High Spatial Resolution Remote Sensing Data, Analysis, and Applications

Preface

1 INTRODUCTION

High spatial resolution data provide a novel data source for addressing environmental questions with an unprecedented level of detail. These remote sensing data are a result of significant advances in image acquisition platforms and sensors, including satellite, manned aircraft, and unmanned aerial vehicle (UAV) platforms. Furthermore, the recent development of commercially operated satellite platforms with high spatial resolution sensors allows for the collection of a large amount of images at regular time intervals, with relatively large footprints (i.e., image swathes). For example, the WorldView series, with the WorldView-1 satellite launched on September 18, 2007, and WorldView-4 launched on November 11, 2016, are capable of resolving objects at 31 cm in the panchromatic band and at 1.24 m in 4 (or 8)-band multispectral over a 13.1-km-wide swath. 

For a specific study site, these image data can be easily searched and acquired at a reasonable cost through service companies, such as DigitalGlobe. In addition, the recent proliferation of UAVs has made it possible to collect images at spatial and temporal scales that would be impossible using traditional platforms. Many recent works have focused on collecting imagery using UAV-equipped multispectral, hyperspectral, and thermal sensors as well as laser scanners at centimeter resolutions. High (meters) and ultra-high (centimeters) spatial resolution images open a door for fine-scale analysis of objects at the Earth’s surface. 

A number of scientific journal articles have highlighted the usefulness of high spatial resolution remote sensing, including the use of remote sensing in studying the physical environmental system, the human system, and the interactions between them. Examples in physical environmental studies include fine-scale forest inventory (Mora et al., 2013), wetland plant community identification (Zweig et al., 2015), grassland mapping (Lu and He, 2017), and water resources (Debell et al., 2016). In terms of the human system, high spatial resolution remote sensing has been used to study urban impervious surfaces (Yang and He, 2017), public health (Hartfield et al., 2011), and epidemics (Lacaux et al., 2007). 

As for human-environment interactions, high-resolution remote sensing has been used for land degradation (Wiesmair et al., 2016), precision farming (ZarcoTejada et al., 2013), water and air pollution (Yao et al., 2015), and natural hazards (e.g., earthquakes, typhoons, floods, landslides) (Joyce et al., 2009). This increased spatial resolution exasperates the intraclass variability found in an image. For example, in a grassland scene, vegetation leaves, gaps, shadows, and stems are all visible in the pixels of a high spatial resolution image. 

While this information is potentially useful for mapping purposes, the added details in a high-resolution image pose challenges for image segmentation and feature selection. Furthermore, the number of detectable entities or classes increases with spatial resolution. Traditional information extraction techniques may not operate well at high spatial resolutions due to large data volume and heterogeneous spectral information (Wulder et al., 2004), spurring the need for the development of innovative image processing techniques. To this end, techniques operated in the spatial domain have the potential for successfully extracting information from high spatial resolution images (Culvenor, 2003). 

To effectively utilize information contained in high spatial resolution imagery, some key questions must be addressed, including 

1. What are the challenges of using new sensors and new platforms? 

2. What are the cutting-edge methods for fine-level information extraction from high spatial resolution images? 

3. How can high spatial resolution data improve the quantification and characterization of physical-environmental or human patterns and processes? 

This book intends to address the above-mentioned questions in three separate parts: (1) data acquisition and preprocessing, (2) algorithms and techniques, and (3) case studies and applications. This chapter aims to provide an overview of the book and an outline of each chapter. Section 2 discusses the opportunities and challenges of using new sensors and platforms and of using high spatial resolution remote sensing data. 

It also introduces the chapters that address recent developments in platforms, sensors, and data with a focus on UAVs. Section 3 addresses the issues related to high spatial image processing and introduces cutting-edge methods. Section 4 summarizes state-of-the-art high spatial resolution applications and demonstrates how high spatial resolution remote sensing can support the extraction of detailed information needed in different systems. 

Readers should bear in mind that platforms, sensors, data, methods, and applications are related to each other. Although we intend to organize the chapters based on their primary focuses, we also acknowledge that the authors of each chapter may tell a complete story, sometimes from introducing an image acquisition system all the way to a specific application.

