Mapping Tree Canopy in Urban Environments Using Point Clouds from Airborne Laser Scanning and Street Level Imagery

Area of Science:

• Urban forestry and remote sensing.
• Ecological resilience and urban planning.
• Geospatial data analysis and computer vision.

Background:

• Urban trees are crucial for ecosystem resilience, requiring accurate inventories.
• Traditional urban tree inventory methods are labor-intensive and time-consuming.
• Automated methods are needed to efficiently update urban tree data.

Purpose of the Study:

• To develop and validate a multi-stage methodology for the automatic updating of urban tree inventories.
• To compare the effectiveness of Airborne Laser Scanning (ALS) point clouds and Google Street View (GSV) imagery for tree detection.
• To assess methods for reducing false positives in automated tree identification.

Main Methods:

• Individual tree detection using ALS point clouds and computer vision techniques on GSV imagery.
• A two-stage approach involving initial tree detection followed by false positive reduction.
• False positive reduction using street-level image checking and machine learning classification with orthophotograph spectral data.

Main Results:

• Initial tree detection achieved a high recall rate (85.07%-86.42%), with subsequent false positive reduction lowering recall to approximately 75% (71.43%-78.18%).
• Both ALS and GSV data sources provided robust and accurate results for urban tree inventory updates.
• The primary difference between data sources lies in accessibility and coverage: ALS covers all areas, while GSV is limited to streets.

Conclusions:

• Automated methodologies using ALS and GSV data can effectively update urban tree inventories.
• ALS combined with orthophotographs offers broader spatial coverage and often freely available data compared to GSV.
• The developed multi-stage methodology offers a viable solution for efficient and accurate urban tree management, supporting resilient city initiatives.