Hyperspectral Remote Sensing: Theory and Applications

By Prashant K. Srivastava and George P. Petropoulos

Hyperspectral Remote Sensing: Theory and Applications offers the latest information on the techniques, advances and wide-ranging applications of hyperspectral remote sensing, such as forestry, agriculture, water resources, soil and geology, among others. The book also presents hyperspectral data integration with other sources, such as LiDAR, Multi-spectral data, and other remote sensing techniques. Researchers who use this resource will be able to understand and implement the technology and data in their respective fields. As such, it is a valuable reference for researchers and data analysts in remote sensing and Earth Observation fields and those in ecology, agriculture, hydrology and geology.

• Includes the theory of hyperspectral remote sensing, along with techniques and applications across a variety of disciplines
• Presents the processing, methods and techniques utilized for hyperspectral remote sensing and in-situ data collection
• Provides an overview of the state-of-the-art, including algorithms, techniques and case studies

• Language: English
• Publisher: Elsevier Science
• Release date: Aug 5, 2020
• ISBN: 9780081028957

Book preview

States.

Preface

This book aims to provide an all-inclusive overview of state-of-the-art hyperspectral remote sensing and its applications in different research areas. The book is designed in such a way that it will be used by the people in their respective research domains who will realize that hyperspectral technology may offer a solution to their application area. Readers will have a better understanding of how to incorporate and evaluate different approaches to hyperspectral analyses, as well as which approaches may or may not work for the applications of interest. Hyperspectral Remote Sensing: Theory and applications is the first volume of the Series of Earth Observation by Elsevier. The purpose of this book is not to provide a tutorial in hyperspectral remote sensing; rather, its aim is to provide an illustration of the potential applications and analysis techniques that can be used, addressing the unique challenge in different applications across the globe.

The aim of the book is to make researchers aware of and enrich their understanding of the concept of hyperspectral data processing. This book has 21 chapters addressing the principles, techniques, and applications of hyperspectral remote sensing under different research themes with field spectroscopy and airborne and spaceborne imaging spectroscopy.

The first section of the book explains the basic concept and underlying principles of errors and their correction methods. This covers all the major aspects of hyperspectral data, source of errors, types of errors, and their correction methods. Advanced classification techniques such as radial basis function neural network (RBFN) and Support Vector Machine (SVM) for feature selection are also provided.

Further sections of book will take readers through all the major applications of hyperspectral remote sensing in vegetation, water, soil, and minerals, as well as pollution detection. Data fusion with other remote sensor images and utilization of spectral indices for different applications are also presented. This section will also demonstrate narrow band and selected bands of hyperspectral data to detect and interpret the level of hydrocarbon pollution in water resources as well as in forest regions. The book describes case studies that have applied this information to the use of hyperspectral remote sensing in forestry, agriculture, water, soil, and mineral applications. The final chapter deals with the future perspective and challenges in hyperspectral remote sensing community.

The case studies in each chapter illustrate how hyperspectral remote sensing is being used to solve many of the disturbing environmental issues of our society. These include identification of functionally distinct plants, chengal tree abundance, precision agriculture, tropical grassland discrimination under different inorganic fertilizers, snow, inland water and wetland mapping, meteorological studies, soil contamination and hydrocarbon pollution detection, and monitoring, detection of crop parasites, and image fusion for cocoa bean fermentation analysis. For water application, topics include snow mapping and parameter retrieval (such as snow grain size, contamination), inland water quality mapping and a case study on a wetlands ecosystem. For soil and land applications, topics include heavy-metal contamination in soil, soil parameters, soil properties presenting the efficacy of hyperspectral data and multiimage fusion datasets.

We are grateful to the reviewers who made the time to review the chapter manuscripts and Elsevier editorial acquisition members including Morse Redding, Marisa Lafleur, Honest Joy, Robertson Naomi for their constant support and help. Last but not the least, the editors thank the publisher for providing the opportunity to set down the thoughts of several contributors to produce this book.

