چكيده لاتين
Urban vegetation provides essential ecosystem services both directly and indirectly. It regulates the urban climate through shading and cooling, reduces air pollution and noise pollution. Trees offer important ecosystem services to society and are vital for human well-being and wildlife in human settlements. The tree canopy, in particular, represents the interface where most of the fundamental interactions between vegetation and the atmosphere occur. Therefore, special attention must be given to urban vegetation, which is an essential component of good quality of life.
Assessing forest conditions in urban and suburban areas is essential to support ecosystem service planning and management, as most ecosystem services provided are the result of forest characteristics. However, data collection for assessing forest conditions is time-consuming and labor-intensive. A viable approach to address this issue is to establish relationships between tree feature measurements and airborne laser (LiDAR) data. In this study, we evaluated the volume and above-ground biomass of vegetation in urban areas using metrics derived from LiDAR data.
One of the important sources for estimating biomass is LiDAR data, which allows for estimation on an individual tree basis. In general, LiDAR data can provide accurate estimates of biomass. Various features related to tree canopy, which can be extracted from LiDAR data due to its three-dimensional measurements, are used to estimate tree biomass. However, extracting trees and estimating biomass in urban areas poses multiple challenges due to the complexity of the scene and the presence of various artificial structures such as buildings. The main objective of this paper is to propose a machine learning-based framework for extracting tree canopy in urban areas and subsequently estimating biomass. In the proposed framework, the PointCNN network along with computer vision techniques was used to extract the canopy of each tree. Then, canopy geometry-related features were extracted, and tree biomass was calculated using various machine learning techniques, including Random Forest (RF), Support Vector Regression (SVR), LightGBM, and XGBoost. The results showed that the proposed framework was highly capable of detecting tree canopies. Among the different machine learning models, the Random Forest model, with R² = 0.77 and RMSE = 464.14 kg, provided the best estimate of tree biomass. Finally, using the developed model, the biomass of all trees in the studied urban area was calculated, and a map of urban biomass per hectare was produced. Qualitative assessments showed that the generated map had a high level of agreement with the vegetation cover status in the study area, indicating the success of the proposed framework in extracting tree canopies and estimating urban biomass from LiDAR data.
