A comparative mapping of plant species diversity using ensemble learning algorithms combined with high accuracy...

Area of Science:

• Ecology
• Remote Sensing
• Computational Biology

Background:

• Plant species diversity (PSD) is crucial for ecosystem functions and services.
• Accurate spatial distribution mapping of PSD is challenging due to data acquisition and prediction algorithm limitations.
• Advancements in remote sensing and machine learning provide new opportunities for PSD mapping.

Purpose of the Study:

• To develop and evaluate machine learning algorithms combined with high-accuracy surface modeling (HASM) for predicting PSD.
• To map PSD in the Xinghai region of the Qinghai-Tibetan Plateau using remote sensing and environmental data.
• To identify the key factors influencing PSD prediction accuracy.

Main Methods:

• Utilized Landsat 8 spectral reflectance and vegetation indices, along with environmental variables.
• Applied machine learning models: Lasso, Ridge, XGBoost, and Random Forest.
• Integrated predictions with HASM and in situ data for high-accuracy PSD mapping.

Main Results:

• Plant species diversity showed a strong association with vegetation indices, followed by spectral reflectance and environmental factors.
• The HASM-XGBoost model demonstrated the best performance, achieving the lowest Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).
• Fusion of heterogeneous data and ensemble of models significantly improved PSD prediction accuracy, especially in areas with limited field samples.

Conclusions:

• Machine learning algorithms, particularly XGBoost, combined with HASM, are effective for high-accuracy PSD mapping.
• The integration of remote sensing data and machine learning offers a powerful approach to overcome challenges in PSD spatial distribution simulation.
• This approach has the potential to revolutionize large-area PSD prediction, particularly in data-scarce regions.