2 Theoretical Framework
2.1 Understanding satellites: definition and common features
A satellite is an object that orbits around a larger celestial body due to gravitational forces. Natural satellites, such as the Earth orbiting the Sun or the Moon orbiting the Earth, are formed through natural processes and are an integral part of the universe’s structure. These celestial objects differ from artificial satellites, which are human-engineered and serve human-centric purposes.
The term satellite is most commonly associated with artificial satellites mechanisms designed and constructed by humans to perform a variety of tasks. These artificial satellites are launched into orbit around Earth or other celestial bodies to support diverse applications, including communication, Earth observation, navigation, and scientific exploration (Adams, 2017). Their versatility and functional diversity make them invaluable tools for advancing scientific knowledge, technological progress, and global connectivity.
Despite their varied applications, all satellites share fundamental operational features. A primary requirement is a reliable power source, typically solar panels, combined with a storage battery to ensure uninterrupted operation during periods without sunlight. As solar irradiance decreases proportionally to the square of the distance from the Sun (or any spherical source), it quickly becomes insufficient for powering satellites far beyond Earth’s orbit. Consequently, satellites operating at greater distances from the Sun depend on radioisotope thermoelectric generators (RTGs), which provide both electricity and heat to maintain equipment at operational temperatures (NASA n.d.). These energy systems enable satellites to sustain their core functions, including data acquisition, communication, and maneuvering in orbit. This combination of power, communication, and navigation systems underpins the essential functionality of satellites, making them adaptable to a broad range of tasks in space.
2.2 Publicly available satellite data and platforms for monitoring coastal shoreline dynamics: a comprehensive overview
2.2.1 Copernicus Browser
The Copernicus Browser is a vital online platform providing free and open access to geospatial data and imagery from the European Union’s Copernicus Earth Observation Program, managed by the European Space Agency (ESA). It grants users access to an extensive range of datasets, particularly from the Sentinel satellite constellation, enabling the detailed exploration of Earth’s surface for scientific, environmental, and commercial applications.
The data is available in standard geospatial formats, such as GeoTIFF for raster imagery, ensuring seamless compatibility with geospatial analysis tools like QGIS, R, and Python. Users can download imagery at various resolutions, including 10m, 20m, and 60m, catering to different project requirements and levels of detail (Sinergise Solutions n.d.).
2.2.2 PlanetScope: free access under specific conditions
PlanetScope is a satellite platform operated by the company Planet, featuring a constellation of approximately 130 satellites. This constellation is capable of imaging the entire land surface of Earth every day, covering an impressive 200 million km² per day. The satellite imagery has a resolution of 3 meters per pixel, making it well-suited for detailed environmental and spatial analyses.
A PlanetScope Scene Product represents a single framed image captured as part of the satellite’s continuous line-scan of the Earth’s surface. These scenes are individual segments within a strip of imagery, overlapping with one another, and are not aligned to a specific tiling grid system. This format allows for flexible use, but requires additional processing for some applications.
One of the most remarkable aspects of PlanetScope is its accessibility. The platform offers freely available data samples and provides free access for students, academic researchers, and humanitarian projects through the Planet Education and Research Program (Planet 2024a).
While PlanetScope boasts an impressive daily collection capacity and frequent revisits, its affordability and accessibility are perhaps its most standout features, ensuring that high-quality satellite data is available for a wide range of users and applications.
2.2.3 RapidEye (discontinued in 2020)
RapidEye, a satellite constellation, was in operation from 2009 to 2020 and was developed by the company Planet, which also operates PlanetScope, as mentioned earlier. The constellation consisted of five high-resolution satellites, each capable of capturing imagery with a spatial resolution of approximately 5 meters per pixel (Planet 2024b).
RapidEye’s sensors captured imagery across five spectral bands: Red, Green, Blue, Red Edge, and Near Infrared. This capability made RapidEye highly valuable for applications in agriculture, forestry, and environmental monitoring. Furthermore, more than 70% of the images were acquired with a view angle of less than 10°, with a maximum view angle limited to 20°, ensuring minimal distortion and consistent high-quality data (ESA n.d.).
The extensive archive of RapidEye imagery is available for research and educational purposes upon request, free of charge (Planet n.d.). This historical archive remains a critical resource for studying and analyzing changes in the Earth’s surface over time.
2.3 Free and open-Source software Tools that could be used for monitoring coastal shoreline fluctuations
2.3.1 QGIS: strengths and limitations in geospatial analysis
QGIS (Quantum Geographic Information System) is open-source, cross-platform software for efficient geospatial data management, visualization, analysis, and mapping. It is valued for its accessibility, flexibility, and integration capabilities across various geospatial technologies. One of the primary advantages of QGIS is its cost-effectiveness. As free software, it eliminates the financial barrier associated with many proprietary GIS platforms. It supports a wide range of vector and raster file formats. This compatibility allows users to work with nearly any geospatial data type, including shapefiles and GeoTIFFs (Map-site n.d.).
