Remote Sensing Image Analysis

Satellite and airborne image processing — classification, spectral indices, change detection, and atmospheric correction for Earth observation

10m
Sentinel-2
Resolution
13
Spectral Bands
(Sentinel-2)
16-day
Landsat Revisit
Cycle
50+
Years of Landsat
Archive

Overview

Remote sensing image analysis transforms raw satellite radiance data into meaningful geospatial information about Earth's surface. The workflow encompasses radiometric calibration, atmospheric correction (converting top-of-atmosphere to surface reflectance), spectral index computation (NDVI, NDWI, MNDWI), image classification (supervised/unsupervised), and change detection. For coastal and ocean applications, these techniques map shoreline change, mangrove extent, water turbidity, coral reef health, aquaculture expansion, and land use/cover dynamics. Modern approaches increasingly integrate deep learning (U-Net, DeepLab) and cloud platforms (Google Earth Engine).

Key Concepts

Atmospheric Correction

Converting TOA radiance to surface reflectance by removing atmospheric scattering/absorption effects. Methods range from dark object subtraction (DOS) to radiative transfer models (6S, MODTRAN). Essential for multi-temporal analysis and accurate spectral indices.

ρ_surface = (L_TOA - L_path) × π / (E_sun × cos θ × T↓ × T↑)

Spectral Indices

Band ratio indices enhance specific features: NDVI = (NIR-Red)/(NIR+Red) for vegetation; NDWI = (Green-NIR)/(Green+NIR) for water; MNDWI = (Green-SWIR)/(Green+SWIR) for urban-water discrimination; FAI for floating algae.

Image Classification

Supervised (Random Forest, SVM, MLC) and unsupervised (K-means, ISODATA) methods assign land cover classes. Object-based image analysis (OBIA) segments pixels into objects before classification. Accuracy assessed via confusion matrix, kappa, F1-score.

Change Detection

Post-classification comparison, image differencing, PCA-based change, and spectral change vectors detect LULC transitions. Multi-temporal composites (BFAST, LandTrendr) detect gradual and abrupt changes in long time series.

Deep Learning for RS

CNNs (U-Net, ResNet, DeepLabV3+) achieve state-of-the-art classification and segmentation. Transfer learning from ImageNet to RS imagery. Semantic segmentation maps every pixel; instance segmentation identifies individual objects (ships, buildings).

Cloud Computing (GEE)

Google Earth Engine provides petabyte-scale access to satellite archives with server-side processing. Enables continental and global analyses without downloading data. Python and JavaScript APIs support reproducible workflows.

Common Sensors

SensorResolutionBandsRevisitBest For
Landsat 8/9 (OLI)30m (15m pan)118 days (combined)LULC, long-term change
Sentinel-2 (MSI)10-60m135 daysVegetation, water, urban
MODIS (Terra/Aqua)250-1000m36DailyOcean color, fire, global
Sentinel-1 (SAR)5-20mC-band6 daysFlood mapping, all-weather
Planet (Dove)3-5m4-8DailyHigh-frequency monitoring
WorldView-30.31m (pan)291 dayDetailed mapping

Interactive Visualizations

Spectral Signatures of Common Land Cover Types

Classification Accuracy — Confusion Matrix Heatmap

NDVI Time Series — Mangrove vs. Agriculture vs. Urban

Key References

  1. Lillesand, T.M., Kiefer, R.W. & Chipman, J.W. (2015). Remote Sensing and Image Interpretation, 7th Ed. Wiley.
  2. Gorelick, N. et al. (2017). Google Earth Engine: planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202, 18–27.
  3. Zhu, Z. & Woodcock, C.E. (2014). Continuous change detection and classification of land cover using all available Landsat data. Remote Sensing of Environment, 144, 152–171.
  4. Ma, L. et al. (2019). Deep learning in remote sensing applications: A meta-analysis and review. ISPRS J. Photogrammetry and Remote Sensing, 152, 166–177.
  5. McFeeters, S.K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. Int. J. Remote Sensing, 17(7), 1425–1432.