Unlocking Roadside Carbon Sequestration Potential: Machine Learning Estimation of AGB in Highway Vegetation Belts...

Area of Science:

• Ecology
• Remote Sensing
• Geospatial Analysis

Background:

• Aboveground biomass (AGB) is vital for understanding vegetation productivity and carbon sequestration.
• Roadside vegetation, a significant carbon sink, remains understudied compared to forests and croplands.
• Accurate AGB estimation is critical for ecosystem service assessment and carbon cycle research.

Purpose of the Study:

• To evaluate remote-sensing predictors and machine-learning algorithms for estimating AGB in highway roadside vegetation.
• To compare the performance of various regression models in AGB estimation.
• To develop a spatially explicit AGB map for roadside corridors.

Main Methods:

• Field-measured AGB samples were integrated with GF-2 high-resolution satellite imagery.
• Six remote-sensing variables (vegetation indices and band ratios) were used as predictors.
• Five regression models (MLR, PLSR, RF, SVR, XGBoost) were developed and compared using five-fold cross-validation.

Main Results:

• The Random Forest (RF) model, using multiple variables, demonstrated superior performance (training R² of 0.83, testing RMSE of 0.84 kg·m⁻²).
• The optimal RF model was applied to GF-2 imagery to map AGB along a 32 km highway segment.
• An estimated total aboveground biomass of 566.97 t was calculated for the studied roadside corridor.

Conclusions:

• Combining high-resolution remote sensing with machine learning effectively enhances AGB estimation for linear roadside vegetation.
• This approach provides valuable technical support for ecological monitoring, roadside greening management, and carbon accounting in transportation infrastructure.
• Further research into underexplored vegetation types like roadside corridors is essential for comprehensive carbon cycle understanding.