Hyperspectral Remote Sensing in Global Change Studies

Authored by: Jiaguo Qi , Yoshio Inoue , Narumon Wiangwang

Fundamentals, Sensor Systems, Spectral Libraries, and Data Mining for Vegetation

Print publication date:  December  2018
Online publication date:  December  2018

Print ISBN: 9781138058545
eBook ISBN: 9781315164151
Adobe ISBN:

10.1201/9781315164151-3

 

Abstract

Hyperspectral imaging spectroscopy data will become increasingly available and important in advancing global studies pertaining to agriculture, water, food security, and climate variability. This chapter provides an overview of the significant advances that can be made by using hyperspectral sensors for reducing uncertainties in modeling, mapping, and monitoring global change studies. However, challenges exist in data access, data storage, data visualization, and analytical methodologies. Nevertheless, with the dawn of cloud computing, machine learning, and artificial intelligence, these challenges are slowly overcome. The chapter discusses these challenges and suggests solutions. Many hyperspectral imaging spectrometers are going to be launched into space (e.g. HyspIRI of NASA) that makes it feasible to collect extensive data of the Planet Earth in 100s of hyperspectral narrowbands along various portions of the electromagnetic spectrum.

 Add to shortlist  Cite

Hyperspectral Remote Sensing in Global Change Studies

3.1  Introduction

A few decades ago, hyperspectral imagery data and processing software were available to only spectral remote sensing experts. Nowadays, one of the fastest growing technologies in the field of remote sensing has been investing on the research and development of hyperspectral sensors for data acquisition and software for data analysis [1]. Unlike multispectral imaging systems (e.g., Landsat, SPOT, IKONOS) that capture reflected or emitted incoming radiation from the Earth's surface in a few broad wavelength bands across the electromagnetic spectrum, hyperspectral imagers measure reflected radiation at numerous narrow, contiguous wavelength channels. The substantially finer spectral resolution data from hyperspectral sensors enhance the capability to characterize the Earth's surface more effectively than do the broadband multispectral data [2].

The distinction between hyperspectral and multispectral is based on the narrowness and contiguous nature of the measurements, not the “number of bands” [3]. For example, a sensor that measures only 20 spectral bands can be considered hyperspectral if those bands are narrow (e.g., 10 nm) and contiguous where there is useful content to be collected. On the other hand, if a sensor measures 20 wider spectral bands (e.g., 100 nm), or is separated by non-measured wavelength ranges, the sensor is no longer considered hyperspectral [1]. The detailed contiguous range of spectral bands of a hyperspectral sensor provides an ability to produce a contiguous “spectrum,” which is one of the characteristics that distinguishes it from multispectral sensors.

Radiances measured by multispectral sensors are generally adequate for rough discrimination of surface cover into categories; however, they are rather limited in the amount of quantitative information that can be inferred from the spectral content of the data. The spectra, or spectral reflectance curves, from hyperspectral remote sensors provide much more detailed information about absorption regions of the surface of interest, very much like the spectra that would be measured in a spectroscopy laboratory. This unique characteristic of hyperspectral data is useful for a wide range of applications, such as mining, geology, forestry, agriculture, and environmental assessment.

This chapter is to focus on existing hyperspectral remote sensing systems, global change requirements, application examples, and challenges ahead.

3.2  Hyperspectral Sensors and Characteristics

In the 1970s, space-based multispectral remote sensors were launched and produced images of the Earth's surface. Even with only a few broad wavelength bands, the images greatly improved our understanding of our planet's surface. The idea of developing hyperspectral, imaging sensors, also known as imaging spectroscopy, emerged in the early 1980s to improve our ability to better characterize the Earth's surface. The Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) developed at the NASA Jet Propulsion Laboratory in California was the first spectrometer being used on moving platforms such as aircraft.

In the early imaging spectroscopy era, most of the hyperspectral sensors were mounted on aircraft (e.g., AVIRIS). After decades of research and development, hyperspectral technology was expanded to space-based remote sensing systems and several satellite hyperspectral sensors were proposed and subsequently launched. The very first spaceborne hyperspectral sensors were NASA's Hyperion sensor on the Earth Observing-1 (EO-1) satellite, and the U.S. Air Force Research Lab's Fourier Transform Hyperspectral Imager (FTHSI) on the MightySat II satellite. With more satellite-based sensors being planned, more hyperspectral imagery will be available to provide near global coverage at regular repeated cycles [1] suitable for global change studies.

3.2.1  Spaceborne Systems

Spaceborne hyperspectral sensors aboard satellites may provide continuous acquisition of the Earth's surface images at lower relative cost. However, wide spatial coverage by a hyperspectral sensor is often compromised with its spatial resolution or ground sampling interval and other challenges. Consequently, repeated hyperspectral images are not widely available. Table 3.1 lists currently available spaceborne hyperspectral instruments, with a wide range in the number of spectral bands, spectral range, and swath width.

Table 3.1   Operational and Planned Satellite Hyperspectral Instruments

Instrument (Satellite)

Altitude (km)

Pixel Size (m)

Number of Bands

Spectral Range (nm)

Spectral Resolution (nm)

IFOV (μrad)

Swath (km)

Hyperion (EO-1)

705

30

220

400–2500

10

42.5

7.5

FTHSI (MightySat II)

575

30

256

450–1050

10–50

50

13

CHRIS (PROBA)

580

25

19

400–1050

1.25–11.0

43.1

17.5

OMI (AURA)

705

13,000

780

270–500

0.45–1.0

115

2600

HICO

∼390

92

102

380–900

5.7

<20

42 × 192

COIS (NEMO)

605

30

210

400–2500

10

49.5

30

HIS (SIMSA)

523

25

220

430–2400

20

47.8

7.7

Warfighter-1 (OrbView-4)

470

8

200

450–2400

11

20

5

470

8

80

3000–5000

25

20

5

EnMAP (Scheduled 2014)

650

30

94

420–1000

5–10

30

30

650

30

155

900–2450

10–20

30

30

HypSEO (MITA)

450

20

210

400–2500

10

40

20

MSMI (SUNSAT)

660

15

200

400–2350

10

22

15

PRISMA

695

30

250

400–2500

10

43

30

Global Imager (ADEOS-2)

803

250–1000

36

380–1195

10–1000

310–1250

1600

WFIS

705

1400

630

360–1000

1–5

2000

2400

The FTHSI program initiated in 1995, was successfully launched in July 2000, maybe the only Department of Defense (DoD) space-based hyperspectral imager to discern spectrally unique objects with the Fourier transform technique. The width of the image footprint was 13 km with a spatial resolution 30 m, covering a spectral range from 450 to 1050 nm by 256 spectral bands (Table 3.1).

In November 2000, NASA launched the Hyperion sensor onboard the EO-1 satellite as part of a 1-year technology validation/demonstration mission. The Hyperion imaging spectrometer had a 30 m spatial resolution, 7.7 km swath width, and 10 nm contiguous spectral resolution (Table 3.1). With its high radiometric accuracy of 220 spectral bands, complex landscapes of the Earth's surface could be imaged and spectrally characterized.

The Compact High Resolution Imaging Spectrometer (CHRIS) was launched in October 2001 on board the PROBA platform (Table 3.1). The sensor was developed by in the United Kingdom with support from the British National Space Centre. The sensor was designed for the study of atmospheric aerosols, land surfaces, and coastal and inland waters with its 62 spectral bands ranging from 400 to 1050 nm and a spatial resolution of 17 m. Despite the fact that the mission was designed for a 1-year life span, the sensor was in operation for more than ten years [4].

The Global Imager (GLI) was part of the ADEOS II mission, an international satellite mission led by the Japanese Aerospace Exploration Agency (JAXA) with participations from the United States (NASA) and France (CNES—Centre Nationale d'Etudes Spatiales; Table 3.1). Its spectral range was from 250 to 1000 nm and its image size ws very large (1600 km). The GLI mission was to collect data to aid better understandings of water, energy, and carbon circulations in order to contribute to global environmental change studies. ADEOS II was launched on December 14, 2002, but unfortunately, the mission ended 10 months later, due to a failure of the solar panel on October 24, 2003.

The Ozone Monitoring Instrument (OMI) flown on the EOS Aura spacecraft was designed to measure atmospheric composition (Table 3.1). The sensor was launched on July 15, 2004, into an ascending node 705 km sun-synchronous polar orbit. The OMI was a nadir pushbroom hyperspectral imaging sensor that observed solar backscatter radiation in the UV and visible wavelengths (264–504 nm). It had 780 spectral bands with a swath large enough to provide global coverage in one day (14 orbits) at a spatial resolution of 13 × 24 km at nadir. The key air quality components include NO2, SO2, and aerosol characteristics, as well as ozone profiles.

