Miller, J. R., B. Hu, and R. Fernandes. 2001. BOREAS Follow-On DSP-06 CASI LAI and Canopy Closure of Conifer Flux Tower Sites. Data set. Available on-line [http://www.daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, U.S.A.
1.2 Data Set Introduction
The leaf area index (LAI) and canopy
closure images over conifer tower sites in the Southern Study Area (SSA)
and Northern Study Area (NSA) were produced from the Compact Airborne Spectrographic
Imager (CASI) high spatial resolution winter data using the Forest-Light
Interaction Model (FLIM) (Rosema et al., 1992). The CASI images used have
been calibrated (Gray et al., 1997), atmospherically corrected (O'Neill
et al., 1997), and geo-referenced. For application of a canopy BRDF model,
such as FLIM, the original high-resolution CASI images were spatially degraded
to a resolution of 30 m by 30 m, in order to ensure that a tree crown and
its shadow occur in the same pixel.
Additional LAI and canopy closure images
were produced over the SSA-OBS and NSA-OBS sites using the FLIM-CLUS algorithm
(Fernandes et al., 1999). These supplementary images, which delineate open
areas at 2m resolution and canopy areas at a resolution ranging from 2m
to 30m, are provided in an effort to map the large number of open gaps
in the OBS areas. Visual evaluation indicates that the spatial patterns
in the canopy LAI and canopy closure images are consistent with those exhibited
by the color composite CASI images for the same sites. The canopy LAI images
were further validated by field-measured LAI along specific transects.
The R-squared correlation between the retrieved canopy LAI and the field-measured
LAI ranges from 0.51 to 0.86 for the sites investigated.
1.3 Objective/Purpose
Experts in forest ecosystem functioning
can exploit the LAI and canopy closure images. The images may also be used
together with CASI summer images to map the spatial distribution of the
understory vegetation, and for correcting other remote sensing data products
(such as snow water equivalent estimates) that rely on estimates of canopy
cover or leaf area.
1.4 Summary of the Parameters
The parameters include canopy leaf area
index and canopy closure derived from FLIM (Hu et al., 1999) and FLIM-CLUS
(Fernandes et al., 1999). The canopy closure takes into account both within-
and between-crown gap probability.
1.5 Discussion
The open overstory and spatial and temporal
variations of the understory vegetation in the boreal forests pose special
challenges to the extraction of the canopy LAI and canopy closure using
optical indices (such as NDVI and SR). For the data set submitted, the
canopy LAI and canopy closure were retrieved from CASI images by canopy
model inversion (Hu et al., 1999), and by combining canopy model inversion
with multispectral clustering (Fernandes et al., 1999).
The FLIM model is a stochastic model
developed to describe the bi-directional reflectance properties of forest
canopies. It accounts for both the effects of shadowing of the overstory
and crown transmittance (Rosema et al., 1992). This model has been applied
to Thematic Mapper (TM) data of the Kootwijk Forest in the Netherlands
to derive LAI and canopy coverage (Rosema et al., 1992). Gemmell and Varjo
(1999) investigated the inversion of FLIM using simulated red and near-infrared
reflectance data for a wide variety of stand characteristics in the boreal
forest. Their results illustrate that the FLIM can be successfully inverted
to retrieve the overstory crown coverage and LAI, given understory reflectance
and the physical parameters of the tree crown. The research of Gemmell
and Varjo (1999) also shows that the at-canopy reflectance is sensitive
to the understory reflectance for TM image viewing geometry and indicates
that an important limitation to the success of model inversion may be the
variations of the understory reflectance. Miller et al. (1997) have shown
that the understory reflectance exhibits considerable variability both
spatially (along measurement transects at the flux tower sites) and temporally
(during the growing season). As an example, the standard deviation of the
understory reflectance for the SSA-OBS site is 0.035 the red band and 0.065
in the near-infrared band, and for the NSA-YJP site, 0.047 in the red band
and 0.055 in the near-infrared band. The CASI winter images were therefore
used to derive the canopy closure and canopy LAI for the data set submitted,
exploiting a relatively uniform and spectrally featureless snow background.
1.6 Related Data Sets
This data set is related to the surface
reflectance imagery acquired by CASI in the winter of 1994 (RSS-19). Other
data sets mapping leaf area index over the study sites include:
BOREAS RSS-04 1994 Southern Study Area Jack Pine LAI & FPAR Data
BOREAS RSS-07 LAI, Gap Fraction, and FPAR Data
BOREAS RSS-07 Landsat TM LAI Images of the SSA and NSA Data
BOREAS RSS-07 Regional LAI and FPAR Images From Ten-Day AVHRR-LAC Composites
BOREAS TE-06 Biomass and Foliage Area Data
BOREAS TE-06 Multiband Vegetation Imager Data
BOREAS TE-23 Canopy Architecture and Spectral Data From Hemispherical
Photos
2.2 Title of Investigation
Direct Methods of Characterization of
Overstory and Understory at BOREAS Flux Towers: Retrievals of Crown Closure
and LAI
2.3 Contact for Data Production Information
Contact 1:
Dr. John R. Miller
York University
Toronto, ON, Canada
(416) 736-2100 ext. 77729
(416) 736-5626 (Fax)
jrmiller@yorku.ca
Contact 2:
Dr. Baoxin Hu
York University
Toronto, ON, Canada
(416) 736-2100 ext. 33854
(416) 736-5626 (Fax)
baoxin@yorku.ca
Contact 3:
Dr. Richard Fernandes
Center for Research in Earth and Space Technology (CRESTECH)
York University
Toronto, ON, Canada
fernande@geog.utoronto.ca
Channel Center Wavelength (nm) Half Bandwidth (nm) ------- ---------------------- ------------------- 1 485.62 36.01 2 543.23 22.83 3 583.26 18.43 4 635.92 15.81 5 666.37 15.84 6 798.87 22.34 7 865.23 25.20Table 1: CASI multi-spectral band set of winter 1994 imagery
The CASI data have been calibrated (Gray et al., 1997), atmospherically corrected (O'Neill et al., 1997) and geo-referenced before being used to retrieve the canopy LAI and canopy closure. The algorithms to retrieve the canopy LAI and canopy closure are described in section 9.1.1.
