ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57
Contents lists available at SciVerse ScienceDirect
ISPRS Journal of Photogrammetry and Remote Sensing journal homepage: www.elsevier.com/locate/isprsjprs
Estimation of forest canopy structural parameters using kernel-driven bi-directional reflectance model based multi-angular vegetation indices Ram C. Sharma ⇑, Koji Kajiwara, Yoshiaki Honda Center for Environmental Remote Sensing (CEReS), Chiba University, 1-33 Yayoi-cho, Inage-ku, Chiba, Japan
a r t i c l e
i n f o
Article history: Received 26 April 2012 Received in revised form 27 December 2012 Accepted 31 December 2012 Available online 16 February 2013 Keywords: Forest canopy Canopy 3D structure Near-surface bi-directional reflectance Multi-angular vegetation indices BRDF parameters Canopy structural index
a b s t r a c t Near-surface bi-directional reflectance and high-spatial resolution true-color imagery of several forested canopies were acquired using an unmanned helicopter. The observed reflectance from multiple viewzenith angles were simulated with a kernel-driven bidirectional reflectance model, and the BRDF parameters were retrieved. Based on the retrieved BRDF parameters, kernel-derived multi-angular vegetation indices (KMVIs) were computed. The potential of KMVI for prediction of canopy structural parameters such as canopy fraction and canopy volume was assessed. The performance of each KMVI was tested by comparison to field measured canopy fraction and canopy volume. For the prediction of canopy fraction, the KMVI that included the nadir-based NDVI performed better than other KMVI emphasizing the importance of nadir observation for remote estimation of the canopy fraction. The Nadir BRDF-adjusted NDVI was found to be superior for the prediction of canopy fraction, which could explain 77% variation of the canopy fraction. However, none of the existing KMVI predicted the canopy volume better than Nadir BRDF-adjusted NDVI and Nadir-view NDVI. The Canopy structural index (CSI) was proposed with the combination of normalized difference between dark-spot near infrared reflectance and hot-spot red reflectance. The CSI could establish an improved relationship with the canopy volume over Nadir BRDF-adjusted NDVI and Nadir-view NDVI, explaining 72% variation in canopy volume. In addition, MODIS based KMVI were evaluated for the prediction of canopy fraction and canopy volume. MODIS based KMVI also showed similar results to the helicopter based KMVI. The promising results shown by the CSI suggest that it could be an appropriate candidate for remote estimation of three-dimensional canopy structure. Ó 2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved.
1. Introduction The forest canopy- an important component of earth–atmospheric interaction which influences the movement of carbon, water, trace gases, and energy between the biosphere and atmosphere; and a special habitat which covers approximately one third of the earth’s land surface – has been defined as the whole above ground forest volume (Bongers, 2001). Several remote sensing methods have been developed for assessment of canopy structural parameters such as canopy fraction, leaf area index and canopy height. The advancement in the field of multi-angular remote sensing techniques, i.e., observations from multiple viewing angles have made it possible to recognize the anisotropic reflectance pattern of the canopy (Kimes, 1983, 1986; Deering et al., 1994, 1999; Sandmeier et al., 1998). The reflectance of a canopy as seen by the sensor at the given illuminating sun and viewing sensor geometry have been modeled ⇑ Corresponding author. Tel.: +81 43 290 3832; fax: +81 43 290 3857. E-mail address:
[email protected] (R.C. Sharma).
as the areal proportions of sunlit crown, sunlit ground, shadowed crown, and shadowed ground multiplied by each component’s reflectance (Li and Strahler, 1986; Gerard and North, 1997). Since the observed reflectance is dependent on the physical structure of the canopy, the anisotropic reflectance of the canopy is expected to contain canopy structural information. Bi-directional reflectance models have been developed by linking the surface reflectance to the canopy structural parameters considering the view and sun geometry (Li and Strahler, 1986; Roujean et al., 1992; Wanner et al., 1995). In this model, the bi-directional reflectance (R) for a sun-zenith angle (SZA), view-zenith angle (VZA), and relative azimuth angle (RAA) is given by the following equation:
h b Rðh; t; D/Þ ¼ fiso þ fvol K vol ðh; t; D/Þ þ fgeo K geo h; t; D/; ; b r
ð1Þ
In Eq. (1), K vol and K geo are the kernels for radiative transfertype volumetric scattering and geometric-optical surface scattering respectively. The K vol and K geo are trigonometric functions of
0924-2716/$ - see front matter Ó 2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.isprsjprs.2012.12.006
R.C. Sharma et al. / ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57
SZA (h), VZA (t), and RAA (D/); crown relative height (hb, height of crown center/vertical crown radius) and crown shape (br, vertical crown radius/horizontal crown radius) parameters are also included in Kgeo. The fvol and f geo are the kernel weights for volumetric scattering and geometric-optical surface scattering respectively. The fiso is a constant called isotropic scattering. The Ross-Thick (RT) and Li-Sparse-Reciprocal (LSR) kernels are usually used to describe the K vol and K geo respectively (Wanner et al., 1995; Lucht et al., 2000), and the resulting BRDF is called the kernel-based BRDF model. The isotropic parameter is equivalent to nadir-view nadirsun reflectance. The volumetric parameter is in theory exponentially related to leaf area index, while the geometric parameter is a function of crown/object density and size, related to surface roughness or gapiness of the landscape (Strahler et al., 1999). The RTLSR kernel combination is currently used in the operational MODIS BRDF/Albedo algorithm (Schaaf et al., 2002). The RTLSR based BRDF model has been found to perform well over a wide range of surface covers (Privette et al., 1997), and even with a small sampling of quality input observations (Lucht, 1998). The linear combination of the RTLSR based model has been conveniently inverted for deriving canopy bio-physical parameters by fitting with observation data (Kimes et al., 2000; Li et al., 2001; Armston et al., 2007), and the model has been validated in several land cover types (Hu et al., 1997; Liang et al., 2002; Disney et al., 2004; Salomon et al., 2006; Susaki et al., 2007; Samain et al., 2008; Knobelspiesse et al., 2008; Hill et al., 2008; Liu et al., 2009; Román et al., 2009; Jiao et al., 2011; Wang et al., 2012). Different approaches have been attempted to retrieve canopy structural information using multi-angular remote sensing such as radiative transfer modeling (North, 1996, 2002), geometric-optical modeling (Chopping et al., 2011, 2012) and