Change Detection
The process of identifying differences in the state of an area by comparing satellite images from different dates. Used to monitor deforestation, urban growth, flood extent, glacier retreat, and other environmental changes over time.
Overview
Change detection identifies differences in the state of a landscape by analyzing images acquired at different times. It is one of the most valuable EO capabilities — satellites provide consistent, repeatable coverage that no other monitoring system can match. The increasing availability of dense time series from Landsat (50+ years) and Sentinel-2 (5-day revisit) has transformed change detection from bi-temporal comparison into continuous monitoring.
How It Works
Methods group into three categories. Image algebra operates directly on pixel values — subtracting one date's index from another and thresholding. Post-classification comparison independently classifies each date then compares maps pixel by pixel, producing from-to change matrices. Time-series methods like BFAST and CCDC analyze the full temporal trajectory, distinguishing abrupt changes (fire, clearing) from gradual ones (degradation).
Preprocessing is critical: images must be geometrically co-registered to sub-pixel accuracy and atmospherically corrected to surface reflectance.
Key Facts
- Radiometric consistency between dates is the single most important factor for reliable change detection.
- The Landsat archive provides over 50 years of consistent 30m imagery for long-term change analysis.
- CCDC detects both timing and type of change by fitting harmonic models to every pixel's time series.
- Sub-pixel co-registration (better than 0.5 pixel RMSE) is required to avoid false change signals.
- Change vector analysis captures both magnitude and spectral direction of change.
Applications
Deforestation Monitoring
Systems like DETER, PRODES, and Global Forest Watch detect forest loss at 30m resolution globally.
Urban Growth Analysis
Multi-decadal change detection from Landsat archives reveals spatial patterns of urban expansion.
Disaster Damage Assessment
Rapid pre/post-event comparison enables damage mapping within hours to days.
Agricultural Monitoring
Detecting changes in crop condition through growing-season time series for yield forecasting and drought warning.
Limitations & Considerations
Seasonal vegetation cycles can produce spectral changes larger than actual land cover transitions. Different sensors have different spectral response functions that must be harmonized. Cloud cover creates data gaps. Gradual changes may fall below detection thresholds. The choice of change threshold is inherently subjective.
History & Background
Digital change detection emerged with Landsat 1 in 1972. Bi-temporal methods were systematically developed in the 1980s-90s. The free Landsat archive (2008) enabled pixel-level time-series analysis. Algorithms like LandTrendr (2010), BFAST (2010), and CCDC (2014) exploit dense temporal records. Google Earth Engine made planetary-scale change detection computationally accessible.
Analyze Change Detection data with LYRASENSE
Use our agentic notebook environment to work with satellite data and compute indices like Change Detection — no setup required.