4.1.1 Collection Environment
The CASI sensor was flown aboard the Chieftain Navaho aircraft operated by the Ontario Provincial Remote Sensing Office (PRSO). Data was collected at 2332.65 meters AGL with the viewing angle at nadir. The sensor was operated at the spatial mode (see Table 1 for the band set).4.1.2 Platform
The CASI sensor was flown aboard the Chieftain Navaho aircraft operated by the Ontario Provincial Remote Sensing Office (PRSO). Data were collected at 2332.65 meters AGL.4.1.3 Mission objective
CASI was used to generate reflectance from radiance images collected during the BOREAS field campaigns. These images will be used along with other sensors, such as Airborne Visible and Infrared Imaging Spectrometer (AVIRIS), Special Sensor Microwave/Imager (SSM/I), Landsat Thematic Mapper (TM), Geostationary Operational Environmental Satellite (GOES), Advanced Very High Resolution Radiometer (AVHRR) and Satellite Pour l'Observation de 1a Terre (SPOT) images to determine: (1) model-based and image-based methods to obtain surface reflectance from airborne optical imagery, (2) the role of seasonal changes in understory components to changes in reflectance of open boreal canopies, and (3) the seasonal/temporal variation in closed canopy reflectance as a function of canopy architecture, species composition, canopy biophysical parameters of LAI and biomass, and phenologic development and chemistry of foliar components.4.1.4 Key Variables
The CASI sensor directly measures the following quantities: at-sensor radiance, downwelling and upwelling irradiance.4.1.5 Principles of Operation
CASI is a pushbroom imaging spectrograph in the visible and near-infrared spectrum with a reflection grating and a two-dimensional CCD (charge coupled device) solid-state array detector. The instrument operates by looking down in a fixed direction and imaging successive lines of the scene under the platform, building up a two-dimensional image as the platform moves forward (Anger et al. 1990). The CASI instrument has been used in a variety of applications from forest cover mapping to pollution monitoring.4.1.6 Instrument Measurement Geometry
Data were collected at various altitudes from 160 m AGL to 2600 m AGL. The CASI sensor can be tilted forward or aft of the aircraft. Data has been collected at a number of different sensor pitch angles from -45 to + 45 degrees for investigations of BDRF effects.4.1.7 Manufacturer of Instrument
Itres Instruments
#155, 2635-37th Avenue N.E.
Calgary, Alberta, Canada
T1Y 5Z6
Tel: (403)-250-9944
Fax: (403)-250-9916
4.2 Calibration
The CASI instrument has been calibrated
in the Instrument Services Laboratory at Center for Research in Earth and
Space Technology, CRESTECH (formerly ISTS). A two-stage approach was used
to determine the absolute and relative radiometric response of the instrument.
An integrating sphere (Thomas et. al. 1991) was used to determine the relative
response of the CCD array. For an absolute radiometric calibration, a calibrated
Spectralon reflectance panel (Labsphere) and a NRC traceable calibrated
FEL lamp were used. Dark current data were taken to remove any instrument
anomalies. A more detailed description of the calibration can be found
in (Harron et. al. 1995, Babey and Soffer 1992 and Gray et al. 1997).
Mercury, Argon, etc. narrow line lamps
were used to determine the spectral calibration of the instrument. A narrow-bandwidth
tiltable filter was used to determine the spectral band pass of the instrument
at selected wavelengths. The transmission of the window used during data
collection was characterized at a range of view angles to enable correction
of CASI data to at-sensor radiance.
Radiance Imagery Units are W/(m2 µm sr) * 100
Reflectance Imagery Units are %refl * 500
Radiance imagery is converted to at-ground measured reflectance using a variant of the 5S radiative transfer code, referred to as CAM5S (O'Neill et al., 1997). This procedure allows correction of the imagery at aircraft altitudes for each pixel in the scene. Further details can be found in Gray et al. 1997.