spectral invariant (Schull et al., 2007, 2011; Huang et al., 2007; Lewis et al., 2007; Knyazikhin et al., 2011; Wang et al., 2011) as some examples. Another approach for retrieval of canopy structural parameters is the use of multi-angular vegetation indices as a combination of reflectance from multiple view angles (Sandmeier et al., 1998; Sandmeier and Deering, 1999; Lacaze et al., 2002; Chen et al., 2003, 2005; Pocewicz et al., 2007; Hasegawa et al., 2010; Hill et al., 2011). A number of spectral vegetation indices, calculated as the combinations of multiple spectral bands derived from nadir viewing remote sensing have been broadly used in the prediction of various canopy bio-physical parameters (e.g., Tucker, 1979; Kaufman and Tanre, 1992; Huete et al., 1988, 2002). The simplicity and cost-effectiveness involved with the acquisition and computation of vegetation indices, and thus suitability for mapping over global scales have promoted an extensive utilization of vegetation indices. Improvement of spectral vegetation indices with multiangular remote sensing is expected to advance the retrieval of canopy structural parameters. Since the observed reflectance from multiple sun and view geometry are very limited in practice, the reflectance at the desired geometry could be retrieved by fitting the observation data using the bi-directional reflectance model (e.g., RTLSR based BRDF model); and the vegetation indices could be computed for any desired geometric condition. On the other hand, since the RTLSR based BRDF model is constructed using the physical structure of the surface; BRDF model parameters themselves ðfiso ; f vol and f geo Þ should be related to the canopy structural parameters. The question of whether there is a semi-empirical bio-physical interpretation of the model parameters is of great interest (Strahler et al., 1999). Vegetation indices as a combination of multiple BRDF parameters have been developed to establish a relationship with canopy bio-physical parameters (d’Entremont et al., 1999; Gao et al., 2003). In both cases, since the vegetation indices are based on a kernel-derived BRDF model, they will be called kernel-derived multi-angular vegetation indices (KMVIs). The list of KMVI which are reported in the literature and used in this study are summarized in Table 1.
51
It has been difficult to measure the in situ high-angular resolution reflectance for the BRDF analysis especially from forested canopies and the availability of such datasets are very limited to certain field campaigns (e.g., Deering et al., 1992; Gamon et al., 2004; Román et al., 2011). Recently, unmanned helicopters have become increasingly used in monitoring and mapping of canopy parameters in several ecosystem types (Sugiura et al., 2005; Rango et al., 2006; Hunt et al., 2010; Breckenridge et al., 2011; Xiang and Tian, 2011; Laliberte and Rango, 2011). Since unmanned helicopters can potentially operate with complicated areas to be accessed, the widespread use of unmanned helicopters to measure near-surface bi-directional reflectance is expected to contribute tremendously to better understanding of the BRDF characteristics of different forest types, and to enhance the satellite based predictions. In this context, near-surface bi-directional reflectance and highspatial resolution true-color imagery of several forested canopies were acquired using an unmanned helicopter; and the potential of KMVI were assessed for the prediction of canopy structural parameters such as canopy fraction and canopy volume. 2. Methods 2.1. Study sites The study sites consist of four different forests: Hokkaido Forest, Kouchi Forest, Shiga Forest, and Yamanashi Forest in Japan. These forests are comprised of conifer forest, broadleaf forest, and mixed conifer and broadleaf forest. The major species under the conifer sites are Japanese larch (Larix kaempferi), Japanese red pine (Pinus densiflora), and Japanese cypress (Chamaecyparis obtuse); the broadleaved sites are mainly composed of Japanese white birch (Betula platyphylla); and the mixed broadleaf and conifer forest are composed of Japanese cypress (C. obtuse) and Japanese white birch (B. platyphylla). At each forested site, unmanned-helicopter based observations were conducted at several plots. The plots were spaced at least 500 m apart to represent an area of a MODIS pixel (500 m 500 m), and located on relatively flat terrain. The details on each forest site are described in Table 2, providing their geolocation (latitude and longitude), forest types, and number of plots. Overall, data from twenty plots were acquired in this study. 2.2. Measurement of bidirectional reflectance factor A portable spectroradiometer (MS-720, Eko Co. Ltd., Japan) and a digital camera (Canon EOS; Canon Inc., Japan) were deployed on an unmanned helicopter (Yamaha-RMAX radio controlled helicopter, Yamaha Motor Co., Ltd., Japan) and used to obtain multi-angular data. Both instruments were arranged in such a way to view the same target area using a differential geographic positioning system (DGPS) with ±10 cm accuracy and an attitude and heading reference system (AHRS) to provide orientation and location. Canopy upwelling radiance and high-spatial resolution true-color imagery were obtained from multiple view-zenith angles (VZAs) viewing at the center of the same target canopy while maintaining the position of the helicopter at a hemispheric axis of 60 m radius above ground level. Data were acquired in the principal plane usually from the following VZA (negative VZA refers to backward scattering direction and positive VZA to forward scattering direction): 0°, ±10°, ±20°, ±30°, ±40°, ±45°, ±50°, ±55°, and ±60°. The multiangular observation method used in this study is illustrated in Fig. 1. Another similar spectroradiometer was positioned about 1.5 m above the ground level to measure the upwelling radiance reflected by a reference panel (White Spectralon; Labsphere
52
R.C. Sharma et al. / ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57
Table 1 Overview of the kernel-derived multi-angular vegetation indices (KMVIs) used under the study, their formulation, and references. fiso ; f vol and f geo are retrieved isotropic, volumetric, and geometric kernel weights respectively. HS, N, and DS are the hot-spot (45, 45, 0), nadir (45, 0, 0), and dark-spot (45, 45, 180) views respectively. KMVI
Formula
Reference
Nadir BRDF-adjusted NDVI (NDVI0,0,0)
nir red fiso fiso nir þf red fiso iso
Strahler et al. (1999)
Nadir-view NDVI (NDVI45,0,0)
N nir Nred N nir þNred HSred DSred HSnir DSnir ANIXred ANIXnir ANIXred þANIXnir HSred DSred DSred HSnir DSnir HSnir þDSnir HSnir DSred HSnir þDSred
Tucker et al. (1979)
NDVIN ð1 HSred Þ NDVIN HDSnir
Pocewicz et al. (2007) Hasegawa et al. (2010) This study
Anisotropy index (ANIXRed) Anisotropy index (ANIXNir) Normalized difference anisotropic index (NDAX) Hot-spot dark-spot index (HDSred) Normalized difference between hot-spot and dark-spot index (NDHDnir) Hot-spot dark-spot NDVI (NDVIHD) Hot-spot incorporated NDVI (NDVIHS) Normalized hot-spot signature vegetation index (NHVI) Canopy structural index (CSI)