The Hyperspectral Imager for the Coastal Ocean (HICO), the Navy's “Sea Strike” was designed and built by the Naval Research Laboratory (NRL) to be the first spaceborne imaging spectrometer optimized for scientific investigation of the coastal ocean and nearby land regions with high signal-to-noise ratios in the blue spectral region and full coverage of water-penetrating wavelengths (Table 3.1). Due to the fact that water absorbs most of the light in the electromagnetic spectrum, visible light is the only part of the spectrum that sufficiently penetrates the water column to sense the water and the bottom surface properties. The sensor was launched on the H-2 Transfer Vehicle (HTV) and was rendezvoused with the International Space Station (ISS) in September 2009. The sensor has been serving as a spaceborne hyperspectral method to detect submerged objects, to provide environmental data products to Naval forces, and to develop the coupled physical and bio-optical models of coastal ocean regions globally.

A few more sensors have been proposed for the near future launches, including PRecursore IperSpettrale della Missione Applicativa (PRISMA) under development by the Italian Space Agency (ASI). Listed in Table 3.1, PRISMA is a pushbroom hyperspectral sensor and is optimized to derive information about land cover, soil moisture and agricultural land uses, quality of inland waters, status of coastal zones, pollution, and the carbon cycle [5]. The hyperspectral instrument is to acquire images in 250 spectral bands at 30 m spatial resolution. When combined with a panchromatic camera, a higher spatial resolution (5 m) could be produced.

Environmental Mapping and Analysis Program (EnMAP) was a German pushbroom hyperspectral satellite (Table 3.1) with a pointing feature for fast target revisits (4 days), providing high-quality hyperspectral image data on a timely and frequent basis. Aboard EnMAP were Hyperspectral Imagers (HSI) sensors designed to derive surface physical parameters on a global scale, with accuracy not achievable by currently available spaceborne sensors. Data from the 249 spectral channels, with a 30 m pixel size, were assimilated in physically based ecosystem models, and ultimately provided information products on the status of various terrestrial ecosystems [6].

The wide field-of-view imaging spectrometer (WFIS) was a sensor that represented another aspect of imaging spectroscopy (Table 3.1), by providing a possibility for entire global surface study with frequent revisits like the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard Terra and Aqua satellites. WFIS was a pushbroom sensor designed to operate in the visible and near-infrared spectral region (360–1000 nm) with an approximately 1-nm sampling interval [7].

3.2.2  Airborne Systems

Airborne hyperspectral imagery has become more accessible due to the increasing number of companies operating hyperspectral spectrometers. Similar to the scanning mechanism of a spaceborne hyperspectral sensor, an airborne sensor generates hundreds (often image columns) of individual pixels along the scan line direction (often perpendicular to the flight direction). As the aircraft moves forward, a new array of pixels is generated along the flight direction. As such, geometric quality of airborne images can be affected by environmental conditions such as wind and/or by flight operations such as aircraft speed and alignment. Therefore, airborne hyperspectral image processing/calibration may be complex and occasionally bring in errors to the analysis [8].

Airborne hyperspectral sensors are more flexible than those satellite-based sensors in terms of acquisition schedules adjustable to weather conditions, spectral and spatial resolution requirements, and flight line arrangements. However, airborne hyperspectral images can be costly for large area coverage, due to limited swath width and slower speed of the carrier as compared to spaceborne examples.

In further comparison with spaceborne hyperspectral sensors, the configuration of airborne systems varies widely in terms of spectral range, number of spectral bands, manufacturers, and spatial and temporal coverage. A survey of existing airborne hyperspectral sensors and their spectral characteristics is presented in Table 3.2. This is not an exhausted list but represents what is available for airborne systems at the time of publication.

Table 3.2   Current Airborne Hyperspectral Sensors and Data Providers

Airborne Sensors

Manufacturer

Number of Bands

Spectral Range (μm)

AISA EAGLE (Airborne Imaging Spectrometer)

Spectral Imaging

up to 488

0.40–0.97

AISA EAGLET (Airborne Imaging Spectrometer)

Spectral Imaging

up to 410

0.40–1.00

AISA HAWK (Airborne Imaging Spectrometer)

Spectral Imaging

254

0.97–2.50

AISA DUAL (Airborne Imaging Spectrometer)

Spectral Imaging

up to 500

0.40–2.50

AISA OWL (Airborne Imaging Spectrometer)

Spectral Imaging

up to 84

8.00–12.00

AVIRIS (Airborne Visible/Infrared Imaging Spectrometer)

NASA Jet Propulsion Lab

224

0.40–2.50

CASI-550 (Compact Airborne Spectrographic Imager)

ITRES Research

288

0.40–1.00

CASI-1500 Wide-Array (Compact Airborne Spectrographic Imager)

ITRES Research

288

0.38–1.05

SASI-600 (Compact Airborne Spectrographic Imager)

ITRES Research

100

0.95–2.45

MASI-600 (Compact Airborne Spectrographic Imager)

ITRES Research

64

3.00–5.00

TASI-600 (Compact Airborne Spectrographic Imager)

ITRES Research

32

8.00–11.5

DAIS 7915 (Digital Airborne Imaging Spectrometer)

GER Corporation

32 8 32 1 6

0.43–1.05 1.50–1.80 2.00–2.50 3.00–5.00 8.70–12.3

DAIS 21115 (Digital Airborne Imaging Spectrometer)

GER Corporation

76 64 64 1 6

0.40–1.00 1.00–1.80 2.00–2.50 3.00–5.00 8.00–12.0

EPS-H (Environmental Protection System)

GER Corporation

76 32 32 12

0.43–1.05 1.50–1.80 2.00–2.50 8.00–12.5

HYDICE (Hyperspectral Digital Imagery Collection Experiment)

Naval Research Lab

210

0.40–2.50

HyMap

Analytical Imaging and Geophysics

32 32 32 32

0.45–0.89 0.89–1.35 1.40–1.80 1.95–2.48

HySpex

Norsk Elektro Optikk

128 (VIS/NIR1) 160 (VIS/NIR2) 160 (SWIR1) 256 (SWIR2)

0.40–1.00 0.40–1.00 0.90–1.70 1.30–2.50

PROBE-1

Earth Search Sciences Inc.

128

0.40–2.50

3.2.3  Ground-Based Systems

Ground-based hyperspectral systems are currently available from a few commercial companies and their spectral characteristics are listed in Table 3.3. These systems can be mounted on low-elevation platforms such as trucks or be handheld, due to their light weight and small size. In general, the spectral range of these systems is from ultraviolet to middle infrared (200–2500 nm) with varying bandwidth or spectral resolution. The ground sampling interval or the footprint of these sensors varies depending on the height of the sensor and their total field-of-view (FOV). One of the advantages of the ground-based sensing systems is that they are flexible in deployment and can often be used for both field-based and in-lab spectral measurements. Specifications on the operating system requirements, software support, as well as data storage, also vary greatly from sensor to sensor, and from manufacturer to manufacturer.

Table 3.3   Current Handheld Hyperspectral Sensors and Data Providers

Handheld Sensors

Manufacturer

Spectral Resolution

Spectral Range (μm)

FieldSpec 3 Hi-Res Portable Spectroradiometer

ASD Inc. (Analytical Spectral Devices)

3 nm @700 nm 8.5 nm @1400 nm 6.5 nm @2100 nm

0.35–2.50

FieldSpec 3 Max Portable Spectroradiometer

ASD Inc. (Analytical Spectral Devices)

3 nm @700 nm 10 nm @1400/2100 nm

0.35–2.50

Handheld 2 Portable Spectroradiometer

ASD Inc. (Analytical Spectral Devices)

<3 nm @700 nm

0.325–1.075

UV-VIS Spectrometers (USB4000-UV-VIS)

Ocean Optics Inc.

∼1.5

0.20–0.85

VIS-NIR Spectrometers (USB4000-VIS-NIR)

Ocean Optics Inc.

∼1.5

0.35–1.00

UV-NIR Spectrometers (HR4000CG)

Ocean Optics Inc.

0.75

0.20–1.10

UV-VIS Hyperspectral USB Spectrometer

Edmund Optics Inc.

1.5

0.20–0.72

VIS-NIR Hyperspectral USB Spectrometer

Edmund Optics Inc.

2

−1.05

3.3  Hyperspectral Remote Sensing Methods

Many of the methods developed for multispectral imagery analysis and processing can be adopted for hyperspectral images. However, the following are more specifically developed and optimized for hyperspectral analysis.

