alexa Fusion of Hyperspectral and L-Band SAR Data to Estimate Fractional Vegetation Cover in a Coastal California Scrub Community | Open Access Journals
ISSN: 2469-4134
Journal of Remote Sensing & GIS
Make the best use of Scientific Research and information from our 700+ peer reviewed, Open Access Journals that operates with the help of 50,000+ Editorial Board Members and esteemed reviewers and 1000+ Scientific associations in Medical, Clinical, Pharmaceutical, Engineering, Technology and Management Fields.
Meet Inspiring Speakers and Experts at our 3000+ Global Conferenceseries Events with over 600+ Conferences, 1200+ Symposiums and 1200+ Workshops on
Medical, Pharma, Engineering, Science, Technology and Business

Fusion of Hyperspectral and L-Band SAR Data to Estimate Fractional Vegetation Cover in a Coastal California Scrub Community

Shuang Li1, 2, Christopher Potter1*, Cyrus Hiatt3 and John Shupe3
1NASA Ames Research Centre Mail Stop 242-4, Moffett Field, CA 94035, USA
2Henan University, College of Environment and Planning, Kaifeng, Henan 475004, China
3California State University Monterey Bay, Seaside, CA, USA
Corresponding Author : Christopher Potter
NASA Ames Research Centre Mail Stop 242-4
Moffett Field, CA 94035, USA
Tel: 650-604-6164
Fax: 650-604-4680
E-mail: [email protected]
Received May 05, 2012; Accepted May 25, 2012; Published May 30, 2012
Citation: Li S, Potter C, Hiatt C, Shupe J (2012) Fusion of Hyperspectral and L-Band SAR Data to Estimate Fractional Vegetation Cover in a Coastal California Scrub Community. J Geophys Remote Sensing 1:104. doi:10.4172/2169-0049.1000104
Copyright: © 2012 Li S, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Visit for more related articles at Journal of Remote Sensing & GIS

Abstract

A study was carried out to investigate the utility of airborne hyperspectral and satellite L-band Synthetic Aperture Radar (SAR) data for estimating fractional coverages of herbaceous, coastal scrub, and bare ground cover types on the central California coast. Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) imagery collected in September of 2008 and Phased Array L-band SAR (PALSAR) (HH- and HV-polarizations) captured in April and July of 2008 were combined for vegetation cover mapping. Hyperspectral features, computed as AVIRIS indices (NDVI, TCARI/OSAVI, and PRI), and textural information (energy, contrast, homogeneity, and fractal dimension) produced by L-band SAR were fused together to generate a new feature space. We used global Ordinary Least Squares (OLS) linear regression to integrate and decompose the new feature space for fractional vegetation mapping. Ground measurements of fractional cover were collected from plots located within the U.S. Forest Service’s Brazil Ranch study site for validation of the OLS model predictions. Significant linear relationships were found between fractional cover mapping from remote sensing and the ground-truth data. The estimation accuracy of fractional coverage mapping from remote sensing in terms of Root Mean Square Error (RMSE) was 17%, 12%, and 10%, for the herbaceous, coastal scrub, and bare ground covers, respectively. Decomposition results showed that textural information from L-band SAR strongly supported herbaceous and coastal scrub fractional mapping, while indices features from AVIRIS significantly improved mapping of herbaceous cover and bare ground.