Nir45;45;180 Red45;45;0 Nir45;45;180 þRed45;45;0
Sandmeier et al. (1998) Sandmeier et al. (1998) Sandmeier and Deering (1999) Lacaze et al. (2002) Chen et al. (2005) Pocewicz et al. (2007)
NDVI0;0;0
Table 2 Description of the sites under the study, their geo-location (latitude and longitude), forest types, and number of plots. Study sites
Geo-location (latitude, longitude)
Forest type
No. of plots
Hokkaido Forest (HF) Kouchi Forest (KF) Shiga Forest (SF) Yamanashi Forest (YF)
43°330 5500 N, 32°440 2900 N, 35°070 5100 N, 35°270 1700 N,
Conifer forest Mixed broadleaf and conifer forest Broadleaf forest Conifer forest
4 8 4 4
144°390 0200 E 132°590 5800 E 136°120 2600 E 138°450 4700 E
Both of the instruments shared a similar field of view (FOV): 45° (horizontal) 45° (vertical); and the area covered on the ground, given by 2 tanðFOV=2Þ altitide of helicopterð60 mÞ, was calculated to be 49.70 m (length) 49.70 m (width). For each plot, data from all VZA were completed within 30 min; for such a short period, environmental conditions (solar position, solar irradiance, weather conditions, atmospheric aerosols etc.) were assumed to be invariant. The reflectance along the principal plane was used for this study because the directional effects on reflectance are greatest along the principal plane and the marked hot-spot and dark-spot reflectance can be observed in this plane. Data were taken under natural light during clear sky days, and the data taken from a short altitude (60 m) were assumed to be free from atmospheric scattering. For each plot, data were taken during vegetation growing month when the leaves were mature enough so that distinct difference between canopy and ground reflectance could be captured. Data quality was assured through repeated aviation for each plot. Aviation with unexpected changes in the attitude of the helicopter, resulting in non-overlapping images of the same region of interest, were discarded in the field. Fig. 1. Near-surface bi-directional reflectance observation method used under the study. Three instances of measurement at backward scattering direction (45° view-zenith angle), nadir scattering direction (0° view-zenith angle), and forward scattering direction (+45° view-zenith angle) in the principal plane are shown.
Company, USA), exposed directly to the sun. The spectroradiometers measured the radiance (W m2 lm1) at 1 nm spectral resolution, between 350 and 1050 nm; and both spectroradiometers were inter-calibrated before the measurement. Bi-directional reflectance factor (BRF) of the target canopy was calculated by normalizing the upwelling radiance received from a canopy to the upwelling radiance measured over the reference panel. The helicopter was autonomously aviated along a planned path by an experienced operator aided with DGPS and AHRS system. Specific VZA were obtained by rotating the pan-tilt unit attached to the spectroradiometer and camera using a portable computer onboard the helicopter and a computer program to control the pan-tilt unit.
2.3. Measurement of canopy structural parameters The high-spatial resolution nadir-viewing true-color imagery taken from the helicopter was processed for extraction of green canopy fraction. Using the digital number values of red, green, and blue bands of an image; normalized difference between green and red (NG-R) index (Pérez et al., 2000) images were created. The visual interpretation of NG-R index images showed that the NG-R index value of each sub-canopy component varied in this order: sunlit ground and shadowed ground < sunlit crown < shadowed crown. The threshold between green canopy (sunlit crown and shadowed crown) and canopy ground (sunlit ground and shadowed ground) could be easily determined while working with NG-R index images; and thus using the threshold value, the canopy fraction was measured for each plot. The NG-R index, which is similar to normalized difference vegetation index (NDVI) that uses near infrared and red reflectance, may be less sensitive to illumina-
R.C. Sharma et al. / ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57 Table 3 Description of the measured canopy structural parameters for each plot under the study. The unmanned helicopter based observation dates, canopy fraction, average canopy height (m), and canopy volume (m3) are provided. Plots
Observed dates
Canopy fraction
Canopy height
Canopy volume
HF1 HF2 HF3 HF4 KF1 KF2 KF3 KF4 KF5 KF6 KF7 KF8 SF1 SF2 SF3 SF4 YF1 YF2 YF3 YF4
2008.09.05 2008.09.05 2008.09.05 2008.09.05 2009.09.09 2009.09.09 2009.09.09 2009.09.09 2009.09.10 2009.09.10 2009.09.10 2009.09.10 2009.09.07 2009.09.07 2009.09.14 2009.09.14 2010.09.17 2010.09.17 2010.09.22 2010.09.22
0.85 0.77 0.71 0.93 0.83 0.84 0.87 0.91 0.67 0.54 0.82 0.73 0.58 0.55 0.87 0.78 0.56 0.48 0.51 0.62
14 22 17 16 22 19 21 21 15 13 19 21 19 21 17 19 18 16 15 17
11.9 16.94 12.07 14.88 18.26 15.96 18.27 19.11 10.05 7.02 15.58 15.33 11.02 11.55 14.79 14.82 10.08 7.68 7.65 10.54
tion variations, and may have the potential to work well for different background conditions (Pérez et al., 2000). Several studies have used visual method for discriminating green canopies for validation of satellite based estimation (e.g., Purevdorj et al., 1998; Gitelson et al., 2002). For each plot, average canopy height was estimated based on measurement of trees within a 3 m radius circular area inside the footprint of helicopter based observation. Since the target canopy (plot) was selected over more or less homogenous forest stands, it was assumed that the measured canopy fraction and the canopy height represent an area of a MODIS pixel (500 m 500 m). The product of the canopy fraction and average canopy height (m) was assumed as canopy volume (m3 per squared meter forest area). The measured structural parameters for each plot are described in Table 3. 2.4. Computation of KMVI The unmanned helicopter based reflectance was converted into MODIS red (620–670 nm) and near-infrared band (841–876 nm) by normalizing the measured hyper-spectral data with the band-specific relative spectral response (RSR) of MODIS. Then, the normalized red band and near infrared band at each plot was simulated using the Ross-Thick (RT) and Li-Sparse-Reciprocal (LSR) kernel based BRDF model (Eq. (1)) choosing the values for hb and br as 2.0 and 1.0 similar to the MODIS BRDF/Albedo algorithm. Thus, the BRDF parameters for the red and near infrared band were retrieved for each plot. For the retrieval of the BRDF parameters at each plot, usually measurements with seventeen view-zenith angles on the principal plane were used. The retrieved BRDF parameters were further simulated with the RTLSR kernel to obtain the reflectance on the principal plane for the hot-spot (HS), nadir (N) and dark-spot (DS). At the principal plane, with a SZA of 45°; backward scattering (45° VZA), nadir scattering (0° VZA), and forward scattering (+45° VZA) were assumed as the hot-spot, nadir, and dark-spot views, respectively. Using the convention to represent each observation point with three sets of angles (SZA, VZA, and RAA); HS, N, and DS are represented as (45, 45, 0), (45, 0, 0), and (45, 45, 180) respectively. Based on the simulated reflectance at hot-spot, nadir, and dark-spot; the kernel-derived multi-angular vegetation indices (KMVIs) (Table 1) were computed. In addition, the BRDF model parameters of the red (620– 670 nm) and near infrared (841–875 nm) bands were obtained