I hope Hyperspectral Remote Sensing will provide insight into the breadth of the hyperspectral application-related topics. Users of this book are encouraged to adapt it and use it the way it best fits their own needs that would help them in understanding the capabilities and potentials of hyperspectral remote sensing and applications.

Editors

Prem Chandra Pandey, Greater Noida, India

Prashant K. Srivastava, Varanasi, India

Heiko Balzter, Leicester, United Kingdom

Bimal Bhattacharya, Ahmedabad, India

George P. Petropoulos, Athens, Greece

Section I

Introduction to Hyperspectral Remote Sensing and Principles of Theory and Data Processing

Outline

1 Revisiting hyperspectral remote sensing: origin, processing, applications and way forward

2 Spectral smile correction for airborne imaging spectrometers

3 Anomaly detection in hyperspectral remote sensing images

4 Atmospheric parameter retrieval and correction using hyperspectral data

5 Hyperspectral image classifications and feature selection

1

Revisiting hyperspectral remote sensing: origin, processing, applications and way forward

Prashant K. Srivastava¹,², Ramandeep Kaur M. Malhi¹, Prem Chandra Pandey³, Akash Anand¹, Prachi Singh¹, Manish Kumar Pandey¹ and Ayushi Gupta¹,    ¹Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, India,    ²DST-Mahamana Centre for Excellence in Climate Change Research, Banaras Hindu University, Varanasi, India,    ³Center for Environmental Sciences & Engineering, School of Natural Sciences, Shiv Nadar University, Greater, Noida, India

Abstract

After several years of research and development in hyperspectral imaging systems that enriched our knowledge and enhanced our capacity to explore the Earth, these systems have been widely accepted by the remote sensing community. They have evolved as major techniques and have now entered the mainstream of the earth observation data users. This chapter discusses the origin of hyperspectral remote sensing, its importance, preprocessing, inversion models suitable for hyperspectral datasets, as well as several possible applications, including but not limited to, vegetation analysis, agriculture, urban, water quality, and mineral identification. The chapter concludes by looking at the way forward for hyperspectral remote sensing.

Keywords

Inversion models; atmospheric corrections; hyperspectral applications; challenges; ultraspectral

1.1 Introduction

Multispectral remote sensors with a few broad spectral bands were evolved during the 1970s to monitor natural resources. Looking at the existing limitations of multispectral images in different applications and recognizing the demand for better and more advanced images with higher spectral resolution, there was an urgent need for research and development of hyperspectral imaging systems. Hyperspectral remote sensing is an output of imaging spectroscopy that is facilitated by rapid advancement in technologies and the development of detectors, optical design and components, atmospheric radiative transfer and processing capability. Imaging spectroscopy, in turn, utilizes two sensing techniques, namely spectroscopy and imaging. An imaging system captures the spatial distribution of a scene and measures the relative concentration of the objects, while spectroscopy offers the ability to differentiate the elusive absorption features of divergent materials for a scene. The initial or conventional approaches of using multispectral or broadband sensors has the limitation of dividing a discontinuous spectral coverage into numerous broadbands. Hyperspectral remote sensing gained growth and popularity because it collects data that span over a vast region in umpteen contiguous narrow spectral bands of the electromagnetic spectrum, ranging from visible (VIS)–near-infrared (NIR) to shortwave infrared (SWIR) and can achieve a spectral resolution of 10−2λ. Based on platform type, hyperspectral remotely sensed data can be classified as non-imaging or imaging in situ measurements, airborne images, and space-borne images.

1.2 Origin of hyperspectral remote sensing

A landmark step was achieved in 1979 when hybrid array detectors, mercury cadmium telluride on silicon charge-coupled devices was made available for the first time leading to the construction of an imaging spectrometer that operated at wavelengths beyond 1.0 of μm. The airborne imaging spectrometer (AIS) was developed at the National Aeronautics and Space Administration (NASA) Jet Propulsion Laboratory (JPL) in 1983 and operated at wavelengths between 0.8 and 2.5 μm. AIS was replaced by the airborne visible/infrared imaging spectrometer (AVIRIS) in the early 1990s that covered the entire spectrum from 0.4 μm to 2.45 μm at a high spectral rate plus high spatial resolution over an 11 km swath. The primary objectives of these missions were identifying and assessing the characteristics of surface materials. It was with AVIRIS data that the first vegetation analysis was carried out through near-infrared spectroscopy (NIRS) analysis by John Aber and Mary Martin of the University of New Hampshire. Although AVIRIS has offered the bulk of high-quality hyperspectral data, it lacks regularity.

