Environmental control of land-atmosphere CO2 fluxes from temperate ecosystems: a statistical approach based on homogenized time series from five land-use types

Abstract

We assembled homogenized long-term time series, up to 19 years, of measurements of net ecosystem exchange of CO2 (NEE) and its partitioning between gross primary production (GPP) and respiration (Reco) for five different ecosystems representing the main plant functional types (PFTs) in France. Part of these data was analyzed to determine the influence of the main environmental variables on carbon fluxes between temperate ecosystems and the atmosphere, and to investigate the temporal patterns of their variations. A multi-temporal statistical analysis of the time series was conducted using random forest (RF) and wavelet coherence approaches. The RF analysis showed that, in all ecosystems, the incident solar radiation was highly correlated with GPP and that GPP was better correlated with the temporal variations of NEE than Reco. The air temperature was the second most important driver in ecosystems with seasonal foliage, i.e., deciduous forest, cropland and grassland; whereas variables related to air or soil drought were prominent in evergreen forest sites. The environmental control on CO2 fluxes was tighter at high frequency suggesting an increased resilience to environmental variations at longer time spans. The spectral analysis performed on three of the five sites selected revealed contrasting temporal patterns of the cross-coherence between CO2 fluxes and climate variables among ecosystems; these were related to the respective PFT, management and soil conditions. In all PFTs, the power spectrum of GPP was well correlated with NEE and clearly different from Reco. The spectral correlation analysis showed that the canopy phenology and disturbance regime condition the spectral correlation patterns of GPP and Reco with the soil moisture and atmospheric vapour deficit.

Keywords

• net co2 ecosystem exchange
• ecosystem respiration
• gross primary production
• spectral analysis
• random forest analysis