2 IMAGE ACQUISITION SYSTEMS AND PREPROCESSING

An image acquisition system, which includes a platform and one or more sensor(s), plays an important role in the sense of how efficiently the areas under study can be surveyed. Traditional remote sensing platforms include satellite, aircraft, and space shuttle platforms, while a more recent addition is the UAV remote sensing system. Each of these platforms has its own technological and operational specifications (Table 1). 

Satellite platforms can survey large geographical areas, but their spatial resolutions are relatively coarse, the sensors are fixed, and the image acquisition plan is out of the end user’s control. In other words, most high spatial resolution satellites can only acquire data after they have been programmed or paid to do so (Joyce et al., 2009). One exception is that during a major disaster event, satellite operators usually schedule imagery collection without needing an approved request. 

Different from satellite platforms, the use of both manned and unmanned aircraft can be tailored to meet the needs of an end user, and acquisition sensors and parameters can be adjusted to ensure the best possible imagery. In terms of surveying area, manned aircraft can survey in kilometer squares with ideal sensors, while UAVs are only good for studying relatively small areas due to their limited battery life. However, UAVs have attracted increased attention in recent years due to their high

flexibility, low operational costs, and ability to collect very high spatial resolution images. In comparison with manned aircraft, which can mount large sensors, UAVs with limited payloads can only carry compatible sensors. With the advancement in sensing and computing technologies, sensors have become more compatible. It is hoped that multiple sensors, such as multiple cameras and/or LiDAR sensors, can be mounted on UAVs in the near future (Asner et al., 2012). However, aerial image acquisitions involve additional cost and efforts for image preprocessing because they are a function of the camera or sensor optics. 

The image preprocessing efforts include, but are not limited to, obtaining ground control data, performing specific radiometric correction, and image mosaicking. Clearly, the optimal platform depends not only on the size of the study area and the objective of study, but also on the available budget and resources. Many studies have reviewed challenges in relation to UAV image collection and preprocessing (e.g., Whitehead and Hugenholtz, 2014). It is almost impossible to process UAV images using the existing image preprocessing tools because they were developed mainly for well-calibrated platforms and sensors. 

A UAV acquires images with inconstant scales, large overlaps, variable image orientations, and high amounts of relief displacement (Hardin and Jensen, 2011). In spite of these challenges, UAV remote sensing systems have proved useful for many different applications (see Chapters 1, 2, and 3). In Chapter 1, UAV remote sensing technology with high spatial and temporal resolutions is considered critical for high-throughput phenotyping and precision agriculture. 

In the field experiments, the authors used a variety of UAV platforms and sensors, including PrecisionHawk Lancaster 5 with a Nikon 1 J4 visual camera, Tuffwing UAV Mapper with a Sony A6000 visual camera, and DJI Matrice 100 UAV with a MicaSense RedEdge multispectral camera. Many considerations were applied during image acquisition and preprocessing steps. The authors conclude that UAV has great potential for obtaining imagery that can provide timely and accurate information for decision support, but further research regarding UAV engineering configurations that enable navigation, and communication and data transfer capabilities are needed. 

The authors point out a variety of issues related to UAV imagery preprocessing, diagnostic analysis of environmental and plant conditions, and artificial intelligence for decision support. Similarly, Chapters 2 and 3 also describe the need to build a UAV-based imaging system, but with a focus on hyperspectral imagers. The author highlights the need for a UAV-based hyperspectral system, factors to be considered when building and operating a UAV-based hyperspectral system, and the methods for preprocessing UAV-based hyperspectral data. Other high spatial resolution data from a variety of sensors, including optical, LiDAR, and synthetic-aperture radar (SAR), are introduced in Chapter 4, 5, and 6 with a focus on data preparation and preprocessing. In particular, Chapter 4 provides a brief overview of data preparation needed for integrating high-resolution multispectral and LiDAR data through a case study of mapping coastal wetlands. 