Keywords
AVIRIS; ALOS-PALSAR; L-band SAR; Fractional coverage; Coastal scrub
Abbreviations
HBV: Herbaceous; CS: Coastal Scrub; BG: Bare Ground
Introduction
Coastal shrub ecosystems in California have a high degree of biological diversity and endemism, and provide critical habitat for a large number of rare, endangered, and threatened animal and plant species [1]. These mixed herbaceous-shrub communities are of interest because they dominate the central and southern coastal regions of California, but have been largely overlooked as a key biomass carbon component. Coastal scrub covers 7% of the region and is the fourth most extensive vegetation class in the state.
Our study focused on fractional vegetation coverage mapping for herbaceous-shrub ecotypes on the Big Sur coast in Monterey County, CA. The regional-scale products of such remote sensing can uniquely support wildlife habitat mapping and biogeochemical cycling studies. Visible and Near-infrared (V/NIR) imagery from airborne and spaceborne remote sensing have been used in previous studied to investigate the herbaceous-shrub ecotype [2,3]. Hyperspectral imagery has been widely used for vegetation cover mapping [4]. There has been increasing interest in using Synthetic Aperture Radar (SAR) data [5-7], and in combining hyperspectral and SAR data for improved vegetation mapping and estimation of vegetation structural variables [8-11].
Previous studies with satellite SAR for observing vegetation coverage have indicated that backscatter intensity is of little use for forest and shrubland detection at C-band (6 cm) and S-band (10 cm) wavelengths [10-12]. Yatabe [12] compared the research applications of SAR at different radar frequencies (C-, S- and L-band) and found that L-band (24 cm wavelength) was the most effective for discriminating forest and shrubland. Hyperspectral imagery provides many attributes complementary to vegetation canopy information from SAR data and can be used to detect vegetation health status based on spectral characteristics [13,14]. Jouan [9] reported on fusion of SAR and hyperspectral imagery to map land cover by using the evidential fusion method. Blaschke [15] extracted information from SAR and hyperspectral data by an object-based approach. Huang [10] fused Air SAR, AVIRIS and Landsat data for fractional cover mapping in Yellowstone National Park.
In this study, we investigated the fusion of hyperspectral imagery and L-band SAR data for fractional coverage mapping of herbaceous-shrub ecotype in central California. L-band SAR data are available from the Phased Array L-band SAR (PALSAR) instrument, which was installed on the Advanced Land Observing Satellite (ALOS). Hyperspectral imagery from the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) was collected by NASA Jet Propulsion Laboratory (JPL) in 2008. A new feature space was created by combining spectra information from a standard spectral library, vegetation indices from hyperspectral imagery, and textural information from L-band SAR data. Fractional cover products for herbaceous, coastal scrub, and bare ground were produced by decomposing the new feature space.
Study Site
The primary research site was located at the Brazil Ranch (center coordinates: latitude 36.35° N, longitude 121.88° W) near Big Sur, California (Figure 1). The Brazil Ranch is named after Tony and Margaret Brazil and the pioneer family that worked to establish the land as a farm, ranch, as well as a dairy operation in the early 20th century. Today, the property serves as a primary research site for the U. S. Forest Service to monitor and manage vegetation, wildlife, water quality, and sensitive coastal habitats.
The Big Sur region is characterized by a Mediterranean climate with rounded ridges, steep sides, and narrow canyons. The terrain is rugged and undulating with the steepest elevation gradients on the Pacific U. S. coast, ranging (over just several km inland) from sea-level up to 1570 meters. Rainfall varies from 40 to 150 cm throughout the range, with the most on the higher mountains in the north. The majority of all precipitation falls in the winter (November-March). During the summer, fog and low clouds are frequent along the coast. Mean annual temperature ranges from 10 to 15°C.
Drier, southeast-facing slopes share a relatively equal distribution of coyote bush (Baccharis pilularis) and California coffeeberry (Rhamnus californica) along with some California sagebrush (Artemisia californica) (Ecological Subregions of California, 2011). The coastal scrub community is usually a successional plant community that, in the absence of fire, gradually moves into herbaceous cover where the soil depth transitions from the shallowest to intermediate depth. The herbaceous plant community includes California annual grassland series and California Oat grass series. Coastal sage scrub and chaparral are known as secondary pioneer plant in California grasslands, which invade grassland and increase in the absence of fire or grazing. We noted a propagation of the introduced Cape ivy (Delairea odorata) during our field work. Cape ivy, a vine native to South Africa, has become a significant threat to coastal scrub.
Remote Sensing Datasets