3.3.1  Support Vector Machines

Support Vector Machines (SVMs) are new methods that have been successfully used for hyperspectral data classification [9–13], and can efficiently work with large inputs, handle noise-attached data in a robust way, and produce sparse or efficient solutions [14,15]. SVMs are based on kernel methods that map data from the original input feature space to a kernel feature space of higher dimensionality and then solve for a linear problem solution in that space [15]. These methods allow an interpretation of learning algorithms geometrically in the kernel space (which is nonlinearly related to the input space), and thus combining statistics and geometry in an effective way [15] to take advantages of hyperspectral imagery.

3.3.2  Kernel Fisher Discriminant Analysis

The Kernel Fisher Discriminant (KFD) analysis is another new and effective method for hyperspectral data classification [16]. The KFD method adopts the same concept of kernel used in SVMs to obtain nonlinear solutions, however, KFD minimizes a different function than the SVMs do, and thus the solution is expressed in a different way for potentially more accurate classification.

3.3.3  Matched Filtering

Some hyperspectral applications are only focused on searching for the existence or fractional abundance of one or a few single target materials. Matched filtering (MF) is a type of unmixing procedure that identifies only targets of interest [17]. It is sometimes called a Partial Unmixing because spectra of all endmembers in an image are not required. Matched filtering was originally developed to compute abundances of targets of interest that were relatively rare in the image [1].

Matched filtering algorithms perform a mathematical transformation to maximize the contribution of the target spectrum while minimizing the background [1,18]. Therefore, the approaches perform best when target material is rare and does not contribute significantly to the background signature [1]. A modified version of matched filtering uses derivatives of the spectra to enhance the differentiation ability [18]. The output image presents the fraction of the pixel that contains the target material [1].

3.3.4  Libraries Matching Techniques

Spectral libraries are collections of reflectance spectra measured from materials of known composition, usually in the field or laboratory [1] that are highly desirable for hyperspectral image analysis. Laboratory-derived spectra may be found at the ASTER Spectral Library, which contains the NASA Jet Propulsion Laboratory Spectral Library and the U.S. Geological Survey Spectral Library. Other publicly accessible reference spectral libraries are also available, such as those in digital image processing software (e.g., ENVI, ERDAS, PCI Geomatica) and other sources [19–25].

Absorption features of specific materials within a given IFOV are typically present in the spectra. These absorption features provide the information needed for the Earth's surface characterization based on the surface spectral absorption locations, relative depths, and widths [26,27]. Characterization and automatic detection of such absorption features on the basis of the spectral similarity between the pixel and target spectra is the fundamental principle of library matching techniques [26]. A measured spectrum may be divided into several spectral regions before absorption features in each of the regions are detected and matched with those from the spectral library. The algorithm assigns the pixel to the class that its spectrum most closely resembles [26]. Library matching techniques perform best when the scene includes extensive areas of pure materials that have corresponding reflectance spectra in the reference library.

Comparing the spectral properties in a hyperspectral image with the one stored in libraries can take a significant amount of time and computing resources. Therefore, coding techniques (e.g., Binary Spectral Encoding) are developed to represent a pixel spectrum, which has high degree of redundancy, in a simple and effective manner [27]. Library matching techniques may not perform well when mixtures of targets are present in one pixel, or some materials have very similar spectral characteristics. Sample mixed spectra can be included in the library to improve the accuracy; however, it is not likely that all possible mixtures (and all mixture proportions) can be included in the reference library [18].

3.3.5  Derivative Spectroscopy

Among the techniques developed in remote sensing analysis, derivative spectroscopy is particularly promising for use with hyperspectral image data. Differentiation of a spectral curve estimates the slope at each wavelength over the entire spectral range. A first-order derivative is the rate of change of the absorbance with respect to wavelength. Although differentiation of the spectra does not provide more information than the original spectra, however, it can emphasize the potentially unique target features while suppressing other unwanted information [26]. Second-order derivative spectra, which are insensitive to substrate reflectance, have been used to mitigate soil background influences in vegetation studies [28,29].

Second or higher derivatives are relatively insensitive to variations in illumination (due to cloud cover), solar angle variance, or topographic effects [26]. Although some of the researchers have used high-order derivatives, first- and second-order derivatives have been the most common [30,31]. Talsky [32] suggested that the signal-to-noise ratio (SNR) decreases as derivative order increases. Spectral derivatives have successfully been used in remote sensing applications for decades [28,33,34]. In addition, several studies used this method directly toward specific applications, such as water quality assessment [35–37].

3.3.6  Narrow Band Spectral Indices

Hyperspectral indices have been developed for quantification of biophysical parameters, based on specific absorption features that best describe the biophysical indicators. Examples of such indices are provided below [26].

3.3.6.1  Normalized Difference Vegetation Index: NDVI

The traditional normalized difference vegetation index or NDVI has been modified or computed with narrow spectral bands such as the one below to emphasize the sensitivity to green vegetation density.

3.1 NDVI n a r r o w b a n d = ρ ( 860 nm ) ρ ( 660 nm ) ρ ( 860 nm ) + ρ ( 660 nm )

3.3.6.2  Yellowness Index: YI

The “yellowness” index (YI) is sensitive to decreased chlorophyll content or leaf chlorosis and, therefore, is an indicator of stresses in plant leaves. The YI measures the change in shape of reflectance spectra between the 550 nm (maximum green reflectance band) and the 650 nm (maximum red absorption band). The YI uses only wavelengths in the visible spectrum, a region that is relatively insensitive to change in leaf water content and structure [26,33,38].

3.3.6.3  Normalized Difference Water Index: NDWI

NDWI is used to determine vegetation liquid water content, and can be derived from narrow spectral bands that are sensitive to water content.

3.2 NDWI n a r r o w b a n d = ρ ( 860 nm ) ρ ( 1240 nm ) ρ ( 860 nm ) + ρ ( 1240 nm )

3.3.6.4  Red-Edge Position Determination: REP

The REP is defined as the point of maximum slope on a vegetation reflectance spectrum between the red and near-infrared wavelengths. The REP is strongly correlated with foliar chlorophyll content and, therefore, can be a sensitive indicator of vegetation stress [26]. A linear method, based on narrow spectral bands features, was proposed by Clevers [39] to highlight the red-edge changes in a given spectrum:

3.3 REP = 700 + 40 [ ρ ( 860 nm ) ρ ( 1240 nm ) ρ ( 860 nm ) + ρ ( 1240 nm ) ]
where
ρ ( r e d e d g e ) = ρ ( 670 nm ) + ρ ( 780 nm ) 2

3.3.6.5  Crop Chlorophyll Content Prediction

This narrowband vegetation index, developed by Haboudane et al. [40], integrates the capabilities of indices that minimize soil background affects and indices that are sensitive to chlorophyll concentration. The commonly used indices include the Transformed Chlorophyll Absorption in Reflectance Index (TCARI) [41] and the Optimized Soil-Adjust Vegetation Index (OSAVI) [42].

Crop Chlorophyll Content = TCARI OSAVI
3.4 TCARI = 3 [ ( ρ 700 ρ 670 ) 0.2 ( ρ 700 ρ 550 ) ( ρ 700 ρ 670 ) ]
OSAVI = ( 1 + 0.16 ) ( ρ 800 ρ 670 ) ( ρ 800 + ρ 670 + 0.16 )

3.3.7  Neural Network

The Neural Network (NN) is one of the promising feature selection methods. NNs are mathematical models that simulate brain dynamics [43] that are supposed to be quite powerful in remote sensing imagery analysis, especially in image classification, due to their nonlinear properties. It should be noted that NN is highly sensitive to the Hughes phenomenon (the curse of dimensionality), which is particularly a problem for hyperspectral images, and may not work effectively when dealing with a high number of spectral bands [15]. Moreover, the use of NN for hyperspectral image classification has been limited primarily because of the lengthy computational time required for the training process. Nonetheless, several researches have successfully used NN algorithms to estimate vegetation types and biophysical parameters, such as in coastal and ocean waters [44–46].

3.4  Global Change Requirements and Applications

Hyperspectral imagery has been used to assess, analyze, detect, and monitor several key global environmental change variables. For example, hyperspectral data have been used to estimate sediments, chlorophyll a, and algal type information in oceans and inland waters [37,47–50], identify vegetation species [51], study plant canopy chemistry [52–56], detect vegetation stress [57,58], monitor biogeochemical and greenhouse gas cycles [59–63], improve land cover classification accuracy, and more details in plant species recognition [64]. Geologists also use imaging spectroscopy for soil organic matter estimation, salinity and moisture content detection, and mineral mapping [51,65].

