[Effect of the Sampling Scale and Number on the Heterogeneity of Soil Respiration in a Mixed Broadleaf-conifer...

By upscaling the observed results at the plot scale, the carbon efflux from soils in a region can be estimated. Therefore, it is very important to investigate the spatial relations of soil respiration (R_s) and its environment and to evaluate the effect of the sampling scale and number on the accuracy of R_s measurement at the spatial scale. Based on field observation data for a mixed broadleaf-conifer forest in the Pangquangou Nature Reserve of the Shanxi Province, two analysis methods, that is, traditional statistics and geostatistics, were used to analyze the influence of the soil water content (W_s), soil temperature (T₁₀), litter mass (L_w), litter moisture content (L_m), soil total carbon (C), total nitrogen (N), and ratio of C/N and sulfur (S) on the R_s heterogeneity at 4, 2, and 1 m sampling scales. The results show no significant differences between the average R_s values for the three sampling scales, but the degree of variation of R_s, which was evaluated based on the coefficient of determination, increases with increasing sampling scales, ranging from 16% to 22%. At the 4 m sampling interval, the correlations between R_s and W_s, L_w, C, and C/N are highly significant (P<0.01) and significant for N (P <0.05). At the 2 m sampling interval, R_s shows a highly negative significant correlation with T₁₀ (P<0.01) and insignificant correlations with the other factors. At the 1 m sampling interval, significant relations between R_s and all other factors were not observed. With the decrease of the sampling interval scale, the spatial autocorrelation of R_s decreases gradually, ranging from high to weak autocorrelations.This indicates that the role the structural factors play decreases with the decrease of the sampling scale, but that of the random factors increases gradually. At the same confidence level for a certain sampling number, the estimated error in R_s decreases with decreasing sampling scale. The analysis of the effect of the sampling number at different sampling scales on the accuracy of R_s shows that the error of R_s at both the 2 and 1 m sampling scales is approximately±12% at the 95% confidence interval and±16% at the 4 m sampling scale. At the 90% confidence interval, the error of R_s at both the 2 and 1 m sampling scales is less than ±10%; at the 4 m scale, it is ±13%. Our results provide insights into how to arrange the sampling sites at the plot scale to measure the seasonal R_s.