Hyperspectral AVIRIS imagery provided information related to the biochemical state of the herbaceous-coastal scrub ecotype. AVIRIS collects data in 224 continuous channels of approximately 10-nanometer band pass over the spectral-wavelength range of 0.35-2.50 μm (from visible light to Near-infrared). A nominal pixel size of 3.5 m was collected by NASA/JPL on September 24, 2008, at approximately 9:40 a.m. local time. The AVIRIS imagery was Orthocorrected by NASA/JPL using a full three-dimensional ray tracing method [16]. Each pixel in the image was individually ray traced using the best-estimate of sensor location and attitude until it intersected the DEM. The spatial fidelity of the data was much improved from previous datasets, especially in areas of rugged and variable across-track terrain, resulting in an accuracy of one pixel.
The AVIRIS imagery was captured mainly for the purpose of assessing the burn severity of Big Sur wildfires that occurred in 2008 (but which did not spread into the Brazil Ranch property; Table 1). Landsat 5 Thematic Mapper (TM) data were used to generate boundaries of the 2008 Big Sur wildfires and to clip the burnt area from the AVIRIS data set. Two cloud-free TM scenes (Path 43, Row 35) were selected from May 13, 2008 (pre-fire) and September 18, 2008 (post-fire).
An IKONOS image (acquired on March 08, 2007) was used to visually select 8 vegetation plots, since cliffs and steep slopes at the site made these areas otherwise inaccessible for in-situ survey assessment. Brovey transform (resolution merge) was applied on the selected IKONOS image to merge multispectral and panchromatic bands [17], and improve the spatial resolution to nearly 1 meter in the VIS/NIR bands.
PALSAR measurements were analyzed for sensitivity to the surface geometry and the dielectric constant of the illuminated surface. ALOS was launched in January 2006 by Japan Aerospace Exploration Agency (JAXA), which offered a Quad-polarization operation mode. We acquired PALSAR data over the research area for April 14, 2008 with a Fine Beam Single polarization (FBS, look angle 34.3°, HHpolarization, and a 6.25 m × 6.25 m ground resolution) and for July 18, 2008 with Fine Beam Double Polarization (FBD, look angle 34.3°, HV-polarization, ground resolution approximately 12.5 m × 12.5 m). In FBS mode, ALOS/PALSAR was operated in HH-polarization with a bandwidth of 28 MHz. In FBD mode, the polarization option was HH/HV at 14-MHz bandwidth. The operating sensor frequency is 1.27 GHz, which corresponds to a wavelength of 23.6 cm (L-band).
We acquired PALSAR data from one of the ALOS data nodes at the Alaska Satellite Facility (ASF). The SAR dataset was preprocessed to a 1.5 product level. The ASF performed the following steps: range compression using Fast Fourier Transform (FFT), secondary range compression using range migration compensation, range migration curvature corrections, azimuth compression, multi-look processing, and conversion from slant to ground range [18].
Methods
Field survey methods
From September through December of 2010, we inventoried 43 vegetation plots within Brazil Ranch, including 8 plots that were visually selected for high bare ground cover from the 1-m IKONOS imagery (due to steep slopes that made the survey plots otherwise inaccessible) [19]. We established circular plots area with a radius of 17 m. Plots were set up by marking a center point and estimating vegetation percentage within a 17 m radius around the center point. Each plot was divided into four quads to improve the precision. Four field-crew members each independently estimated vegetation fraction percentages (herbaceous, coastal scrub and bare ground) in each quad by ocular estimation. Ocular estimation is an accurate and widely employed method for vegetation evaluation [11,20-22]. We compiled and averaged four quads to arrive at final vegetation cover estimation for each plot. The center location of each plot was positioned by Garmin GPSMAP 60CX unit in carrier phase (set to maximize spatial accuracy). The coordinates of plot center were differentially corrected by the National Geodetic Survey using the network of base station data (NGS, http://www.ngs.noaa.gov/CORS/Data.html).
Wildfire boundaries from TM imagery
We used a remotely sensed burn severity index called Differenced Normalized Burn Ratio (dNBR) derived from TM data to delineate the 2008 wildfire boundary. The TM sensor is appropriate for burn severity analysis because it records Near Infrared (NIR) and Middle Infrared (MIR) reflectance in bands 4 and 7, respectively. TM4 is primarily dependent on the refractive index of leaf morphology and discontinuities within the leaf [23] while TM7 is sensitive to water content in both soils and vegetation [24].
The TM images were converted into radiance and then at-sensor reflectance using instrument gains and offsets. The MODTRAN4 model was further used for atmosphere correction [25]. The spectral index of NBR was calculated from TM4 and TM7 (with central wavelength of 0.83 and 2.22 μm, respectively) bands according to [Eq. (1)] [26,27].
NBR = (NIR - MIR) / (NIR + MIR)          (1)
dNBR is the multi-temporal difference of pre- and post-fire NBR [28], defined as:
dNBR = NBRprefire – NBRpostfire       (2)
Hyperspectral image processing
The AVIRIS dataset was clipped to the 2008 dNBR fire boundary. We converted AVIRIS radiance to reflectance via atmospheric correction using the FLAASH algorithm. The FLAASH method is based on observations by Kaufman [29], of a nearly fixed ratio between the reflectances of pixels at 660 nm and 2100 nm. It performs a second and final MODTRAN4 calculation loop over water.
Atmospherically corrected AVIRIS data were used to calculate vegetation indices. Four indices (NDVI, OSAVI, TCARI, and PRI) were selected in this study to generate spectral space for extraction of vegetation information [Eq. (3),(4),(5),(6)). These indices have been related to Leaf Area Index (LAI) and vegetation biochemical state, including chlorophyll absorption or other specific features.
