Remote Sensing and Image Interpretation

History, principles, electromagnetic spectrum

Remote sensing is the acquisition of information about objects or areas from a distance, typically from aircraft or satellites. The foundation of all remote sensing is the electromagnetic (EM) spectrum — the continuum of energy from gamma rays to radio waves. Different portions of the spectrum interact uniquely with Earth's surface materials, creating spectral signatures that enable identification and analysis.

The electromagnetic spectrum spans wavelengths from less than 10⁻¹² m (gamma rays) to greater than 10³ m (radio waves). Remote sensing primarily uses: visible light (0.4-0.7 μm) — what human eyes see; near-infrared (0.7-1.3 μm) — sensitive to vegetation health; shortwave infrared (1.3-3 μm) — distinguishes moisture content and minerals; thermal infrared (3-14 μm) — measures emitted temperature; and microwave (1 mm-1 m) — penetrates clouds and vegetation.

The interaction of EM energy with matter follows three processes: absorption, transmission, and reflection. Every material has a unique spectral reflectance curve — healthy vegetation reflects strongly in NIR (the "red edge"), water absorbs in NIR, and bare soil has a gradual increase across the visible spectrum. These differences form the basis for classification, vegetation monitoring, and mineral exploration.

Remote sensing systems are characterized by four types of resolution: spatial (pixel size on the ground), spectral (number and width of spectral bands), temporal (revisit frequency), and radiometric (sensitivity to energy differences, measured in bits). Higher resolution in one dimension often comes at the cost of another — Landsat 8 offers 30m spatial resolution with 16-day revisit, while commercial systems like WorldView achieve 0.3m resolution but with less frequent coverage.

• The EM spectrum is the foundation — different wavelengths reveal different surface properties
• Spectral signatures enable material identification and classification
• Four resolution types (spatial, spectral, temporal, radiometric) define sensor capabilities
• Every remote sensing application exploits specific wavelength-material interactions

• Plot spectral reflectance curves for vegetation, water, soil, and urban surfaces
• Compare Landsat, Sentinel-2, and MODIS in terms of all four resolution types
• Identify the optimal spectral bands for distinguishing water from vegetation

Cameras, film, aerial photography

Aerial photography, the oldest form of remote sensing, remains relevant for high-resolution mapping. Modern digital cameras have replaced film but the geometric and radiometric principles remain the same. Understanding camera optics, focal length, and ground coverage is essential for planning aerial surveys.

The geometry of vertical aerial photographs follows the central projection model — rays of light pass through the camera lens to form an inverted image on the focal plane. Scale varies across the photograph due to terrain relief and camera tilt. The scale relationship S = f/H (focal length / flying height) applies only at the center of a truly vertical photograph over flat terrain.

• Aerial photography provides the highest spatial detail for local mapping
• Central projection geometry causes scale variation across photographs
• Digital sensors have replaced film but geometric principles remain unchanged

Stereo viewing, parallax, measurements

Photogrammetry extracts 3D measurements from 2D photographs using overlapping stereo pairs. When two photographs of the same area are taken from different positions, the apparent shift of objects (parallax) is proportional to their height above the reference surface. This principle enables extraction of digital elevation models, building heights, and precise measurements from aerial imagery.

Modern digital photogrammetry uses Structure from Motion (SfM) algorithms that automatically identify matching features in overlapping images, compute camera positions, and generate dense 3D point clouds. This technology has been democratized by drone platforms — a consumer drone with a standard camera can now produce centimeter-accurate 3D models using software like Agisoft Metashape, Pix4D, or the open-source OpenDroneMap.

