Self-learning growth simulator for modelling forest stand dynamics in changing conditions

Abstract

Finnish forest structures vary from even-aged planted forests to two- and multi-storied mixed stands. Also, the range of silvicultural systems in use has increased because thinning from above and continuous cover management are gaining popularity. The data currently available for modelling stand dynamics are insufficient to allow the development of unbiased and reliable models for the simulation of all possible transitions between various current and future stand conditions. Therefore, the models should allow temporal and regional calibration along the accumulation of new information on forest development. If the calibration process is automated, the simulators that use these models constitute a self-Learning system that adapts to the properties of new data on stand dynamics. The current study first developed such a model set for stand dynamics that is technically suitable for simulating the stand development in all stand structures, silvicultural systems and their transitions. The model set consists of individual-tree models for diameter increment and survival and a stand-Level model for ingrowth. The models were based on the permanent sample plots of the 10th and 11th national forest inventories of Finland. Second, a system for calibrating the models based on additional data was presented. This optimization-based system allows different types and degrees of calibration, depending on the intended use of the models and the amount of data available for calibration. The calibration method was demonstrated with two external datasets where a set of sample plots had been measured two times at varying measurement intervals.