4.2.1 SpecificationsReturn to top of document.
(As given by instrument manufacturer ITRES)Field of View: 37.8 degrees across track, 0.076 degrees along-track
Spectral Range: 545 nm spectral window between about 400 and 1000 nm
Spectral Samples: 288 max at 1.9 nm intervals
Spectral Resolution: 2.2 nm FWHM @ 650 nm
Aperture: f/2.8 to f/11.0
Dynamic Range: 12 bits
Noise Floor: 1.4 DN
Signal-to-Noise Ratio: 420:1 peak
Data Throughput: 420 Kbytes or greaterOperating Modes
Spatial Mode:
512 spatial pixels, 19 spectral band maximumSpectral Mode:
288 spectral pixels, 101 adjacent looks
48 spectral pixels, 511 adjacent looksEnhanced Spectral Mode (Hyperspectral Mode)
72 spectral pixels, 405 adjacent looksFull Frame Mode:
288 spectral pixels, 512 spatial pixels.Environmental Operating Conditions
Temperature: 5 to 40 C operating, -20 to +60 storage
Relative Humidity: 20% to 80% non-condensing, 10% to 90% non-condensing storage
Maximum Altitude: 3048 m above sea level (unpressurized)4.2.1.1 Tolerances
None given.
4.2.2 Frequency of Calibration
Calibration data was collected for the CASI before each IFC. While only one set of calibration data was used to calculate the Radiance Scale Factors, the other calibration data allows tracking of any changes in the instrument response.4.2.3 Other Calibration Information
The CASI sensor viewed the NASA/GSFC large integrating sphere during IFC-1 during BOREAS 1994. Also, an inter-agency calibration experiment was conducted at the Instrument Services Laboratory involving ISTS, NASA/GSFC, and CCRS (Harron et. al. 1995), which served to compare the calibration radiance sources used at these institutions. See Harron et al (1995) for details.
6.2 Field Notes
CASI operator comments are recorded
in flight logs stored as part of the ISTS BOREAS CASI image database. Examination
of the incident light probe (ILP) data suggests that the atmospheric conditions
varied by more than 10% but not more than 20% for the images used.
7.1.1 Spatial Coverage
The upper-right and lower-left corner coordinates of the data set are:Site ID Corner UTM Geographic location (meters) (degrees) ----------------------------------------------------------------------- SSA-OBS NorthWest 491695 E, 5982773 N 105.127 W, 53.993 N SouthEast 493023 E, 5981577 N 105.106 W, 53.982 N SSA-YJP NorthWest 522629 E, 5971029 N 104.656 W, 53.887 N SouthEast 524565 E, 5966641 N 104.627 W, 53.848 N SSA-OJP NorthWest 517441 E, 5975460 N 104.734 W, 53.927 N SouthEast 521209 E, 5969610 N 104.677 W, 53.875 N NSA-OJP NorthWest 522925 E, 6198622 N 98.633 W, 55.932 N SouthEast 524253 E, 6197314 N 98.612 W, 55.921 N NSA-YJP NorthWest 543998 E, 6196706 N not available SouthEast 545798 E, 6189306 N not available NSA-OBS NorthWest 531781 E, 6193687 N 98.492 W, 55.888 N SouthEast 533403 E, 6192033 N 98.612 W, 55.873 N7.1.2 Spatial Coverage Map
Not Available.7.1.3 Spatial resolution
The canopy LAI and canopy closure images are provided at 30 m spatial resolution for all study sites using the FLIM algorithm. In recognition of the substantial number of gaps present at the OBS sites, canopy LAI and canopy closure were also gridded at a 2m spatial resolution at these sites using the FLIM-CLUS algorithm. Note that the FLIM-CLUS algorithm has a variable spatial scale when estimating canopy cover at each (2m) pixel. With FLIM-CLUS, open areas are delineated at 2m scale. Canopy areas are delineated using a constrained moving window average that can range in scale from 2m to 30m depending on the presence of open areas in neighbouring pixels. Both the FLIM-CLUS and FLIM images were derived from CASI images with the spatial resolution of 2m by 2m for the SSA-OBS, SSA-OJP, SSA-YJP, NSA-OBS, NSA-OJP sites, and 1m by 1m for the NSA-YJP site.7.1.4 Projection
The canopy LAI and canopy closure images are derived from CASI images that were geocorrected to UTM 13 and UTM 14 (WGS84), for SSA and NSA sites, respectively, based on GPS and attitude data. Because of the limitations of the GPS and attitude data, the absolute positional accuracy of the imagery is not high. However, the relative positional accuracy is good. Therefore, in order to obtain higher absolute positional accuracy, the position of the flux towers were identified in the CASI imagery and the difference between the tower position in the geocorrected image and its actual position was calculated. The position of all the pixels in each image were then shifted by the calculated difference.7.1.5 Grid Description
The pixel size of the images in this data set are 30.0 meters square for FLIM products and a square pixel size of 2.0 meters for FLIM-CLUS products.
7.2 Temporal Characteristics
7.2.1 Temporal Coverage
The CASI images used for this data set were obtained in February of 1994.7.2.2 Temporal Coverage Map
Site Date SSA-OJP
SSA-YJP
SSA-OBS7-Feb, 1994
8-Feb, 1994
8-Feb, 1994NSA-OJP
NSA-YJP
NSA-OBS10-Feb, 1994
11-Feb, 1994
10-Feb, 19947.2.3 Temporal Resolution
CASI imagery represents the measured instantaneous irradiance and reflected at-sensor radiance. Frequently, multiple image acquisitions over a target were obtained during one day.
