Skip to content

Chapter 17: Earth Observation and Remote Sensing

Chapter Overview Earth observation (EO) from space has transformed our ability to monitor and understand the planet at global scale with consistent, repeatable measurements. This chapter provides a comprehensive treatment of the physical principles underlying remote sensing, the design and characterization of optical and synthetic aperture radar (SAR) sensors, interferometric and polarimetric analysis techniques, hyperspectral and thermal imaging, and the rapidly growing commercial EO sector. Algorithms for geophysical parameter retrieval, artificial intelligence applications, and major archival data systems are also addressed.

17.1 Fundamentals of Remote Sensing Physics

Remote sensing is the science of acquiring information about an object or phenomenon without direct physical contact, exploiting the interaction of electromagnetic radiation (EMR) with matter. From a satellite platform, the atmosphere, land surface, ocean, and cryosphere all interact with incoming solar radiation and emit their own thermal radiation in ways that carry unique geophysical signatures.

17.1.1 Electromagnetic Radiation and the Solar Spectrum

The Sun approximates a blackbody at \(T_\odot \approx 5778\) K. The spectral radiance of a blackbody is described by the Planck function:

\[B_\lambda(T) = \frac{2hc^2}{\lambda^5} \frac{1}{\exp\!\left(\dfrac{hc}{\lambda k_B T}\right) - 1}\tag{17.1}\]

where \(h = 6.626 \times 10^{-34}\) J s is Planck’s constant, \(c = 2.998 \times 10^8\) m s\(^{-1}\) is the speed of light, \(k_B = 1.381 \times 10^{-23}\) J K\(^{-1}\) is the Boltzmann constant, \(\lambda\) is wavelength, and \(T\) is absolute temperature. The wavelength of peak emission is given by Wien’s displacement law:

\[\lambda_{\max} \cdot T = 2897.8\ \mu\text{m}\cdot\text{K}\tag{17.2}\]

For the Sun, \(\lambda_{\max} \approx 0.50\,\mu\)m (visible green), whereas for the Earth at \(T \approx 288\) K, \(\lambda_{\max} \approx 10\,\mu\)m (thermal infrared). This spectral separation forms the basis for distinguishing reflected solar radiation from thermally emitted radiation.

17.1.2 Surface Reflectance and Emittance

The bidirectional reflectance distribution function (BRDF) \(f_r(\theta_i, \phi_i; \theta_r, \phi_r)\) describes how surface reflectance varies with illumination and viewing geometry:

\[f_r(\theta_i, \phi_i; \theta_r, \phi_r) = \frac{dL_r(\theta_r, \phi_r)}{dE_i(\theta_i, \phi_i)}\tag{17.3}\]

where \(L_r\) is the reflected radiance and \(E_i\) is the incident irradiance. For operational remote sensing, the hemispherical-directional reflectance factor (HDRF) and bidirectional reflectance factor (BRF) are commonly used (Schaepman-Strub, G., Schaepman, M.E., Painter, T.H., Dangel, S., Martonchik, J.V. 2006). Surface reflectance \(\rho_s\) is retrieved from top-of-atmosphere (TOA) radiance after atmospheric correction.

The surface energy balance links emittance to surface temperature via Kirchhoff’s law: at thermal equilibrium, emissivity \(\varepsilon_\lambda\) equals absorptivity \(\alpha_\lambda\):

\[L_\lambda^{\text{emit}} = \varepsilon_\lambda \cdot B_\lambda(T_s)\tag{17.4}\]

For most natural surfaces, \(\varepsilon_\lambda \approx 0.95\)–\(0.99\) in the 8–14 \(\mu\)m window.

17.2 Spatial Resolution and Sensor Design

17.2.1 Ground Sampling Distance

The ground sampling distance (GSD), which defines the smallest resolvable spatial element on the ground, is given by:

\[\text{GSD} = \frac{p \cdot H}{f}\tag{17.5}\]

where \(p\) is the detector element (pixel) size, \(H\) is the orbital altitude, and \(f\) is the effective focal length of the optical system. For a typical commercial satellite at \(H = 500\) km with \(p = 8\,\mu\)m and \(f = 12.9\) m (e.g., WorldView-3), GSD \(\approx 0.31\) m.

