Regional scale terrace mapping in fragmented mountainous areas using multi-source remote sensing data and sample...

Area of Science:

• Earth Observation
• Agricultural Science
• Soil Science

Background:

• Terraces are crucial soil erosion control structures on hill-slopes, altering surface runoff and reducing nutrient loss.
• Accurate regional-scale terrace mapping is essential for effective soil conservation, sustainable agriculture, and ecological planning.
• Optical remote sensing alone is insufficient for terrace identification in mountainous regions due to weather-related data limitations.

Purpose of the Study:

• To develop a robust method for high-resolution (10m) terrace mapping in plateau mountainous regions.
• To improve terrace identification accuracy by integrating multi-spectral optical and SAR data with terrain, texture, and time-series information.
• To assess the effectiveness of a novel pixel-based supervised classification with a sample purification strategy.

Main Methods:

• Incorporation of multi-spectral optical and Synthetic Aperture Radar (SAR) data.
• Utilized terrain, texture, and time-series features for classification.
• Employed a pixel-based supervised classification method with a sample purification strategy.
• Validated results using 610 sample data points and 10-fold cross-validation.

Main Results:

• Achieved stable Overall Accuracy (OA), Producer's Accuracy (PA), and User's Accuracy (UA) above 90%.
• F1 score and Kappa coefficient remained stable (0.90-0.93 and 0.81-0.87, respectively), indicating reliable classification.
• Time-series and texture features were identified as key factors for terrace recognition, more so than terrain.
• Uncertainty in mapping was concentrated in specific areas, often due to land cover heterogeneity and spectral similarity.

Conclusions:

• The proposed sample purification strategy and data integration yield a more reliable regional terrace map than existing products.
• Time-series and texture features are critical for accurate terrace mapping, challenging previous assumptions about terrain importance.
• The developed approach is potentially transferable to other mountainous regions for rapid and robust terrace identification.