With known multispectral specifications and properties, one wants to gain information from spectral information, and there must be several narrow spectral channels in the same electromagnetic spectrum (400–2500 nm) compared to a few broad spectral channels in multispectral images (Pandey et al., 2019b). Thus the idea behind hyperspectral imaging systems is straightforward, while offering even more information to understand our environment better. This led to further innovations and advancement in hyperspectral imaging systems in the early 1980s. Hyperspectral remote sensing or imaging spectrometry has evolved from ERTS-1 [1] (Earth Resources Technology Satellite)-1 (Landsat-1) 4-band images, to AIS, AVIRIS. AVIRIS was the first hyperspectral imaging system developed at NASA JPL in the United States as a result of innovations in spectrometers technologies, then AVIRIS-NG (Airborne Visible Infrared Imaging Spectrometer-Next Generation) and HyspIRI (Hyperspectral Infrared Imager) were introduced.

The first hyperspectral mission Earth Observation-1 (EO-1) Hyperion sensor was launched by NASA’s Earth Observing-1 satellite in November 2000. This space-borne mission brought significant and dramatic changes in the applications on forestry, agriculture, mineral identifications and land-use or land-cover classifications (Pandey et al., 2018, 2019a). Enhanced spectral resolution in the 400–2500 nm range was extensively and widely used by researchers, and provided more reliable and accurate results (Kramer, 2002). In order to achieve high spectral resolution, multispectral sensors were effectively enhanced in their image acquisition in the spectral range, reduction in the spectral sampling rate, and several contiguous spectral channels. Later on, the Compact High-Resolution Imaging Spectrometer (CHRIS) for the European Space Agency (ESA) Project for On-Board Autonomy-1 (PROBA-1) in 2001 was launched by ESA for the global mapping of natural resources. Hyperion has been decommissioned and CHRIS is currently working and operating in the orbit. It has been designed for a life operational time period of a one year only and has 62 contiguous channels with a spatial resolution of 17 m.

These hyperspectral imaging satellites are accompanied by several other missions later on (Puschell, 2000). The information on current and future space-borne hyperspectral missions were compiled with references to several published articles (Nieke et al., 1997; Puschell, 2000; Kramer, 2002; Staenz and Held, 2012). There is a focus to dramatically increase contiguous spectral bands in upcoming space-based hyperspectral imaging systems (Nieke et al., 1997). A medium spectral resolution hyperspectral imaging system called PRecursore IperSpettrale della Missione Applicativa (PRISMA) was launched by Italy’s ASI on March 22, 2019 (Longo, 2020). PRISMA’s hyperspectral sensors will be capable to acquire images in ~235 contiguous spectral channels in visible (VIS)–near-infrared (NIR)–shortwave infrared (SWIR) ranges. The Environmental Mapping and Analysis Program (EnMAP) for hyperspectral sensors was launched by Deutsche Zentrum für Luft-and Raumfahrt (DLR) (The German Aerospace Center) the German Research Centre for Geosciences [GeoForschungsZentrum (GFZ)] in 2015 to acquire images in over 200 narrow contiguous bands. In the meantime, NASA has scheduled the HyspIRI mission launch for 2022, which will acquire images with 237 spectral bands with 60 m spatial resolution. Detailed information on present and future hyperspectral missions are presented in Table 1–1. This also presents the major space-borne missions launched or planned by several countries or space agencies. Apart from those listed in Table 1–1, there are HISUI (The Japanese Hyperspectral Imager Suite) onboard the Advanced Land Observing Satellite-3 (ALOS-3), and the French HYPXIM hyperspectral missions (Briottet et al., 2011; Matsunaga et al., 2013) were also in orbit. HISUI is a space-borne instrument suite designed to work both as hyperspectral as well as multispectral imagers. Therefore the importance of these planned hyperspectral missions or initiatives should be understood to ensure guaranteed, uninterrupted hyperspectral image acquisition and coverage beyond 2020.