7.3 Data Characteristics
7.3.1 Parameter/Variable
The image parameters are canopy LAI and canopy closure.7.3.2 Variable Description
Canopy closure is defined as the fraction of viewed area from nadir (30 m by 30 m pixel or smaller) comprised of overstory vegetation, and it takes into account both the within- and between-crown gap probability.
Canopy LAI here refers to effective LAI (Chen et al., 1997) which takes into account the clumping effect of the canopy.7.3.3 Unit of Measurement
The canopy closure is a percentage. The canopy LAI is dimensionless and the digital number in the image data is 10 * LAI.7.3.4 Data Source
The canopy LAI and canopy closure images were produced through the application of the FLIM model to BOREAS CASI winter images (RSS-19). The field-measured LAI values that were used to validate the derived LAI were from Chen et al. (1997).7.3.5 Data Range
The digital numbers for the canopy LAI range from 0 to 50 (10 * LAI). The digital numbers for canopy closure range from 0 to 100 (in percentage).
7.4 Sample Data Record
Not applicable for image data.
8.2 Data Format
Image data created by FLIM are supplied
as Band Interleaved by Line (BIL) format. The image files are binary data
stored in 2-byte pixels. The files are named as follows:
SSA-OJP Canopy Cover: 94-02-07_ssa-ojp_cc2.img, 125 columns, 195 lines SSA-OJP Canopy LAI: 94-02-07_ssa-ojp_lai2.img, 125 columns, 195 lines SSA-YJP Canopy Cover: 94-02-07_ssa-yjp_cc2.img, 64 columns, 146 lines SSA-YJP Canopy LAI: 94-02-07_ssa-yjp_lai2.img, 64 columns, 146 lines SSA-OBS Canopy Cover: 94-02-07_ssa-obs_cc2.img, 44 columns, 39 lines SSA-OBS Canopy LAI: 94-02-07_ssa-obs_lai2.img, 44 columns, 39 lines NSA-OJP Canopy Cover: 94-02-10_nsa-ojp_cc2.img, 44 columns, 43 lines NSA-OJP Canopy LAI: 94-02-10_nsa-ojp_lai2.img, 44 columns, 43 lines NSA-OBS Canopy Cover: 94-02-10_nsa-obs_cc2.img, 54 columns, 55 lines NSA-OBS Canopy LAI: 94-02-10_nsa-obs_lai2.img, 54 columns, 55 lines NSA-YJP Canopy Cover: 94-02-11_nsa-yjp_cc2.img, 60 columns, 246 lines NSA-YJP Canopy LAI: 94-02-11_nsa-yjp_lai2.img, 60 columns, 246 linesData Files produced by FLIM-CLUS are stored in 1-byte pixels. They are:
NSA-OBS Canopy Cover: 94-02-10_nsa-obs_cc1.img, 811 columns by 827 lines NSA-OBS Canopy LAI: 94-02-10_nsa-obs_lai1.img, 811 columns by 827 lines SSA-OBS Canopy Cover: 94-02-07_ssa-obs_cc1.img, 664 columns by 598 lines SSA-OBS Canopy LAI: 94-02-07_ssa-obs_lai1.img, 664 columns by 598 linesAn independent ASCII header file is associated with each of these files. These header files contain the same information as in the following sample.
*** BOREAS CASI IMAGE HEADER *** -------------------------------- Comment : sitelo, OJP BOREAS Site Identifier : OJP Image Identifier : Tape : 110 File :15 DATE AND MISSION INFORMATION ----------------------------- Date (DD-MM-YYYY) (GMT) : 07 Feb 1994 Start Time (GMT) (HH:MM:SS) : 21:46:07 End Time (GMT) (HH:MM:SS) : 21:48:01 Study Area : SSA Altitude (m) (ASL) : 2057.9 Heading (Deg CW from North) : 27.4804 Ground Speed (m/s) : 65.581 Site Name : OJP Frame rate (1/s) : 33.29 Platform : C-GCJX Navaho Image Mode : Imaging_mode Start Latitude, Longitude : 53.8639412 -104.737335 End Latitude, Longitude : 53.923027 -104.685806 CALIBRATION INFORMATION ------------------------ Is data calibrated : YES Calibration Filename : /casi/coeffs/e223f80.rad Calibration Date (DD-MM-YYYY) : 31-03-1997 Calibration Scale Factor : 10 Calibration Units : LAI (dimensionless) Along track look angle (pitch) : 0 Across track look angle (deg) : 18.9 Center pix size across track(m): 30 Center pix size along track(m) : 30 Wavelengths (nm) : N/A Fstop : 8 Bandset : Not Specified Lens Identifier : Not Specified IMAGE SIZE AND FORMAT INFORMATION --------------------------------- Num lines (geocorrected image) : 195 Num pixels (geocorrected image): 125 Number of bands : 1 Data type : 2 byte, big endian, integer Interleave mode : BSQ Bands Present : LAI Tape Record size : Image record size : 250 Number of Header Records : 0 Number of Bytes/Header Record : 0Return to top of document.
