A groundbreaking study by a research team from Tsinghua University, led by Professor Bing Xu, has introduced the Seasonal Tree Height Neural Network (STHNN), a machine learning model that sets a new standard for measuring urban tree heights. Published in the Journal of Remote Sensing, this study marks a significant advancement in urban ecology by providing a method that combines LiDAR and satellite imagery with machine learning to achieve an 80% accuracy rate with a margin of error as low as 1.58 meters.
The STHNN model's development was driven by the need for more accurate and cost-effective forest monitoring data in urban areas. By employing SHapley Additive exPlanations (SHAP) technology, the team was able to refine their analysis by removing 23 non-essential variables from an initial pool of 52, thereby enhancing the model's predictive accuracy and reducing computational demands. This innovative approach not only improves the precision of tree height measurements but also offers insights into seasonal variations, with data from 2018 to 2023 showing that tree heights in Shenzhen vary between 6 and 14 meters, with winter canopies consistently lower than those in summer.
The research compared several machine learning techniques, including multiple linear regression, support vector machines, random forests, XGBoost, and artificial neural networks, ultimately demonstrating STHNN's superior performance and adaptability across different regions and seasons. Supported by the National Key Research and Development Program of China, this study not only contributes to the scientific understanding of urban forests but also has the potential to influence global efforts in ecological conservation and sustainable urban development. For more details on the study, visit https://www.journalofremotesensing.com.
