 vs. NDVI, (b) GDVI^3 vs. NDVI and (c) GDVI^4 vs. NDVI).
Figure 1. Characteristics of GDVI vs. NDVI ((a) GDVI^2 vs. NDVI, (b) GDVI^3 vs. NDVI and (c) GDVI^4 vs. NDVI).
Remotesensing 06 01211f1
Figure 2. Location of the test area.
Figure 2. Location of the test area.
Remotesensing 06 01211f2
Figure 3. Distribution of the sampling areas taking the Landsat-MODIS image pair of 1 May 2007 as an example.
Figure 3. Distribution of the sampling areas taking the Landsat-MODIS image pair of 1 May 2007 as an example.
Remotesensing 06 01211f3
Figure 4. Logarithmic relationships between vegetation indices and LAI (image pair dated 12 July 2010).
Figure 4. Logarithmic relationships between vegetation indices and LAI (image pair dated 12 July 2010).
Remotesensing 06 01211f4
Figure 5. Logarithmic relationships between vegetation indices and LAI (image pair dated 1 May 2007. (a) GDVI vs. LAI, (b) VIs with positive range vs. LAI, and (c) VIs with negative range vs. LAI).
Figure 5. Logarithmic relationships between vegetation indices and LAI (image pair dated 1 May 2007. (a) GDVI vs. LAI, (b) VIs with positive range vs. LAI, and (c) VIs with negative range vs. LAI).
Remotesensing 06 01211f5aRemotesensing 06 01211f5b
Figure 6. Sensitivity of GDVI vs. other vegetation indices for different biomes (image dated 1 May 2007. (a) GDVI^2 vs. VIs, (b) GDVI^3 vs. VIs, and (c) GDVI^4 vs. VIs).
Figure 6. Sensitivity of GDVI vs. other vegetation indices for different biomes (image dated 1 May 2007. (a) GDVI^2 vs. VIs, (b) GDVI^3 vs. VIs, and (c) GDVI^4 vs. VIs).
Remotesensing 06 01211f6aRemotesensing 06 01211f6b
Figure 7. Relationships between MODIS Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI) and GDVI and MODIS LAI (1 May 2007).
Figure 7. Relationships between MODIS Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI) and GDVI and MODIS LAI (1 May 2007).
Remotesensing 06 01211f7
Table 1. Frequently applied vegetation indices developed in the past decades.
Table 1. Frequently applied vegetation indices developed in the past decades.
IndexFormulaFull NameReferences
SRρNIRRSimple Ratio Index[46]
NDVI(ρNIRρR)/(ρNIR + ρR)Normalized Difference Vegetation Index[1,7]
PVIsin(α)(ρNIR) − cos(α)(ρR)
α = the angle between the soil line and NIR axis
Perpendicular Vegetation Index[9]
TVI(NDVI + 0.5)1/2Transformed Vegetation Index[8]
a = the slope of the soil line
Weighted Difference Vegetation Index[10,11]
SAVI(1 + L)(ρNIRρR)/(ρNIR + ρR + L)
Low vegetation, L = 1, intermediate, 0.5, and high 0.25
Soil-Adjusted Vegetation Index[17]
TSAVIa(ρNIRR − b)/[(NIR + ρR − ab + X(1 + a2)]
a = slope of the soil line, b = soil line intercept, and X = adjustment factor to minimize soil noise
Transformed Soil Adjusted Vegetation Index[18]
OSAVI(ρNIRρR)/(ρNIR + ρR + 0.16)Optimized Soil-Adjusted Vegetation Index[20]
ρRB = ρR − γ × (ρR − ρB)
γ = 1, ρB = reflectance of blue band
Atmospherically Resistant Vegetation Index[24]
SARVI(1 + L)(ρNIRρRB)/(ρNIR + ρRB + L)
ρRB is the same as that in ARVI, L is a correction factor similar to those of SAVI
Soil Adjusted and Atmospherically Resistant Vegetation Index[24]
SARVI2 or EVIG × ((ρNIRρR)/(ρNIR + C1 × ρR − C2 × ρB + L))
ρB = reflectance of blue band, G = 2.5, C1 = 6, C2 =7.5 and L = 1
Soil Adjusted and Atmospherically Resistant Vegetation Index 2 or Enhanced Vegetation Index[25,26]
EVI22.5(ρNIRρR)/(ρNIR + 2.4ρR + L)
L = 1
Enhanced Vegetation Index 2[3]
( ρ NIR 2 ρ R ) / ( ρ NIR 2 + ρ R )
Non-Linear Vegetation Index[12]
( 1 + L ) ( ρ NIR 2 ρ R ) / ( ρ NIR 2 + ρ R + L )

