AGOG 484/584/ APLN 551 Fall 2018
Concept definition Applications Instruments and platforms Techniques to process hyperspectral data A problem of mixed pixels and spectral unmixing Reading Textbook, Chapter 15 Textbook, Chapter 10 (only 10.6)
Imaging Spectrometry = Hyperspectral Imaging The acquisition of images in hundreds of registered, contiguous spectral bands such that for each picture element (pixel) of an image it is possible to derive a complete reflectance spectrum (Goetz et al., 1985).
The Hyperspectral Imaging concept
Multispectral vs Hyperspectral Band Passes
Kaolinite at different instrument resolutions LANDSAT TM 6 bands GERIS Geophysical Environmental Research Imaging Spectrometer 63 bands HYDICE Hyperspectral Digital Imagery Collection Experiment 224 bands AVIRIS Airborne Visible Infrared Imaging Spectrometer 224 bands Lab Spec - >800 bands
Multispectral vs Hyperspectral Hyperspectral benefits: Able to detect mixtures of materials within the same pixel Able to identify specific materials with high degree of accuracy Get some measure of relative abundance based on depth of absorption features More able to produce quantitative (rather than qualitative) results
Signal-to-noise Ratio Defined as the amplitude of the signal divided by the RMS (Root Mean Square) amplitude of the noise Signal factors include: Energy flux (higher for sunny days) Dwell time of detector over GIFOV IFOV Spectral bandwidth of detector Information content increases linearly with increasing SNR For mapping spectrally distinct materials, SNR is more important than the spatial resolution
AVIRIS Signal-to-Noise Ratio Improvements through time
Pushbroom Dispersive Hyperspectral System Each instrument type provides a different view of the spectral character of a scene
Hyperspectral Instruments AVIRIS (Airborne Visible Infrared Imaging Spectrometer ) The benchmark of hyperspectral instruments Developed and operated by NASA (JPL) VNIR and SWIR (reflective) = 224 bands (210 after processing) Operated on ER-2 (high altitude) and Twin Otter (low altitude) GIFOV from 4m to 20m
CASI (Compact Airborne Spectrographic Imager) Canadian Sensors from ITRES CASI 550, CASI 1500, SASI 600, TABI 320, TASI 600, MASI 600 up to 288 bands in 380-1050nm HyMap VNIR and SWIR Built by Integrated Spectronics (Australia) Operated by HyVista Corp (Australia) Probe operated by ESSI (U.S.)
Small size for small unmanned aircrafts Micro Hyperspec by Headwall Photonics Weight = 1.4 lb (0.63 kg) Designed for Unmanned Aircraft Systems (UAS), or drones Three options VNIR, Extended VNIR, NIR or SWIR microcasi-1920 by ITRES Weight = <1.5 kg 288 programmable channels, and wide imaging array
Spaceborne hyperspectral Hyperion Aboard the NASA s EO-1satellite, operated by USGS The only operational US commercial HSI instrument to make it to orbit VNIR and SWIR capability (reflective), 256 bands Older technology, somewhat noisy and some calibration issues CHRIS European VNIR sensor, experimental Collection of up to 19 of 62 bands available 18 meter pixels, 13 km swath Part of the PROBA-1 mission
Hyperion images of Khirban en-nahas (Jordan) Natural color False color
Planned missions HyspIRI NASA Hyperspectral reflective, 224 10 nm bands 50-60 m pixels, 145 km swath Separate multispectral thermal IR system EnMAP (Germany) EnMAP (Environmental Mapping and Analysis Program) will sample areas of 30 x 30 km2 with a ground sampling distance (GSD) of 30 m Mean spectral sampling distance and resolution is of 6.5 nm at the VNIR, and of 10 nm at the SWIR. Off-nadir pointing capability of up to 30 enables a target revisit time of 4 days.
Plant Sciences Determination of amounts of chlorophyll, lignin, cellulose, water, nutrient content, etc. Agricultural Precision agriculture Geological Lithology and mineralogy, hydrocarbon exploration, etc
Difficulty of dealing with many bands Quazi-continuous spectral curves Comparison of Spectral Curves Orbital (Atmospheric attenuation and spatial resolution) Laboratory Library Field
Must do rigorous correction because you compare satellite-derived reflectance to ground measurements Typically corrected using radiative transfer models (e.g., MODTRAN) Hyperspectral data also allow exploration of atmospheric effects using the sensor data themselves (rather than in situ atmospheric measurement)
Extraction of spectral curves from a hyperspectral remote sensing image Visual comparison of spectral curves of individual pixels Spectral Libraries Jet Propulsion Laboratory (JPL) 0.4 2.5 microns for 160 minerals United States Geological Survey (USGS) 0.4 2.5 microns for 500 minerals and vegetation types John Hopkins University s library ASTER library other..