53
from MODIS BRDF/Albedo product (MCD43A1) (ORNL DAAC, 2011). This is an 8-day level 3 global product (version 005) provided at 500 m resolution on sinusoidal grid projection (Schaaf et al., 2002). Based on the geo-location of the helicopter based observation plots, the BRDF parameters close to the helicopter based observation date were extracted for a single MODIS pixel (500 m 500 m). The quality of the BRDF parameters (MCD43A2) used in the study was either ‘best quality full inversion’ or ‘good quality full inversion’. Then based on the MODIS BRDF parameters, similar to the helicopter based computation, the red and near infrared reflectance for hot-spot, nadir, and dark-spot were obtained, and the KMVI were computed. 2.5. Proposal of canopy structural index While the multi-angular vegetation indices are expected to provide improved prediction of the canopy structural parameters, the acquisition of reflectance in the field was constrained by available view-zenith angles and real solar positions. Therefore, simulation of the measured reflectance from multiple sun-sensor geometries with the BRDF model offers an opportunity to retrieve the reflectance under the desired geometric condition. For example, the Nadir BRDF-adjusted reflectance (fiso) provides an opportunity to model nadir-sun nadir-view reflectance which avoids the distracting spectral signatures caused by the canopy shadows. In contrast, the reflectance at hot-spot and dark-spot provides the reflectance anisotropy with opportunity to access canopy structural information. In this study, the observation points at (45, 45, 0) and (45, 45, 180) are taken as the hot-spot and dark-spot respectively, assuming they represent better positions to classify the reflectance anisotropy related to the canopy structure. The fractional area covered by the sub-canopy components (sunlit crown, shadowed crown, sunlit background, and shadowed background) changes with the variation of the sun and sensor geometry. In the principal plane, the pronounced reflectance anisotropy caused by significant changes in the fractional canopy components could be observed. While the scene viewed from the hot-spot is mostly composed of sunlit ground and sunlit crown, the shadowed crown and shadowed ground dominate the scene viewed from the dark-spot. For a constant canopy height, the appearance of shadowed ground and shadowed crown from the dark-spot would be higher in sparse forest. In forested canopies, since the near infrared region is highly sensitive to the canopy shadows; the decrease in near infrared reflectance at the dark-spot implies higher shadows caused by a more open canopy. For a constant canopy fraction, the occlusion of the sunlit ground at hot-spot would be higher in tall forest and the decrease in red reflectance at the hot-spot would be faster. Therefore, the normalized difference between the dark-spot near infrared reflectance and hot-spot red reflectance is combined with Nadir BRDF-adjusted NDVI (NDVI0,0,0) to formulate the Canopy structural index (CSI) (Eq. (2)) as an indicator of canopy three-dimensional structure.
CSI ¼
Nir45;45;180 Red45;45;0 NDVI0;0;0 Nir45;45;180 þ Red45;45;0
ð2Þ
3. Results and discussion 3.1. Performance of helicopter based KMVI The performance of helicopter based kernel-derived multiangular vegetation indices (KMVIs) for the prediction of canopy fraction and canopy volume were assessed using the coefficient of determination (R2) and percentage root mean square error (RMSE%). The RMSE% was calculated using the following equation:
54
R.C. Sharma et al. / ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn 2 i¼1
RMSE% ¼
ðPi Oi Þ
Pn
n
O i¼1 i
100
ð3Þ
n
In Eq. (3), Pi is the predicted canopy fraction or canopy volume based on kernel-derived multi-angular vegetation index (KMVI) for each plot (i). Oi is the observed/measured canopy fraction or canopy volume for each plot (i) and n is the total number of plots. For the prediction of canopy fraction, only the KMVI that includes nadir viewing NDVI, viz. Nadir BRDF-adjusted NDVI (NDVI0,0,0), Nadir-view NDVI (NDVI45,0,0), Hot-spot incorporated NDVI (NDVIHS), Normalized hot-spot signature vegetation index (NHVI) and Canopy structural index (CSI) performed satisfactorily (R2 greater than 0.50) (Table 4). However, NDVI0,0,0 was found to be superior (R2 = 0.77; RMSE% = 9.47) (Fig. 2) followed by NDVI45,0,0 (R2 = 0.56; RMSE% = 13.10), and NDVIHS (R2 = 0.54; RMSE% = 13.40) over other KMVI for the prediction of canopy fraction. In this study, the remaining KMVI that did not include the nadir-viewing reflectance, viz. Normalized difference anisotropic index (NDAX), Anisotropy index (ANIXRed), Anisotropy index (ANIXNir), Hot-spot dark spot index (HDSred), Normalized difference between hot-spot and dark-spot index (NDHDnir), and Hotspot dark-spot NDVI (NDVIHD) were found to be less important for the prediction of canopy fraction (R2 lesser than 0.50). The KMVI based on hot-spot and dark-spot reflectance and that uses only the red band such as ANIXRed and HDSred were found to be insensitive to the canopy fraction. However, the performance of hot-spot and dark-spot based KMVI that uses only the near infrared band such as ANIXNir, NDHDnir, and NDVIHD was intermediate. The NDAX though uses both the near infrared and red band reflectance but only from the hot-spot and dark-spot, and does not include the nadir-based reflectance, did not perform. Therefore, it seems that nadir-viewing observation cannot be avoided for remote estimation of the canopy fraction. On the other hand, for the prediction of canopy volume, Nadir BRDF-adjusted NDVI (NDVI0,0,0), Nadir-view NDVI (NDVI45,0,0), Hot-spot incorporated NDVI (NDVIHS), and Canopy structural index (CSI) performed well (R2 greater than 0.50) (Table 4). The performance by Hot-spot dark-spot NDVI (NDVIHD), Anisotropy index (ANIXNir), and Normalized difference between hot-spot and darkspot index (NDHDnir) was intermediate while the remaining KMVI: Anisotropy index (ANIXRed), Normalized difference anisotropic index (NDAX), and Hot-spot dark spot index (HDSred) did not perform. Among the KMVI tested, CSI performed best (R2 = 0.72; RMSE% = 14.55) (Fig. 3) followed by NDVI0,0,0 (R2 = 0.62;
Fig. 2. Performance of unmanned helicopter based KMVI for the prediction of canopy fraction using Nadir BRDF-adjusted NDVI.
