論文ID: 22.003
In this work we discuss possibilities and challenges in utilization of several statistical methods for assessment of forest resources related to forest inventories, especially question of dataset size where the time and resources required for data collection are often in contrast to sample size and analysis of all potential parameters of potential models. The combination of a priori knowledge of the phenomena being studied (tree number, wood volume…) and understanding of behavior of individual variables provided by remote sensing instruments (different predictor variables) is crucial for production of reliable models for forest resource assessment. Using our dataset we compared two regression techniques and one machine learning for predictor analysis for wood volume estimation. All techniques in general provided similar results in terms of variable importance and accuracy, but in more detailed analysis differences appeared, indicating that if possible biological knowledge and understanding of variables should not be neglected.
