Artificial forests are widely created in northwest China, and their biomass estimation is fundamental for carbon trading. With the development of drone technology, Unmanned Aerial Vehicles (UAVs) have been extensively used for biomass estimation, enabling the acquisition of three-dimensional information of trees.

In a study published in International Journal of Applied Earth Observation and Geoinformation, a research team led by Prof. BAO Anming from the Xinjiang Institute of Ecology and Geography (XIEG) of the Chinese Academy of Sciences, estimated the biomass of young Picea crassifolia forests in Qinghai Province, China.

The researchers combined UAV LiDAR data with multispectral data to identify individual Picea crassifolia. The fused data effectively resolved the issue of accurately extracting the crown widths and heights of young Picea crassifolia.

"The extraction of single-tree crown widths and heights significantly improves data accuracy and acquisition efficiency, and effectively reduces errors caused by manual operations," said TAO Zefu, first author of this study.

To identify the boundaries of a Picea crassifolia forest, the researchers tested the performance of three deep convolutional neural networks (RepLKNet, GoogLeNet, and ResNet), and found that the RepLKNet model performed the best.

Besides, the researchers used the Random Forest and XGBoost models to predict biomass, and found that the accuracy for biomass estimation will change with different imagery resolutions. To be specific, the accuracy initially increased and then decreased as the imagery resolution decreased, reaching the highest accuracy at a resolution of 50 m.

The study validates the feasibility of UAV sampling as a replacement for traditional manual surveys, and calls for careful examination of performance differences when applying UAV for biomass estimation in other ecosystems.

Artificial forests around Xining, Qinghai Province (Image by XIEG)