17.2.2 Diffraction Limit: Rayleigh Criterion

The angular resolution of a circular aperture of diameter \(D\) is limited by diffraction to:

\[\theta_{\min} = 1.22 \frac{\lambda}{D}\tag{17.6}\]

The corresponding ground resolution at nadir is \(\Delta x = \theta_{\min} \cdot H = 1.22 \lambda H / D\). For \(\lambda = 0.55\,\mu\)m, \(H = 500\) km, and \(D = 1.1\) m, \(\Delta x \approx 0.30\) m, consistent with the optical GSD of sub-metre commercial sensors.

17.2.3 Nyquist Sampling and MTF

To avoid aliasing, the Nyquist criterion requires that the detector sampling frequency exceed twice the optical cutoff frequency. The system modulation transfer function (MTF) combines contributions from the optics, detector, electronics, and platform motion:

\[\text{MTF}_{\text{sys}}(f) = \text{MTF}_{\text{opt}}(f) \cdot \text{MTF}_{\text{det}}(f) \cdot \text{MTF}_{\text{elec}}(f) \cdot \text{MTF}_{\text{motion}}(f)\tag{17.7}\]

A minimum MTF of 0.20 at Nyquist is a common specification for imaging satellites.

17.3 Radiometric Calibration

17.3.1 Digital Number to Radiance Conversion

Raw sensor output is expressed in digital numbers (DN). Conversion to physical radiance units follows:

\[L = G \cdot \text{DN} + B\tag{17.8}\]

where \(G\) is the absolute gain (W m\(^{-2}\) sr\(^{-1}\) \(\mu\)m\(^{-1}\) DN\(^{-1}\)) and \(B\) is the offset, determined from pre-launch and in-orbit calibration. The signal-to-noise ratio (SNR) at reference radiance \(L_{\text{ref}}\) is:

\[\text{SNR} = \frac{L_{\text{ref}}}{L_{\text{NE}\Delta L}}\tag{17.9}\]

where \(L_{\text{NE}\Delta L}\) is the noise-equivalent change in radiance. Typical specifications: Landsat 9 OLI SNR \(>\) 200:1 at reference radiance; WorldView-3 SNR \(\approx\) 100:1.

17.3.2 Calibration Methods

Pre-launch calibration uses integrating spheres and flat-field sources to characterize gain, offset, linearity, and inter-channel uniformity. Vicarious calibration uses ground truth measurements at pseudo-invariant calibration sites (PICS) such as Railroad Valley Playa (Nevada, USA), La Crau (France), and Algodones Dunes (California) (Thome, K.J. 2001). Cross-calibration exploits simultaneous or near-simultaneous imaging of the same target by two sensors in compatible spectral bands. Solar diffuser on-board calibration (e.g., Landsat 9 OLI, Sentinel-2 MSI) illuminates the detector array with a known solar-illuminated panel.

17.4 Major Optical Earth Observation Sensors

Table 17.1. Comparison of major optical Earth observation satellites (2024).
Satellite Operator GSD (m) Swath (km) Bands Revisit
Landsat 9 OLI USGS/NASA 30 (15 pan) 185 9 VNIR/SWIR 16 days
Sentinel-2A/B MSI ESA 10/20/60 290 13 5 days
Planet Dove Planet 3–5 24 4 (8 SuperDove) Daily
WorldView-3 Maxar 0.31 pan, 1.24 multi 13.1 29 1 day
Pleiades 1A/1B Airbus DS 0.5 pan, 2 multi 20 5 1 day
Gaofen-7 CNSA 0.65 stereo 20 2 panchromatic 5 days
SPOT-6/7 Airbus DS 1.5 pan, 6 multi 60 5 1 day

17.4.1 Landsat 9: OLI-2 and TIRS-2

Landsat 9, launched in September 2021, carries the Operational Land Imager-2 (OLI-2) and Thermal Infrared Sensor-2 (TIRS-2). OLI-2 covers 9 spectral bands (0.43–2.29 \(\mu\)m) at 30 m GSD with a 185 km swath, and includes a panchromatic band at 15 m GSD. TIRS-2 provides two thermal bands (10.6 \(\mu\)m and 12.0 \(\mu\)m) at 100 m resolution. The 16-day repeat cycle is halved to 8 days when combined with Landsat 8. The archive, extending back to 1972 with earlier Landsat missions, provides an unparalleled 50+ year record of land surface change (Wulder, M.A., Roy, D.P., Radeloff, V.C., et al. 2022).

