[DC-03-031] Hyperspectral Remote Sensing

Qi, Y. (2025). Hyperspectral Remote Sensing.  The Geographic Information Science & Technology Body of Knowledge (2025 Edition), John P. Wilson (ed.), DOI: 10.22224/gistbok/2025.1.16

1. Hyperspectral Remote Sensing (HRS) Defined
2. Physical Foundation of HRS: Spectroscopy
3. Non-imaging HRS: Field Spectroscopy
4. Imaging HRS: Imaging Spectroscopy
5. HRS Data Analysis
6. HRS Applications
7. Future of HRS

1. Hyperspectral Remote Sensing (HRS) Defined

Hyperspectral remote sensing (HRS) is a specialized technique that collects and processes hundreds of contiguous spectral bands across a broad range of electromagnetic wavelengths. Unlike multispectral remote sensing, which records only a few discrete broad bands (e.g., 30 nm), HRS acquires continuous spectra in narrow bandwidths (e.g., 5 nm). This capability provides a richer set of spectral signatures, enabling more precise identification of surface materials, vegetation, and minerals, among other applications (Figure 1). Although HRS applies many of the same physical principles as multispectral remote sensing, it requires specialized instruments, data collection procedures, and analytical methods.

Hyperspectral remote sensing (HRS) originated from the principles of spectroscopy, which examines how matter interacts with electromagnetic radiation across a range of wavelengths. HRS encompasses both non-imaging spectroscopy (Curran, 1989)—typically conducted in laboratory or close-range settings to measure small samples—and imaging spectroscopy (Goetz et al., 2009), which acquires hyperspectral images of larger landscapes or objects. A variety of HRS instruments capture spectral information across the visible, near-infrared, shortwave infrared, mid-wave infrared, and thermal infrared regions. Among the most commonly used systems are passive remote sensing devices designed to measure reflectance within solar wavelengths (0.4–2.5 µm). The discussion here focuses on this specific HRS system.

2. Physical Foundation of HRS: Spectroscopy

Spectroscopy is the study of how electromagnetic radiation interacts with matter as a function of wavelength. Depending on the type of interaction and the wavelengths examined, the field includes sub-disciplines focusing on absorption, emission, reflectance, and fluorescence. The core principle of spectroscopy is that each material interacts with electromagnetic radiation in a unique way at different wavelengths, producing a characteristic spectral signature that reveals its composition, structure, and physical properties (Kokaly et al., 2009). This versatility has led to broad applications of spectroscopy in physics, chemistry, environmental science, and the food industry. Spectroscopy provides the fundamental physical basis of HRS by linking the electromagnetic spectrum to the vibrational and electronic resonances of molecules, as well as to microscopic surface and volumetric properties. In HRS that measures solar-reflective spectra, most natural materials exhibit diagnostic absorption features primarily driven by two spectroscopic processes at different wavelengths. First, in the visible to near-infrared ranges, electronic transitions and charge transfer processes—often influenced by transition metal ions such as Fe, Ti, Cr, Co, and Ni—determine the positions of characteristic absorption features. Second, in the shortwave and mid-infrared regions, vibrational processes involving water (H₂O) and hydroxyl (OH⁻), as well as carbonate and sulfate groups, govern the absorption features (Figure 2). These diagnostic absorptions alter the reflectance spectrum recorded by HRS, thereby revealing the chemical composition and physical properties of the target materials.

3. Non-imaging HRS: Field Spectroscopy

Field spectroscopy (non-imaging hyperspectral remote sensing) involves measuring the reflectance spectra of objects in laboratories or close-range environments (Milton et al., 2009). It emerged in the 1970s with the development of field spectrometers designed to capture accurate spectral reflectance data under field conditions. Today, various portable spectrometers from different manufacturers operate at different spectral ranges and resolutions. For instance, the ASD FieldSpec 4 Hi-Res Spectrometer (Malvern Panalytical) records wavelengths between 350 and 2500 nm at resolutions of 2 nm (700 nm), 8 nm (1400 nm), and 6 nm (2100 nm).

Field spectroradiometers can be handheld, mounted on fixed supports (e.g., towers), or installed on mobile platforms (e.g., tractor-mounted systems). Their instantaneous measurements depend on factors such as sky conditions, illumination–instrument viewing geometry, instrument settings, field techniques, measurement methods, and the parameter of interest (e.g., radiance or reflectance). Field spectroscopy protocols aim to minimize ambient scattered light and human interference on measured surfaces, while standardized data processing ensures reproducible, comparable results for calibration and reference (Figure 3). Field spectroscopy must account for the anisotropic angular distribution of solar irradiance and the variations in target reflectance due to viewing geometry (described by the bidirectional reflectance-distribution function, BRDF). Most field measurements collect radiance at a specific viewing angle and field of view, known as the hemispherical-conical reflectance factor (HCRF). Consequently, field methods, instruments, and data processing procedures are designed to convert these angular radiance measurements into hemispherical reflectance products using BRDF corrections.