3.4.1  Global Change Requirements

Because the Earth is a dynamic system, sufficient understanding of the complex interactions among physical and ecological processes is needed for global change studies. To achieve this, both long- and short-term observations are required to quantify, analyze, and subsequently understand the spatial and temporal variability, trend, and magnitudes of changes in ecosystems dynamics. Multispectral remote sensing systems such as NASA's Landsat and NOAA's AVHRR sensors have been instrumental and inspirational in providing global coverage for systematic analysis of the Earth's dynamics [61,63]. However, the spectral characteristics and the sensor design of these systems limit their applications in the areas that require specific and more accurate assessment of, for example, nutrient deficiencies in plants, algal information of lakes and streams, invasive species identification and detection [66], soil composition, and specific atmospheric gas concentrations. For example, reflectance spectra of agro-ecosystems and the seasonal changes of the spectra in a paddy rice field, shown in Figure 3.1, are much better spectrally characterized by hyperspectral signatures than by that of multispectral data [67]. Broadband spectral signatures would not be able to detect such subtle changes, but hyperspectral measurements enable the detection and quantification of plant, soil, and ecosystem variables due to their high spectral resolution and continuity. These requirements lead to the exploration of hyperspectral sensing systems where spectral information is much richer for enhanced and new applications in global change studies than multispectral data.

Typical reflectance spectra in agro-ecosystem surfaces (upper), and seasonal change of spectra in a paddy rice field (lower).

Figure 3.1   Typical reflectance spectra in agro-ecosystem surfaces (upper), and seasonal change of spectra in a paddy rice field (lower).

3.4.2  Global Change Applications

Hyperspectral data have been used in numerous applications to specifically address global environmental issues. The following are not meant to be an exhausted application list; rather they are examples demonstrating the type of issues one can address with hyperspectral data.

3.4.2.1  Water Quantity and Quality

Imaging spectrometry is a cost-effective technique for water quality studies over large areas of aquatic systems, such as lakes, coastal areas, bays, estuaries, and even oceans. For example, spaceborne and airborne hyperspectral images have been used to assess the trophic status of lakes and to map the spatial distribution of water quality parameters, such as temperature, chlorophyll a, turbidity, and total suspended solids, over large areas [37,47,68–70]. These studies are possible because of narrow, unique spectral absorption features of water bodies that are only detectable by hyperspectral data [37]. Owing to the unique spectral signatures of algal pigments, compositions of algal populations in aquatic systems can be detected by analyzing absorption properties in the region between 400 and 700 nm [49]. Various methods are effective in mapping water quality, including pigment-specific absorption algorithms, spectral angle mapping algorithms, spectral libraries comparison, principle component analysis, derivative spectroscopy, regression techniques, and other spectral indicators like band ratios [30,31,37,71,72].

3.4.2.2  Carbon Sequestration and Fluxes

Carbon sequestration is a process of removing carbon from the atmosphere and depositing it in a reservoir or carbon sink. The main natural sinks are photosynthesis by terrestrial plants, and physicochemical and biological absorption of carbon dioxide by the oceans. The oceans are the largest active carbon sinks on Earth, absorbing more than a quarter of the carbon dioxide released from human activities [4].

Hyperspectral data proved to be useful to estimate and map primary production in the oceans and other open surface waters [48–50] while multispectral data are equally suitable for terrestrial primary production estimation. The use of hyperspectral remote sensing to measure chlorophyll a from space has been a highly successful technique for mapping phytoplankton distribution on a global basis, which could be used to estimate the amount of carbon sequestration in the oceans.

3.4.2.3  Greenhouse Gas Emissions

Greenhouse gases, as defined by UNFCCC, are “the atmospheric gases responsible for causing global warming and climate change. The major greenhouse gases are carbon dioxide (CO2), methane (CH4) and nitrous oxide (N2O). Less prevalent—but very powerful—greenhouse gases are hydrofluorocarbons (HFCs), perfluorocarbons (PFCs) and sulphur hexafluoride (SF6).” Hyperspectral remote sensing has shown to be successful in detecting methane gas concentrations [59,60,62], despite some technical challenges. Although methane is transparent in the visible part of the electromagnetic spectrum, it does contain a number of spectral features at longer wavelengths that can be detected. Three significant absorption features, at 3.31, 3.28, and 3.21 μm, were detected in the methane spectrum [59]. Some of the spectral features may appear to be obscured by regions of water vapor absorption; however, the obscured spectral features are still detectable at high methane concentrations [73]. The methane absorption features at 0.88 and 7.7 μm have also been proposed for atmospheric studies in the presence of significant water vapor content [73]. Once the spectral bands associated with absorption features are determined, band ratio techniques could be used develop indicators of methane concentration.

3.4.2.4  Atmospheric Chemistry

Although multispectral remote sensing has been used for atmospheric monitoring, hyperspectral imagery proved to be more suitable due to its fine spectral resolution and sampling intervals. Space measurements of the ozone column have been conducted since the 1970s with the series of Solar Backscatter Ultraviolet (SBUV) and Total Ozone Mapping Spectrometer (TOMS) sensors [74]. These spectrometers have been used to monitor the ozone layer, measure stratospheric dynamics, and detect tropospheric ozone pollution on a regional scale [75]. The SBUV and TOMS, with 1-nm spectral resolution, measure backscattered radiance at two wavelengths and estimate the ozone column [76,77]. In 1995, the Global Ozone Monitoring Experiment (GOME) was launched and operated as the first space instruments that measured the ultraviolet and visible part of the spectrum with a high spectral resolution [75]. In 2002, the Scanning Imaging Absorption Spectrometer for Atmospheric Cartography (SCIAMACHY) was launched onboard the Environmental Satellite (ENVISAT) to facilitate the measurement of atmospheric absorptions from the ultraviolet to the near-infrared spectral range (240–2380 nm), providing knowledge about the composition, dynamics, and radiation balance of the atmosphere.

The OMI hyperspectral sensor aboard the EOS Aura satellite was successfully launched in 2004, as a continuation of the long-term global total ozone monitoring from satellite measurements that began in 1970 with SBUV and TOMS. Data from OMI were used to derive ozone columns using Differential Optical Absorption Spectroscopy (DOAS) [75,78–80]. In comparison, TOMS estimated the spatial distribution of the total ozone by observing changes of solar radiation in several near UV spectral bands. DOAS derived the ozone column by fitting a reference ozone absorption cross section to the measured sun-normalized radiance. The main advantages of DOAS compared to the original SBUV/TOMS techniques are that DOAS was less sensitive to changes in variability of the periodic radiometric calibration of the instrument and less sensitive to disturbing factors like absorbing aerosols [75].

OMI can distinguish between aerosol types, such as smoke, dust, and sulfates, and can measure cloud pressure and coverage. Other instruments on the Aura satellite, such as the Tropospheric Emission Spectrometer (TES) may provide global measurements of tropospheric ozone and its photochemical precursors. With the entire spectrum from 3.2 to 15.4 μm at high spectral resolution, many other gases such as carbon monoxide, ammonia, and organics can be retrieved [81]. Aura satellite is providing the next level of atmospheric measurements in the stratospheric and tropospheric layers in order to understand the recovering process of the stratospheric ozone layer, the composition of chemistry in the troposphere, and the roles of upper tropospheric aerosols, water vapor, and ozone in climate change. The four instruments on Aura provide valuable data to global change studies as well as continuing important atmospheric composition monitoring that began earlier with other satellites such as TOMS [81,82,110].

3.4.2.5  Vegetation Ecology

Vegetation is a key attribute of land-use and land cover change dynamics not only for its role in food production but also its role in land-atmosphere interactions. The exchange of biochemicals between terrestrial ecosystems and the atmosphere could be determined by the properties of vegetation. Therefore, an accurate assessment of vegetation properties and temporal dynamics is very important in Earth system science [83]. Due to the extended spectral dimension, hyperspectral remote sensing has improved modeling, monitoring, and understanding of vegetation canopies [52,67] and enhanced quantitative mapping of key vegetation properties [54–56]. For example, data from hyperspectral sensors have been used to derive plant species composition, and biological and biochemical properties of the forests [54,111], such as chlorophyll and nitrogen concentrations. These studies suggest that hyperspectral remote sensing brings new capabilities to estimate vegetation properties that otherwise would not be possible with traditional multispectral sensors. Still, interest has increased in using hyperspectral remote sensing for biodiversity monitoring [84,85] and invasive species mapping [66].