RMSE% = 16.96), NDVI45,0,0 (R2 = 0.52; RMSE% = 19.06), and NDVIHS (R2 = 0.51; RMSE% = 19.25). In this study, none of the existing KMVI could predict the canopy volume better than by Nadir BRDF-adjusted NDVI (NDVI0,0,0) and Nadir-view NDVI (NDVI45,0,0). However, Canopy structural index (CSI) established an improved relationship with the canopy volume over NDVI0,0,0 and NDVI45,0,0. The CSI was formulated with the inclusion of dark-spot near infrared reflectance and hot-spot red reflectance. 3.2. Performance of MODIS based KMVI The performance of the MODIS based kernel-derived multiangular vegetation indices (KMVIs) for the prediction of canopy fraction and canopy volume were also assessed using the coefficient of determination (R2) and percentage root mean square error (RMSE%). The MODIS based KMVI also performed similar to the helicopter based KMVI for the prediction of canopy fraction and canopy volume (Table 5). For the prediction of canopy fraction, the Nadir BRDF-adjusted NDVI (NDVI0,0,0) was found to be superior (R2 = 0.67; RMSE% = 11.35) (Fig. 4) followed by Nadir-view NDVI (NDVI45,0,0). (R2 = 0.55; RMSE% = 13.25), and Hot-spot incorporated NDVI (NDVIHS) (R2 = 0.52; RMSE% = 13.68) over other KMVI.
Table 4 Prediction of canopy structural parameters: canopy fraction and canopy volume using unmanned helicopter based kernel-derived multi-angular vegetation indices (KMVIs). The performance of each KMVI was evaluated using coefficient of determination (R2) and percentage root mean square error (RMSE%). The value inside the parenthesis refers to RMSE%. The satisfactory results (R2 greater than 0.5) are displayed in bold. Helicopter-based KMVI
Canopy fraction
Canopy volume
Nadir BRDF-adjusted NDVI (NDVI0,0,0) Nadir-view NDVI (NDVI45,0,0) Anisotropy index (ANIXRed) Anisotropy index (ANIXNir) Normalized difference anisotropic index (NDAX) Hot-spot dark spot index (HDSred) Normalized difference between hot-spot and dark-spot index (NDHDnir) Hot-spot dark-spot NDVI (NDVIHD) Hot-spot incorporated NDVI (NDVIHS) Normalized hot-spot signature vegetation index (NHVI) Canopy structural index (CSI)
0.77 (9.47) 0.56 (13.10) 0.02 (19.55) 0.45 (14.65) 0.03 (19.45) 0.02 (19.55)
0.62 (16.96) 0.52 (19.06) 0.04 (26.95) 0.36 (22.00) 0.01 (27.37) 0.05 (26.81)
0.45 0.38 0.54 0.51 0.52
0.33 0.28 0.51 0.42 0.72
(14.65) (15.55) (13.40) (13.83) (13.68)
(22.51) (23.34) (19.25) (20.95) (14.55)
R.C. Sharma et al. / ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57
55
Fig. 3. Performance of unmanned helicopter based KMVI for the prediction of canopy volume using Canopy structural index.
Fig. 5. Performance of MODIS based KMVI for the prediction of canopy volume using Canopy structural index.
Table 5 Prediction of canopy structural parameters: canopy fraction and canopy volume using MODIS based kernel-derived multi-angular vegetation indices (KMVIs). The performance of each KMVI was evaluated using coefficient of determination (R2) and percentage root mean square error (RMSE%). The value inside the parenthesis refers to RMSE%. The satisfactory results (R2 greater than 0.5) are displayed in bold.
However, for the prediction of canopy volume, the Canopy structural index (CSI) performed best (R2 = 0.63; RMSE% = 16.73) (Fig. 5) followed by Nadir BRDF-adjusted NDVI (NDVI0,0,0) (R2 = 0.55; RMSE% = 18.45), Nadir-view NDVI (NDVI45,0,0) (R2 = 0.51; RMSE% = 19.25), and Hot-spot incorporated NDVI (NDVIHS) (R2 = 0.50; RMSE% = 19.45). Nevertheless, the helicopter based Nadir BRDF-adjusted NDVI (NDVI0,0,0) and Canopy structural index (CSI) could better predict the canopy fraction (R2 = 0.77; RMSE% = 9.47) and canopy volume (R2 = 0.72; RMSE% = 14.55) respectively (Table 4) than with similar MODIS based NDVI0,0,0 (R2 = 0.67; RMSE% = 11.35) and CSI (R2 = 0.63; RMSE% = 16.73) (Table 5). Though the footprint of helicopter (49.70 m 49.70 m) based observation was different with MODIS based pixel (500 m 500 m), the results are still comparable.