9.1.1 Derivation Techniques and Algorithms
FLIM is a stochastic model. It expresses the reflectance of the scene viewed by the sensor as the weighted sum of the ground reflectance and the reflectance of a homogeneous and infinitely deep forest canopy (hereafter simply "homogeneous and infinite" canopy reflectance). Their corresponding weights, called "crown factor" and "ground factor," are affected by the size of the crown, tree density, and the leaf area index and thus are the functions of the crown coverage and transmittance through the crowns. The "ground factor" and "crown factor" can be expressed by the two variables, Cv and Tv. Cv is the crown coverage in the observation direction, and Tv is the crown transmittance, the transmittance of light through the tree crown from the observer's point of view. Please refer to Rosema et al. (1992) for the equations. Table 2 summarizes the input and output parameters and intermediate variables of the forward FLIM.Input Parameters ------------------------- Tree density Average tree crown radius Average tree height Average crown leaf area index Geometric factor Solar angles View angles Understory reflectance "Homogeneous and infinite" canopy reflectance Intermediate Variables ------------------------- Crown coverage in the view direction, Cv Transmittance in the view direction, Tv Output Parameter ------------------------- Canopy reflectanceTable 2: Input and output parameters and intermediate variablesTo invert FLIM for canopy LAI and canopy closure, the at-canopy reflectance in two spectral bands are needed, along with the physical parameters in Table 2, except for tree density and average crown leaf area index.
For the canopies of the BOREAS conifer flux tower sites, tree crowns can be assumed ellipsoids with the horizontal radius r and vertical radius b. The ellipticity (b/r) is large for the crowns of these canopies, especially for the old black spruce. As a result, FLIM was modified to account for the effects of crown shape on the shadows on the ground and on the crown transmittance (Hu et al., 1999).
To apply the FLIM model to the CASI images obtained in the winter of 1994 over the flux tower sites (which have high solar zenith angle, due to the latitude of these sites), the original high spatial resolution CASI images were spatially degraded to a spatial resolution of 30 m by 30 m, in order to ensure that a tree crown and its shadow occur in the same pixel. To invert the FLIM model, the at-canopy reflectance in the near-infrared band (798.87 nm) and the red band (666.37 nm) were used. The physical parameters of the crown, such as the average radius, average tree height and ellipticity, were estimated based on the information provided on the BOREAS home page (http://www.daac.ornl.gov/BOREAS/bhs/BOREAS_Home.html [Internet Link]) and from Leblanc et al. (1999). The foliage angle distributions (geometric factors) for the canopies investigated were estimated based on the gap fraction measurements with the Li-Cor LAI-2000 instrument (provided by J. Chen of University of Toronto). The high spatial resolution CASI images were examined to locate regions of dense canopy and to estimate its crown coverage at a 30 m spatial resolution, and thereby to estimate the values for the reflectance of the "homogeneous and infinite" canopy. Although the snow (understory) reflectance is relatively uniform, it still varies with factors such as snow age and depth. Furthermore, the apparent sunlit snow reflectance (i.e. seen from above canopy) is a function of canopy closure and height (Soffer et al., 1995). To obtain the typical snow reflectance for each site, the high spatial resolution images were used to locate enough sunlit snow pixels to calculate the mean value of the reflectance of these pixels. Given the input parameters in Table 2, except for the tree density and LAI, a look-up-table was established for each site with the intermediate variables in FLIM as input and the at-canopy reflectance in the red and NIR bands as output. Based on the look-up-table, the intermediate variables Cv and Tv, canopy LAI and canopy closure were retrieved from the measured at-canopy reflectance in the red and NIR bands (Hu et al., 1999).FLIM-CLUS LAI and Canopy Cover
FLIM-CLUS represents a combination of the FLIM algorithm with multi-spectral clustering and ray-tracing pre-processing to identify open areas separate from canopy regions. FLIM-CLUS is intended for site-specific LAI mapping as it requires user interaction. FLIM-CLUS consists of three stages of processing: multi-spectral clustering and aggregation, ray-tracing, and application of the FLIM model in an identical manner as described above.
Multi-spectral clustering is applied to the original reflectance images to label each pixel as one of the following surface cover categories:The 2m pixel resolution implies that gaps within canopies (reported as smaller than 1m from LAI transect data collected by Chen and Cihlar, 1995) are unlikely to be identified as sunlit or mixed snow areas. Furthermore, the unknown categories minimize the impact of subjective judgement of the operator performing cluster labeling. The multi-spectral clustering consists of two passes of the K-Means algorithm (Hartigan, 1975) with 20 clusters followed by visual aggregation after each pass. The operator identifies sunlit snow clusters after the first pass. The remaining spatial regions are then clustered and labeled into any one of the above categories. The blue (channel 1), red (channel 5) and near-infrared channels (channel 6) are used as inputs to the clustering algorithm. The blue channel is selected because of the low variability of snow reflectance in this spectral region (Wiscombe and Warren, 1980). The red and near-infrared channels are used because of their relationship to vegetation cover and leaf area index (Chen and Cihlar, 1996).
- Sunlit snow - snow illuminated by direct sunlight visible by the sensor.