• Parallax in stereo pairs enables 3D measurement from 2D images
• Structure from Motion has democratized 3D mapping via drones
• Photogrammetric products include DEMs, orthomosaics, and 3D models

Sensor systems, Landsat, SPOT, MODIS

Multispectral scanners collect data in multiple discrete spectral bands simultaneously, enabling spectral analysis impossible with panchromatic photography. The Landsat program, initiated in 1972, provides the longest continuous satellite observation record of Earth's surface. Landsat 8/9's OLI sensor collects 11 bands from coastal aerosol (0.43 μm) through thermal infrared (12.5 μm) at 15-30m resolution.

The European Space Agency's Sentinel-2 constellation (two satellites) provides 13 spectral bands at 10-60m resolution with a 5-day revisit time. Its free, open data policy has revolutionized earth observation by making high-quality multispectral data accessible to everyone. Sentinel-2 bands include three red-edge bands (centered at 705, 740, and 783 nm) specifically designed for vegetation analysis — a major advantage over Landsat.

MODIS (Moderate Resolution Imaging Spectroradiometer) on NASA's Terra and Aqua satellites provides daily global coverage in 36 spectral bands at 250m-1km resolution. While its spatial resolution is coarse compared to Landsat, MODIS's temporal resolution enables monitoring of rapidly changing phenomena: wildfire progression, algal blooms, snow cover dynamics, and vegetation phenology.

• Landsat provides 50+ years of continuous Earth observation data
• Sentinel-2's free data and red-edge bands have transformed vegetation monitoring
• MODIS enables daily global monitoring despite lower spatial resolution
• Spectral band selection determines what phenomena can be detected

Thermal properties, urban heat islands

Thermal infrared (TIR) remote sensing detects emitted radiation rather than reflected radiation, enabling temperature mapping of Earth's surface. All objects above absolute zero emit thermal radiation, with the intensity and spectral distribution governed by their temperature and emissivity. Kirchhoff's law states that good absorbers are good emitters — blackbodies have emissivity of 1.0, while real surfaces range from 0.85-0.99.

Urban Heat Island (UHI) studies are a major application of thermal RS. Cities can be 2-8°C warmer than surrounding rural areas due to impervious surfaces absorbing solar radiation, reduced evapotranspiration from vegetation loss, waste heat from buildings and vehicles, and canyon effects between tall buildings. Landsat thermal band data at 100m resolution enables mapping of surface temperature patterns across urban landscapes.

• Thermal RS measures emitted radiation, revealing surface temperature patterns
• Urban Heat Islands demonstrate the environmental impact of urbanization
• Emissivity must be accounted for to derive accurate surface temperatures

SAR, interferometry, polarimetry

Synthetic Aperture Radar (SAR) is an active remote sensing system that transmits microwave pulses and records the returned signal. SAR's key advantage is all-weather, day-and-night operation — microwaves penetrate clouds, smoke, and light rain. This makes SAR indispensable for monitoring tropical regions (where persistent cloud cover limits optical sensing) and for disaster response (when timely information is critical).

SAR interferometry (InSAR) compares the phase of radar signals from two or more acquisitions to measure surface deformation with millimeter precision. Applications include monitoring volcanic inflation, earthquake-induced ground displacement, land subsidence from groundwater extraction, and glacier flow velocities. Differential InSAR (DInSAR) has revealed that many Asian megacities are sinking at alarming rates — Jakarta, for example, subsides up to 25 cm per year.

Polarimetric SAR transmits and receives in multiple polarization states (HH, VV, HV, VH), providing information about the physical structure and orientation of scattering elements. Quad-pol SAR data enables decomposition into surface scattering (bare soil, water), volume scattering (vegetation canopy), and double-bounce scattering (buildings, tree trunks standing in water). This capability is particularly valuable for flood mapping — distinguishing open water, flooded vegetation, and urban flooding.

For Bangladesh, SAR is critically important for flood monitoring during the monsoon season when cloud cover makes optical imagery unusable for months. Sentinel-1 SAR data, available every 6-12 days, enables near-real-time flood extent mapping. The Flood Forecasting and Warning Centre (FFWC) increasingly incorporates SAR-derived flood maps into its operational workflow.

