Home – Home – Jornals – Siberian Journal of Forest Science 2026 number 1

In recent decades, there has been a rapid increase in the use of deep machine learning tools based on artificial neural networks in various fields of science. Deep neural networks vary in their architecture, for example, in a convolutional neural network, different layers can use convolutional kernels to extract key features from an image and pool the layers to generalize these features. Recurrent neural networks process sequential data series and retain memory of past data by returning the output of a layer back to the same layer. Training a neural network involves optimizing the weights of connections in the network to minimize the prediction error. Deep learning has the potential to leverage information hidden in large datasets to provide innovative solutions to complex environmental challenges. Big data consists of images, audio, videos, or unstructured text, which can be challenging to analyze using traditional statistical methods. With an exponential increase in publications on the methods and results of deep learning on neural networks in various fields of knowledge, this review attempts to analyze some of its applications in the field of forest ecology. In particular, it presents the results of using artificial neural networks to solve certain problems in Russian forestry, such as combining heterogeneous data to estimate forest phytomass, mapping and predicting forest cover dynamics, and identifying plant roots in minirizotron images. The final section describes some of the achievements, challenges, and uncertainties of deep machine learning in ecosystem ecology.