MODIS-based KMVI
Canopy fraction
Canopy volume
Nadir-BRDF adjusted NDVI (NDVI0,0,0) Nadir-view NDVI (NDVI45,0,0) Anisotropy index (ANIXRed) Anisotropy index (ANIXNir) Normalized difference anisotropic index (NDAX) Hot-spot dark spot index (HDSred) Normalized difference between hot-spot and dark-spot index (NDHDnir) Hot-spot dark-spot NDVI (NDVIHD) Hot-spot incorporated NDVI (NDVIHS) Normalized hot-spot signature vegetation index (NHVI) Canopy structural index (CSI)
0.67 (11.35) 0.55 (13.25) 0.01 (19.65) 0.32 (16.29) 0.04 (19.35)
0.55 (18.45) 0.51 (19.25) 0.06 (26.67) 0.25 (23.82) 0.00 (27.51)
0.01 (19.65)
0.07 (26.53)
0.34 0.37 0.52 0.50
0.25 0.28 0.50 0.33
(16.05) (15.68) (13.68) (13.97)
0.51 (13.83)
(23.82) (23.34) (19.45) (22.51)
0.63 (16.73)
4. Conclusion Since the kernel-derived multi-angular vegetation indices (KMVIs), which includes nadir based reflectance such as Nadir BRDF-adjusted NDVI (NDVI0,0,0) and Nadir-view NDVI (NDVI45,0,0) performed best for the prediction of canopy fraction, this study emphasized the importance of nadir-viewing observation for remote estimation of the canopy fraction. However, the Canopy structural index (CSI) could establish an improved relationship over NDVI0,0,0 and NDVI45,0,0 for the prediction of canopy volume suggesting that the inclusion of dark-spot near infrared reflectance and hot-spot red reflectance are necessary for better prediction of canopy three-dimensional structure. In this study, 77% variation of the canopy fraction and 72% variation of the canopy volume could be explained by helicopter based NDVI0,0,0 and CSI respectively; while 67% variation of canopy fraction and 63% variation of canopy volume could be explained by MODIS based NDVI0,0,0 and CSI respectively. The promising results shown by the CSI suggests that it could be an appropriate candidate for remote estimation of the canopy three-dimensional structure. In future, strength of the CSI will be tested over diverse forested canopies. Acknowledgements
Fig. 4. Performance of MODIS based KMVI for the prediction of canopy fraction using Nadir BRDF-adjusted NDVI.
Authors would like to appreciate for invaluable comments received from two anonymous reviewers which were vital in shaping
56
R.C. Sharma et al. / ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57
the manuscript to its present form. Trade names are provided for reference purposes only, and do not mean endorsement by author.
References Armston, J.D., Scarth, P.F., Phinn, S.R., Danaher, T.J., 2007. Analysis of multi-date MISR measurements for forest and woodland communities, Queensland, Australia. Remote Sensing of Environment 107 (1–2), 287–298. Bongers, F., 2001. Methods to assess tropical rain forest canopy structure: an overview. Plant Ecology 153 (1), 263–277. Breckenridge, R.P., Dakins, M., Bunting, S., Harbour, J.L., White, S., 2011. Comparison of unmanned aerial vehicle platforms for assessing vegetation cover in sagebrush steppe ecosystems. Rangeland Ecology and Management 64 (5), 521–532. Chen, J.M., Liu, J., Leblanc, S.G., Lacaze, R., Roujean, J.-L., 2003. Multi-angular optical remote sensing for assessing vegetation structure and carbon absorption. Remote Sensing of Environment 84 (4), 516–525. Chen, J.M., Menges, C.H., Leblanc, S.G., 2005. Global mapping of foliage clumping index using multi-angular satellite data. Remote Sensing of Environment 97 (4), 447–457. Chopping, M., North, M., Chen, J., Schaaf, C.B., Blair, J.B., Martonchik, J.V., Bull, M.A., 2012. Forest canopy cover and height from MISR in topographically complex Southwestern US landscapes assessed with high quality reference data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 5 (1), 44–58. Chopping, M., Schaaf, C.B., Zhao, F., Wang, Z., Nolin, A.W., Moisen, G.G., Martonchik, J.V., Bull, M., 2011. Forest structure and aboveground biomass in the southwestern United States from MODIS and MISR. Remote Sensing of Environment 115 (11), 2943–2953. Deering, D.W., Eck, T.F., Banerjee, B., 1999. Characterization of the reflectance anisotropy of three boreal forest canopies in spring–summer. Remote Sensing of Environment 67 (2), 205–229. Deering, D.W., Middleton, E.M., Eck, T.F., 1994. Reflectance anisotropy for a sprucehemlock forest canopy. Remote Sensing of Environment 47 (2), 242–260. Deering, D.W., Middleton, E.M., Irons, J.R., Blad, B.L., Walter-Shea, E.A., Hays, C.J., Walthall, C., Eck, T.F., Ahmad, S.P., Banerjee, B.P., 1992. Prairie grassland bidirectional reflectances measured by different instruments at the fife site. Journal of Geophysical Research 97 (D17), 18887–18903. d’Entremont, R.P., Schaaf, C.B., Lucht, W., Strahler, A.H., 1999. Retrieval of red spectral albedo and bidirectional reflectance using AVHRR HRPT and GOES satellite observations of the New England region. Journal of Geophysical Research 104 (D6), 6229–6239. Disney, M., Lewis, P., Thackrah, G., Quaife, T., Barnsley, M., 2004. Comparison of MODIS broadband albedo over an agricultural site with ground measurements and values derived from earth observation data at a range of spatial scales. International Journal of Remote Sensing 25 (23), 5297–5317. Gamon, J.A., Huemmrich, K.F., Peddle, D.R., Chen, J., Fuentes, D., Hall, F.G., Kimball, J.S., Goetz, S., Gu, J., McDonald, K.C., 2004. Remote sensing in BOREAS: lessons learned. Remote Sensing of Environment 89 (2), 139–162. Gao, F., Schaaf, C.B., Strahler, A.H., Jin, Y., Li, X., 2003. Detecting vegetation structure using a kernel-based BRDF model. Remote Sensing of Environment 86 (2), 198– 205. Gerard, F.F., North, P.R.J., 1997. Analyzing the effect of structural variability and canopy gaps on forest BRDF using a geometric-optical model. Remote Sensing of Environment 62 (1), 46–62. Gitelson, A.A., Kaufman, Y.J., Stark, R., Rundquist, D., 2002. Novel algorithms for remote estimation of vegetation fraction. Remote Sensing of Environment 80 (1), 76–87. Hasegawa, K., Matsuyama, H., Tsuzuki, H., Sweda, T., 2010. Improving the estimation of leaf area index by using remotely sensed NDVI with BRDF signatures. Remote Sensing of Environment 114 (3), 514–519. Hill, M.J., Averill, C., Jiao, Z., Schaaf, C.B., Armston, J.D., 2008. Relationship of MISR RPV parameters and MODIS BRDF shape indicators to surface vegetation patterns in an Australian tropical savanna. Canadian Journal of Remote Sensing 34 (S2), S247–S267. Hill, M.J., Román, M.O., Schaaf, C.B., Hutley, L., Brannstrom, C., Etter, A., Hanan, N.P., 2011. Characterizing vegetation cover in global savannas with an annual foliage clumping index derived from the MODIS BRDF product. Remote Sensing of Environment 115 (8), 2008–2024. Hu, B., Lucht, W., Li, X., Strahler, A.H., 1997. Validation of kernel-driven semiempirical models for the surface bidirectional reflectance distribution function of land surfaces. Remote Sensing of Environment 62 (3), 201–214. Huang, D., Knyazikhin, Y., Dickinson, R.E., Rautiainen, M., Stenberg, P., Disney, M., Lewis, P., Cescatti, A., Tian, Y., Verhoef, W., Martonchik, J.V., Myneni, R.B., 2007. Canopy spectral invariants for remote sensing and model applications. Remote Sensing of Environment 106 (1), 106–122. Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83 (1–2), 195–213. Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment 25 (3), 295–309. Hunt, E.R., Hively, W.D., Fujikawa, S., Linden, D., Daughtry, C.S., McCarty, G., 2010. Acquisition of NIR–green–blue digital photographs from unmanned aircraft for crop monitoring. Remote Sensing 2 (1), 290–305.