L is a correction factor similar to those of SAVI
Modified Non-linear Vegetation Index[23]
VARI(ρGρR)/(ρG + ρRρR)
ρG = reflectance of the green band
Visible Atmospherically Resistant Index[13]
WDRVI(a × ρNIRρR)/(a × ρNIR + ρR)
a = 0.05–1, usually, 0.1–0.2
Wide Dynamic Range Vegetation Index[15]
Table 2. Atmospherically corrected and reflectance-based Simple Ratio Index (SR) and its derivatives (n > 1).
Table 2. Atmospherically corrected and reflectance-based Simple Ratio Index (SR) and its derivatives (n > 1).
NWheatForestsWoodlandsOlive PlantationsRangelandsBare Soil

Table 3. Satellite data used in the study.
Table 3. Satellite data used in the study.
Acquisition Date of Landsat ImagesMODIS Production Date (Including Day Of Year (DOY))
(Path/Row: 174/35, 30 m)LAI (C005)
(MOD15A2, 8-day, 1,000 m)
(MOD13Q1, 16-day, 250 m)
27 March 2003 (ETM+)22 March (81), and 29 March (89), 200381 (22 March), 97 (7 April)
1 May 2007 (TM)23 April (113), 1 May (121) and 9 May (129), 2007113 (23 April), 129 (09 May)
21 August 2007 (TM)13 August (225), 21 August (233), and 29 August (241), 2007225 (13 August), 241 (29 August)
12 July 2010 (TM)4 July (185), 12 July (193), and 20 July (201), 2010193 (12 July)
29 August 2010 (TM)21 August (233), 29 August (241), and 6 September (249), 2010241 (29 August)
Table 4. Sampling areas before pixel resizing and random sample number in each land cover type taking the Landsat-MODIS pair of 1 May 2007 as an example.
Table 4. Sampling areas before pixel resizing and random sample number in each land cover type taking the Landsat-MODIS pair of 1 May 2007 as an example.
Land Cover TypesSampling Areas (in Pixels)
Randomly Sampling No
Landsat TM/ETM+(30 m)MOD15A2 LAI (1,000 m)MOD13Q1 NDVI/EVI (250 m)
White Soil Olive7,2031312150
Irrigated Wheat161,6272302,726100
Barley for Harvesting128,5001782,167100
Barley for Grazing24,9793442150
Stone Mining79311310
Bare Soil71,078941,19950
Table 5. Calibrated logarithmic relationships between vegetation indices (VIs) and Leaf Area Index (LAI) from 890 random samples across all land cover types taking the image pair of 1 May 2007 as an example.
Table 5. Calibrated logarithmic relationships between vegetation indices (VIs) and Leaf Area Index (LAI) from 890 random samples across all land cover types taking the image pair of 1 May 2007 as an example.
Logarithmic FunctionEstimation ErrorMultiple R2
GDVI^2 = 0.251ln(LAI) + 0.639±0.1130.793
GDVI^3 = 0.228ln(LAI) + 0.764±0.1100.769
GDVI^4 = 0.188ln(LAI) + 0.836±0.1010.729
NDVI = 0.242ln(LAI) + 0.514±0.1190.762
SAVI = 0.094ln(LAI) + 0.212±0.0640.626
SARVI = 0.151ln(LAI) + 0.213±0.0790.739
OSAVI = 0.132ln(LAI) + 0.276±0.0700.734
MNLI = 0.095ln(LAI) − 0.071±0.0530.711
EVI = 0.183ln(LAI) + 0.352±0.1150.660
WDRVI = 0.217ln(LAI) − 0.32±0.1190.721
Table 6. Calibrated biome mean-based logarithmic relationships between the vegetation indices and LAI.
Table 6. Calibrated biome mean-based logarithmic relationships between the vegetation indices and LAI.
Image DateLogarithmic FunctionMultiple R2
27 March 2003NDVI = 0.2146ln(LAI) + 0.44480.7426
GDVI^2 = 0.2431ln(LAI) + 0.69410.7881
GDVI^3 = 0.1965ln(LAI) + 0.820.7600
GDVI^4 = 0.1438ln(LAI) + 0.88780.7036
EVI = 0.1912ln(LAI) + 0.38850.5178
SAVI = 0.1083ln(LAI) + 0.27320.4915
SARVI = 0.1915ln(LAI) + 0.29190.6642