For temporal or dynamic data, QGIS provides tools like TimeManager, which enables users to visualize changes over time. This is particularly useful for monitoring environmental changes (Map-site n.d.).
QGIS is also highly extensible, offering a vast library of plug-ins to enhance its functionality. These plug-ins allow users to perform specialized tasks, such as georeferencing raster data, creating temporal animations, and automating processes through batch operations. The active development community continuously contributes new plug-ins and updates, ensuring that QGIS remains a dynamic and evolving tool. Another strength of QGIS is its user-friendly interface, which lowers the learning curve for beginners. Tutorials and extensive online resources make it accessible even for users with little to no prior experience in GIS. The software also supports advanced functionality, such as 3D visualization, time-series analysis, and complex spatial queries, making it suitable for both basic and advanced geospatial tasks.
While QGIS is highly versatile, it does have limitations. The software integrates well with external tools, however, machine/deep learning and some other advanced geospatial tasks require additional software. For instance, QGIS does not natively support models like Convolutional Neural Networks (CNNs) or the Segment Anything Model (SAM), which are often essential for tasks like automated object detection and image segmentation in remote sensing (Gillian n.d.).
2.3.2 Geospatial analysis using python libraries
Python provides a comprehensive ecosystem for geospatial analysis, supporting a wide range of workflows, from preprocessing and analyzing raster and vector data to implementing advanced machine learning techniques. Its libraries, such as rasterio, geopandas, and xarray, are designed to handle geospatial data efficiently. For raster data, rasterio allows users to read, write, and manipulate formats like GeoTIFF, while geopandas extends Pandas to support spatial operations for vector data, including spatial joins and reprojections. Additionally, xarray enables the analysis of multidimensional raster datasets, such as time-series or climate data.
Python’s integration with deep learning frameworks, such as TensorFlow and PyTorch, makes it particularly suited for complex geospatial tasks. Convolutional Neural Networks (CNNs) are widely implemented for tasks like land cover classification, object detection, and semantic segmentation. Models such as UNet, optimized for image segmentation, are frequently applied to delineate detailed land use patterns or identify structures in satellite imagery. Python also supports tools like the Segment Anything Model (SAM), which generalizes segmentation tasks across diverse datasets using minimal user input. SAM’s transformer-based architecture allows it to identify features like buildings, vegetation, or land boundaries in high-resolution imagery efficiently. These frameworks leverage GPU acceleration, allowing for the processing of large datasets (Setu et al. 2024).
Visualization is another area where Python offers strengths. Libraries such as folium and plotly enable the creation of interactive maps, while matplotlib and cartopy provide robust tools for static visualizations with geospatial overlays. Python’s ability to connect with cloud-based platforms, such as Google Earth Engine (GEE), further enhances its capacity to process and analyze large-scale geospatial datasets without the need for significant local infrastructure.
However, Python’s complexity can be a drawback. Its workflows often require combining multiple libraries, which can increase development time and lead to steeper learning curves. Additionally, while Python supports high-performance tasks like deep learning, setting up the required environment, including GPU dependencies, can be resource-intensive and technically demanding. For visualization, while Python provides dynamic tools, creating highly customized or publication-quality outputs may require additional effort compared to other geospatial tools (Priyadharshini 2015).
In summary, Python’s geospatial libraries are well-suited for handling raster and vector data, integrating machine learning and deep learning techniques, and supporting cloud-based workflows. While its flexibility and advanced tools provide significant capabilities, they also introduce complexity that may require technical expertise and time to manage effectively.
2.3.3 R for remote sensing: capabilities and limitations
R has become a widely used tool for remote sensing analysis due to its extensive library of packages, its flexibility, and its ability to handle large spatial datasets. Remote sensing involves deriving valuable insights from satellite imagery or other remote platforms, often requiring extensive preprocessing, analysis, and visualization. With specialized packages like terra, sf, and stars, R provides robust solutions for many aspects of remote sensing workflows, from raster data handling to vector geometry operations.
The terra package is a key tool for raster data processing in R, succeeding the raster package with enhanced efficiency for large datasets. Raster data, commonly derived from satellite imagery like Sentinel-2 or Landsat, represents spatially continuous information. With terra, users can manage multi-band rasters, crop to areas of interest, normalize reflectance values, and perform raster algebra for tasks such as NDVI (Normalized Difference Vegetation Index) calculations (Ghosh and Hijmans 2023). Its built-in functions for processing large files without exhausting memory make it particularly suitable for handling data from modern satellite missions, where file sizes can be enormous. It is also Supporting formats like GeoTIFF and JP2 (Wasser 2017).
The sf package complements terra by focusing on vector data, such as points, lines, and polygons, which represent boundaries, infrastructure, or sampling locations. It adheres to the Simple Features standard, enabling seamless work with vector geometries and integration with raster datasets. This allows for workflows like extracting raster values for polygons or performing spatial overlays. The package integrates well with R’s data manipulation and visualization tools, such as dplyr and ggplot2, making it easy to create maps and conduct spatial analysis.