Table 1–1

aWide View CCD Cameras (WVC).

bHyperspectral Imager (HSI).

cInfrared Multispectral Scanner (IRMSS).

dhttps://hyspiri.jpl.nasa.gov/.

ehttps://www.unoosa.org/documents/pdf/copuos/2019/copuos2019tech11E.pdf.

fTIR measures both day and night data with 1 daytime image and 1 night-time image every 5 days.

gAfter launch, mission control was handed over from SunSpace to SAC (Satellite Applications Centre) at Hartebeetshoek near Pretoria in South Africa.

The focus of the HyspIRI mission will be on the world’s ecosystem studies offering information on future ecosystems, disasters, and the carbon cycle. This mission will offer VIS–NIR and SWIR hyperspectral data and multispectral thermal data regularly. The combination of spatiotemporal and spectral data will offer a challenge to researchers.

1.3 Atmospheric correction: a primary step in preprocessing hyperspectral images

Passive remote sensing is basically the study of interaction between light source and Earth surface features, in which every feature has an unique spectral response. The spectral resolution and continuous wavelength range of an image play an important role in distinguishing the feature; the higher the spectral resolution the higher will be its feature-detection capability (Okada and Iwashita, 1992; Clark, 1999). Hyperspectral imaging holds the potential to distinguish surface features more precisely because of its high-spectral resolution and narrow bandwidths. Hyperspectral images have numerous narrow bands covering ultra-violet, VIS–NIR, and SWIR regions of the electromagnetic spectrum, which makes it more susceptible to atmospheric distortions. Atmospheric gases and aerosols absorb and transmit the incoming light in regards to its wavelength and this causes distortions in the image. Therefore atmospheric correction is required in all hyperspectral data in which the raw radiance image is converted to a reflectance image considering all spectra is shifted to a similar albedo.

The techniques for accurate removal of atmospheric absorption and scattering from a hyperspectral image is mainly divided into two categories, namely relative and absolute (also termed as empirical) atmospheric corrections (San and Suzen, 2010). Further the technique of relative atmospheric correction is subdivided into three methods: (1) Internal average reflectance correction (Kruse, 1988; Ben-Dor and Kruse, 1994), (2) flat field correction (Gao et al., 2009), and (3) empirical line correction (Gao et al., 2009). In the relative atmospheric correction technique there is no need of providing a priori information about surface or atmosphere as it uses the data statistics and runs mathematical operations to correct the image. On the other hand absolute atmospheric correction techniques take a priori information such as water vapor, atmospheric gas content, and topography effects into consideration to run radiative transfer codes. On a per-pixel basis, the difference between radiation leaving Earth and received at the sensor along with the a priori inputs are used by the radiative transfer codes to correct the atmospheric distortions. Absolute atmospheric correction has been proven to be the better technique over the relative correction technique because it considers the local geographical and atmospheric a priori inputs for correcting the data (Nikolakopoulos et al., 2002).

Several radiative transfer codes have been introduced by researchers over the past decades for atmospheric corrections. Many of these codes are developed for a satellite specific imaging system and for a specific spectral and spatial range. Some of these radiative transfer codes include, Atmospheric Correction Now (ACORN) that is based on MODerate resolution atmospheric TRANsmission-4 (MODTRAN-4) (Miller, 2002). Atmospheric Correction (ATCOR) is also a MODTRAN-4 based code and is integrated with ERDAS Imagine (Earth Resource Development Assessment System) software (Adler-Golden et al., 1999), Atmospheric Removal (ATREM) (Gao et al., 1993), High Accuracy Atmospheric Correction for Hyperspectral data (HATCH) (Qu et al., 2003) which is an improved version of ATREM, and Fast Line-of-sight Atmospheric Analysis of Spectral Hypercube (FLAASH) (Cooley et al., 2002) which is integrated in ENVI (Environment for Visualizing Images) software. The major drawbacks of these radiative transfer codes are their complex algorithms, precalculated lookup tables, inputs based on interpolation, and the requirement of a priori data related to pixel-wise atmospheric and topographic information.

