Comparing simple respiration models for eddy flux and dynamic chamber data.
Authors:
Richardson, A.D. Braswell, B.H. Hollinger, D.Y. Burman, P. Davidson, E.A. Evans, R.S. Flanagan, L.B. Munger, J.W. Savage, K. Urbanski, S.P. USDA, FS
Source:
Agricultural and forest meteorology. 2006 Dec. 20, v. 141, no. 2-4, p. 219-234 .
NALT Subjects:
plants cell respiration ecosystems soil respiration mathematical models soil-plant-atmosphere interactions turbulent flow carbon dioxide grasslands deciduous forests coniferous forests air temperature soil temperature seasonal variation neural networks estimation
Other Subjects:
gross ecosystem exchange net ecosystem exchange eddy covariance measurements
Issue Date:
20-Dec-2006
Abstract:
Selection of an appropriate model for respiration (R) is important for accurate gap-filling of CO2 flux data, and for partitioning measurements of net ecosystem exchange (NEE) to respiration and gross ecosystem exchange (GEE). Using cross-validation methods and a version of Akaike's Information Criterion (AIC), we evaluate a wide range of simple respiration models with the objective of quantifying the implications of selecting a particular model. We fit the models to eddy covariance measurements of whole-ecosystem respiration (R(eco)) from three different ecosystem types (a coniferous forest, a deciduous forest, and a grassland), as well as soil respiration data from one of these sites. The well-known Q(10) model, whether driven by air or soil temperature, performed poorly compared to other models, as did the Lloyd and Taylor model when used with two of the parameters constrained to previously published values and only the scale parameter being fit. The continued use of these models is discouraged. However, a variant of the Q(10) model, in which the temperature sensitivity of respiration varied seasonally, performed reasonably well, as did the unconstrained three-parameter Lloyd and Taylor model. Highly parameterized neural network models, using additional covariates, generally provided the best fits to the data, but appeared not to perform well when making predictions outside the domain used for parameterization, and should thus be avoided when large gaps must be filled. For each data set, the annual sum of modeled respiration (annual ΣR) was positively correlated with model goodness-of-fit, implying that poor model selection may inject a systematic bias into gap-filled estimates of annual ΣR.
