Description: Results of adjusting the Simple Geometric Model against MISR red band HDRF data (orbit: 014073, path 34, MISR blocks 57-60, mapped to a 250 m grid in UTM zone 13N WGS84). The area covered is about 211,490 km2 with some missing data where MISR aerosol/surface retrievals failed. A dynamic background contribution is predicted prior to model adjustment using regression with the isotropic, geometric, and volume scattering kernel weights of a Li-Ross BRDF model as independent variables (nadir camera spectral BRFs do not help and tend to result in overfitting). The regression coefficients for each of the four Walthall model parameters were obtained by extracting backgrounds for a number of calibration sites: i.e., provided with MISR red band HRDFs, estimates of upper canopy crown cover and approximate b/r ratio, and a GO model, an optimization algorithm can extract the best-matching background for each site. The cover estimates were obtained from CLPX orthophotographs. The results were obtained using several different dynamic backgrounds, with the different runs denoted with CLPX-1, -2, -3...
Note that no corrections for topography were effected; there is also no flagging of results affected by topography.
Dynamic Background Calibration
January 2009 update: close to optimal MISR/GO results (from analysis of multiple runs vs CLPX lidar data) are available here:
This map set supersedes those linked to below. The zip archive includes an aboveground biomass map produced via regression on fractional cover and mean canopy height, using US Forest Service estimates from the Interior West map. CAUTION: these estimates may not be valid in absolute terms, even if the spatial distribution is reasonable. If you have estimates of aboveground woody biomass for a set of sites you can perform your own regression using the fractional cover and mean canopy height retrievals.
If necessary, use gzip -d sgm_misr_14073*.gz or a utility to unpack the .gz file. You can then open the raster with Imagine, ArcGIS, HyperCube, or your favorite IP package. If usinh Imagine you'll probably want to rebuild pyramid layers for rapid panning and zooming (if these were not included in the archive), If you open the layers as pseudocolor image in Imagine you'll get layer 5 (mean canopy height) in a ROYGBIV scale. The layers are:
1 - RMSE (model fitting) 2 - mean crown radius (m) 3 - crown aspect ratio (b/r, where b and r are vertical and horizontal crown radii) 4 - fractional crown cover (invalid for grassland!) 5 - mean crown center height within each 250 m2 map cell Calculate h+b to get top-of-canopy height estimates (using b=b/r*r).
Missing data are indicated by an RMSE layer value of -1.0. Note that mean crown radius and crown aspect ratio are effective parameters (plant # density and h/b ratio are fixed on model inversion).
A manuscript been submitted that explains how it was determined that
background=CLPX-1, r_start=3.0, b/r_start=1.0 is likely to be one of the better
model inversion runs. Briefly, the results of a series of ensemble model
inversion runs with systematically-varying backgrounds and starting points was
performed to obtain sets of results. These were then compared with the
lidar-derived canopy statistics from the April and September CLPX flights. Top
to bottom series: April (abs), September (abs), April, September, where (abs)
indicates that the absolute value of the MISR retrieval was taken -- it's
believed to be sometimes arbitrarily negated. The plots below show RMSE(MISR/GO
height, lidar heights) for various combinations of backgrounds, r, and b/r
start values. There are seven backgrounds: #5 and #7 are intentionally bad.
Within each background there are four r start values (3.0, 4.0, 5.0, 6.0).
Within each background/r start value combination there are five b/r start
values (0.5, 1.0, 1.5, 2.0, 2.5). This gives 140 combinations in all. The
plots show that with a half-decent background and making sure that the b/r
start value is less than 2.0 we can guarantee an RSME much less than 5.0
meters.
Other, earlier sample mapped results are in gzipped ERDAS Imagine raster (.img)
files: CLPX-1 run, CLPX-4 run , CLPX-5 run
Vegetation canopy heights were calculated by creating 2 m resolution raster grids containing the XYZ vegetation elevation values (April 2003; with snow cover) and substracting from these the corresponding ground elevations from interpolated XYZ data. The min, max, mean, range, and st.dev. for all "pixels" inside MISR "pixels" were then extracted for comparison. Note that the lidar canopy height statistics may be biased because of outliers and / or because the April data were used: the "ground" elevations are thus "snowpack" elevations.
Note that the slope of the XY scatterplots in the spreadsheets is highly dependent on the lidar data used (with or without snow) and on the way in which the lidar mean canopy heights are calculated (e.g., all values used, > 1 m used, > 3 m used).
Cover values are also compared with either crown cover data extracted from the Forest Service Interior West maps (2005 version), and/or with estimates from CLPX orthophotos.
Profile Examples (sites 1-6 are grassland; 7-14 are forest):
Lidar/MISR (CLPX-11 run) Height Comparison: Calibration Sites (April lidar)
(sites 1-6 are grassland/shrubland; 7-14 are forest)
Lidar/MISR (CLPX-11 run)/IW Cover Comparison: Calibration Sites (note: IW maps are
forest only)
(sites 1-6 are grassland/shrubland; 7-14 are forest)
Lidar/MISR (CLPX-1 run) Height Comparison: All Sites
(sites 1-37 are grassland/shrubland; remainder are forest)
Lidar/FS-IW Height Comparison: All Sites (note: IW maps are forest only)
(sites 1-37 are grassland/shrubland; remainder are forest)
CLPX Data Used
The lidar data and orthoimagery were obtained from the National Snow and Ice Data Center archive for the CLPX-Airborne IR Orthophotography and Lidar Topographic Mapping campaign, part of the Cold Land Processes Field Experiment (CLPX). The sites are:
FF = Fraser, Fool Creek FS = Fraser, St.Louis Creek NI = North Park, Illinois River NM = North Park, Michigan River NP = North Park, Potter Creek RB = Rabbit Ears, Buffalo Pass RS = Rabbit Ears, Spring Creek RW = Rabbit Ears, Walton Creek LSOS = Fraser, Local Scale Observation Site
USFS Data Used
The US Forest Service data used are available here. Note that this is a preliminary release not intended for validation (metadata: "The version of this dataset is a draft, intended for review by FIA and other interested parties. The release of this dataset is not intended for use beyond these purposes. Future versions of this dataset will be provided with more complete accuracy assessment as well as documentation of the modeling and analysis procedures"). These data are used here for comparative purposes.
| 
| 
|