The Normalized Difference Vegetation Index (NDVI) is a measure of vegetation greenness cover [30,31], and can be used to discriminate vegetated from bare ground. AVIRIS Optimized Soil Adjusted Vegetation Index (OSAVI) represented improvements in the dynamic range or decreased sensitivity to differences in soil backgrounds [32]. The chief advantages of OSAVI are its simplified formulation and the lack of a requirement for a priori knowledge of the soil type. This index is suitable for vegetation applications since the residual variation in OSAVI is evenly spread across the full range of vegetation index response.
AVIRIS Transformed Chlorophyll Absorption in Reflectance Index (TCARI) provides information to estimate the active radiation absorbed for photosynthesis. The combination of TCARI/OSAVI permits a qualitative estimation of the chlorophyll content of leaves [33]. AVIRIS Photochemical Reflectance Index (PRI) measures xanthophyll activity, which is usually applied to vegetation detection prior to senescence [34].
NDVI = (R831 – R638) / (R831 + R638) (3)
OSAVI = (1+ 0.16) × (R800 – R670) / (R800 + R670 + 0.16) (4)
TCARI= 3 × [(R700 – R670) – 0.2 × (R700 – R550) × (R700 / R670)] (5)
PRI = (R531 – R570) / (R531 + R570) (6)
PALSAR processing
Radiometric calibration: Radiometric calibration of PALSAR data was carried out using the following method [Eq. (7)]. Digital numbers (the amplitude of the backscattered signal) of PALSAR data were transformed into a backscattering coefficient (in decibels).
image (7) [18]
Where the calibration constant for PALSAR L1.5 products is Kdb = -83 dB.
Radiometric terrain correction: Radar backscatter is significantly impacted by terrain undulations. Slope-induced distortions can have a direct impact on radiometric quality [6,35,36]. The correction of these effects becomes important when quantitative image analysis is performed with respect to geo- and biophysical parameters [37]. A 10-m resolution Digital Elevation Model (DEM) was used to correct terrain induced distortions (United States Geological Survey (USGS), National Elevation Dataset (NED)). Based on a lookup table describing the transformation between the radar and map geometry, Ulander [38] developed an approach [Eq. (8) and (9)] to minimize the dependence on terrain undulation [37]. The lookup table was generated using the NED DEM and the orbital information of the PALSAR data. The normalized, terrain-corrected radar backscattering coefficient is defined as:
image (8)
where is image the averaged radar brightness and ψ is the projection angle between the surface normal and the image plane normal, which varies between 0°and 90°, and θloc is the local incidence angle. We note that ψ is the complementary angle to the smallest angle between the surface normal and the image plane.
Calculation of cosψ is given by Eq. (9)
cos(ψ ) = sin(θ ).cos(u) + cos(θ ).sin(u).sin(v) (9)
Where θ is the local incidence angle of a horizontal surface patch (i.e. ellipsoidal incidence angle), and u,v are terrain slope and aspect of the surface relative the vertical and azimuth directions, which are calculated from the DEM [38].
We note that, in the correction for slope-induced backscattering distortion, the backscattering coefficient tends to vary (decrease) with increasing local incidence angle (θ ) in the non-radiometric corrected (but terrain-corrected) curve, while the radiometric-corrected data shows a stable backscattering coefficient over the local incidence angle (Figure 2).
Radiometric corrected PALSAR backscattering coefficients (HH and HV) were co-registered with the AVIRIS imagery and resampled at the same spatial resolution as AVIRIS (Figure 3). Co-registration accuracy was estimated to be 0.5 pixels. The speckle noise in PALSAR backscattering coefficients was minimized by applying Lee-Sigma filters with a 5 × 5 moving-window before textural information extraction [39].
Textural feature extraction from PALSAR data: Three texture features called energy, contrast, homogeneity were extracted from PALSAR backscattering coefficients by using the method of cooccurrence matrices (GLCM). The fractal dimension of PALSAR backscattering coefficients was extracted using the triangular prism surface area method (TPSAM) [40]. These four features were generated from HH- and HV-polarizations comprising a set of 2 × 4 bands.
Energy is also called the angular second moment [41], which measures textural uniformity. Contrast is the spatial frequency that represents the amount of the local variation in the scene. Local homogeneity is called the inverse difference moment. For a specific vegetation area, local homogeneity and contrast are inversely correlated, while energy is kept constant. On the other hand, local homogeneity and energy are inversely correlated, while contrast remains constant. Fractals measure the roughness attributes in the SAR data.
Classification scheme
Endmember selection: Image pixels within the Brazil Ranch study area were visually separated into two types based on their land cover composition: pure pixels covered entirely by a single cover class (herbaceous, coastal scrub, or bare ground), and mixed pixels composed of combinations of the above-mentioned classes. Three endmembers were identified from the pure pixels and classified by using a priori information from a spectral library [24].
