A Novel Forest Dynamic Growth Visualization Method by Incorporating Spatial Structural Parameters Based on Convolutional Neural Network

• Published in: IEEE journal of selected topics in applied earth observations and remote sensing 2024-01, Vol.17, p.1-20
• Main Authors: Wang, Linlong, Zhang, Huaiqing, Lei, Kexin, Yang, Tingdong, Zhang, Jing, Cui, Zeyu, Fu, Rurao, Yu, Hongyan, Zhao, Baowei, Wang, Xianyin
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; convolutional neural network; Convolutional neural network (CNN); Cunninghamia lanceolata; Data analysis; Data mining; Data models; Forest growth; forest growth model; forest growth model (FGM); Forestry; Forests; Growth models; Machine learning; Mathematical models; Morphology; Neural networks; Parameters; Physical characteristics; Plantations; Predictions; Predictive models; Root-mean-square errors; Solid modeling; spatial structure; Stocking density; three-dimensional (3-D) visualization; three-dimensional visualization; Vegetation; Visualization
• ISSN: 1939-1404; 2151-1535
• DOI: 10.1109/JSTARS.2023.3342445

Current visual methods of forest dynamic growth mostly focus on the plot or stand level, which cannot express the morphological and structural characteristics of individual trees, as well as their statistical linkages, and causes each tree in the stand growing at the same rate. Additionally, these visual growth models still have some space for improvement in terms of prediction accuracy and multi-relational data mining. In our study, uneven-aged Chinese fir ( Cunninghamia lanceolata ) plantations were chosen as our study subject and proposed a novel method of forest dynamic growth visualization modeling by incorporating spatial structure parameters and using convolutional neural network technique (FDGVM-CNN-SSP) to explore the effect of spatial structure on the morphological growth and to develop a prediction growth model of Chinese fir plantations by introducing a convolutional neural network (CNN) model. The results show that, (1) Spatial structural parameters C and U have a certain contribution to the forest growth, which C and U can explained 21.5%, 15.2%, and 9.3% of the variance in DBH, H, and CW growth models, respectively. (2) CNN model outperformed machine learning algorithms SVR, MARS, Cubist, RF, and XGBoost in terms of prediction performance. (3) Based on FDGVM-CNN-SSP, we simulated Chinese fir plantations at individual tree level and stand level from 2018-2022 and found that DBH and H' fitting performance in measured and predicted data was highly consistent with R 2 and RMSE of 86.8%, 2.06cm in DBH and 79.2%, 1.11m in H, but CW's R2 and RMSE of 72.2%, 0.65m caused crowding (C) inconsistency.