Jiao, Z., Woodcock, C., Schaaf, C.B., Tan, B., Liu, J., Gao, F., Strahler, A., Li, X., Wang, J., 2011. Improving MODIS land cover classification by combining MODIS spectral and angular signatures in a Canadian boreal forest. Canadian Journal of Remote Sensing 37 (2), 184–203. Kaufman, Y.J., Tanre, D., 1992. Atmospherically resistant vegetation index (ARVI) for EOS-MODIS. IEEE Transactions on Geoscience and Remote Sensing 30 (2), 261– 270. Kimes, D.S., 1983. Dynamics of directional reflectance factor distributions for vegetation canopies. Applied Optics 22 (9), 1364–1372. Kimes, D.S., Knyazikhin, Y., Privette, J.L., Abuelgasim, A.A., Gao, F., 2000. Inversion methods for physically-based models. Remote Sensing Reviews 18 (2–4), 381– 439. Kimes, D.S., Newcomb, W.W., Nelson, R.F., Schutt, J.B., 1986. Directional reflectance distributions of a hardwood and pine forest canopy. IEEE Transactions on Geoscience and Remote Sensing GE-24 (2), 281–293. Knobelspiesse, K.D., Cairns, B., Schmid, B., Román, M.O., Schaaf, C.B., 2008. Surface BRDF estimation from an aircraft compared to MODIS and ground estimates at the Southern Great Plains site. Journal of Geophysical Research 113 (D20), 105. Knyazikhin, Y., Schull, M.A., Xu, L., Myneni, R.B., Samanta, A., 2011. Canopy spectral invariants. Part 1: A new concept in remote sensing of vegetation. Journal of Quantitative Spectroscopy and Radiative Transfer 112 (4), 727–735. Lacaze, R., Chen, J.M., Roujean, J.-L., Leblanc, S.G., 2002. Retrieval of vegetation clumping index using hot spot signatures measured by POLDER instrument. Remote Sensing of Environment 79 (1), 84–95. Laliberte, A.S., Rango, A., 2011. Image processing and classification procedures for analysis of sub-decimeter imagery acquired with an unmanned aircraft over arid rangelands. GIScience and Remote Sensing 48 (1), 4–23. Lewis, P., Disney, M., 2007. Spectral invariants and scattering across multiple scales from within-leaf to canopy. Remote Sensing of Environment 109 (2), 196–206. Li, X., Gao, F., Wang, J., Strahler, A., 2001. A priori knowledge accumulation and its application to linear BRDF model inversion. Journal of Geophysical Research 106 (D11), 11925–11935. Li, X., Strahler, A.H., 1986. Geometric-optical bidirectional reflectance modeling of a conifer forest canopy. IEEE Transactions on Geoscience and Remote Sensing GE24 (6), 906–919. Liang, S., Fang, H., Chen, M., Shuey, C.J., Walthall, C., Daughtry, C., Morisette, J., Schaaf, C., Strahler, A., 2002. Validating MODIS land surface reflectance and albedo products: methods and preliminary results. Remote Sensing of Environment 83 (1–2), 149–162. Liu, J., Schaaf, C., Strahler, A., Jiao, Z., Shuai, Y., Zhang, Q., Roman, M., Augustine, J.A., Dutton, E.G., 2009. Validation of moderate resolution imaging spectroradiometer (MODIS) albedo retrieval algorithm: dependence of albedo on solar zenith angle. Journal of Geophysical Research 114 (D1), 106. Lucht, W., 1998. Expected retrieval accuracies of bidirectional reflectance and albedo from EOS-MODIS and MISR angular sampling. Journal of Geophysical Research 103 (D8), 8763–8778. Lucht, W., Schaaf, C.B., Strahler, A.H., 2000. An algorithm for the retrieval of albedo from space using semiempirical BRDF models. IEEE Transactions on Geoscience and Remote Sensing 38 (2), 977–998. North, P.R.J., 1996. Three-dimensional forest light interaction model using a Monte Carlo method. IEEE Transactions on Geoscience and Remote Sensing 34 (4), 946–956. North, P.R.J., 2002. Estimation of fAPAR, LAI, and vegetation fractional cover from ATSR-2 imagery. Remote Sensing of Environment 80 (1), 114–121. Oak Ridge National Laboratory Distributed Active Archive Center (ORNL DAAC), 2011. MODIS subsetted land products, Collection 5. ORNL DAAC, Oak Ridge, Tennessee, U.S.A
. (accessed 18.08.11). Pérez, A.J., López, F., Benlloch, J.V., Christensen, S., 2000. Colour and shape analysis techniques for weed detection in cereal fields. Computers and Electronics in Agriculture 25 (3), 197–212. Pocewicz, A., Vierling, L.A., Lentile, L.B., Smith, R., 2007. View angle effects on relationships between MISR vegetation indices and leaf area index in a recently burned ponderosa pine forest. Remote Sensing of Environment 107 (1–2), 322– 333. Privette, J.L., Eck, T.F., Deering, D.W., 1997. Estimating spectral albedo and nadir reflectance through inversion of simple BRDF models with AVHRR/MODIS-like data. Journal of Geophysical Research 102 (D24), 29529–29542. Purevdorj, T.S., Tateishi, R., Ishiyama, T., Honda, Y., 1998. Relationships between percent vegetation cover and vegetation indices. International Journal of Remote Sensing 19 (18), 3519–3535. Rango, A., Laliberte, A., Steele, C., Herrick, J.E., Bestelmeyer, B., Schmugge, T., Roanhorse, A., Jenkins, V., 2006. Using unmanned aerial vehicles for rangelands: current applications and future potentials. Environmental Practice 8 (3), 159– 168. Román, M.O., Gatebe, C.K., Schaaf, C.B., Poudyal, R., Wang, Z., King, M.D., 2011. Variability in surface BRDF at different spatial scales (30–500 m) over a mixed agricultural landscape as retrieved from airborne and satellite spectral measurements. Remote Sensing of Environment 115 (9), 2184–2203. Román, M.O., Schaaf, C.B., Woodcock, C.E., Strahler, A.H., Yang, X., Braswell, R.H., Curtis, P.S., Davis, K.J., Dragoni, D., Goulden, M.L., Gu, L., Hollinger, D.Y., Kolb, T.E., Meyers, T.P., Munger, J.W., Privette, J.L., Richardson, A.D., Wilson, T.B., Wofsy, S.C., 2009. The MODIS (Collection V005) BRDF/albedo product: assessment of spatial representativeness over forested landscapes. Remote Sensing of Environment 113 (11), 2476–2498.