- Mixed snow - snow in open areas illuminated by indirect sunlight or shadowed by micro-topography visible by the sensor.
- Shaded snow - snow shadowed or obscured by the canopy.
- Sunlit canopy - canopy illuminated by direct sunlight visible by the sensor.
- Mixed canopy - canopy in indirect sunlight visible by the sensor.
- Shaded canopy - canopy shadowed or obscured by other canopy areas.
- Unknown sunlit - sunlit areas not visibly distinguishable as canopy or snow.
- Unknown shadow - shadowed areas not visibly distinguishable as canopy or snow
Ray-tracing is applied to the cluster maps to delineate unknown shadowed regions that are likely to correspond to open rather than canopy areas. It is assumed that the canopy has negligible lateral transmissivity and a uniform height. The first assumption is reasonable for black spruce crowns with the low solar zenith angles during the winter period. The second is an assumption of FLIM and represents a first approximation. It is also assumed that the surface has negligible slope at scales corresponding to the size of the shadows cast. This is valid for the OBS sites processed in this data set. Finally, it is assumed that only direct illumination is present. This assumption is reasonable during the clear sky winter overpasses when the CASI images were acquired. Given these assumptions, and knowledge of the position, it is possible to infer the position of the canopy edge required to produce a boundary between mixed or open snow and the other cover classes. Specifically, the position of the canopy edge is estimated by placing the solar azimuth opposite its true position and assuming the open and mixed snow regions are actually dense canopy areas. The SEENAREA ray-tracing algorithm (PCI Inc., 1999) is then applied to estimate the cast shadow from these hypothetical canopy regions. The region of cast shadow then corresponds to the hypothesized region of additional shadowed open areas. The unknown areas that fall within this hypothesized shadowed open area were then assigned to shadowed snow. All clusters corresponding to canopy areas and remaining regions corresponding to unknown snow or canopy areas are then grouped as belonging to the canopy region. In addition, pixels bordering the canopy region are also included to minimize the likelihood of mixed pixels being identified as open areas.
The FLIM algorithm is applied to every pixel designated as belonging to the canopy area. This includes mixed pixels at edges of canopy clusters and unknown pixels not identified as open areas by ray-tracing. This strategy is seen as conservative since, theoretically, FLIM should map low or zero LAI and canopy cover values for these included regions if their reflectance corresponds to sparse or open areas. The major difference between the application of FLIM within the FLIM-CLUS framework in comparison to the stand-alone application described above is the method used to average pixel reflectance. FLIM assumes that the field of view of the reflectance measurement being inverted corresponds to a statistically representative mixture of sunlit and shadowed canopy and background present within a canopy region. This assumption requires adequate averaging of high resolution pixels to ensure that each inversion is performed over a representative average of canopy area around each pixel (and not just a sunlit or shadowed area the pixel may happen to fall within). Based on the canopy heights and solar zenith angle, a 30m averaging window was identified as corresponding to the length of a cast shadow. The averaging window is applied to each canopy area pixel. The open areas identified by the clustering and ray-tracing algorithm are considered to be statistical outliers and are not included when applying the averaging window. In this sense, the effective spatial scale of each LAI and canopy cover estimate may range from a single 2m pixel surrounded by open areas, to a 30m region composed entirely of canopy area centered on the pixel being processed.
The parameter values and look-up-table methodology used with FLIM-CLUS are defined in Fernandes et al.(1999).
9.2 Data Processing Sequence
9.2.1 Processing StepsFLIM
- Estimate the physical parameters of the tree crown for each flux tower site.
- Determine the reflectance of the "homogeneous and infinite" canopy and the snow reflectance.
- Establish a Look-Up-Table (LUT) for each flux tower site.
- Based on the LUT, retrieve the intermediate variables Cv and Tv from the measured at-canopy reflectance in the red and NIR bands.
- Calculate the canopy LAI and canopy closure from Cv and Tv.
FLIM-CLUS9.2.2 Processing Changes
- Estimate the physical parameters of the tree crown for each flux tower site.
- Determine the reflectance of the "homogeneous and infinite" canopy and the snow reflectance.
- Establish a look-up-table for each flux tower site.
- Apply multispectral clustering and aggregation to identify candidate canopy and open areas at high resolution.
- Perform ray-tracing to identify unknown shadowed area that is likely shadowed snow area.
- Apply a 30m moving-average window, restricted to canopy areas.
- Based on the the look-up-table, retrieve the intermediate variables Cv andTv from the measured at-canopy reflectance in the red and NIR bands.
- Calculate the canopy LAI and canopy closure from Cv and Tv.
None.
9.3 Calculations
Canopy closure = Cv * (1-Tv)
Canopy LAI = Cv * crown LAI
where, crown LAI is calculated from Tv (Hu et al., 1999).
9.4 Graphs and Plots
None.