The following steps were performed to reach aforementioned outcome: First, we used rotation transforms on AVIRIS imagery. Minimum Noise Fraction Transform (MNF) is an algorithm designed to determine the inherent dimensionality of hyperspectral imagery, segregate noise in the data, and reduce the computational requirements for subsequent processing [42]. A threshold value of 0.27 was selected based on the plots of spatial coherence and eigenvalues. This threshold identified the first 39 bands in MNF space to be used for spectral information extraction, while the remaining bands were considered noise.
Next, we used the Pixel Purity Index (PPI) [43] to find the most spectrally pure pixels in the MNF space. The PPI was computed by repeatedly projecting n-D scatter plots onto a random unit vector. Each resulting value corresponded to the number of times that a pixel was recorded as extreme. A threshold value of 20 was selected as critical value for space partitioning in our study area.
Lastly, the pure pixels were separated into three classes. The spectral library provided by Elvidge [24] was selected as a priori information to perform supervised Bayes Maximum Likelihood classification [44]. The Elvidge spectral library was measured as hemispherical reflectance using a Beckman UV-5240 spectrophotometer at Jasper Ridge in central California, which included all the vegetation species (in both dry and wet seasons) also found in the Brazil Ranch study site. Each pure pixel was categorized as one of the three classes, namely, herbaceous, coastal scrub, and bare ground (Figure 4).
Construction of the new feature space: Pixels with a PPI value less than 20 were identified and a new 3-D feature space was constructed to combine spectra information, index features, and L-band textural information for fractional decomposition. The new feature space was composed of 50 bands (39 bands generated by MNF from AVIRIS imagery, 3 bands from AVIRIS vegetation indices, and 2×4 texture bands from PALSAR data).
For the first dimension, basic spectral information was provided by the 39 MNF bands. Examination of the vegetation and soil information contained in the MNF-transformed data, together with the associated eigenvalues, indicated that 94% of the total statistical variance in the AVIRIS imagery was contained in the first 39 MNF bands of the image. For the second dimension, phenological characteristics of vegetation were observed in all three indices. Annual herbaceous species dominate the great majority of the grasslands in Big Sur. Most of the herbaceous materials had senesced, and coastal scrub was still green or lightgreen when our AVIRIS data was acquired. Peak green season in the central coast region of California occurs from around February 15 to March 20, after which herbaceous vegetation cover gradually turns brown. TCARI/OSAVI and PRI provided the needed information for phenological vegetation discrimination. The ratio of TCARI/OSAVI is especially useful when herbaceous vegetation, coastal scrub, and bare ground were co-located in a pixel. For the third dimension, the 2 × 4 texture features from SAR data were used [45, 46].
Decomposition from OLS: OLS regression is widely used to infer linear regression model parameters in the remote sensing literature [47]. We applied OLS regression on the 3-D feature space constructed from 50 bands. Signatures of the pure pixels (assigned 100% coverage of either herbaceous, coastal scrub, or bare ground) were used as explanatory variables in the OLS regression with each mixed pixel’s spectral signature as the response variable.
The generalized decomposition model is described as Eq. (10):
image (10)
where image is the modeled mixed pixel feature vector in the new combined feature space, βn s are coefficients, Rns refer to herbaceous vegetation, coastal scrub, and bare ground respectively, and ε is random error term.
Three continuous raster layers were generated by OLS analysis as the proportion of each cover type represented in the mixed pixels. A composite fractional coverage map was generated from the three mixed raster layers merged with the pure pixels. This overall approach is summarized in a general flow chart (Figure 5).
Results
Fractional vegetation mapping
The mean value of herbaceous cover was higher at Brazil Ranch than the other two cover types at around 51.6%, and was generally highest on the south-facing ridge top locations. The percentage of herbaceous cover declined gradually with decreasing elevation toward the northeast (Figure 6). Coastal scrub was distributed more across the valley and the gently sloping areas. The mean cover value for coastal scrub at the site was 21.9%. Bare ground often coexisted with herbaceous cover on ridge tops, but we also detected scattered patches of coastal scrub on the steep slopes.
An Iso-clustering algorithm [44,48] was further used to characterize the landscape pattern. Five ecotypes were determined by the combination of different cover percentages (Figure 8). The first three ecotypes were dominated by the individual coverages of herbaceous, coastal scrub, and bare ground with percentage of 86%, 89%, and 77% respectively. The fourth mixed ecotype was composed of coastal scrub (more than 50%), herbaceous (around 30%), and bare ground (less than 20%). The fifth ecotype was dominated by a combination of coastal scrub (56%) and bare ground (41%).
Fractional coverage accuracy assessment
We compared field survey estimations of vegetation fractional cover to remote sensing predictions from a fusion of SAR and hyperspectral imagery (Figure 9). Linear regression results produced coefficients of determination for herbaceous, coastal scrub, and bare ground cover of R2 = 0.80, 0.89, and 0.92, respectively. The estimated accuracy of fractional coverage mapping from remote sensing was calculated in terms of Root Mean Square Error (RMSE) at 17%, 12%, and 10% for herbaceous, coastal scrub, and bare ground, respectively.