Radiative transfer codes based on atmospheric correction was introduced by Gao et al. (1992, 1993) in the 1990s and is termed ATREM. Using this algorithm, from AVIRIS data scaled surface reflectance is retrieved assuming a horizontal surface having Lambertian reflectance. The ATREM radiative transfer code is formulated by deriving the pixel-wise water vapor absorption bands at 0.94 and 1.4 µm, the incoming solar radiation, azimuth angle, and narrow bandwidth. Accordingly the transmission spectrum of atmospheric gases namely, carbon dioxide (CO2), carbon monoxide (CO), methane (CH4), nitrogen oxide (NO2) and oxygen (O2) is simulated. The scattering effect caused by suspended atmospheric molecules and aerosols is modeled using the Simulation of the Satellite Signal in Solar Spectrum (5S) model (Tanré et al., 1990). In early 2000 a line-by-line atmospheric transmittance model (Gao and Davis, 1997) was introduced into ATREM code which uses 6S module (Vermote et al., 1994) that included the effect of NO2 in the 0.4–0.8 µm spectral range. Several other radiative transfer codes having advanced features like topographic corrections, spectral smoothing, and feature adjacency correction such as ACRON, FLAASH, and HATCH have recently been introduced.

Taking into consideration the advancements in atmospheric correction techniques used for hyperspectral images, FLAASH is one of the most accurate and easy to use radiative transfer model. FLAASH is a software package introduced by Air Force Research Laboratory, Space Vehicles Directorate (AFRL/VS) and is currently integrated with ENVI software for commercial use. FLAASH is able to more accurately analyze the VIS to SWIR region of the electromagnetic spectrum. FLAASH is a MODTRAN-4 based model with physics-based derivations of different atmospheric interaction properties that include atmospheric pressure, water vapor column, and aerosol and cloud cover, which is further used as a priori information for converting radiance to reflectance (Berk et al., 1998, 2000). Hyperspectral data, its sensor type, and metadata (including sensor altitude, viewing, and solar angle) is used as input for FLAASH to retrieve preliminary water vapor column data at a per-pixel basis. Then a water vapor Look-Up Table (LUT) is generated using MODTRAN simulation to retrieve reflectance from radiance. FLAASH also generates a cloud mask to classify the pixels having cloud cover and flag those pixels to be removed before calculation. Considering the variables are all wavelength dependent, image polishing and renormalization is performed last to generate a new reflectance cube (Boardman, 1998).

Despite these atmospheric correction models, there are still possibilities for further improvement in radiative transfer codes. The built-in smoothing modules need improvement in order to identify any artificial broad absorption features in the retrieved reflectance spectra. A hybrid approach using image-based empirical methods and radiative transfer codes can increase the retrieval accuracy and lower the computational complexities.

1.4 Empirical and radiative transfer models

Inverse modeling plays a vital role for quantitative remote sensing, which mainly uses physical- or empirical-based models to estimate unknown parameters (Wang, 2012). Over the past few years various types of models have been developed related to atmosphere, vegetation, and radiation and a model-based inverse problem was also introduced. The introduction of multispectral and hyperspectral remote sensors with greater spectral and spatial information came into the account to solve model-based inverse problems (Liang, 2005; Wang et al., 2009). Inverse modeling is mainly useful to retrieve various types of vegetation parameters using physical or empirical models. These models also can be used in direct (forward) mode to compute canopy-reflectance values after providing leaf and canopy traits [chlorophyll content, water content, Leaf Area Index (LAI)]. Moreover, such physical-based radiative transfer models can be inverted from reflectance or EO data for retrieval of biophysical and biochemical variables like LAI and fraction of photosynthetically active radiation, which are mainly used to monitor the health status and enumerate the vegetation influence. Narrow-band information in hyperspectral sensors can aid in providing quantitative estimates of canopy biochemical properties (such as chlorophyll and nitrogen concentrations) when compared to multispectral (broadband) sensors (Goodenough et al., 2004, 2006).