10.2 Quality Assessment
10.2.1 Data Validation by SourceReturn to top of document.
The derived canopy LAI were quantitatively evaluated using the field-measured canopy LAI with the LAI-2000 by Chen et al. (1997). The canopy LAI values at the six flux tower sites were measured every 10 m along transects A, B, and C during the BOREAS FFC in 1993. Transect B is oriented at 135 degrees (relative to true North) originating at the tower of each site. Transects A and C are parallel to transect B with a separation 10 m from each other (See Chen et al, 1997 for details). To compare the field-measured LAI values and the retrieved LAI values (with a spatial resolution of 30 m by 30 m), the field-measured LAI values were averaged. First the field-measured LAI values along each transect were subjected to a sliding average of three measurements (separated by 20 meters), and then the resulting values were averaged between the tree parallel transects.
As mentioned previously, due to the large solar zenith angles in the winter CASI imagery, the original 2m by 2m CASI images were spatially degraded to 30m by 30m (an upper bound for FLIM-CLUS) in order to ensure that a tree crown and its shadow occur in the same pixel. It is hard to locate the transects in the spatially degraded images. Therefore, we performed the validation as follows. (1) Locate transect B in the 2m by 2m reflectance images. (2) Calculate the average reflectance of a 15 by 15 pixel area with the center at each point along the transect. (3) For each point along the transect, retrieve the LAI based on the averaged reflectance in the red and near-infrared band. (4) Compare the retrieved LAI with the averaged field-measured LAI.
We are aware that there are other LAI data sets measured during BOREAS for potential validation, such as the LAI data set in the NSA-YJP site of Queen's University, Canada and the LAI data set in the SSA-OBS site of University of Wisconsin, Madison, USA. However, for consistency, only the LAI data sets from a single investigator, Chen et al. (1997), were used.10.2.2 Confidence Level/Accuracy Judgement
The correlation coefficient (R-squared) between the field-measured LAI values and the retrieved LAI values are as follows.FLIM Site ID R-squared ------- --------- SSA-OBS 0.52 SSA-YJP 0.64 NSA-OJP 0.86 NSA-YJP 0.51 SSA-OJP 0.57 FLIM-CLUS Site ID R-squared ------- --------- SSA-OBS 0.67 NSA-OBS 0.14Note that all of the surface LAI transects were limited to "canopy areas" so that a quantitative evaluation of the performance of the LAI and canopy cover maps in open areas was not possible. The low R-square at the NSA-OBS may be due to the low range of surface LAI values against which the FLIM-CLUS algorithm was evaluated. The relative standard error of the FLIM-CLUS LAI estimates (defined as standard error divided by site mean surface LAI) was under 10% at both OBS sites.10.2.3 Measurement Error for Parameters
The look up table used for FLIM has a non-linear relationship between reflectance and LAI and canopy cover. As such, bins corresponding to uniform intervals of reflectance tend to correspond to larger intervals of LAI and canopy cover as LAI and canopy cover increase. For example, with the OBS sites, any LAI value above 3.0 falls within the same bin given the precision error of reflectance estimates. A weighted interpolation scheme (Fernandes et al., 1999) is used to estimate the LAI and canopy cover for these pixels.10.2.4 Additional Quality Assessments
None.10.2.5 Data Verification by Data Center
The imagery was viewed to verify data format, and size, and value ranges.
11.2 Known Problems with the Data
None given.
11.3 Usage Guidance
None given.
11.4 Other Relevant Information
None given.
ORNL DAAC User Services
Oak Ridge National Laboratory
(865) 241-3952
ornldaac@ornl.gov
ornl@eos.nasa.gov
15.2 Procedures for Obtaining Data
BOREAS data may be obtained through
the ORNL DAAC World Wide Web site at http://www.daac.ornl.gov/
[Internet Link] or users may place requests for data by telephone or
electronic mail.
15.3 Output Products and Availability
Requested data can be provided electronically
on the ORNL DAAC's anonymous FTP site or on various media including, CD-ROMs,
8-MM tapes, or diskettes.
16.2 Film Products
None.
16.3 Other Products
The data are avialable on CD-ROM media.
17.2 Journal Articles and Study Reports
Babey, S., and R. J. Soffer, 1992, "Radiometric Calibration of the
Compact Airborne Spectrographic Imager (CASI)," Canadian Journal of Remote
Sensing, (Special issue on Imaging Spectrometery), Vol 18, No. 4, Oct.
1992, pp. 233-242.
Chen, J. M., P. M. Rich, S. T. Gower, J. M. Noeman, and S. T. Plummer, 1997, "Leaf area index of boreal forests: Theory, techniques, and measurements", Journal of Geophysical Research, Vol. 12, No. D24, pp. 29,429-29,443.
Fernandes, R.A., Hu, B., Miller, J.R and Rubinstein, I.G., 1999, "A multi-scale approach to mapping effective leaf area index in Boreal Picea mariana stands using high-resolution CASI imagery", Submitted to Int. J. Rem. Sens.
Freemantle, J. R., J. R. Miller and A. B. Hollinger, 1991, "Improvements in Spectral Feature Extraction after Image Based Refinement of Spectral Calibration of Imaging Spectrometer Data." Proceedings of the 14th Canadian Symposium on Remote Sensing. Calgary, Alberta. pp. 347-349.
Gemmell, F. and J. Varjo, 1999, "Utility of reflectance model inversion versus two spectral indices for estimating biophysical characteristics in a boreal forest test site", Remote Sens. Enviorn., 68, 95-111.