R.C. Sharma et al. / ISPRS Journal of Photogrammetry and Remote Sensing 78 (2013) 50–57 Roujean, J.-L., Leroy, M., Deschamps, P.-Y., 1992. A bidirectional reflectance model of the earth’s surface for the correction of remote sensing data. Journal of Geophysical Research 97 (D18), 20455–20468. Salomon, J.G., Schaaf, C.B., Strahler, A.H., Gao, F., Jin, Y., 2006. Validation of the MODIS bidirectional reflectance distribution function and albedo retrievals using combined observations from the aqua and terra platforms. IEEE Transactions on Geoscience and Remote Sensing 44 (6), 1555–1565. Samain, O., Kergoat, L., Hiernaux, P., Guichard, F., Mougin, E., Timouk, F., Lavenu, F., 2008. Analysis of the in situ and MODIS albedo variability at multiple timescales in the Sahel. Journal of Geophysical Research 113 (D14), 119. Sandmeier, S., Müller, C., Hosgood, B., Andreoli, G., 1998. Physical mechanisms in hyperspectral BRDF data of grass and watercress. Remote Sensing of Environment 66 (2), 222–233. Sandmeier, S.R., Deering, D.W., 1999. A new approach to derive canopy structure information for boreal forests using spectral BRDF data. In: IEEE IGARSS Proceedings Germany, vol. 1. pp. 410–412. Schaaf, C.B., Gao, F., Strahler, A.H., Lucht, W., Li, X., Tsang, T., Strugnell, N.C., Zhang, X., Jin, Y., Muller, J.-P., Lewis, P., Barnsley, M., Hobson, P., Disney, M., Roberts, G., Dunderdale, M., Doll, C., d’Entremont, R.P., Hu, B., Liang, S., Privette, J.L., Roy, D., 2002. First operational BRDF, albedo nadir reflectance products from MODIS. Remote Sensing of Environment 83 (1–2), 135–148. Schull, M.A., Ganguly, S., Samanta, A., Huang, D., Shabanov, N.V., Jenkins, J.P., Chiu, J.C., Marshak, A., Blair, J.B., Myneni, R.B., Knyazikhin, Y., 2007. Physical interpretation of the correlation between multi-angle spectral data and canopy height. Geophysical Research Letters 34 (18), L18405. Schull, M.A., Knyazikhin, Y., Xu, L., Samanta, A., Carmona, P.L., Lepine, L., Jenkins, J.P., Ganguly, S., Myneni, R.B., 2011. Canopy spectral invariants, Part 2: Application
57
to classification of forest types from hyperspectral data. Journal of Quantitative Spectroscopy and Radiative Transfer 112 (4), 736–750. Strahler, A.H., Muller, J.P., Lucht, W., Schaaf, C.B., Tsang, T., Gao, F., Li, X., Lewis, P., Barnsley, M.J., 1999. MODIS BRDF/albedo product: algorithm theoretical basis document version 5.0. MODIS documentation. Sugiura, R., Noguchi, N., Ishii, K., 2005. Remote-sensing technology for vegetation monitoring using an unmanned helicopter. Biosystems Engineering 90 (4), 369– 379. Susaki, J., Yasuoka, Y., Kajiwara, K., Honda, Y., Hara, K., 2007. Validation of MODIS albedo products of paddy fields in Japan. IEEE Transactions on Geoscience and Remote Sensing 45 (1), 206–217. Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment 8 (2), 127–150. Wang, Z., Schaaf, C.B., Chopping, M.J., Strahler, A.H., Wang, J., Román, M.O., Rocha, A.V., Woodcock, C.E., Shuai, Y., 2012. Evaluation of moderate-resolution imaging spectroradiometer (MODIS) snow albedo product (MCD43A) over tundra. Remote Sensing of Environment 117, 264–280. Wang, Z., Schaaf, C.B., Lewis, P., Knyazikhin, Y., Schull, M.A., Strahler, A.H., Yao, T., Myneni, R.B., Chopping, M.J., Blair, B.J., 2011. Retrieval of canopy height using moderate-resolution imaging spectroradiometer (MODIS) data. Remote Sensing of Environment 115 (6), 1595–1601. Wanner, W., Li, X., Strahler, A.H., 1995. On the derivation of kernels for kerneldriven models of bidirectional reflectance. Journal of Geophysical Research 100 (D10), 21077–21089. Xiang, H., Tian, L., 2011. Development of a low-cost agricultural remote sensing system based on an autonomous unmanned aerial vehicle (UAV). Biosystems Engineering 108 (2), 174–190.