3.4.2.6  Vegetation Biochemical Properties

Foliar chemistry measurements of plant canopies allows a better understanding of ecosystem function and service, since many biochemical processes, such as photosynthesis, respiration, and litter decomposition, are related to the foliar chemistry of plants. Important chemical components of vegetation foliage that could be used as bio-indicators are the concentration of nitrogen and carbon, and the content of water [86]. For example, hyperspectral remote sensing combined with canopy radiative transfer models provided consistent and accurate information of these chemical compositions [54]. Forest biomass and aboveground carbon stocks have been estimated using hyperspectral imagery from AVIRIS using partial least squares (PLS) regressions[87].

Plant pigments, such as chlorophyll and carotenoids, have specific absorption spectra, which play essential roles in the photochemical cycle in plant leaves. These specific absorption features are only detectable remotely with hyperspectral sensors, as demonstrated by various studies. For example, Inada [88] showed that the narrow spectral band ratio of 800 nm/550 nm was the most effective index for estimating the leaf chlorophyll content of rice. The reflectance at 675 and 550 nm has been used to determine chlorophyll content as well as for nitrogen stress detection.

The total nitrogen content of a canopy can be estimated using narrowband width measurements in visible and near-infrared wavelengths such as 480, 620, and 840 nm [89], and the estimate accuracy can be further improved by using sharp absorption features in the shortwave infrared wavelengths, 1650 and 2200 nm [90].

3.4.2.7  Invasive Plant Species Detection

Species invasion is recognized as a significant threat to global biodiversity and ecosystem health. In some cases, invasive species could irreversibly change the structure and functioning of entire ecosystems and result in biological diversity loss [66,91]. Ecosystem research suggests that invasive and aggressive plant species may be the result of general ecosystem stress related to changes in the frequency of landscape disturbance, such as road construction, deforestation, land-use conversion to agricultural or urban development, or other hydrologic alterations [91,92]. Identification of the extent of landscapes being stressed by invasive plant species using spectral signatures can help target vulnerable areas in need of restoration or protection [91,111]. Contrary to multispectral remote sensing, which can only detect invasions when the effects have spread out in a wide area, hyperspectral imagery offers a unique potential for analyzing the signals of ecological changes at an early stage and, therefore, an indication of biological invasion and biogeochemical change [66,92]. For example, non-native species ice plant, jubata grass, fennel, and giant reed grass (Phragmites ssp.) from a range of habitats in California were mapped with AVIRIS imagery at relatively high accuracy [93] and using CASI for reed grasses in the Laurentian Great Lakes [91,101], and pure pixels of Brazilian pepper was detected with hyperspectral imagery [94]. It was also demonstrated that an early detection of invasive weeds (spotted knapweed and baby's breath) was also possible [95] using hyperspectral sensors.

3.4.2.8  Vegetation Health

Information about vegetation health, such as disease, fire disturbance, and insect attack, is crucial in ecosystem protection and management and hyperspectral remote sensing can provide diagnostic indicators for early detection. For example, Lawrence and Labus [96] successfully used high spatial resolution imagery to identify different levels of tree stress resulting from Douglas-fir beetle attack. Koetz et al. [97] also mapped spatially distributed fuel moisture content and fuel properties with inversion of radiative transfer models to serve as input for forest fire spread and mitigation models. Diagnostic analysis of specific disease, however, has been a challenging issue with hyperspectral remote sensing, as sensors only “see” plant symptoms rather than the causes. Discrimination of diseases may be possible with knowledge of the physiological effect of the disease on leaf and canopy elements. For example, necrotic diseases can cause a darkening of leaves in the visible spectrum and a cell collapse that would decrease near-infrared reflectance. Chlorosis induced diseases (mildews and some viruses), for example, cause marked changes in the visible reflectance (similar to nitrogen deficiency). Other diseases, however, may be detected by their effects on canopy geometry.

3.5  Hyperspectral Remote Sensing Challenges

There are numerous challenges facing hyperspectral remote sensing, ranging from system design and data processing to methodological developments. Because little is available from literature, the following encompass general issues that can be addressed for broader and improved applications of hyperspectral remote sensing imagery.

3.5.1  System Design Challenges

From global change perspective, the design of a hyperspectral sensor entails the configuration of the following key parameters and requirements: spectral regions, number of spectral bands, spectral bandwidth, spatial resolution, swath width, revisit cycle, signal-to-noise ratios, onboard storage, and data downlinks to ground stations. These system parameters will determine the appropriateness of a specific global change application.

There is a challenge in balancing the spectral bandwidth and signal-to-noise ratio. Existing hyperspectral sensors operate in the spectral region from approximately 200–2500 nm. The spectral bandwidth widens from ultraviolet to shortwave infrared, in order to maintain an acceptable signal-to-noise ratio in the longer wavelength region. This widening presents a challenge in methodological development of hyperspectral remote sensing as significant sharp absorption features will likely not be detectable with wider spectral bandwidth.

The signal-to-noise ratio needs to be balanced with spatial resolution requirements as well. As global change studies increasingly require quantitative information about the Earth's surface properties, there is a need to acquire hyperspectral images at high spatial resolution and over large coverage areas. High spatial resolution requires small instantaneous field of view (IFOV) of the sensor, but smaller IFOV results in lower signal-to-noise ratio and compromises the sensor's ability to have large geographic area coverage.

There is also a conflict in the number of spectral bands, spatial resolution, swath width, and onboard data storage. The requirements for a large number of contiguous spectral bands and high spatial resolution of large geographic coverage undoubtedly increases the data volume, which presents challenges for both onboard storage capacity and the time required for downlinking to ground stations. For global change studies, the priority ought to be on the geographic extent of coverage as this allows a broader and diverse ecosystem analysis and at the same time increases global access to hyperspectral imagery.

3.5.2  Processing and Visualization Challenges

For a given geographical area imaged, the data can be viewed as a two-dimensional image that represents spatial location and spectral information. Displaying hyperspectral data is more challenging than it is for multispectral data. Hyperspectral images contain far more spectral bands than can be displayed with a standard red-green-blue (RGB) display. A convenient visualization approach is to reduce the dimensionality of the image (from tens to hundreds of dimensions) to three dimensions at the expense of information losses [98,99]. The optimal hyperspectral display methods for quantitative and qualitative analysis of the data should enhance natural colors, preserve natural edges or contours of the features, highlight target features of interest, and enable simple and quick computational processing [99].

Several different techniques have been proposed and implemented for useful dimensionality reduction of hyperspectral images. Color matching functions (CMF) are one of these methods that specify how much of each of three primary colors must be mixed to create the “color sensation” in the form of monochromatic light at a particular wavelength in support of human image interpretation [98]. The technique linearly projects hyperspectral data in the visible range onto the color matching functions to determine the amount of the three primary colors that would create the same color sensation as viewing the original spectrum. It creates consistent images where hue, brightness, and saturation have interpretable and relevant meaning [98]. A disadvantage of the CMF is that there might be a decrease in sensitivity of human vision at the edges of the visible spectrum [98].

Principal component analysis (PCA) is also used to reduce hyperspectral data dimensionality by assigning the first three principal components to RGB [100,101]. Recent work found that the use of wavelets reduced noise in spectra before applying PCA could improve visualization [102]. Disadvantages of PCA include the difficulty to interpret the displayed image because the displayed colors represent principal components that do not typically represent natural colors of the features. The colors change drastically, depending on the data, and they do not correlate strongly with data variation. The standard saturation used in PCA display leads to simultaneous contrast problems and the computational complexity is high [98].

A number of linear methods have been used to optimize the hyperspectral imagery display [72,99,101,103–105]. Jacobson and Gupta [105] used fixed linear spectral weighting envelopes to create natural looking palettes while other information could still easily be added using highlight colors. The method maximized usefulness for human analysis while maintaining the natural look of the imagery [98]. Another data dimensionality reduction method is artificial neural networks (ANN). After the neural network training process, images can be processed very quickly, making it reasonable to use this approach for real-time analysis. However, a disadvantage of ANN is that it is unclear how the neurons handle new spectral inputs that were not in the training dataset.

3.5.3  Data Volumes and Redundancy

Hyperspectral images are composed of a large number of spectral bands in order to generate fine enough spectral resolution needed to characterize the spectral properties of surface materials. As a result, the volume of data in a single scene can be overwhelming. Although hyperspectral imagery provides the potential for more accurate and detailed information extraction than possible with other types of remotely sensed data, it can be spectrally over-determined or over-calculated. A tremendous amount of the data in a scene are redundant and much of the additional data do not add to the inherent information content for a particular application [27]. Spectral redundancy means that the information content of one spectral band can be fully or partly predicted from other bands within the scene [27]. The adjacent spectral bands are often found to be highly correlated to one another and their reflectance values, therefore, appear nearly identical. One way to identify spectral redundancy is by computing the correlation matrix for the image where high correlation values between bands indicate high degrees of redundancy or dimensionality [27]. This is another manifestation of inter-band redundancy known as the Hughes Phenomenon.

