A Practical Method for Red-Edge Band Reconstruction for Landsat Image by Synergizing Sentinel-2 Data with Machine...

Area of Science:

• Remote Sensing
• Earth Observation
• Geospatial Analysis

Background:

• Red-edge spectral bands are crucial for accurate vegetation monitoring using multispectral remote sensing.
• Landsat OLI data lacks red-edge bands, limiting its application in detailed vegetation health assessments.
• Existing Landsat data is widely used, making its spectral enhancement highly valuable for long-term ecological studies.

Purpose of the Study:

• To develop and validate an innovative method for reconstructing red-edge bands for Landsat OLI.
• To enhance the spectral resolution of Landsat data for improved vegetation monitoring.
• To assess the feasibility of extending this method to historical Landsat TM/ETM+ data.

Main Methods:

• Investigated consistency between Landsat OLI and Sentinel-2 MSI bands using various resampling and atmospheric correction techniques.
• Employed machine learning algorithms (ridge regression, GBRT, random forest) to model and reconstruct red-edge bands.
• Validated reconstructed bands and derived vegetation indices against Sentinel-2 MSI data.

Main Results:

• Bilinear interpolation and LaSRC atmospheric correction yielded high band consistency (R² > 0.88).
• The Gradient Boosted Regression Tree (GBRT) algorithm accurately reconstructed three OLI red-edge bands (R² > 0.96, RMSE < 0.0122).
• Reconstructed Landsat red-edge indices showed strong agreement with Sentinel-2 indices (R²: 0.78–0.95).

Conclusions:

• The proposed method effectively extends the spectral domain of Landsat OLI, significantly enhancing its utility for vegetation remote sensing.
• This approach offers a valuable tool for improving regional and global vegetation monitoring using historical and current Landsat data.
• Provides a pathway for enhancing historical Landsat TM/ETM+ data for improved time-series vegetation analysis.