Inversion of the canopy Radiative Transfer Model (RTM) is known as one of the best approaches for the retrieval of biophysical and biochemical parameters. During the past few years various types of studies have been done on RTMs. Most useful RTMs are the leaf PROSPECT (leaf optical properties model) and canopy-based SAIL models (canopy bidirectional reflectance model) that combine the PROSAIL model and are useful for inverse modeling (Verrelst et al., 2019). Advance methods such as PROSPECT (COSINE) and PROSAIL have maximum computational power and numbers of inversion algorithms but are still time consuming to provide efficient results. Further optimization and the robustness of LUT inversion application have also been explored against spectroscopic data (Banskota et al., 2013, 2015; Bao et al., 2017). LUT-based inversion approaches provide fast and efficient results because of before-hand completion of the inversion itself which is assumed to be the most computationally expensive part of the inversion procedure. The LUT-based inversion toolbox is a part of the ARTMO (Automated Radiative Transfer Models Operator) software package that can be run on MATLAB and also provides essential tools and information for running as well as inverting a collection of plant RTMs, both at the leaf (PROSPECT, DLM, Liberty, and Fluspect) and canopy level (4SAIL, INFORM, FLIGHT). Leaf and canopy RTMs are used as an input for the generation of class-based LUTs (Fig. 1–1). Moreover ARTMO is capable of evaluating LUT-based inversions on land-cover classification, which allows reasonable retrieval of biophysical parameters over land surface (Rivera et al., 2013; Verrelst et al., 2013).

Figure 1–1 Basic principle of the Look-Up Table-based inversion toolbox (Rivera et al., 2013).

A study has demonstrated the use of CASI (Compact Airborne Spectrographic Imager) hyperspectral reflectance data for monitoring some forest sites with dense canopies (LAI >4) and has also shown the achievability of RTMs for pigment estimation. The RTM inversion technique exhibited promising results with RMSE values from 3.0 to 5.5 µg/cm² for a leaf chlorophyll range of 19.1–45.8 µg/cm² (Zarco-Tejada et al., 2001).

This study has confirmed that the success of LUT-based inversion strongly depends on the retrieval parameters and applied regularization options. Performance of LUT-based inversion is directly related to the used-cost functions (Verrelst et al., 2013). LUT-based inversion on other types of hyperspectral sensors such as AVIRIS and Hyperion are also useful to retrieve vegetation parameters and pigments for the reorganization of the plant and crop species.

1.5 Applications of hyperspectral remote sensing

There are a plethora of applications possible using hyperspectral remote sensing, but herein only a number of important areas such as vegetation, urban, mineral, water and agriculture are provided with respect to hyperspectral applications.

1.5.1 Vegetation analysis

Terrestrial vegetation is considered as one of the Earth’s most significant natural resources and is spread over a large surface. It encompasses different landscapes namely forest, agriculture, rangeland, wetland, and urban vegetation (Jensen, 2009). Monitoring terrestrial vegetation at spatio-temporal scale is crucial for studying different terrestrial resource management applications. Field measurements of foliar and canopy vegetation’s biophysical and biochemical parameters are the best employed approaches for its monitoring (Mutanga et al., 2004; Chmura et al., 2007; Tilling et al., 2007). Conventional field estimations of these parameters involve time, money, and labor and it cannot be easily extended to big geographical areas.