Unique materials that make up the hyperspectral remote sensing image. The idea is to try to identify regions in the hyperspectral remote sensing image that have the most uniform material, pure pixels Field Pure Pixel Laboratory Pure Pixel Image Pure Pixel Endmember pixels
Spectral Angle Mapper (SAM) Matched Filtering Linear Spectral Unmixing
Comparing spectral curves extracted from the hyperspectral remote sensing images to those measured in the field or extracted from the spectral libraries. Calculating the spectral angle between the spectral curves of the hyperspectral remote sensing image and the field or library spectral curves. Similarity identifier in that it measures the similarity between an unknown material present in the hyperspectral remote sensing image and a reference spectral curve.
The use of hyperspectral data to identify specific materials on the ground Relies on highly resolved spectral curves Comparison of the spectral curves from the satellite sensor to spectral libraries measured in labs or field Requires careful image calibration (radiance to reflectance conversion)
Maximizes the response of a known endmember and suppresses the response of the composite unknown background by matching the spectral signature of a pixel in the hyperspectral remote sensing data with that of the known endmember. Knowledge of all endmembers is not required
Estimate the abundance of endmember materials in each pixel in the hyperspectral remote sensing image Materials with different spectral characteristics occur within a single pixel Endmember pixels
19.1% Trees 43.0% Road 24.7% Grass/GV 13.2% Shade
19.1% Trees 43.0% Road 24.7% Grass/GV 13.2% Shade Each pixel contains different materials, many with distinctive spectra.
19.1% Trees 43.0% Road 24.7% Grass/GV 13.2% Shade Some materials are commonly found together. These are mixed resulting in mixed or integrated spectra
Proportion of each constituent in the pixel
Linear Spectral Mixture Analysis (SMA) r mix, b m ( femrem, b em 1 ) e b m em 1 f em 1 There can be at most m=n+1 endmembers or else you cannot solve for the fractions f uniquely RMS 1 n n b 1 e 2 b r mix,b f em r em,b e b = Reflectance of observed (mixed) image spectrum at each band b = Fraction of pixel filled by endmember em = Reflectance of each endmember at each band = Reflectance in band b that could not be modeled (error) n = number of image bands m = number of endmembers
In order to analyze an image in terms of mixtures, you must somehow estimate the endmember spectra and the number of endmembers you need to use Endmember spectra can be pulled from the image itself, or from a reference library (requires calibration of DNs to reflectance). To get the right number and identity of endmembers, trial-and-error usually works. Very often, shade will be an endmember shade : a spectral endmember (often the null vector) used to model darkening due to terrain slopes, tall buildings, and unresolved shadows
The goal of the spectral mixture analysis (SMA) is usually to solve the inverse problem - to find the spectral endmember fractions that are proportional to the amount of the physical endmember component in the pixel.
Landsat TM image of part of the Gifford Pinchot National Forest
Mature regrowth Old growth Shadow Immature regrowth Burned Broadleaf Deciduous Clearcut Green vegetation = = Non-photosynthetic vegetation Grasses
Spectral mixture analysis (SMA) from the Gifford Pinchot National Forest NPV Green vegetation Shade R = NPV G = green veg B = shade light tones in fraction images = high abundance
As a rule of thumb, the number of useful endmembers is 4-5 for Landsat TM data. It rises to about 8-10 for imaging spectroscopy There are many more spectrally distinctive components in many images, but they are rare or don t mix, so they are not useful endmembers.
Hyperspectral data are airborne or satellite data with very high spectral resolution (and usually with high radiometric resolution too) Requires extensive calibration to allow the calculation of ground reflectance from satellite radiance Allows detailed mapping of subtle constituents of the earth, ocean and atmosphere
Spectral Mixture Analysis (SMA) is useful because: It produces fraction images that are closer to what you want to know about abundance of physically meaningful scene components It helps in reducing dimensionality of data sets to manageable levels without throwing away much data By isolating topographic shading, it provides a more stable basis for classification and a useful starting point for GIS analysis