17.4.2 Sentinel-2 MSI

The European Space Agency Sentinel-2 mission comprises two satellites (2A launched 2015, 2B launched 2017), each carrying a Multi-Spectral Instrument (MSI) with 13 bands ranging from 443 nm to 2190 nm. Spatial resolutions of 10 m (visible and near-infrared), 20 m (red-edge and SWIR), and 60 m (atmospheric correction bands) are achieved with a 290 km swath. With 2A and 2B combined, the revisit time at mid-latitudes is 5 days. Open data policy and high spatial/temporal resolution have made Sentinel-2 the workhorse of operational land monitoring in Europe (Drusch, M., Del Bello, U., Carlier, S., et al. 2012).

17.4.3 Planet Dove and SkySat

Planet Labs operates the world’s largest constellation of Earth observation satellites, with over 200 Dove 3U CubeSats (each \(\sim\)5 kg) providing daily global coverage at 3–5 m GSD. Each Dove carries a 4-band (RGB + NIR) imager; newer SuperDove satellites extend to 8 bands. Planet’s SkySat constellation (21 satellites, \(\sim\)100 kg each) provides 50 cm pan, 1 m multispectral imagery with same-day revisit over major cities (Safyan, M. 2020).

17.4.4 Commercial Very-High-Resolution (VHR) Sensors

WorldView-3 (Maxar, 2014) achieves 0.31 m panchromatic GSD and 1.24 m in 8 multispectral bands at 617 km altitude, with 16 SWIR bands (3.7 m) and 12 CAVIS atmospheric compensation bands. Pleiades HR 1A and 1B (Airbus DS) offer 0.5 m pan and 2 m multi with a 20 km swath, enabling stereo DEMs at 1 m vertical accuracy. Gaofen-7 (CNSA, 2019) delivers 0.65 m stereo optical imagery combined with a laser altimeter, enabling 1:10,000 scale topographic mapping.

17.5 Synthetic Aperture Radar: Fundamentals

17.5.1 Radar Equation

The received power for a monostatic radar (transmitter and receiver co-located) is given by the radar range equation:

\[P_r = \frac{P_t G_t G_r \lambda^2 \sigma}{(4\pi)^3 R^4}\tag{17.10}\]

where \(P_t\) is transmitted power, \(G_t\) and \(G_r\) are transmit and receive antenna gains, \(\lambda\) is wavelength, \(\sigma\) is target radar cross-section (RCS) in m\(^2\), and \(R\) is slant range. For distributed targets (extended scenes), the RCS is expressed via the backscattering coefficient:

\[\sigma^0 = \frac{\sigma}{A} \quad [\text{m}^2/\text{m}^2]\tag{17.11}\]

In decibels: \(\sigma^0_{\text{dB}} = 10 \log_{10}(\sigma^0)\). Typical values range from \(-30\) dB (calm sea) to \(+5\) dB (urban corner reflectors).

17.5.2 SAR Image Formation: Range and Azimuth Processing

A SAR transmits linear frequency-modulated (LFM) chirp pulses with bandwidth \(B_r\) over pulse duration \(\tau_p\). The chirp rate is \(k_r = B_r / \tau_p\). Pulse compression (matched filtering) in range achieves range resolution:

\[\delta r = \frac{c}{2 B_r}\tag{17.12}\]

For Sentinel-1 IW mode with \(B_r = 56\) MHz, \(\delta r \approx 2.7\) m in slant range. In azimuth, the satellite motion synthesizes a long aperture of length \(L_a = \lambda R / D_a\) (where \(D_a\) is the physical antenna length). In stripmap operation, in which the antenna beam is held at a fixed pointing angle, this achieves the azimuth resolution:

\[\delta az = \frac{D_a}{2}\tag{17.13}\]

Notably, for stripmap SAR the azimuth resolution is independent of range and wavelength, determined solely by the physical antenna length (this does not hold for radar in general, nor for the scanning modes below). For Sentinel-1 with \(D_a = 12.3\) m the stripmap limit is \(\delta az \approx 6.1\) m; the wide-swath TOPS (IW) mode scans the beam in azimuth to cover a wider swath and therefore gives a coarser azimuth resolution of about 20 m.