Field spectroscopy also enables small-scale property estimation: for instance, reflectance spectra of plant leaves can be used to determine biochemical and biophysical traits. In addition, field spectroscopy is essential for calibrating airborne and satellite HRS sensors and supports the scaling of measurements from small areas (e.g., individual leaves) to broader scenes (e.g., vegetation canopies), and ultimately to image pixels.  

4. Imaging HRS: Imaging Spectroscopy

Imaging spectroscopy is an HRS technique that acquires large-scale airborne or spaceborne hyperspectral images of land surfaces or other targets. The instrument used for this purpose is known as an imaging spectrometer. NASA’s Jet Propulsion Laboratory (NASA/JPL) pioneered the development of the first airborne imaging spectrometer (AIS) in the 1970s. In 1984, NASA/JPL launched the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) program, which has been recognized as one of the most accurate and well-calibrated airborne imaging sensors (Thompson et al., 2017).

In 2009, NASA/JPL began developing the Next Generation Airborne Visible/Infrared Imaging Spectrometer (AVIRIS-NG), which became operational in 2013 (Figure 4). AVIRIS-NG covers wavelengths from 380 to 2510 nm with a 5 nm sampling interval and has been deployed on multiple airborne platforms, including the Twin Otter aircraft (flight altitudes of 600–3000 m with 1–4 m spatial resolution) and the ER-2 high-altitude research plane (flight altitude of 20,000 m with 17–20 m spatial resolution). A third instrument (AVIRIS-3) is currently under development at NASA/JPL and will measure wavelengths from 380 to 2500 nm with a 7.4 nm sampling interval.

Over the past few decades, additional airborne imaging spectrometers have been developed by various manufacturers, including the Compact Airborne Spectrographic Imager (CASI) by ITRES Research Ltd. (Alberta, Canada), as well as systems from Headwall, Specim, and HySpex.  

Spaceborne imaging spectrometer development began in the early 1980s. In November 2000, NASA launched Hyperion aboard the Earth Observing-1 (EO-1) satellite, widely recognized as the first spaceborne imaging spectrometer. Hyperion is a pushbroom instrument that acquires images 7.65 km across track and 185 km along track, encompassing 242 spectral bands (400–2500 nm) with a spatial resolution of 30 m. After Hyperion was decommissioned in 2007, NASA and other international space agencies introduced new spaceborne imaging spectrometers. Table 1 summarizes these instruments and their key characteristics.

Table 1. A list of past, current, and planned spaceborne hyperspectral imaging spectrometers. Landsat-Next is included in this table as its broad coverage from visible to short-wave infrared wavelengths.  Adapted from Ustin and Middleton, 2024.

5. HRS Data Analysis

HRS data analysis encompasses a range of algorithmic techniques aimed at detecting, classifying, identifying, quantifying, and characterizing objects or features of interest in hyperspectral datasets. By leveraging the large number of narrow, contiguous spectral bands, researchers have adapted existing multispectral methods and developed specialized approaches that often yield superior performance. This section provides a concise overview of several common methods used in HRS data analysis.

5.1 Narrow-band Ratios

Band ratios derived from reflectance spectra have long been used to enhance spectral contrast and characterize objects. However, broad bandwidths in multispectral data can limit their ability to capture narrow absorption features. In contrast, HRS data offer hundreds of narrow bands, allowing for the development of narrow-band ratios that better isolate specific absorption features of target materials and biochemical constituents. For instance, narrow-band ratios in the 350–1,050 nm range outperform broad-band ratios in estimating agricultural crop biophysical traits (Thenkabail et al., 2000).  

5.2 Spectral Unmixing Method

One persistent challenge in remote sensing is that a single pixel often encompasses multiple surfaces or materials, producing a composite spectrum that may not directly match the pure spectra of its constituent materials. Because HRS provides extensive spectral libraries of pure components, spectral mixture analysis (SMA) seeks to estimate the proportion of each material within the mixed pixel. A widely used and straightforward SMA approach assumes that the observed pixel spectrum is a linear combination of the pure substance (endmember) spectra, weighted by their fractional abundances. This linear unmixing technique has, for example, been employed to map specific vegetation species in the Santa Monica Mountains (Roberts et al., 1998).

5.3 Statistical Methods

The extensive number of bands in HRS data enables the quantification of physical and chemical properties (e.g., leaf pigments) via spectral measurements grounded in spectroscopic principles. These methods originated from laboratory-based research on animal feed and forage quality, where statistical models were employed to link laboratory spectra with biochemical measurements under controlled conditions (Norris et al., 1976). Traditional statistical techniques, such as stepwise multiple linear regression (SMLR), can lead to overfitting when too many independent variables are used to predict a single dependent variable. Consequently, there has been a growing emphasis on methods that leverage the entire hyperspectral range, such as partial least squares regression (PLSR), rather than relying on a small subset of highly correlated bands, as is typical in SMLR (Qi et al., 2014).

