Alpine Meadow Fractional Vegetation Cover Estimation Using UAV-Aided Sentinel-2 Imagery

Area of Science:

• Ecology
• Remote Sensing
• Environmental Science

Background:

• Fractional Vegetation Cover (FVC) is vital for assessing ecosystem health, but its estimation in alpine meadows is challenging due to sparse vegetation and limitations of traditional pixel dichotomy models with Sentinel-2 imagery.
• Accurate FVC data is essential for monitoring alpine meadow ecosystems, particularly in regions like the Qinghai-Tibet Plateau.

Purpose of the Study:

• To enhance the accuracy of Fractional Vegetation Cover (FVC) estimation in alpine meadows by integrating Sentinel-2 and Unmanned Aerial Vehicle (UAV) data.
• To compare the performance of traditional pixel dichotomy models with various machine learning algorithms (RF, XGBoost, LightGBM, DNN) for FVC estimation.

Main Methods:

• Preliminary FVC estimation using a pixel dichotomy model with nine vegetation indices on Sentinel-2 imagery.
• Evaluation of preliminary estimates against reference FVC data derived from centimeter-level UAV data.
• Application and optimization of four machine learning models (RF, XGBoost, LightGBM, DNN) using Sentinel-2 and UAV data for accurate FVC inversion.

Main Results:

• Machine learning algorithms significantly improved FVC estimation accuracy in alpine meadows when using Sentinel-2 and UAV data.
• The Deep Neural Network (DNN) model demonstrated the best performance, achieving a coefficient of determination of 0.82 and a Root Mean Square Error (RMSE) of 0.09.
• Different vegetation indices showed varying effectiveness based on FVC levels: GNDVI for high coverage (FVC > 0.7, RMSE=0.08) and NIRv/SR for low coverage (FVC < 0.4, RMSE=0.10).

Conclusions:

• The integration of Sentinel-2 and UAV data with machine learning, particularly DNN, offers a robust approach for accurate FVC estimation in challenging alpine meadow environments.
• This method enhances FVC estimation accuracy with reduced fieldwork, supporting effective monitoring of alpine meadows on the Qinghai-Tibet Plateau.
• Understanding the performance variations of vegetation indices across different FVC levels is crucial for optimizing estimation models.