[M-IS05] Environmental, Socio-economic, and Climatic Changes in Northern Eurasia

[MIS05-P08] Modeling peatlands net ecosystem exchange using machine learning techniques based on eddy covariance observations

*Egor Dyukarev^1,2, Evgeniy Zarov², Artyom Kulik², Eugenia Golovatskaya¹, Elena Lapshina² (1.Institiute of Monitoring of Climatic and Ecological Systems Siberian Branch Russian Academy of Sciences, 2.Yugra State University, Khanty-Mansiysk, Russia )

Keywords:net ecosystem exchange, eddy-covariance, peatlands, machine learning

Accurate estimates of net ecosystem CO2 exchange (NEE) would improve the understanding of natural carbon sources and sinks and their role in the regulation of global atmospheric carbon. We use the random forest, artificial neural network, gaussian process regression machine learning (ML) methods for predicting NEE, CH4, latent and sensible fluxes at West Siberian peatlands. The CO₂ fluxes measured using the eddy-covariance technique at the Mukhrino (MUH) field station (Khanty-Mansi Autonomous Okrug-Yugra, Russia) and Plotnikovo (PLT) station (Tomsk Region, Russia) in 2022-2024 were used at this study. The observed meteorological variables were used as predictors. Comparison of model efficiency allowed to select the best model for NEE prediction and partition into ecosystem respiration and gross primary production. The gaussian process regression model with an exponential kernel was more accurate in reproduction observed NEE values. Incoming shortwave radiation, soil temperature, water table depth and atmospheric pressure were the best predictors for NEE, while outgoing longwave radiation, soil temperature and water table depth was better for methane fluxes. Gap-filled time series were used for estimation of annual total energy balance and greenhouse gases fluxes. All studied peatland ecosystems are sink for atmospheric carbon, but the intensity of this sink is determined to a greater extent by vegetation characteristics and weather conditions.