Using these data, the authors quantified important structural metrics for different wetland compositional types, and suggested that the combined use of LiDAR, multispectral remote sensing, and terrain metrics achieved the best model performance. The authors suggest new research directions to address the challenges often experienced in the integration of LiDAR and multispectral data. Chapter 5 discusses multiview image matching steps, and introduces advanced image matching techniques. In addition, multiview image matching accuracy and related challenges, limitations, and opportunities are also discussed. 

Chapter 6 focuses on high-resolution radar data acquisition, preprocessing, and processing. The chapter starts with a nonmathematical presentation of the fundamentals of SAR that are relevant to the high-resolution imaging of natural environments, followed by basic SAR image preprocessing, image processing, four SAR satellite systems, and case studies of SAR application for identifying manmade and natural features.

3 HIGH SPATIAL RESOLUTION IMAGE PROCESSING

The availability of enormous amounts of high spatial resolution remote sensing data has necessitated the development of appropriate image processing methods. This book presents a few algorithms and workflows, including structure from motion (SfM) techniques, stepwise procedures, spectral unmixing, object-based image classification, and convolutional neural networks to extract information from different high spatial resolution data. 

SFM reconstructs 3-D geometry and camera position from a sequence of 2-D images captured from multiple viewpoints (Ullman, 1979). This algorithm was developed decades ago, but has recently become popular for processing UAV remote sensing images. Using a consumer-grade DJI Phantom 3 Professional Quadcopter with a RGB camera, the authors in Chapter 7 collected UAV images from multiple positions and analyzed them using various SfM programs to establish a workflow for generating reliable estimates of wood chip volume.

While optical image processing exploits mostly spectral properties, in the case of airborne laser scanning (ALS) data, methods should be developed to take advantage of both range and spectral information from a single source. Chapter 8 proposes a stepwise procedure including land-cover classification, tree crown segmentation, and regression modeling to examine the feasibility of multispectral ALS data for tree carbon stock estimation. 

The proposed workflow provides a benchmark for processing emerging multispectral ALS data in land-cover mapping. Linear spectral unmixing has captured growing attention for its ability to extract fine-scale features (Yang et al., 2014). When endmembers are identified in a 2-D spectral mixing space, all pixels in the images can be processed as a linear combination of the identified endmembers. However, the choice of an appropriate triangle structure (from a spectral mixing space) is yet to be automated. To address this gap, Chapter 9 introduces an indicator that is capable of selecting suitable feature space objectively, and thus holds great potential for automated high-resolution remote sensing image unmixing.

High spatial resolution image classification has generally advanced from pixel-based to object-based approaches, with the latter delineating real-world objects composed of many pixels (Wulder et al., 2004). The objects thus add several new possible layers to image analysis ranging from spectral descriptive statistics to textural and geometric information. However, object-based image analysis also introduces a number of analytical issues to consider. 

In the image segmentation step, an inappropriate scale can negatively affect classification results, and therefore researchers must decide which segmentation scale to apply to the image (Ma et al., 2015). Chapter 10 provides a review on five scale-selection methods, and compares their advantages and disadvantages using WorldView-2 imagery. The authors suggest that scale-selection methods be explored before choosing an optimal scale to apply for segmenting objects. 

Computer vision aims to mimic the human visual system for automatic image extraction, analysis, and interpretation. Chapter 11 summarizes the relevant computer vision technologies with a focus on convolutional neural networks. Specifically, the authors discuss how computer vision can be adapted to work in ecological studies, and introduce experiments to assess the effectiveness of convolutional networks in scenarios similar to the camera trap scenario. The chapter ends with several considerations for the use of these techniques in ecological studies.

4 CASE STUDIES AND APPLICATIONS

Using various high spatial resolution data, Part 3 of this book covers a range of unique applications. For grasslands, UAV-based multispectral images were used in Chapter 12 to investigate vegetation biophysical and biochemical properties in a tall grassland, while manned-aircraft-based hyperspectral images were used in Chapter 13 to invert a radiative transfer model for mapping leaf chlorophyll in a mixed grassland. Chapter 12 demonstrates that canopy leaf area index (LAI) and chlorophyll content can be accurately retrieved from UAV-acquired imagery using spectral indices. 

