Accurate quantification of Net Ecosystem Exchange (NEE) in forests is essential to understand their role in sequestering anthropogenic CO2 emissions and supporting climate mitigation. The CO2 flux measured by the above-canopy Eddy-Covariance (EC) provides ground-truth observations worldwide. However, the applicability of above-canopy EC is affected by below-canopy CO₂ accumulation due to decoupling that frequently occurs under nocturnal or weak-turbulence conditions, often resulting in underestimates of ecosystem respiration. An above-canopy EC, complemented by below-canopy wind and CO₂ concentration profile measurements, was integrated to investigate the contribution of below-canopy accumulation to above-canopy CO₂ flux. This study evaluates the effectiveness of the standard deviation of vertical wind speed (σw) as an alternative turbulence indicator to the conventional friction velocity (u) threshold. Our results indicate that σw serves as a direct proxy for vertical mixing efficiency and offers superior detection of decoupling events. Optimal thresholds were objectively identified using knee-point analysis to maximize the trade-off between the Success Rate (SR), defined as the percentage of decoupling events relative to all events, and the Data Retention (DR), defined as the percentage of remaining data relative to all data. The determined σw threshold of 0.22 m s− 1 outperformed the u threshold of 0.20 m s− 1, achieving higher SR (79.8% vs. 72.0%) and better DR (81.4% vs. 74.6%) values. This confirms that σw effectively minimizes the inclusion of decoupled data while preserving more valid observations, a finding further corroborated by analyzing vertical CO₂ profiles. To address gaps in the filtered dataset, five gap-filling strategies were evaluated using Monte Carlo simulations. The Support Vector Regression (SVR) algorithm proved to be most effective, outperforming traditional Marginal Distribution Sampling (MDS) and other machine learning algorithms by capturing nonlinear interactions between environmental drivers and CO2 flux. The integration of σw-based filtering (σw = 0.22 m s− 1) with the SVR gap-filling produced an annual CO₂ flux estimate that was 17.9% lower than the case without turbulence filtering. These findings highlight that standard filtering protocols are essential to prevent overestimating CO₂ flux from below-canopy accumulation under weak-turbulence conditions.

Investigating below-canopy CO₂ accumulation with turbulence-based filtering in a subtropical forest

期刊论文
出版模式：

开放获取
发表日期：

2026年7月7日

查看全文

Kuan-Ying Chen,
Ming-Hsu Li

摘要

Accurate quantification of Net Ecosystem Exchange (NEE) in forests is essential to understand their role in sequestering anthropogenic CO₂ emissions and supporting climate mitigation. The CO₂ flux measured by the above-canopy Eddy-Covariance (EC) provides ground-truth observations worldwide. However, the applicability of above-canopy EC is affected by below-canopy CO₂ accumulation due to decoupling that frequently occurs under nocturnal or weak-turbulence conditions, often resulting in underestimates of ecosystem respiration. An above-canopy EC, complemented by below-canopy wind and CO₂ concentration profile measurements, was integrated to investigate the contribution of below-canopy accumulation to above-canopy CO₂ flux. This study evaluates the effectiveness of the standard deviation of vertical wind speed (σ_w) as an alternative turbulence indicator to the conventional friction velocity (u*) threshold. Our results indicate that σ_w serves as a direct proxy for vertical mixing efficiency and offers superior detection of decoupling events. Optimal thresholds were objectively identified using knee-point analysis to maximize the trade-off between the Success Rate (SR), defined as the percentage of decoupling events relative to all events, and the Data Retention (DR), defined as the percentage of remaining data relative to all data. The determined σ_w threshold of 0.22 m s^− 1 outperformed the u* threshold of 0.20 m s^− 1, achieving higher SR (79.8% vs. 72.0%) and better DR (81.4% vs. 74.6%) values. This confirms that σ_w effectively minimizes the inclusion of decoupled data while preserving more valid observations, a finding further corroborated by analyzing vertical CO₂ profiles. To address gaps in the filtered dataset, five gap-filling strategies were evaluated using Monte Carlo simulations. The Support Vector Regression (SVR) algorithm proved to be most effective, outperforming traditional Marginal Distribution Sampling (MDS) and other machine learning algorithms by capturing nonlinear interactions between environmental drivers and CO₂ flux. The integration of σ_w-based filtering (σ_w = 0.22 m s^− 1) with the SVR gap-filling produced an annual CO₂ flux estimate that was 17.9% lower than the case without turbulence filtering. These findings highlight that standard filtering protocols are essential to prevent overestimating CO₂ flux from below-canopy accumulation under weak-turbulence conditions.