17.5.3 Noise Equivalent Sigma Zero (NESZ)

The NESZ quantifies the minimum detectable \(\sigma^0\) and is the figure of merit for SAR sensitivity:

\[\text{NESZ} = \frac{(4\pi)^3 R^3 v_s k_B T_s F B_n}{P_t \bar{G}^2 \lambda^3 c_0 \tau_p f_{\text{PRF}} N_l}\tag{17.14}\]

where \(v_s\) is satellite velocity, \(T_s\) is system noise temperature, \(F\) is noise figure, \(B_n\) is noise bandwidth, \(f_{\text{PRF}}\) is pulse repetition frequency, and \(N_l\) accounts for system losses (Curlander and McDonough 1991). Typical NESZ values: Sentinel-1 IW \(\approx -22\) dB, ICEYE \(\approx -18\) dB.

17.6 SAR Imaging Modes

Table 17.2. SAR imaging modes and typical performance parameters.
Mode Swath (km) Az. Res. (m) Rg. Res. (m) Example
Stripmap 80–250 3–15 2–10 Sentinel-1 SM (5 imes 5)
ScanSAR 200–500 30–100 5–20 ALOS-2 WD1
TOPSAR (IW) 250 20 5 Sentinel-1 IW
Spotlight 5–10 0.25–1 0.25–1 TerraSAR-X SL
Sliding Spotlight 10–50 1–3 0.25–1 PAZ, COSMO-SkyMed-2

17.6.1 TOPSAR: Terrain Observation by Progressive Scan

TOPSAR, employed by Sentinel-1 in Interferometric Wide (IW) mode, is an evolution of ScanSAR that steers the beam electronically in azimuth from aft to fore during each burst. This progressive scan ensures uniform scene illumination and eliminates the scalloping artefact present in traditional ScanSAR. The IW mode acquires 3 sub-swaths (IW1, IW2, IW3), each with 9 burst images, achieving 250 km total swath at \(5 \times 20\) m spatial resolution in Single Look Complex (SLC) format (De Zan, F., Guarnieri, A.M. 2006).

17.6.2 Spotlight Mode

In spotlight mode, the antenna beam is steered to illuminate a fixed scene continuously as the satellite passes, increasing the synthetic aperture length and thus azimuth resolution. TerraSAR-X achieves \(\delta az = 0.25\) m and \(\delta rg = 0.25\) m in High Resolution Spotlight (HS) mode, with scene sizes of \(\sim5 \times 5\) km (Moreira et al. 2013). ICEYE X-series SAR CubeSats achieve 0.5 m Spotlight resolution from a \(\sim 100\) kg platform.

17.7 SAR Polarimetry

17.7.1 Scattering Matrix

A fully polarimetric (quad-pol) SAR measures the complete \(2\times 2\) complex scattering matrix:

\[\mathbf{S} = \begin{bmatrix} S_{HH} & S_{HV} \\ S_{VH} & S_{VV} \end{bmatrix}\tag{17.15}\]

where subscripts denote transmit/receive polarization (H: horizontal, V: vertical). For reciprocal media, \(S_{HV} = S_{VH}\). The Pauli decomposition expresses the scattering vector as:

\[\mathbf{k}_P = \frac{1}{\sqrt{2}} \begin{bmatrix} S_{HH} + S_{VV} \\ S_{HH} - S_{VV} \\ 2S_{HV} \end{bmatrix}\tag{17.16}\]

The three components correspond to odd-bounce (sphere/flat surface), even-bounce (dihedral), and volume (random) scattering mechanisms, which are directly mapped to RGB in Pauli false-colour composites.

17.7.2 Coherency Matrix and Decompositions

The \(3\times3\) Hermitian coherency matrix is \(\mathbf{T} = \langle \mathbf{k}_P \mathbf{k}_P^{*T} \rangle\) (ensemble-averaged). The Cloude–Pottier eigendecomposition (Cloude, S.R., Pottier, E. 1997) yields:

\[\mathbf{T} = \sum_{i=1}^{3} \lambda_i \mathbf{e}_i \mathbf{e}_i^{*T}\tag{17.17}\]

Key parameters extracted are:

  • Entropy: \(H = -\sum_{i=1}^{3} p_i \log_3 p_i\), \(p_i = \lambda_i / \sum_j \lambda_j\) (\(0 \le H \le 1\))

  • Anisotropy: \(A = (\lambda_2 - \lambda_3)/(\lambda_2 + \lambda_3)\)

  • Mean scattering angle: \(\bar{\alpha} = \sum_{i} p_i \alpha_i\)

The \(H/\alpha\) plane partitions targets into 8 scattering classes (surface, volume, double-bounce dominance regimes) with wide application in land-cover and sea-ice classification. The Freeman–Durden three-component decomposition separates \(\sigma^0\) into surface scattering (\(P_s\)), double-bounce (\(P_d\)), and volume (\(P_v\)) contributions (Freeman, A., Durden, S.L. 1998).

