Application of the Semantic Segmentation Method in the Assessment of Forest Carbon Balance

This article focuses on the problem of human impact on the natural environment and its solution by machine learning methods. The concept of carbon balance is key in assessing climate change on the planet, which is especially important in connection with global warming. The main causes of global warming are associated with anthropogenic emissions of carbon dioxide as a result of agricultural production, land use and burning of fossil fuels. One of the approaches to mitigate this environmental problem is to estimate the carbon balance based on neural network semantic segmentation of forests of terrestrial ecosystems. The proposed convolutional neural network (CNN) model is based on the U-Net architecture. This architecture surpasses traditional machine learning methods in solving computer vision and pattern recognition problems, and is also quite easy to learn. The input data of the model is a set of satellite images of the forest area. The images were obtained using the Sentinel-2 satellite, which provides high spatial resolution and a large number of spectral channels. A software prototype has been developed to implement the CNN model. Training experiments and calculations on changes in the carbon balance in forests led to the conclusion that the results of the study will contribute to the balanced management of reforestation and carbon deposition.

I. The problem of carbon balance

The problem of carbon budget inventory to forests acquires an important applied aspect. UN Framework Convention recognizes the forest functions as a sink of greenhouse gases on Climate Change.