Hyperspectral remote sensing is the most appropriate alternative technique for retrieving vegetation biophysical and biochemical parameters since narrow-band measurements of spectral reflectance in hyperspectral remote sensing allow finer determination of the changes in vegetation’s biophysical and biochemical parameters. Numerous authors have explored the capability of hyperspectral remote sensing in providing valuable information on these parameters. Both ground-level hyperspectral data acquired using hand-held spectroradiometers as well airborne or spaceborne hyperspectral instruments have been widely used in ample vegetation studies. Available airborne hyperspectral instruments include AVIRIS, AVIRIS-NG, and HyMap; whereas spaceborne hyperspectral instruments are Hyperion, CHRIS, EnMAP, and the HJ-1A hyperspectral imager product.

Additionally other vegetation analyses including diversity parameters such as species diversity (Peng et al., 2018; Malhi et al., 2020) and species richness (Psomas et al., 2011; Peng et al., 2019) are also successfully retrieved using this advanced remote sensing technique. A large number of narrow bands of hyperspectral data also encourages the classification and mapping of different vegetation types at varied taxonomic scales, mostly down to the species level. For example, species-level classification with high accuracy is achieved for grapevine varieties (Fernandes et al., 2015) using hyperspectral data (Clark et al., 2005). Classification using hyperspectral data was successfully carried out for different vegetation types such as annual gramineous weeds (Deng et al., 2016), food crops (Mariotto et al., 2013), shrubs of arid zones (Lewis, 2002), and montane or subalpine trees (Sommer et al., 2016) and forest tree species (Hycza et al., 2018). The hyperspectral technique is also used to determine different plant functional types that are functionally similar plant species in terms of resource utility, ecosystem function, and response to environment conditions. Such classification is carried out using methods like spectral mixture analysis (Schaaf et al., 2011). Studies were also performed where hyperspectral remote sensing was used in the detection and classification of the early onset of plant disease and stress (Lowe et al., 2017).

1.5.2 Urban analysis

The world’s population is continuously shifting toward urban centers resulting in the regular modification in the landscape at regional level. The densification of urban areas is making the urban fabric (such as buildings, roads, etc.) more complex and it is also leading to the generation of urban-heat islands (Weber et al., 2018). Hence there is a great need to understand the relation between urban systems with respect to biotic and abiotic components. To understand the dynamics of the urban system, the identification of building materials, impervious surfaces, mineral composition, water quality, vegetation, and fallow lands at very fine spectral resolution is crucial, and can be archived by hyperspectral data. Hyperspectral data provides a very fine spectral resolution that helps in the identification of building and road materials, microclimate model development, vegetation health monitoring, pollution monitoring, water quality assessment, and so forth, because all the elements have a unique absorption feature in their reflectance spectra. There are several studies that have been done using hyperspectral data to enhance the urban management system, including the study done by Karoui et al. (2018) in which they used the hyperspectral data to identify photovoltaic panels within the region using the spectral unmixing technique. Roof-top mapping is a common application of hyperspectral data in urban studies (Chisense et al., 2012) where the endmembers of rooftop materials were generated and classification on HyMap hyperspectral data was performed to identify the rooftop materials. This study is also crucial in urban management as it can help in the identification of buildings that are the major source of urban heat islands. Urban Land Use Land Cover (ULULC) mapping is one of the most important applications of hyperspectral data. Again a large number of spectral bands provides better feature detection capability with respect to multispectral data. There are several classification techniques such as Support Vector Machine (Tratt et al., 2016), Artificial Neural networks (Goel et al., 2003), K-means (Filho et al., 2003; Licciardi et al., 2011), and Principle Component Analysis (PCA) (Licciardi et al., 2011) and Segmented PCA (Pandey et al., 2014) that are commonly used for hyperspectral data. The NIR, thermal and SWIR regions of electromagnetic spectrum in hyperspectral data are sensitive to different gases. This aids in identifying and tracking gaseous emission from compact sources present in urban-industrial areas (Tratt et al., 2016). Hyperspectral data also has a major role in water-quality mapping (Olmanson et al., 2013), identifying soil properties (Hively et al., 2011), and pest detection (Glaser et al., 2009). Apart from the different applications of hyperspectral data for urban studies, there are certain limitations also, including the data complexity and lack of spatiotemporal hyperspectral data as there are limited spaceborne hyperspectral