17.8 SAR Interferometry (InSAR)

17.8.1 Interferometric Phase

When two SAR images acquired from slightly different positions (or times) are combined, the interferometric phase contains information about path length difference \(\Delta R\):

\[\Delta\phi = \frac{4\pi}{\lambda} \Delta R\tag{17.18}\]

For Earth observation, \(\Delta R\) arises from topography (used to generate DEMs) and surface displacement (used in deformation mapping):

\[\Delta\phi_{\text{total}} = \underbrace{\frac{4\pi}{\lambda} \frac{B_\perp}{R\sin\theta} \Delta h}_{\text{topographic}} + \underbrace{\frac{4\pi}{\lambda} d_{\text{LOS}}}_{\text{displacement}} + \phi_{\text{atm}} + \phi_{\text{noise}}\tag{17.19}\]

where \(B_\perp\) is the perpendicular baseline, \(\theta\) is the look angle, \(\Delta h\) is height relative to the reference DEM, and \(d_{\text{LOS}}\) is displacement in the line-of-sight (LOS) direction. The height of ambiguity (one fringe = one \(2\pi\) phase cycle) is:

\[h_a = \frac{\lambda R \sin\theta}{2 B_\perp}\tag{17.20}\]

17.8.2 Coherence

Interferometric coherence is a measure of phase quality:

\[|\gamma| = \frac{\left|\langle s_1 s_2^* \rangle\right|}{\sqrt{\langle |s_1|^2 \rangle \langle |s_2|^2 \rangle}}, \quad 0 \le |\gamma| \le 1\tag{17.21}\]

Coherence degrades with temporal decorrelation (vegetation growth, snow, soil moisture change), geometric decorrelation (\(B_\perp > B_{\perp,\text{crit}}\)), volumetric decorrelation, and thermal noise. For reliable phase unwrapping, \(|\gamma| > 0.25\) is typically required.

17.8.3 Time-Series InSAR

Single-pair InSAR is limited by atmospheric phase noise (\(\sim 1\)–\(10\) cm of apparent displacement). Time-series methods exploit large stacks of interferograms to separate deformation from atmospheric signals and noise (Berardino, P., Fornaro, G., Lanari, R., Sansosti, E. 2002).

SBAS (Small Baseline Subsets): Forms interferograms with short spatial and temporal baselines to maximize coherence and solves for the displacement time-series using singular value decomposition (SVD):

\[\mathbf{B}\mathbf{v} = \boldsymbol{\Delta\phi}\tag{17.22}\]

where \(\mathbf{B}\) is the design matrix encoding temporal baselines and \(\mathbf{v}\) is the velocity vector.

PSInSAR (Permanent Scatterers): Identifies phase-stable scatterers (PS, typically buildings, rock outcrops) that maintain coherence over many years, enabling mm/year subsidence measurement in urban environments (Ferretti, A., Prati, C., Rocca, F. 2001). StaMPS and MintPy are widely used open-source implementations.

Applications of time-series InSAR include: urban subsidence (Mexico City \(>400\) mm/year), mining-induced deformation, earthquake cycle (interseismic, coseismic, postseismic), volcanic inflation/deflation (mm-cm scale), glacier flow velocity, and permafrost thaw settlement.

17.9 Hyperspectral Remote Sensing

Hyperspectral sensors acquire images in hundreds of contiguous narrow spectral bands (typical bandwidth 5–10 nm), enabling spectroscopic identification of surface materials. The data cube \(I(x, y, \lambda)\) contains full spectral information at each pixel.