The greater the number of bands in an image, the more storage and processing time is required for the analysis. Therefore, developing effective tools and approaches to reduce the dimensionality of hyperspectral data, while retaining the information content in the imagery, remains a challenge. When analyzing a hyperspectral image, the focus has been on extracting spectral information within individual pixels, rather than spatial variations within each band. The traditional statistical classification methods that have been developed and used for multispectral image analysis may not be suitable for hyperspectral images unless they are modified to account for the high dimensionality nature of the hyperspectral data.

3.5.4  Radiometric Calibration

One of the most critical steps in hyperspectral data analysis is to convert the measured radiance data to surface reflectance so that individual spectra can be compared directly with laboratory or field data for appropriate interpretation [18,26,106]. A comprehensive conversion method must account for the solar irradiance spectrum, lighting effects due to solar angle and topography, atmospheric transmission, sensor gain and offset, and path radiance due to atmospheric scattering.

Because hyperspectral sensors acquire data at near continuous wavelengths, atmospheric correction should be taken into account the atmospheric absorption properties as shown in Figure 3.2. These absorption regions are dominated by water vapor (1.4 and 1.9 μm) with smaller contributions from carbon dioxide (CO2), ozone (O3), and other gases [18,107]. For example, narrow atmospheric water absorption bands in the visible and near-infrared spectrum are used and include bands at 0.69, 0.72, and 0.76 μm, an oxygen (O2) absorption band at 0.76 μm, and carbon dioxide (CO2) absorption bands in the shortwave infrared region at 2.005 and 2.055 μm and all have been used in atmospheric correction algorithms [106,108].

Plot of atmospheric transmittance versus wavelength for typical atmospheric conditions.

Figure 3.2   Plot of atmospheric transmittance versus wavelength for typical atmospheric conditions.

3.5.5  Methodological Challenges

Hyperspectral images provide rich information about the Earth's surface and therefore are desirable for global change studies. However, several issues should be considered in analysis and interpretation of such data. The large number of spectral bands in hyperspectral imagery and the small number of known target spectra in most image scenes create the problem known as the curse of dimensionality. The use of traditional image classification methods developed for multispectral analysis, such as the Maximum Likelihood Classifier (MLC) and Multiple Linear Regression (MLR or ordinary least squares OLS), without a modification to account for the high dimensionality of the hyperspectral data usually result in low efficiency and accuracy of the classification process. The MLR method assumes no intercorrelation between the independent variables, and the number of samples (endmembers) should be larger than the number of independent variables (spectral bands). Therefore, if independent variables (spectral bands) have significant correlations among each other, which are common for hyperspectral data, the MLR technique will be subjected to a multicollinearity issue [109]. For hyperspectral data that normally have a tremendous number of bands, it would be very difficult to have an adequate number of endmembers to make MLR work effectively. A solution is to engage in feature selection to reduce the dimensionality of the dataset and remove redundant spectral bands and arrived at a dataset with enough bands to address the application but not overwhelm the system with redundancy as discussed in Section 3.5.3.

3.6  Discussion and Future Directions

Accurate and timely information about land cover dynamics is essential for global change studies and hyperspectral data, either from ground-based, airborne, and/or spaceborne systems, has proven to have a greater potential for detailed information extraction than can be achieved from multispectral imagery. Significant progress has been made in the development of new sensors, new technologies for data processing, new methods for analysis, and new models for enhanced information extraction from hyperspectral data over the past decade. However, challenges exist in data access, data storage, data visualization, and analytical methodologies. These challenges are to be addressed by continued research effort and new technology inventions in such areas as sensor design and data compression or data storage or data compression.

Future research is likely to continue to be in the area of methodological developments with a focus on information extraction algorithms from hyperspectral images. At the same time, one would see an increase in hyperspectral image availability for research development, as more and more agencies are planning to launch hyperspectral sensors. Continued progress in sensor design, data availability, and new analytical methods will further promote broader hyperspectral applications in global change studies.