For multidimensional datasets, the stars package offers a specialized approach, particularly for spatiotemporal data. Designed to handle raster cubes with dimensions like time and spectral bands, it is ideal for monitoring changes over time, such as vegetation dynamics or coastal erosion. Stars also integrates with sf for overlaying vector geometries on raster datasets, making it a valuable addition to R’s remote sensing ecosystem, despite being less widely adopted than terra (Ghosh and Hijmans 2023).
When it comes to machine learning, R provides interfaces to frameworks like Keras and TensorFlow, enabling users to build and train neural networks for tasks such as land cover classification or object detection. In addition to these deep learning capabilities, R supports supervised learning models like random forests and support vector machines, as well as unsupervised methods like k-means clustering or hierarchical clustering. These methods are well-suited for tasks such as identifying land use patterns or segmenting satellite imagery based on spectral properties. While R is capable of implementing deep learning models, its adoption for advanced architectures like Convolutional Neural Networks (CNNs) is less common, particularly compared to Python. Unfortunately, tools like SAM are not readily available in R, requiring researchers to rely on Python for these capabilities. Nevertheless, R remains effective for many machine learning tasks in remote sensing, especially when focusing on traditional models and workflows that prioritize statistical and geospatial analysis (Priyadharshini 2015).
Visualization is another area where R shows both strength and limitations. R’s ggplot2 and tmap are exceptional for creating static and high-quality visualizations. However, when it comes to dynamic and interactive visualizations of large datasets, tools like Python’s folium outperform R. While R’s leaflet package does provide interactivity, it lacks some of the advanced features required for visualizing large geospatial datasets interactively (Chege 2024).
In conclusion, R is a powerful tool for remote sensing analysis, offering exceptional capabilities for raster and vector data processing through terra and sf, and advanced spatiotemporal analysis through stars. Its strengths lie in statistical modeling, data visualization, supervised learning, unsupervised learning, and integrating geospatial analysis with dashboards, such as those built with R Shiny. However, tasks involving deep learning workflows or highly specialized models like UNet or SAM are often better handled in Python.
2.4 Panchromatic and pansharpened satellite imagery
Panchromatic and pansharpened satellite imagery are essential tools in remote sensing, offering enhanced capabilities for observing and analyzing Earth’s surface. Panchromatic imagery, often abbreviated as PAN imagery, is a single-channel grayscale image that integrates visible light wavelengths red, green, and blue into a single band. This approach sacrifices spectral detail for improved spatial resolution, making it ideal for capturing fine surface features. The information contained in each pixel of a panchromatic image directly reflects the total intensity of solar radiation reflected by objects within the pixel and detected by the satellite sensor. As a result, PAN imagery provides sharp and detailed images that are highly suitable for spatial analysis (Shanshan 2022).
A panchromatic band by itself produces black and white images with high spatial resolution. For example, satellites like Landsat 7 and 8 offer panchromatic images with a resolution of 15 meters per pixel, which is significantly higher than the 30-meter resolution of their multispectral counterparts. This higher resolution allows panchromatic imagery to assist individual spectral bands by making them "sharper," enhancing the visual and analytical clarity of satellite data (Analytic 2021).
One of the most significant applications of panchromatic imagery is its role in panchromatic sharpening, or pansharpening. This process fuses the high-resolution spatial data from panchromatic images with the spectral information from lower-resolution multispectral images. The resulting pansharpened image combines the best of both worlds: the spatial resolution of the panchromatic image and the spectral richness of the multispectral data. Pansharpening produces high-resolution color images that are more visually detailed and analytically useful for various applications. This fusion process has proven particularly beneficial for mapping land use, monitoring vegetation, and studying urban environments. By enhancing the spatial detail while preserving spectral attributes, pansharpening enables more accurate classification of surface features and clearer delineation of boundaries (McAuliffe 2021).
The process of pansharpening relies on techniques that integrate the complementary strengths of panchromatic and multispectral imagery. Various methods are used, including HSV sharpening and Gram-Schmidt pansharpening. Each method offers specific advantages, depending on the application. For example, HSV sharpening works within the HSV color space, where H stands for hue, S for saturation, and V for value (brightness). In this method, the high-resolution panchromatic data replaces the lower-resolution value component of the multispectral image, while the hue and saturation components remain unchanged. This ensures the resulting image retains its original color characteristics but with improved sharpness and detail derived from the panchromatic band. This approach is straightforward and computationally efficient, making it widely used for applications requiring enhanced visuals (ArcGIS n.d.).
On the other hand, Gram-Schmidt pansharpening is a more complex method that models the panchromatic band as a linear combination of the multispectral bands. This technique simulates a panchromatic band from the spectral information of the multispectral image, aligning it with the actual high-resolution panchromatic data. The simulated and actual data are then fused to create a highly accurate pansharpened image. Gram-Schmidt pansharpening is particularly effective in applications where preserving the spectral integrity of the multispectral data is critical, such as in scientific studies of vegetation health or water quality (ArcGIS n.d.).