1 May 2007NDVI = 0.1784ln(LAI) + 0.40710.8729
GDVI^2 = 0.2346ln(LAI) + 0.64570.8830
GDVI^3 = 0.2209ln(LAI) + 0.76870.8572
GDVI^4 = 0.1873ln(LAI) + 0.83750.8199
EVI = 0.1492ln(LAI) + 0.35280.7410
SAVI = 0.0935ln(LAI) + 0.25140.7302
SARVI = 0.1623ln(LAI) + 0.25740.8521

21 August 2007NDVI = 0.1939ln(LAI) + 0.41850.9307
GDVI^2 = 0.2737ln(LAI) + 0.65870.9547
GDVI^3 = 0.2812ln(LAI) + 0.77680.9432
GDVI^4 = 0.2619ln(LAI) + 0.84040.9135
EVI = 0.1729ln(LAI) + 0.3740.7707
SAVI = 0.1118ln(LAI) + 0.26270.8000
SARVI = 0.1865ln(LAI) + 0.27290.8938

12 July 2010NDVI = 0.1648ln(LAI) + 0.38780.8702
GDVI^2 = 0.2274ln(LAI) + 0.61660.8758
GDVI^3 = 0.2261ln(LAI) + 0.73560.8591
GDVI^4 = 0.2013ln(LAI) + 0.80480.8334
EVI = 0.146ln(LAI) + 0.35410.6227
SAVI = 0.0967ln(LAI) + 0.25790.6405
SARVI = 0.1595ln(LAI) + 0.24350.7598

29 August 2010NDVI = 0.1761ln(LAI) + 0.39550.8983
GDVI^2 = 0.248ln(LAI) + 0.62680.9319
GDVI^3 = 0.2538ln(LAI) + 0.74540.9255
GDVI^4 = 0.2338ln(LAI) + 0.81050.8921
EVI = 0.1527ln(LAI) + 0.34590.7365
SAVI = 0.0989ln(LAI) + 0.24690.7568
SARVI = 0.1698ln(LAI) + 0.24610.8656
Table 7. Values of vegetation indices of some typical biomes (Landsat image dated 1 May 2007).
Table 7. Values of vegetation indices of some typical biomes (Landsat image dated 1 May 2007).
VIsForestWoodlandCitrusWheatBarleyOlivesGrasslandRange-LandBare Land
WDRVI (a =0.20)−0.216−0.302−0.1910.042−0.307−0.526−0.372−0.541−0.587
Table 8. Pearson correlation coefficients (R2) between VIs/Generalized Difference Vegetation Index (GDVI) and fractional vegetation cover (FVC) from Landsat imagery dated 1 May 2007.
Table 8. Pearson correlation coefficients (R2) between VIs/Generalized Difference Vegetation Index (GDVI) and fractional vegetation cover (FVC) from Landsat imagery dated 1 May 2007.
Table 9. Applicability of GDVI.
Table 9. Applicability of GDVI.
GDVILand Cover Types
Forest/MaquisIrrigated CroplandWood-LandsCitrus/OrchardRainfed CroplandOlive PlantationRangelandDesertBare Land
GDVI^2Partly, YesPartly, YesYesYesYesYesYesYesYes
GDVI^4NoNoPartly, YesPartly, YesPartly, YesYesYesYesYes
Back to TopTop