Integrated Airborne LiDAR Data and Imagery for Suburban Land Cover Classification Using Machine Learning Methods

Area of Science:

• Remote Sensing
• Geographic Information Systems (GIS)
• Environmental Science

Background:

• Suburban land use/land cover (LULC) classification is crucial for urban planning and environmental management.
• While fusing Light Detection and Ranging (LiDAR) data with aerial imagery improves LULC classification, further analysis is needed for suburban environments.
• Existing LiDAR-derived models may not fully capture the complex surface characteristics of suburban areas.

Purpose of the Study:

• To explore effective information for enhancing suburban LULC classification through data fusion.
• To propose and evaluate a new LiDAR-derived surface roughness model (RM) for suburban LULC classification.
• To optimize multi-variable integration of aerial imagery and LiDAR data for improved classification accuracy.

Main Methods:

• Simultaneous collection of aerial imagery and LiDAR point clouds.
• Generation of LiDAR-derived models: normalized digital surface model (nDSM), intensity model (IM), and a novel surface roughness model (RM).
• Fusion of aerial imagery with LiDAR-derived models and analysis using Random Forest (RF), K-Nearest Neighbor (KNN), and Artificial Neural Network (ANN) classifiers.

Main Results:

• The fusion of aerial imagery with nDSM, RM, and IM, using the RF classifier, yielded the best LULC classification performance (Overall Accuracy = 84.75%, Kappa = 0.80).
• Variable importance analysis indicated nDSM as the most significant variable, followed by RM, IM, and spectral information.
• The proposed RM demonstrated feasibility and effectiveness in capturing surface fluctuations relevant to suburban LULC classification.

Conclusions:

• The integration of aerial imagery and multiple LiDAR-derived models, particularly nDSM and the novel RM, significantly enhances suburban LULC classification accuracy.
• The Random Forest classifier is highly effective for processing fused data in complex suburban scenarios.
• This research provides valuable methods for applying aerial imagery and LiDAR-derived models to intricate suburban landscapes.