Estimating Land Surface Biophysical Properties using SPOT4 VGT

  • Slides: 20
Download presentation
Estimating Land Surface Biophysical Properties using SPOT-4 VGT Mike Barnsley and Tristan Quaife Environmental

Estimating Land Surface Biophysical Properties using SPOT-4 VGT Mike Barnsley and Tristan Quaife Environmental Modelling & Earth Observation Group, Department of Geography, University of Wales Swansea, UK Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Objectives To derive estimates of key biophysical properties of the land surface in a

Objectives To derive estimates of key biophysical properties of the land surface in a robust and consistent manner from SPOT-VGT image data: Albedo Leaf Area Index (LAI) / Biomass f. APAR NPP Land cover type Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Methods Analysis of variations in detected spectral reflectance as a function of sensor view

Methods Analysis of variations in detected spectral reflectance as a function of sensor view angle and solar illumination angle. Inversion of angular/directional reflectance signal against numerical models of radiation scattering (BRDF models) to yield estimates of surface biophysical properties Validation through analysis of ground data, plant growth models and upscaling. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Angular Sampling from Multiple Orbits Day n+2 Day n+1 Mike Barnsley & Tristan Quaife,

Angular Sampling from Multiple Orbits Day n+2 Day n+1 Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Information Content of MVA Data Left: Nadir-view, multispectral composite (0. 5 mm, 0. 65

Information Content of MVA Data Left: Nadir-view, multispectral composite (0. 5 mm, 0. 65 mm and 0. 85 mm) showing wheat (red), bare soil (blue) and oil seed rape (white). Right: Multi-angle composites (nadir + two oblique angles) at recorded 0. 5 mm, 0. 65 mm and 0. 85 mm, respectively. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

MVA Analysis using SPOT-4 VGT: Northern Spain MVA FCC (NIR Channel) Mike Barnsley &

MVA Analysis using SPOT-4 VGT: Northern Spain MVA FCC (NIR Channel) Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001. Multispectral FCC

MVA Analysis using SPOT-4 VGT: Les Landes Forest, France NIR Mike Barnsley & Tristan

MVA Analysis using SPOT-4 VGT: Les Landes Forest, France NIR Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001. SWIR

Modelling the BRDF Find a mathematical model, preferably one based on the physics shortwave

Modelling the BRDF Find a mathematical model, preferably one based on the physics shortwave radiation transport and specified in terms of measurable biophysical properties, to describe the BRDF. Invert the model to estimate values for the driving parameters. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

BRDF Model Inversion: Constraints Physically-based BRDF models are highly non-linear, requiring numerical (cf. analytical)

BRDF Model Inversion: Constraints Physically-based BRDF models are highly non-linear, requiring numerical (cf. analytical) inversion. Numerical inversion routines are computationally intensive and sensitive to the initial estimates of the model parameters. The models make assumptions about the nature of the surface properties (spatial homogeneity) that are typically not valid for coarse spatial resolution image data. The number of reflectance samples is typically small and they are sub-optimally distributed throughout the viewing and illumination hemispheres. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Angular Sampling UK Spain Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March

Angular Sampling UK Spain Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

BRDF Model Inversion: Solutions? Potential solutions include: Simplify and linearize the BRDF models so

BRDF Model Inversion: Solutions? Potential solutions include: Simplify and linearize the BRDF models so that analytical solutions can be found to the inversion problem (so-called 'linear, semi-empirical, kernel-driven models'); Use alternative numerical inversion strategies, e. g. simulated annealing, to avoid converging on local minima in the merit function; Use 'black box' methods, e. g. , artificial neural networks; Use pre-defined Look-Up Tables (LUTs) and interpolation. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Semi-Empirical BRDF Models rl(qi, fi; qr, fr) = fiso + fgeokgeo + fvolkvol Isotropic

Semi-Empirical BRDF Models rl(qi, fi; qr, fr) = fiso + fgeokgeo + fvolkvol Isotropic Geometric Volumetric Kernels, k, are functions of viewing and illumination geometry. Weights, f, are functions of biophysical properties. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Data Processing Collate 478 images acquired between 17 th May-1 st Oct 1999; Convert

Data Processing Collate 478 images acquired between 17 th May-1 st Oct 1999; Convert DN to TOA radiance; Mask areas of cloud and cloudshadow; Convert from TOA radiances to surface reflectance (SMAC v. 3. 2); Assemble directional spectral reflectance image and ancillary data stacks; Invert data (typically 100 s Gb) against BRDF model. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

16/30 -Day Moving-Window Inversion 16/30 day window Mike Barnsley & Tristan Quaife, UWS. GLC

16/30 -Day Moving-Window Inversion 16/30 day window Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

BRDF Inversions False-colour composite generated from isotropic, volumetric and geometric kernel weights images (NIR

BRDF Inversions False-colour composite generated from isotropic, volumetric and geometric kernel weights images (NIR channel) displayed, respectively, through the red, green and blue colour guns. Simple analytical solution to yield spectral and broadband albedo. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Results: East Anglia, UK Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March

Results: East Anglia, UK Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Results: East Anglia, UK Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March

Results: East Anglia, UK Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Results: Les Landes Forest, Bordeaux, France Mike Barnsley & Tristan Quaife, UWS. GLC 2000,

Results: Les Landes Forest, Bordeaux, France Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Data Validation and Scaling-Up Currently comparing outputs from BRDF models with field measurements of

Data Validation and Scaling-Up Currently comparing outputs from BRDF models with field measurements of biophysical properties. Upscaled using land cover classifications derived from intermediate spatial resolution satellite sensor images (Landsat ETM+). Temporal extrapolation using plant growth models (SUCROS 2, AFRCWheat, etc. ) and standard meteorological observations. Examining spatial and temporal correlations between model outputs and upscaled ground data. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.

Conclusions Demonstrated general utility of the multi-angle approach in general, and the inversion of

Conclusions Demonstrated general utility of the multi-angle approach in general, and the inversion of BRDF models in particular: Directional reflectance variations provide an alternative means of distinguishing surface cover types (e. g. , through MVA analysis); Considerable potential to estimate land surface biophysical properties (e. g. , albedo, LAI and f. APAR) through BRDF model inversion; and Image data from SPOT-VGT are well suited to BRDF studies due to their excellent radiometric calibration and geometric registration. Mike Barnsley & Tristan Quaife, UWS. GLC 2000, Ispra, March 2001.