Table 17.3. Major spaceborne hyperspectral sensors.
Sensor Platform Bands Range (nm) GSD (m) Swath (km)
PRISMA HYC ASI 237 400–2505 30 30
EnMAP VNIR+SWIR DLR 228 420–2450 30 30
EMIT NASA/ISS 384 380–2500 60 75
DESIS DLR/ISS 235 400–1000 30 30
HyperScout-2 ESA 45 400–1000 100 220

17.9.1 Spectral Unmixing

At 30 m spatial resolution, most pixels are mixed, containing contributions from multiple pure materials (endmembers). Linear mixing model:

\[\mathbf{x} = \mathbf{M}\boldsymbol{\alpha} + \mathbf{n}, \quad \sum_{k=1}^{K} \alpha_k = 1, \quad \alpha_k \ge 0\tag{17.23}\]

where \(\mathbf{x} \in \mathbb{R}^B\) is the measured spectrum, \(\mathbf{M} \in \mathbb{R}^{B\times K}\) is the endmember matrix, \(\boldsymbol{\alpha}\) are fractional abundances, and \(\mathbf{n}\) is noise. Endmember extraction algorithms include N-FINDR (maximum simplex volume), Vertex Component Analysis (VCA), and ICE (Iterated Constrained Endmembers). EMIT (Earth Surface Mineral Dust Source Investigation) aboard the ISS has mapped surface mineralogy globally, quantifying dust source regions for radiative forcing studies (Green, R.O., et al. 2022).

17.10 Thermal Remote Sensing

Thermal infrared (TIR) remote sensing exploits the 8–14 \(\mu\)m atmospheric window. Surface temperature retrieval requires separation of emissivity from temperature (the temperature/emissivity separation problem). The split-window algorithm exploits the differential absorption in two adjacent TIR channels (e.g., MODIS bands 31 and 32 at 11 \(\mu\)m and 12 \(\mu\)m):

\[T_s = a_0 + a_1 T_{11} + a_2 (T_{11} - T_{12}) + a_3 (T_{11} - T_{12})^2\tag{17.24}\]

where \(T_{11}\) and \(T_{12}\) are brightness temperatures at 11 \(\mu\)m and 12 \(\mu\)m, and \(a_0, a_1, a_2, a_3\) are regression coefficients dependent on emissivity and water vapour (Wan, Z., Dozier, J. 1996). The ECOSTRESS instrument on the ISS (5 TIR bands, 70 m GSD) provides evapotranspiration and plant temperature stress measurements at sub-field scale.

17.11 Geophysical Applications and Vegetation Indices

17.11.1 Vegetation Indices

Vegetation indices exploit the contrast between chlorophyll absorption in the red and the leaf scattering plateau in the near-infrared. The Normalized Difference Vegetation Index (NDVI) is:

\[\text{NDVI} = \frac{\rho_{\text{NIR}} - \rho_{\text{Red}}}{\rho_{\text{NIR}} + \rho_{\text{Red}}}, \quad -1 \le \text{NDVI} \le 1\tag{17.25}\]

The Enhanced Vegetation Index (EVI) reduces background and atmospheric sensitivity:

\[\text{EVI} = G \cdot \frac{\rho_{\text{NIR}} - \rho_{\text{Red}}}{\rho_{\text{NIR}} + C_1 \rho_{\text{Red}} - C_2 \rho_{\text{Blue}} + L}\tag{17.26}\]

with \(G = 2.5\), \(C_1 = 6\), \(C_2 = 7.5\), \(L = 1\). The Normalized Burn Ratio (NBR) for wildfire burned area mapping uses SWIR:

\[\text{NBR} = \frac{\rho_{\text{NIR}} - \rho_{\text{SWIR},2200}}{\rho_{\text{NIR}} + \rho_{\text{SWIR},2200}}\tag{17.27}\]

17.11.2 Ocean Color and Chlorophyll Retrieval

Ocean color remote sensing retrieves phytoplankton chlorophyll-\(a\) concentration from the water-leaving radiance ratio. The OC3M algorithm (MODIS operational) uses:

\[\log_{10}[\text{Chl-}a] = a_0 + \sum_{i=1}^{4} a_i \left(\log_{10} R_{rs}\right)^i\tag{17.28}\]

where \(R_{rs} = \max(R_{rs}(443), R_{rs}(488)) / R_{rs}(547)\) is the blue-to-green band ratio of remote sensing reflectance and \(a_i\) are polynomial coefficients (O’Reilly, J.E., et al. 1998). The PACE satellite (2024) with the OCI hyperspectral imager enables phytoplankton functional type (PFT) mapping.

17.12 Commercial Earth Observation Sector

The commercial EO sector has undergone a revolution driven by miniaturization, rideshare launch, and New Space economics. Key operators and their capabilities include:

Planet: 200+ Dove CubeSats provide daily global coverage at 3–5 m GSD; 21 SkySats at 50 cm. Planet Data API provides access to \(>10\) million scenes.