5.4 Radiative Transfer Model 

Another widely used approach to quantify physical and chemical properties using HRS data is the radiative transfer model (RTM), a physically based framework for characterizing how light interacts with matter. For instance, the PROSPECT model simulates leaf directional hemispherical reflectance and transmittance based on biochemical constituents such as chlorophyll, water, and dry matter (Jacquemoud & Baret, 1990). The SAIL model extends this concept to canopy reflectance by accounting for leaf area index, leaf angle distribution, and leaf size throughout multiple leaf layers (Verhoef, 1984). The SCOPE model further incorporates the transfer of incident light along with thermal and fluorescence emissions, component temperatures, photosynthesis, and turbulent heat exchange (Van der Tol et al., 2009).

Inversion of an RTM involves identifying the vegetation parameters that reproduce observed reflectance spectra. However, this inversion is often ill-posed: different parameter sets may yield similar modeled spectra, and measurement uncertainties combined with RTM assumptions can introduce inaccuracies. To mitigate these issues, various methods—such as look-up tables, neural networks, and the incorporation of prior information—have been developed to guide the inversion process.

5.5 Image Classification and Object Detection

Image classification and object detection are longstanding tasks in remote sensing. Machine learning (ML) and deep learning (DL) techniques have become especially powerful for analyzing HRS data, owing to their ability to handle high-dimensional inputs and complex spectral features—often surpassing traditional statistical methods designed for multispectral data. For instance, machine learning algorithms such as Support Vector Machines (SVM), Random Forests, k-Nearest Neighbors (k-NN), and gradient boosting, as well as deep learning architectures like Convolutional Neural Networks and the Segment Anything Model, have been employed to extract spatial and spectral features in HRS imagery for classification and object detection (Paoletti et al., 2019).

6. HRS Applications

HRS can support a broad range of remote sensing applications, such as soil property characterization (Ben-Dor et al., 2009), freshwater ecosystem biophysical assessments (Hestir et al., 2015), and snow property analysis (Dozier et al., 2009). By leveraging its high spectral resolution, HRS enables more precise identification of materials and quantitative determination of their physical and chemical properties through specialized analytical methods. HRS is particularly valuable in scenarios where multispectral techniques prove insufficient or cannot capture the level of detail required. This section provides a brief overview of a vegetation trait case study to illustrate both the advantages and challenges of HRS.  

Vegetation physiology and ecology are prominent application areas for HRS, where reflectance spectra have facilitated the quantification of numerous traits and functions, such as leaf area index (LAI), absorbed photosynthetically active radiation (fPAR), canopy temperature, leaf water, chlorophyll, ancillary pigments, cellulose, and lignin (Ustin et al., 2009). Most existing studies, however, are region-specific and rely on limited field or airborne HRS observations. A recent large-scale evaluation (Wang et al., 2023) assessed both physically based (RTM inversion) and empirically based approaches to trait prediction using two extensive HRS datasets (totaling 3,861 foliar samples) from 24 field sites across the eastern United States and southern China.

The results showed that both RTM inversion (using PROSPECT and PROCOSINE) and empirical models (partial least squares regression) accurately estimated vegetation traits. However, while RTM inversion strategies exhibited strong transferability across different datasets, empirical models yielded poorer performance when applied to independent datasets. This finding highlights the importance of generalized trait models, particularly regarding the calibration and refinement of leaf RTMs, as well as the integration of representative samples when training empirical models. Future efforts will focus on developing operational approaches for large-scale retrieval of vegetation and ecological variables using spaceborne HRS data.

7. Future of HRS

The future of HRS is promising, with significant advancements expected in sensor technology, data processing, and applications across diverse fields. Many governmental agencies and private-sector entities worldwide have launched or planned HRS satellites for global-scale studies in the 2020s and 2030s (Table 1). The expansion of these satellites will provide essential, high-resolution spectral observations to advance research on biogeochemical processes, climate change, natural resources, and sustainable development. Data fusion further strengthens HRS applications by integrating hyperspectral data with other remote sensing datasets at varying spectral and temporal resolutions, thus leveraging the strengths of multiple systems. For instance, Landsat and Sentinel imagery can fill temporal gaps in hyperspectral data from satellites such as PRISMA (Italy) and EnMAP (Germany), expanding coverage and enhancing time series analyses.

To address the growing volume of HRS data, machine learning (ML) and deep learning (DL) will be pivotal for automating data processing and analysis, thereby improving accuracy and efficiency in classification, object detection, anomaly detection, and feature extraction (Yuan et al., 2020). Moreover, new analytical tools and IT infrastructures—supported by cloud computing and artificial intelligence—are essential for enhancing the reproducibility, replicability, and reliability of HRS data analyses.