Nitrous oxide fluxes from grassland in the Netherlands .1. Statistical analysis of flux-chamber measurements

Accurate estimates of total nitrous oxide (N2O) losses from grasslands derived from flux-chamber measurements are hampered by the large spatial and temporal variability of N2O fluxes from these sites. In this study, four methods for the calculation of mean N2O fluxes (n = 6) on total N2O losses are compared, namely the arithmetic mean, the geometric mean, the lognormal mean and the mean derived from Finney's method. Mean Auxes were calculated from weekly flux measurements on grassland at four contrasting sites in the Netherlands with three management treatments each. Total losses were calculated by interpolation of the mean fluxes and integration over time. Spatial variation of N2O fluxes was large. The geometric mean was generally much smaller, up to a factor of 7, than the arithmetic mean. The lognormal mean was much larger, up to a factor of 11, than the arithmetic mean, possibly because this estimator is biased for small sample size. Arithmetic means and Finney's method were generally in reasonable agreement. The order in estimated N2O loss increased in the order geometric mean<arithmetic mean less than or equal to Finney's mean<lognormal mean. Because of the small sample size (n = 6), the uncertainty about the precise frequency distribution, the sensitivity of estimators based on logtransformed data, and the problems associated with negative fluxes, the arithmetic mean was preferred as the most appropriate estimator. Evidently, the choice of an estimator of the mean can have great effects on the estimation of total N2O losses.