Contributions assessment
We examined the relative contributions of vegetation index features and textural information to the combined OLS analysis results (Table 2 and Table 3). First, relationships of separate features to measured percent cover data sets were determined by simple regression. These outcomes were then compared to the predictions of percent cover from the combined feature space (labeled as E contributions). A higher R value indicates higher contribution from indices space to the combined feature space, as shown in the rows of Tables 2 and 3 as E contributions.
Features in each of the vegetation indices contributed to discrimination of bare ground to a higher degree than for the other two vegetation classes, while textural information contributed to the discrimination of coastal scrub to a higher degree than for the other two classes. The index features were directly related to the photosynthetic capacity and, hence, the energy absorption of vegetation. This association contributed notably to the discrimination of herbaceous cover from the soil background. For example, PRI index values would be low and declining during the growth phase of the grass canopy and increase rapidly during the senescence period, whereas the PRI would remain relatively constant year-round for bare ground areas.
L-band SAR data contributed useful information to detecting the coverage of coastal scrub (Table 3), with an R valus of 0.62, but made no contribution to the detection of bare ground. The L-band SAR (wavelength ~24 cm) has the advantage of obtaining better coherence than C-band SAR in dense vegetation. This advantage is due to the L-band’s capacity to penetrate the coastal scrub canopy and scatter off trunks and branches, thereby distinguishing woody tissues of shrubs from herbaceous cover.
We summarized the results using OLS methods to decompose the fractional vegetation coverages from the combined feature space using the inputs of HV, HH, NDVI, PRI and TCARI/OSAVI in Table 4. Variance Inflation Factor (VIF) values were lower than 7.5 for all input variables (except for that of TCARI/OSAVI) which indicated no explanatory variable redundancy during the weighted decomposition [49]. VIF quantifies the degree of multicollinearity in an OLS regression analysis. It provides a measure of how much the variance of an estimated regression coefficient (the square of the estimate’s standard deviation) is increased because of collinearity. Robust regression results were included in Table 3 to evaluate the effects of bad leverage outliers that would otherwise bias the parameter estimation with a non-normal distribution. A robust determination down-weights outliers and also accounts for non-normality in sample distributions. The results of these two parameters (t-test and Robust Probability) indicated that the SAR HV or HH explanatory variables were statistically significant (p < 0.1) in the OLS for herbaceous and bare ground covers.
Discussion
Remote sensing is the only practical method to map vegetation types in the steep and inaccessible mountains and valleys of the central Pacific coast. The results presented in our study offer a baseline mapping estimate of vegetation status in an area of the western United States subject to extreme weather events, climate change, and regular wildfires [27]. The methods described above can be replicated in years to come to assess even subtle changes in central California’s coastal vegetation cover.
In ecosystems of the Central Coast, radar backscatter signals received from the terrestrial surface included many pixels with mixed vegetation cover. The backscatter coefficient is contingent on incident angle, wavelength, polarization, surface roughness, and dielectric properties of the surface. The three basic endmembers in our study area have backscattering coefficients that varied significantly. Specular or surface scattering occurred in the area dominated by herbaceous and bare ground. Volume scattering and double bounce scattering occurred in coastal scrub dominated areas. Volume and surface scattering played an important role in the response from coastal scrub plant communities, which increased the overall backscatter magnitude because of the presence of and interaction among different scattering mechanisms.
Our results showed that the fusion of hyperspectral imagery and L-band SAR data can be used for accurate fractional vegetation mapping in the herbaceous-shrub communities of coastal California. The most striking results were obtained with the addition of L-band SAR texture features to help discriminate herbaceous cover from coastal scrub. Textural information from SAR data improved the fractional decomposition significantly. Expanded map products for vegetation fractional cover can next be ingested into biogeochemical cycling models [50,51] for the entire central California coastal region to improve annual plant production and fuel biomass loading predictions.
Acknowledgements
This research was supported by an appointment of the first author to the NASA Postdoctoral Program at the NASA Ames Research Center, administered by Oak Ridge Associated Universities through a contract with NASA. The authors are grateful to the U. S. Forest Service, Los Padres National Forest (Ecosystem Manager Jeff Kwasny), for providing access to the Brazil Ranch property. The authors would like to thank NASA JPL for AVIRIS data acquisition and preprocessing in 2008.
References