References

1
Shippert, P. , Introduction to hyperspectral image analysis, Online Journal of Space Communication, Issue No. 3: Remote Sensing of Earth via Satellite, 2003.
2
Birk, R.J. and McCord, T.B. , Airborne hyperspectral sensor systems, Aerospace and Electronic Systems Magazine, 9(10):26–33, 1994.
3
Goetz, A.F.H. , Imaging spectrometry for earth remote sensing, Science, 228:1147–1153, 1985.
4
Earth Institute News, Oceans’ uptake of manmade carbon may be slowing, Earth Institute News, The Earth Institute at Columbia University, Columbia University, http://www.earth.columbia.edu/articles/view/2586, 2009.
5
Labate, D. , The PRISMA payload optomechanical design, a high performance instrument for a new hyperspectral mission, Acta Astronautica, 65(9–10):1429–1436, 2009.
6
Stuffler, T. , The EnMAP hyperspectral imager—An advanced optical payload for future applications in Earth observation programmes, Acta Astronautica, 61(1–6):115–120, 2007.
7
Haring, R.E. , WFIS: A wide field-of-view imaging spectrometer, Proceeding SPIE, Vol. 3759, 1999.
8
Cetin, H. , Comparison of spaceborne and airborne hyperspectral imaging systems for environmental mapping, Proceeding of ISPRS Congress Istanbul, 2004.
9
Gualtieri, J.A. and Cromp, R.F. , Support vector machines for hyperspectral remote sensing classification, Proceedings SPIE of 27th AIPR Workshop Advance in Computer Assisted Recognition, Vol. 3584:221–232, 1998.
10
Gualtieri, J.A. , Support vector machine classifiers as applied to AVIRIS data, Summaries of the Eighth JPL Airborne Earth Science Workshop, JPL Publication, 99–17:217–227, 1999.
11
Huang, C. , Davis, L.S. , and Townshend, J.R.G. , An assessment of support vector machines for land cover classification, International Journal of Remote Sensing, 23(4):725–749, 2002.
12
Camps-Valls, G. , Robust support vector method for hyperspectral data classification and knowledge discovery, IEEE Transactions on Geoscience and Remote Sensing, 42(7):1530–1542, 2004.
13
Melgani, F. and Bruzzone, L. , Classification of hyperspectral remote sensing images with support vector machines, IEEE Transactions on Geoscience and Remote Sensing, 42(8):1778–1790, 2004.
14
Cristianini, N. and Shawe-Taylor, J. , An Introduction to Support Vector Machines, Cambridge University Press, UK, 2000.
15
Camps-Valls, G. , Kernel-based methods for hyperspectral image classification, IEEE Transactions on Geoscience and Remote Sensing, 43(6):1351–1362, 2005.
16
Mika, S. , Invariant feature extraction and classification in kernel spaces, Advances in Neural Information Processing Systems, 12, MIT Press, Cambridge, MA, 1999.
17
Boardman, J.W. , Kruse, F.A. , and Green, R.O. , Mapping target signatures via partial unmixing of AVIRIS data, Summaries of the Fifth JPL Airborne Earth Science Workshop, 95(1):23–26, 1995.
18
Smith, R.B. , Introduction to Hyperspectral Imaging, MicroImages, Inc., 1–24, 2006.
19
Elvidge, C.D. , Visible and infrared reflectance characteristics of dry plant materials, International Journal of Remote Sensing, 11(10):1775–1795, 1990.
20
Salisbury, J.W. , D'Aria, D.M. , and Jarosevich, E. , Midinfrared (2.5–13.5 micrometers) reflectance spectra of powdered stony meteorites, International Journal of Solar System Studies (Icarus), 92:280–297, 1991.
21
Salisbury, J.W. , Infrared (2.1–25 micrometers) Spectra of Minerals, John Hopkins University Mineral Library, Johns Hopkins University Press, p. 294, 1991.
22
Grove, C.I. , Hook, S.J. , and Paylor, E.D. , Laboratory Reflectance Spectra for 160 Minerals 0.4–2.5 Micrometers, JPL Publication, 1992.
23
Clark, R.N. , The U. S. Geological Survey, digital spectral library, version 1: 0.2 to 3.0 microns, U.S. Geological Survey Open File Report 93-592, p. 1340, 1993.
24
Salisbury, J.W. , Wald, A. , and D'Aria, D.M. , Thermal-infrared remote sensing and Kirchhoff's law 1. Laboratory measurements, Journal of Geophysical Research, 99:11897–11911, 1994.
25
Korb, A.R. , Portable FTIR spectrometer for field measurements of radiance and emissivity, Applied Optics, 35:1679–1692, 1996.
26
Jensen, J.R. , Introductory Digital Image Processing: A Remote Sensing Perspective, Pearson Prentice Hall, Upper Saddle River, New Jersey, p. 526, 2005.
27
Richards, J.A. and Jia, X. , Remote Sensing Digital Image Analysis: An Introduction, Fourth Edition, Springer, p. 439, 2006.
28
Demetriades-Shah, T.H. , Steven, M.D. , and Clark, J.A. , High resolution derivative spectra in remote sensing, Remote Sensing of Environment, 33:55–64, 1990.
29
Li, Y. , Use of second derivatives of canopy reflectance for monitoring prairie vegetation over different soil backgrounds, Remote Sensing of Environment, 44:81–87, 1993.
30
Butler, W.L. and Hopkins, D.W. , Higher derivative analysis of complex absorption spectra, Photochemistry and Photobiology, 12:439–450, 1970.
31
Fell, A.F. and Smith, G. , Higher derivative methods in ultraviolet, visible and infrared spectrophotometry, The Analytical Proceedings, 19:28–32, 1982.
32
Talsky, G. , Derivative Spectrophotometry: Low and Higher Order, VCH Publishers, New York, p. 228, 1994.
33
Philpot, W.D. , The derivative ratio algorithm: Avoiding atmospheric effects in remote sensing, IEEE Transactions on Geoscience and Remote Sensing, 29(3):350–357, 1991.
34
Penuelas, J. , Reflectance indices associated with physiological changes in nitrogen- and water-limited sunflower leaves, Remote Sensing of Environment, 48:135–146, 1994.
35
Dick, K. and Miller, J.R. , Derivative analysis applied to high resolution optical spectra of freshwater lakes, Proceedings of 14th Canadian Symposium on Remote Sensing, Calgary, Alberta, California, 1991.
36
Chen, Z. , Curran, P.J. , and Hansom, J.D. , Derivative reflectance spectroscopy to estimate suspended sediment concentration, Remote Sensing of Environment, 40:67–77, 1992.
37
Wiangwang, N. , Hyperspectral data modeling for water quality studies in Michigan's inland lakes, PhD dissertation, Department of Geography, Michigan State University, p. 243, 2006.
38
Adams, M.L. , Philpot, W.D. , and Norvell, W.A. , Yellowness index: An application of the spectral second derivative to estimate chlorosis of leaves in stresses vegetation, International Journal of Remote Sensing, 20(18):3663–3675, 1999.
39
Clevers, J.G. , Imaging spectrometry in agriculture: Plant vitality and yield indicators, Imaging Spectrometry: A Tool for Environmental Observations, Kluwer Academic, Netherlands, 193–219, 1994.
40
Haboudane, D. , Integrated narrow-band vegetation indices for prediction of crop chlorophyll content for application to precision agriculture, Remote Sensing of Environment, 81(2–3):416–426, 2002.
41
Daughtry, C.S.T. , Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance, Remote Sensing of Environment, 74(2):229–239, 2000.
42
Rondeaux, G. , Steven, M. , and Baret, F. , Optimization of soil-adjusted vegetation indices, Remote Sensing of Environment, 55:95–107, 1996.
43
Demouth, H. and Beale, M. , Neural Network Toolbox User's Guide Version 4, Natick, Massachusetts, 840 p. 2003.
44
Cipollini, P. , Retrieval of sea water optically active parameters from hyperspectral data by means of generalized radial basis function neural networks, IEEE Transactions on Geoscience and Remote Sensing, 39(7):1508–1524, 2001.
45
Zhang, Y. , Application of an empirical neural network to surface water quality estimation in the Gulf of Finland using combined optical data and microwave data, Remote Sensing of Environment, 81(2–3):327–336, 2002.
46
Zhang, Y. , Application of empirical neural networks to chlorophyll-a estimation in coastal waters using remote optosensors, IEEE Sensors Journal, 3(4):376–382, 2003.
47
Malthus, T.J. , An evaluation of the airborne thematic mapper sensor for monitoring inland water quality, Proceeding of the 22nd Annual Conference of the Remote Sensing Society, University of Durham, 317–342, 1996.
48
Mumby, P.J. , A bird's-eye view of the health of coral reefs, Nature, 413:36–37, 2001.
49
Richardson, L.L. , Hyperspectral imaging sensors and the marine coastal zone, Hyperspectral Remote Sensing of the Ocean (Proceedings Volume), 4154:115–123, 2001.
50
Davis, C. , Ocean PHILLS hyperspectral imager: Design, characterization, and calibration, Optics Express, 10(4):210–221, 2002.
51
Clark, R.N. and Swayze, G.A. , Mapping minerals, amorphous materials, environmental materials, vegetation, water, ice, and snow, and other materials: The USGS tricorder algorithm, Summaries of the Fifth Annual JPL Airborne Earth Science Workshop, 95-1(1):39–40, 1995.
52
Curran, P.J. , Remote sensing of foliar chemistry, Remote Sensing of Environment, 30:271–278, 1989.
53
Ustin, S. , Remote Sensing for Natural Resource Management and Environmental Monitoring (Manual of Remote Sensing – Third Edition) Volume 4, John Wiley & Sons, 679–729, 2004.
54
Goodenough, D.G. , Li, J.Y. , and Dyk, A. , Combining hyperspectral remote sensing and physical modeling for applications in land ecosystems, IEEE International Conference on Geoscience and Remote Sensing Symposium, IGARSS'06, 2000–2004, Denver, Colorado, 2006.
55
Koetz, B. , Estimating biophysical and biochemical properties over heterogeneous vegetation canopies—Radiative transfer modeling in forest canopies based on imaging spectrometry and lidar, PhD thesis, Remote Sensing Laboratories, Department of Geography, University of Zurich, Zurich, 2006.
56
Huber, S. , The potential of spectrodirectional CHRIS/PROBA data for biochemistry estimation, Envisat Symposium 2007, Montreux, Switzerland, the Netherlands, p. 6, 2007.
57
Merton, R.N. , Multi-temporal analysis of community scale vegetation stress with imaging spectroscopy, PhD thesis, Geography Department, University of Auckland, New Zealand, p. 492, 1999.
