Abstract. To explore the intrinsic spatial patterns of N2O emissions in agricultural systems, not only should the spatial and temporal variability in N2O emissions be analyzed separately, but the joint spatio-temporal variability should also be explored by applying spatio-temporal semivariogram models and interpolation methods. In this study, we examined the spatio-temporal variability in N2O emissions from a tea-planted soil from 28 April 2014 to 27 May 2014 using 96 static mini chambers in an approximately regular grid on a 40 m2 tea field (sampling 30 times), and the results were compared with long-term observations of the N2O emissions recorded using large static chambers (sampling 5 times). The N2O fluxes observed by the mini chambers during a 30 min snapshot (10:00–10:30 a.m. China Standard Time) ranged from −2.99 to 487.0 mg N m−2 d−1 and were positively skewed with a median of 13.6 mg N m−2 d−1. The N2O flux data were then log-transformed for normality. After detrending the influences from the chamber placement positions (Position) and the precipitation accumulated over two days (Rain2), the log-transformed N2O fluxes (FLUX30t) exhibited strong spatial, temporal and joint spatio-temporal autocorrelations, which were used as three components of spatio-temporal semivariogram models and were characterized by models based on Stein's parameterized Matérn (Ste) function, exponential function and again the Ste function, respectively. The spatio-temporal experimental semivariogram of the N2O fluxes was fitted using four spatio-temporal semivariogram models (separable, product-sum, metric and sum-metric). The sum-metric model performed the best and provided meaningful effective ranges of spatial and temporal dependence, i.e., 0.41 m and 5.4 days, respectively. Four spatio-temporal regression-kriging interpolations were applied to estimate the spatio-temporal distribution of N2O emissions over the study area. The cross-validation results indicated that the four interpolations exhibited similar performances (r = 0.817–0.824, RMSE = 0.456–0.486, p < 0.001), and outperformed the multiple linear regression prediction (r = 0.735, RMSE = 0.560, p < 0.001). The predictions of the four kriging interpolations for the total N2O emissions from the 40 m2 tea field ranged from 18.3 to 18.5 g N; these values were approximately 25 % higher than the results predicted using the observations of large static chambers. Furthermore, compared with the other three models, the metric model exhibited weak sensitivity for peak prediction, although the cross-validation results indicated that they had same prediction capabilities. Our findings suggested: (i) that the size of large static chambers used for long-term observations of N2O fluxes should be no less than 0.4 m and the time interval for gas sampling should be constrained to approximately 5 days; and (ii) that more efficient testing methods should be adopted to replace the conventional cross-validation methods for evaluating the performance of spatio-temporal kriging.
Wet-season spatial variability of N<sub>2</sub>O emissions from a tea field in subtropical central China
