МОДЕЛИРОВАНИЕ НЕТТО-ЭКОСИСТЕМНОГО ОБМЕНА  $\mathrm{CO_2}$ С ПОМОЩЬЮ МЕТОДОВ МАШИННОГО ОБУЧЕНИЯ

In the context of global climate change, the assessment of carbon balance at the regional level is becoming increasingly important for developing effective strategies for greenhouse gas emissions management and adaptation to current climate trends. Terrestrial ecosystems play a crucial role in the global carbon cycle, making varying contributions to carbon dioxide exchange between the underlying surface and the atmosphere. In this study, a machine learning-based model was developed, specifically utilizing the CatBoost gradient boosting algorithm, for comprehensive assessment of spatiotemporal variability in net ecosystem exchange of CO2 (NEE). The results of model experiments for Sakhalin Island in 2023 demonstrated that this approach effectively accounts for multiple factors affecting carbon exchange and provides spatial distribution of CO2 fluxes at a regional scale with monthly temporal resolution. The developed model showed high prediction accuracy with a coefficient of determination (R²) averaging 0.76 across all ecosystems. The obtained results can be applied for carbon balance assessment in other regions and development of measures to mitigate anthropogenic impact on the climate system.