Tables and Figures at a glance

Table icon Table icon Table icon Table icon
Table 1 Table 2 Table 3 Table 4

 

Figures at a glance

Figure Figure Figure Figure Figure
Figure 1 Figure 2 Figure 3 Figure 4 Figure 5

 

Figure Figure Figure Figure
Figure 6 Figure 7 Figure 8 Figure 9
Select your language of interest to view the total content in your interested language
Post your comment

Share This Article

Article Usage

  • Total views: 4299
  • [From(publication date):
    July-2012 - Sep 27, 2017]
  • Breakdown by view type
  • HTML page views : 483
  • PDF downloads :3816
 

Post your comment

captcha   Reload  Can't read the image? click here to refresh

Peer Reviewed Journals
 
Make the best use of Scientific Research and information from our 700 + peer reviewed, Open Access Journals
International Conferences 2017-18
 
Meet Inspiring Speakers and Experts at our 3000+ Global Annual Meetings

Contact Us

Agri, Food, Aqua and Veterinary Science Journals

Dr. Krish

[email protected]

1-702-714-7001 Extn: 9040

Clinical and Biochemistry Journals

Datta A

[email protected]online.com

1-702-714-7001Extn: 9037

Business & Management Journals

Ronald

[email protected]

1-702-714-7001Extn: 9042

Chemical Engineering and Chemistry Journals

Gabriel Shaw

[email protected]

1-702-714-7001 Extn: 9040

Earth & Environmental Sciences

Katie Wilson

[email protected]

1-702-714-7001Extn: 9042

Engineering Journals

James Franklin

[email protected]

1-702-714-7001Extn: 9042

General Science and Health care Journals

Andrea Jason

[email protected]

1-702-714-7001Extn: 9043

Genetics and Molecular Biology Journals

Anna Melissa

[email protected]

1-702-714-7001 Extn: 9006

Immunology & Microbiology Journals

David Gorantl

[email protected]

1-702-714-7001Extn: 9014

Informatics Journals

Stephanie Skinner

[email protected]

1-702-714-7001Extn: 9039

Material Sciences Journals

Rachle Green

[email protected]

1-702-714-7001Extn: 9039

Mathematics and Physics Journals

Jim Willison

[email protected]

1-702-714-7001 Extn: 9042

Medical Journals

Nimmi Anna

[email protected]

1-702-714-7001 Extn: 9038

Neuroscience & Psychology Journals

Nathan T

[email protected]

1-702-714-7001Extn: 9041

Pharmaceutical Sciences Journals

John Behannon

[email protected]

1-702-714-7001Extn: 9007

Social & Political Science Journals

Steve Harry

[email protected]

1-702-714-7001 Extn: 9042

 
© 2008-2017 OMICS International - Open Access Publisher. Best viewed in Mozilla Firefox | Google Chrome | Above IE 7.0 version
adwords