Vegetation Cover Change Detection via NDVI Trend Analysis on GEE

A Google Earth Engine script for detecting multi-decadal vegetation change using Landsat Collection 2 time series.

Quick Start

1. Open Google Earth Engine Code Editor
2. Copy ndvi_threshold_change.js contents
3. Click Run
4. Click anywhere on the map to inspect points

Features

• 8 Change Classes: Loss, Degradation, Emerging, Maturation, Densification, Establishment, Sparse Accumulation, and Transitional Accumulation
• Dynamic Configuration: Easily adjustable analysis period (Year Start/End) and thresholds
• Hemispheric/Seasonal Adaptation: Configurable START/END months for Northern/Southern hemisphere analysis
• Sensitivity Analysis: Built-in parameter to test threshold stability
• Multi-scale Trend Analysis: Compares long-term trend with recent short-term trend
• Trend Acceleration: Identifies accelerating vs decelerating growth in the Inspector
• Statistical Significance: Trends filtered by Mann-Kendall test (p < 0.05) to reject noise.
• Collection 2 Natively: Uses USGS Collection 2 Level-2 Surface Reflectance directly without legacy harmonization.
• Dynamic Legend: Updates based on active layer and time configuration
• Point Inspector: NDVI, trend, classification, and projection
• Epoch Tracking: When areas first reached dense canopy (dynamically generated epochs)
• Trajectory Projection: Estimated year to reach dense canopy

Output Products

The script generates five distinct map layers and one interactive chart:

Layer Name	Description
Vegetation Change	Primary 8-class thematic map (see Classification below).
Canopy Gain Epoch	Year of first detection for "Establishment" and "Maturation" classes (5-year cohorts).
Years to Dense Canopy	Projected recovery time based on current linear trends (theoretical).
Statistical Trends	Binary mask showing areas with significant Mann-Kendall trends (p < 0.05).
Trend Magnitude	Slope values (NDVI/year) for significant pixels.

Interactive Charts

Clicking on the map generates a 40-year NDVI time series chart for that pixel, overlaying the linear trend line and statistical significance.

Classification

NDVI Class	Range	Trend Class	Slope
Dense	≥ 0.6	Gaining	> +0.005/yr
Transitional	0.4-0.6	Stable	±0.005/yr
Sparse	0.2-0.4	Losing	< -0.005/yr
Bare	< 0.2

Limitations

• Thresholds are approximate: Optimal values vary by region and ecosystem
• Linearity Assumption: The "Years to Dense Canopy" projection is a theoretical signal, not an ecological prediction.
• Validation Status (Visual Only): This tool is experimental. Accuracy has been assessed visually but not quantitatively.
• Sensor Homogeneity: Minor spectral differences (TM vs OLI) are uncorrected but deemed acceptable for Collection 2.
• 30m resolution: May not capture fine-scale patterns

See docs/methodology.pdf for documentation, limitations, and references.

Data Sources

• Landsat 5/7/8/9 Surface Reflectance (Collection 2, Tier 1)
• 1985–2025 analysis period

Citation

If you use this software in your research, please cite:

APA:

Pizzi, G. (2025). Vegetation Cover Change Detection via NDVI Trend Analysis on GEE (Version 2.0.1) [Computer software]. https://doi.org/10.5281/zenodo.17831390
https://github.com/gbrlpzz/ndvi-vegetation-cover-change

BibTeX:

@software{pizzi2025vegetation,
  author       = {Pizzi, Gabriele},
  title        = {Vegetation Cover Change Detection via NDVI Trend Analysis on GEE},
  year         = 2025,
  version      = {2.0.1},
  doi          = {10.5281/zenodo.17831390},
  url          = {https://github.com/gbrlpzz/ndvi-vegetation-cover-change}
}

For the methodological documentation, see docs/methodology.pdf.

License

Apache License 2.0

Contact

Gabriele Pizzi | [email protected]