58
Apan, A. , Detecting sugarcane “orange rust” disease using EO-1 Hyperion hyperspectral imagery, International Journal of Remote Sensing, 25:489–498, 2004.
59
Krier, A. and Sherstnev, V.V. , Powerful interface light emitting diodes for methane gas detection, Journal of Physics D: Applied Physics, 33(2):101–106, 2000.
60
Wei, H. , The seasonal variation of column abundance of atmospheric CH4 and precipitable water derived from ground-based IR solar spectra, Infrared Physics and Technology, 41:313–319, 2000.
61
Roy, J. , Saugier, B. , and Mooney, H.A. , Terrestrial Global Productivity, Academic Press, p. 573, 2001.
62
Barnhouse, W.D. , Methane plume detection using passive hyper-spectral remote sensing, MS thesis, Bowling Green State University, p. 141, 2005.
63
Grace, J. , Role of forest biomes in the global carbon balance, The Carbon Balance of Forest Biomes, Taylor and Francis Group, 19–46, 2005.
64
Goodenough, D.G. , Processing Hyperion and ALI for forest classification, IEEE Transactions on Geoscience and Remote Sensing, 41:1321–1331, 2003.
65
Clark, R.N. , Swayze, G.A. , and Gallagher, A. , Mapping the mineralogy and lithology of Canyonlands, Utah with imaging spectrometer data and the multiple spectral feature mapping algorithm, Summaries of the Third Annual JPL Airborne Geoscience Workshop, 92-14(1):11–13, 1992.
66
Asner, G.P. and Vitousek, P.M. , Remote analysis of biological invasion and biogeochemical change, Proceedings of the National Academy of Sciences, 102:4383–4386, 2005.
67
Ustin, S.L. , Remote sensing of environment: State of the science and new directions, Remote Sensing of Natural Resources Management and Environmental Monitoring, 679–729, 2004.
68
Jupp, D.L.B. , Kirk, J.T.O. , and Harris, G.P. , Detection, identification and mapping of cyanobacteria—Using remote sensing to measure the optical quality of turbid inland waters, Australian Journal of Marine and Freshwater Research, 45:801–828, 1994.
69
Roelfsema, C. , Remote sensing of a cyanobacterial bloom (lyngbya majuscule) in Moreton Bay, Australia, IEEE International Transactions on Geoscience and Remote Sensing Symposium, IGARSS'01 Proceedings, 613–615, 2001.
70
Wiangwang, N. , Water clarity/trophic condition monitoring using satellite remote sensing data, Master's thesis, Department of Geography, Michigan State University, p. 152 , 2003.
71
Tsai, F. and Philpot, W. , Derivative analysis of hyperspectral data, Remote Sensing of Environment, 66:41–51, 1998.
72
Richards, J.A. and Jia, X. , Remote Sensing Digital Image Analysis: An Introduction, Third Edition, Springer, p. 363, 1999.
73
Des Marais, D.J. , Remote sensing of planetary properties and biosignatures on extrasolar terrestrial planets, Astrobiology, 2(2):153–181, 2002.
74
Heath, D.F. and Park, H. , The solar backscatter ultraviolet (SBUV) and Total Ozone Mapping Spectrometer (TOMS) experiment, The Nimbus-7 Users Guide, NASA Goddard Space Flight Center, 175–211, 1978.
75
Veefkind, J.P. , Total ozone from the Ozone Monitoring Instrument (OMI) using the DOAS technique, IEEE Transactions on Geoscience and Remote Sensing, 44(5):1239–1244, 2006.
76
Bhartia, P.K. , OMI ozone product, NASA Goddard Space Flight Center, OMI Algorithm Theoretical Basis Document Volume II, p. 91, 2002.
77
Bhartia, P.K. , Highlights of the version 8 SBUV and TOMS datasets released at this symposium, Proceedings of the XX Quadrennial Ozone Symposium, Athens, Greece, p. 294, 2004.
78
Burrows, J.P. , The global monitoring experiment (GOME): Mission concept and first scientific results, Journal of Atmospheric Sciences, 56(2):151–175, 1999.
79
Piters, A.J.M. , GOME ozone fast delivery and value-added products, version 3.0, KNMI Report, GOFAP-KNMI-ASD-01, De Bilt, the Netherlands, 2000.
80
Spurr, R. , Thomas, W. , and Loyola, D. , GOME level 1 to 2 algorithms description, DLR Technical note ER-TN-DLR-GO- 0025, Oberpfaffenhofen, Germany, 2002.
81
Hilsenrath, E. , Early data from Aura and continuity from UARS and TOMS, Space Science Reviews, 125(1-4):417–430, 2006.
82
Levelt, P.F. , The ozone monitoring instrument, IEEE Transactions on Geoscience and Remote Sensing, 44(5):1093–1101, 2006.
83
Itten, K.I. , APEX—The hyperspectral ESA airborne prism experiment, Sensors, 8(10):6235–6259, 2008.
84
Turner, W. and Spector, S. , Remote sensing for biodiversity science and conservation trends, Ecology and Evolution, 18:306–314, 2003.
85
Schaepman, M. and Malenovsky, Z. , Bridging scaling gaps for the assessment of biodiversity from space, The Full Picture, Group on Earth Observations (GEO), Geneva, Switzerland, 258–261, 2007.
86
Huber, S. , Estimating foliar biochemistry from hyperspectral data in mixed forest canopy, Forest Ecology and Management, 256:491–501, 2008.
87
Goodenough, D.G. , Mapping forest biomass with AVIRIS and evaluating SNR impact on biomass prediction, Natural Resources Canada, internal report (presented at NASA JPL AVIRIS Workshop), 2005.
88
Inada, K. , Spectral ratio of reflectance for estimating chlorophyll content of leaf, Japanese Journal of Crop Science, 154:261–265, 1985.
89
Shibayama, M. and Akiyama, T.A. , Spectroradiometer for field use. VII. Radiometric estimation of nitrogen levels in filed rice canopies, Japanese Journal of Crop Science, 55:433–438, 1986.
90
Inoue, Y. , Moran, M.S. , and Horie, T. , Analysis of spectral measurements in rice paddies for predicting rice growth and yield based on a simple crop simulation model, Plant Production Science, 1:269–279, 1998.
91
Lopez, R.D. , An Ecological Assessment of Invasive and Aggressive Plant Species in Coastal Wetlands of the Laurentian Great Lakes: A Combined Field-based and Remote Sensing Approach, The United States Environmental Protection Agency, Environmental Sciences Division, 2001.
92
Lopez, R.D. , Using Landscape Metrics to Develop Indicators of Great Lakes Coastal Wetland Condition. The United States Environmental Protection Agency, EPA/600/X-06/002, Washington, DC, p. 31 , 2006.
93
Ustin, S.L. , Hyperspectral remote sensing for invasive species detection and mapping, Geoscience and Remote Sensing Symposium, 3:1658–1660, 2002.
94
Lass, L.W. and Prather, T.S. , Detecting the locations of Brazilian pepper trees in the everglades with a hyperspectral sensor, Weed Technology, 18(2):437–442, 2004.
95
Lass L. W. , A review of remote sensing of invasive weeds and example of the early detection of spotted knapweed (Centaurea maculosa) and babysbreath (Gypsophila paniculata) with a hyperspectral sensor, Weed Science, 53(2):242–251, 2005.
96
Lawrence, R. and Labus, M. , Early detection of Douglas-fir beetle infestation with subcanopy resolution hyperspectral imagery, Western Journal of Applied Forestry, 18:202–206, 2003.
97
Koetz, B. , Radiative transfer modeling within a heterogeneous canopy for estimation of forest fire fuel properties, Remote Sensing of Environment, 92:332–344, 2004.
98
Wyszecki, G. and Stiles, W.S. , Color Science: Concepts and Methods, Quantitative Data and Formulae, Second Edition, John Wiley and Sons, p. 968, 2000.
99
Jacobson, N.P. , Gupta, M.R. , and Code, J.B. , Linear fusion of image sets for display, IEEE Transactions on Geoscience and Remote Sensing, 45(10):3277–3288, 2007.
100
Ready, P.J. and Wintz, P.A. , Information extraction, SNR improvement, and data compression in multispectral imagery, IEEE Transactions on Communications, 21(10):1123–1131, 1973.
101
Tyo, J.S. , Principal components-based display strategy for spectral imagery, IEEE Transactions on Geoscience and Remote Sensing, 41(3):708–718, 2003.
102
Kaewpijit, S. , Moigne, J.L. , and El-Ghazawi, T. , Automatic reduction of hyperspectral imagery using wavelet spectral analysis, IEEE Transactions on Geoscience and Remote Sensing, 41:863–871, 2003.
103
Harsanyi, J.C. and Chang, C.I. , Hyperspectral image classification and dimensionality reduction: An orthogonal subspace projection approach, IEEE Transactions on Geoscience and Remote Sensing, 32:779–785, 1994.
104
Tyo, J.S. and Olsen, R.C. , Principal-components-based display strategy for spectral imagery, IEEE Workshop on Advances in Techniques for Analysis of Remotely Sensed Data, 276–281, 2003.
105
Jacobson, N.P. and Gupta, M.R. , Design goals and solutions for display of hyperspectral images, IEEE Transactions on Geoscience and Remote Sensing, 43(11):2684–2692, 2005.
106
Gao, B.C. , Atmospheric correction algorithms for hyperspectral remote sensing data of land and ocean, Remote Sensing of Environment, 113(S1):S17–S24, 2009.
107
Gao, B.C. and Goetz, A.F.H. , Column atmospheric water vapor and vegetation liquid water retrievals from airborne imaging spectrometer data, Journal of Geophysical Research, 95(D4):3549–3564, 1990.
108
Kruse, F.A. , Imaging spectrometer data analysis—A tutorial, Proceedings of the International Symposium on Spectral Sensing Research (ISSSR), San Diego, California, 44–54, 1994.
109
Centner, V. , Comparison of multivariate calibration techniques applied to experimental NIR data sets, Applied Spectroscopy, 54:608–623, 2000.
110
Pierce, R.B. , Impacts of background ozone production on Houston and Dallas, Texas, air quality during the Second Texas Air Quality Study field mission, Journal of Geophysical Research 114, 2009.
111
Lopez, R.D. and Frohn, R.C. , Remote Sensing for Landscape Ecology: New Metric Indicators, 2nd Edition, CRC Press, 269, 2017.
Search for more...
Back to top

Use of cookies on this website

We are using cookies to provide statistics that help us give you the best experience of our site. You can find out more in our Privacy Policy. By continuing to use the site you are agreeing to our use of cookies.