Journal
AGRICULTURAL AND FOREST METEOROLOGY
Volume 256, Issue -, Pages 362-378Publisher
ELSEVIER SCIENCE BV
DOI: 10.1016/j.agrformet.2018.01.023
Keywords
Closed chamber campaigns; Eddy covariance technique; Carbon dioxide fluxes; Agricultural ecosystem; Non-linear regression algorithm; Uncertainty analysis
Categories
Funding
- German Federal Ministry of Food and Agriculture [FKZ 22403212, FKZ 22403312]
- German Federal Ministry of Education and Research (BMBF) [FKZ 01LK1101A]
- BMBF within the framework of the Junior Research Group NITROSPHERE [FKZ 01LN1308A]
Ask authors/readers for more resources
Manual chamber campaigns are a versatile method to study management effects at plot scale in factorial experiments. The eddy covariance technique has the advantage of continuous measurements but it requires large homogeneous areas. By pairing the two techniques, the uncertainties of the CO2 fluxes modeled from the chamber campaigns can be quantified through cross-validation with the continuous eddy covariance data. This is particularly important in managed ecosystems with high temporal dynamics. At our agricultural site in Northern Germany, we installed both techniques in parallel for two crop cultivation periods, winter oilseed rape in 2012/13 and winter wheat in 2013/14. First, we compared measured net CO2 exchange (NEE) obtained from the closed chambers with the corresponding half-hourly fluxes from the eddy covariance technique. Despite largely different footprints and measurement windows, the measured fluxes were highly correlated (R-2 = 0.83 in 2012/13 and R-2 = 0.93 in 2013/14). Interpolating from chamber campaigns to the entire measurement period is commonly performed by modeling half-hourly fluxes based on non-linear regressions for photosynthesis and respiration. These modeled fluxes were compared to the fluxes measured with the eddy covariance technique. To understand the observed differences, we performed five modeling setups: 1) Non-linear regressions based algorithm with default settings, 2) non-linear regressions with expert settings, 3) purely empirical modeling with artificial neural networks, 4) cross-validation using eddy covariance measurements as campaign fluxes on original campaign days, and 5) cross-validation on weekly campaign days. The modeled seasonal course of daily NEE agreed well with the eddy covariance measurements for all five setups (R-2 from 0.77 to 0.92) but with periods of systematic offsets in the range of 5 g C m(-2) day(-1). Though the pattern of the offsets was different, all setups had comparable root mean square errors around 1.5 g C m(-2) day(-1) despite having opposite limitations. Cross-validation by simulating campaigns with artificial gaps from the continuous eddy dataset in setup 4) and 5) resulted in bias errors of around 0.4 g C m(-2) day(-1). This translates to a total uncertainty on annual NEE of around 175 g C m(-2) a(-1) purely from the modeling, i.e. the interpolation in-between campaigns. By leave-one-campaign-out scenarios, the sensitivity to single campaigns was examined. The mean effect on the annual total was higher for setup 4 (30 g C m(-2)) with the original number of campaigns than for setup 5 (9 g C m(-2)) with four times more campaigns. Furthermore, the interpolation in-between the campaigns can be improved by deriving vegetation proxies from the continuous eddy covariance measurements, such as an effective green area index (GAI) presented herein.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available