Gray, L. H., J. R. Freemantle, P. R. Shepherd, J. R. Miller, J. W. Harron, and C. H. Hersom, 1997, "Characterization and Calibration of the CASI Airborne Imaging Spectrometer for BOREAS." Canadian Journal of Remote Sensing, (Special issue on BOREAS), 23, 188-195.
Harron, J. W., J. R. Freemantle, L. H. Gray, P. R. Shepherd, C. H. Hersom, J. R. Miller, and A. B. Hollinger, 1995, "Radiometric calibration measures for the multi-temporal BOREAS projects: results of the inter-agency cross calibration and temporal stability of CASI responsivity", Proceedings of the 17th Canadian Symposium on Remote Sensing, 13-16 June, Saskatoon, Saskatchewan, p202-207.
Hartigan, J.A., 1975, Clustering Algorithms, New York, Wiley.
Hu, B., K. Inannen, and J. R. Miller, 1999, "Retrieval of leaf area index and canopy closure from CASI data over the BOREAS flux tower sites", submitted to Remote Sens. Environ.
Leblanc, S., P. Bicheron, J. M. Chen, M. Leroy, and J. Cihlar, 1999, "Investigation of directional reflectance in boreal forests with an improved four-scale model and airborne POLDER data", IEEE trans. Geosci. Remote Sensing, 37, 1396-1414.
Miller, J. R., J. R. Freemantle, P. R. Shepherd, L. Gray, N. O'Neill, A. Royer and E. Senese, 1995, " Deployment of CASI to meet the Needs of BOREAS Science." Proceedings of the 17th Canadian Symposium on Remote Sensing, 13-16 June, Saskatoon, Saskatchewan, pp 169-175.
Miller, J. R., H. P. white, J. M. Chen, D. R. Peddle, G. Mcdermid, R. A. Fournier, P. Shepherd, I. Rubinstein, J. Freemantle, R. Soffer, and E. LeDrew, 1997, Seasonal change in understory reflectance of boreal forests and influence on canopy vegetation indices, Journal of Geophysical Research, Vol. 12, No. D24, pp. 29,475-29,482.
O'Neill, N. T., F. Zagolski, M. Bergeron, A. Royer, J. Miller and J. Freemantle, 1997, "Atmospheric Correction Validation of CASI Images Acquired over the BOREAS Southern Study Area." Canadian Journal of Remote Sensing, (Special issue on BOREAS), vol. 23, No. 2, pp. 143-162.
PCI Inc., 1999, Using PCI Software Volume 2, p. 316-319.
Rosema, A., W. Verhoef, H. Noorbergen, and J. J. Borgesius, 1992, "A new forest light interaction model in support of forest monitoring", Remote Sens. Environ., 42: 23-41.
Soffer R. J., W. Wanner, J. R. Miller and A. H. Strahler, 1995, "Winter boreal forest canopy BRF results: comparisons between airborne data, laboratory simulations, and geometrical-optical model data", Proceedings of IGARSS'95, 10-14 July, Firenze, Italy.
Thomas, P. J., Hollinger A. B., Chu, K. M. and Harron, J. W., 1991, "The ISTS Array Detector Test Facility", Proceedings of the Society of Photo-Optical Instrumentation Engineers, 834, 91-105.
Wiscombe, W. J. and S.G. Warren, 1980, A Model for the Spectral Albedo of Snow: I. Pure Snow, Journal of Atmospheric Sciences, 37(12):2712-2733.
AGL - Above Ground Level AVIRIS - Airborne Visible and Infrared Imaging Spectrometer AVHRR - Advanced Very High Resolution Radiometer BOREAS - Boreal Ecosystem Atmosphere Study CASI - Compact Airborne Spectrographic Imager CCD - Charge Coupled Device FLIM - Forest-Light Interaction Model FWHM - Full Width, Half Maximum GOES - Geostationary Operational Enviornmental Satellite GPS - Global Positioning System GSFC - Goddard Space Flight Center IFC - Intensive Field Campaign ILP - Incident Light Probe ISTS - Institute for Space and Terrestrial Science LAI - Leaf Area Index PRSO - Provincial Remote Sensing Office NASA - National Aeronautics and Space Administration NDVI - Normalized Difference Vegetation Index NIR - Near Infrared NSA-OBS - Northern Old Black Spruce NSA-OJP - Northern Old Jack Pine NSA-YJP - Northern Young Jack Pine SSA-OBS - Southern Old Black Spruce SSA-OJP - Southern Old Jack Pine SPOT - Satellite Pour l'Observation de 1a Terre SSM/I - Special Sensor Microwave/Imager SR - Simple Ratio SSA-YJP - Southern Young Jack Pine TM - Landsat Thematic Mapper UTM - Universal Transverse Mercator WWW - World Wide WebReturn to top of document.
Miller, J. R., B. Hu, and R. Fernandes. 2001. BOREAS Follow-On DSP-06 CASI LAI and Canopy Closure of Conifer Flux Tower Sites. Data set. Available on-line [http://www.daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, U.S.A.
Keywords:
LAI
Canopy Cover
Aircraft Data
Radiance
BRDF
Remote Sensing
Multispectral
Hyperspectral
Boreal Forest