Maxar Technologies (WorldView constellation): VHR optical imagery at 0.3–0.5 m, licensed through US Government NextView/EnhancedView programs and commercial APIs (SecureWatch).

Airbus Defence & Space: Pleiades 1A/1B (0.5 m) and Pleiades Neo (0.3 m) for tasked VHR; SPOT-6/7 (1.5 m) for wide-area monitoring.

ICEYE: Finland-based SAR constellation of \(>30\) X-band SAR satellites (2019–2024), achieving 1 m Spotlight resolution from 100 kg microsatellites. All-weather, 24/7 capability; focus on maritime, flood, and change detection (Ignatenko, V., et al. 2020).

Capella Space: US-based X-band SAR constellation (0.5 m Spotlight); OMEGA mode for moving target indication.

BlackSky: Tasked multispectral constellation (0.9 m GSD) with analytics platform.

Satellogic: \(>30\) hyperspectral satellites at 1 m resolution.

17.13 Artificial Intelligence for Earth Observation

17.13.1 Deep Learning for Classification and Detection

Convolutional neural networks (CNNs) have become the dominant paradigm for EO image analysis. U-Net, originally developed for biomedical segmentation, is widely applied to land cover, building, and road segmentation:

\[\hat{\mathbf{y}} = f_\theta(\mathbf{X}), \quad \mathcal{L} = -\sum_i \left[y_i \log\hat{y}_i + (1-y_i)\log(1-\hat{y}_i)\right]\tag{17.29}\]

Object detection architectures (RetinaNet, YOLO, Faster R-CNN) are deployed for ship, aircraft, and vehicle detection in optical VHR imagery. Change detection using Siamese networks:

\[d(\mathbf{x}_1, \mathbf{x}_2) = \|f_\theta(\mathbf{x}_1) - f_\theta(\mathbf{x}_2)\|_2\tag{17.30}\]

where \(f_\theta\) is a shared encoder branch and \(d\) is the feature-space change magnitude.

17.13.2 Foundation Models for EO

The Prithvi geospatial foundation model (NASA + IBM, 2023) is a Vision Transformer (ViT) pre-trained on 1 million+ Harmonized Landsat Sentinel-2 (HLS) patches using masked autoencoding. SatCLIP and DeCUR apply contrastive learning for multi-modal (optical + SAR) representation alignment (Reed, C.J., et al. 2023). These models enable few-shot adaptation to new downstream tasks (e.g., flood mapping, crop type classification) with \(<100\) labelled examples.

17.14 Data Archives and Access

Table 17.4. Major Earth observation data archives and access mechanisms.
Archive Agency Access
Copernicus Open Access Hub ESA Free, OpenSearch API
USGS EarthExplorer USGS Free, REST API
NASA Earthdata NASA Free, DAAC portals
ESA EO Catalogue ESA Free, OGC API
Planet Data API Planet Commercial subscription
Maxar SecureWatch Maxar Commercial, NGA-licensed
JAXA EORC / G-Portal JAXA Free/registered

The Copernicus Data Space Ecosystem (2023) provides streamlined access to all Sentinel data with cloud-native OGC API, STAC (SpatioTemporal Asset Catalog) endpoints, and Jupyter notebook processing environments, enabling scalable analysis without bulk data transfer.

17.15 Summary

This chapter has surveyed the breadth of spaceborne Earth observation technology, from fundamental radiation physics to state-of-the-art AI-driven analysis pipelines. Key themes include:

  • Remote sensing retrieval begins with the Planck function and surface radiative transfer; calibration bridges raw DN values to physical radiance and ultimately to geophysical parameters.

  • Optical sensors have achieved sub-metre GSD commercially, with daily global revisit enabled by CubeSat constellations.

  • SAR provides all-weather, day-night imaging with capabilities ranging from 0.25 m Spotlight to 500 km ScanSAR swaths.

  • PolSAR and InSAR extract physical scattering mechanisms and millimetre-precision deformation fields.

  • Hyperspectral and thermal sensors enable material identification and temperature mapping at global scale.

  • Commercial New Space operators have democratized EO data access; AI/ML has transformed information extraction workflows.

  • Open data archives (ESA Copernicus, USGS EarthExplorer, NASA Earthdata) support scientific research and operational applications worldwide.