• SAR operates through clouds, enabling monitoring in persistently cloudy regions
• InSAR measures surface deformation with millimeter precision
• Polarimetric SAR reveals physical structure of ground features
• SAR is essential for flood monitoring in monsoon-affected countries like Bangladesh

• Process a Sentinel-1 SAR image in GEE to map flood extent
• Compare pre- and post-flood SAR images using change detection
• Visualize different polarization channels (VV, VH) for a forested area

Enhancement, filtering, classification

Digital image processing transforms raw satellite data into interpretable information through a systematic workflow: preprocessing (radiometric and geometric correction), enhancement (improving visual interpretability), classification (assigning pixels to categories), and accuracy assessment (validating results).

Radiometric correction removes sensor artifacts, atmospheric effects, and illumination variations. Top-of-atmosphere (TOA) reflectance correction accounts for solar geometry and sensor calibration. Surface reflectance correction removes atmospheric scattering and absorption, enabling comparison of images from different dates, sensors, and atmospheric conditions. Atmospheric correction is essential for time series analysis and change detection.

Image classification assigns each pixel to a thematic category (water, forest, urban, agriculture). Supervised classification requires training samples collected from known locations — the analyst identifies examples of each class, and an algorithm (Maximum Likelihood, Random Forest, Support Vector Machine) learns the spectral patterns and applies them to unclassified pixels. Unsupervised classification (K-means, ISODATA) groups spectrally similar pixels without training data, requiring the analyst to interpret the resulting clusters.

Accuracy assessment quantifies classification reliability using an error matrix (confusion matrix). Overall accuracy, producer's accuracy (errors of omission), user's accuracy (errors of commission), and the Kappa statistic are standard metrics. A minimum of 50 reference samples per class is recommended, collected independently of training data. An overall accuracy of 85% or higher is generally considered acceptable for published land cover maps.

• Image processing follows a systematic pipeline from correction to classification
• Atmospheric correction is essential for multi-temporal analysis
• Classification accuracy must be rigorously assessed with independent reference data
• Minimum 85% overall accuracy is the standard for published maps

• Apply atmospheric correction to a Landsat scene and compare TOA vs surface reflectance
• Classify a Sentinel-2 image into 5 land cover classes using Random Forest
• Build an error matrix and calculate overall accuracy, Kappa, and per-class metrics

Agriculture, forestry, urban, water resources

Remote sensing applications have expanded dramatically with increasing data availability, processing power, and algorithm sophistication. In agriculture, NDVI time series track crop growth stages, identify water stress, and estimate yields at field-to-national scales. The European MARS (Monitoring Agricultural ResourceS) system uses Sentinel data to forecast crop production across the EU.

In forestry, remote sensing enables wall-to-wall forest inventory at landscape scales. Spectral indices detect deforestation, LiDAR measures canopy height and biomass, and radar penetrates canopy to characterize forest structure. Global Forest Watch uses Landsat time series to detect forest loss within days of occurrence, providing near-real-time alerts to conservation organizations.

Urban remote sensing monitors the physical expansion of cities, characterizes urban form, and quantifies environmental impacts. Night-time lights from VIIRS/DMSP satellites serve as proxies for economic activity. High-resolution imagery enables building extraction, impervious surface mapping, and urban green space inventory. For rapidly urbanizing cities in South Asia, remote sensing provides the only cost-effective means of tracking spatial growth.

Water resources management uses remote sensing for surface water monitoring (optical + SAR), groundwater depletion assessment (GRACE satellite gravity measurements), snow cover mapping (MODIS), and water quality estimation (chlorophyll-a, turbidity, suspended sediment concentration from ocean color sensors).

• Each domain exploits specific remote sensing capabilities and data sources
• Time series analysis enables monitoring of dynamic processes
• Integration of multiple sensor types provides comprehensive environmental assessment
• Remote sensing is often the only feasible approach for large-scale monitoring