Chapter 13 concludes that the inverted model is able to estimate leaf chlorophyll content for green canopies with high accuracy, but overestimates both mixed green and brown canopies and mixed green and brown canopies with exposed soil.

Both Chapter 14 and 15 delineate wetlands, with Chapter 14 focusing on wetland mapping using GeoEye-1 images, and Chapter 15 characterizing geomorphic and biophysical properties of wetlands with airborne LiDAR. Chapter 14 indicates that wetland complexity, including varying sizes and shapes and the aquatic vegetation communities found within, can be detected using high spatial resolution optical imagery. 

Airborne LiDAR has gained popularity in wetland science research in the past two decades and Chapter 15 reviews this expanding field with examples from a wide range of wetland ecosystem types. Using a case study conducted in a northern peatland complex, the authors demonstrate concepts related to the accuracy of LiDAR-derived ground surface elevations, as well as geomorphic and hydrologic analysis.

When dealing with karst, one of the most fragile and heterogeneous landscapes, high spatial resolution imagery can provide detailed information that aids in the exploration of mechanisms of vegetation dynamics. In Chapter 16, vegetation cover in a degraded karst area was extracted using multispectral high spatial resolution ALOS imagery. The authors conclude that high spatial resolution imagery, when processed with the multiple endmember spectral mixture analysis approach, is able to successfully extract fine-scale features in karst areas. 

Chapter 17 uses IKONOS multispectral and panchromatic images to estimate cherry orchard acreage. High spatial resolution imagery combined with object-based image analyses was found to be effective in estimating the cherry orchard acreage, within a ±3.1% error margin of the U.S. Department of Agriculture’s census data. The author suggests using a UAV multispectral system to collect fine-resolution images because it may provide detailed information on fruit crop yield and/or potential diseases that may adversely affect orchard trees.

5 SUMMARY

High spatial resolution remote sensing data have proven useful for the extraction of ground features at a level not possible with medium- or coarse-resolution images. However, until now, the novelty of UAV platforms and sensor configurations, high cost of commercial-satellite-based data, inconsistent revisit times, and underdeveloped image processing methods have limited the usefulness of high spatial resolution images for large area applications at an operational level, not to mention real-time assessment and monitoring. 

The best practice in high spatial resolution remote sensing data acquisition and processing has yet to be developed. Spatial resolution, image swath, spectral features, and temporal revisit are all critically important in determining whether a particular platform and/or sensor or data processing algorithm is capable of providing specific assessment and monitoring capabilities. It often takes at least 24 hours, or even several days, before a satellite operator can respond to an image acquisition request. This is because response time is determined by the position of the satellite within its orbit, as well as by weather conditions in the case of optical sensors (Joyce et al., 2009). 

Other than satellite platforms, both manned and unmanned airborne platforms are capable of acquiring images for real-time mapping and monitoring, but are limited to local scales. Further, data preprocessing, processing, and delivery systems are currently being developed

 for manned and unmanned systems and their usefulness is yet to be validated. It is expected that in the near future, end users will be able to receive near-real-time images and mapping products, from manned or unmmaned systems, for an area or a feature of interest. No single platform, sensor, data type, or processing technique will work for all cases. Multisource data integration, multiresolution exploitation, and multitemporal analysis will likely be key future lines of research (Zhang, 2010) for real-time assessment and monitoring. 

The application-driven strategy has been encouraged in past decades, with the hope of finding an effective solution for a specific problem. However, a data-driven analysis is possible in the era of big data, and one may achieve better real-time assessment and monitoring using multisource remote sensing big data.
The chapters in this book provide a snapshot of cutting-edge high spatial resolution remote sensing image collection, preprocessing, processing, and applications. More advances in these areas can be expected in the near future. We hope that the collection of chapters in this book will provide a useful benchmark for the high spatial resolution remote sensing community and inspire more studies that would address important scientific and technical challenges in current and future high spatial remote sensing data acquisition, processing, and applications.

Download Link

High Spatial Resolution Remote Sensing Data, Analysis, and Applications

See Also https://georsgisn.blogspot.com/2019/04/spatial-statistics-geospatial-information-modeling-thematic-mapping.html