Accuracy criteria for measuring carbon and nitrogen concentrations in forest soil and litter samples

Abstract

To establish accuracy criteria for measurements of carbon and nitrogen concentrations in forest litter and mineral soil, we examined the change in the coefficient of variance (CV) with the number of measurement repetitions, the distribution of CVs and the effect of analytical error on the variation of CV to suggest a threshold for detecting outliers. We analyzed the measurement values of 266 soil samples and 73 litter samples collected from various forests throughout Japan. The CV changed with the number of measurement values for soil samples, but not for litter samples. This finding suggested that the accuracy criterion of the CV of two measurement values should be applied only to CVs calculated from two measurement values. When the CV was calculated from two measurement values, variation in the CV did not differ among carbon concentration classes of soil samples; however, the variation did differ among carbon concentration classes of litter samples and nitrogen concentration classes of litter and soil samples. These differences suggested that accuracy criterion should be independently decided in each carbon and nitrogen concentration class. Variation in the CV was not strongly affected by the analytical error because variation in the blank value was very small compared with variation in the CV. The distribution of the CV did not differ from the normal distribution for litter samples, but it did differ for soil samples. Thus, we used the robust Huber’s m-estimator outlier test to decide a threshold to represent the accuracy criterion. The threshold value was almost constant among carbon concentration classes both for soil and litter samples, with mean values of 3.6% and 2.6%, respectively. Conversely, the threshold of outliers of soil and litter samples varied among nitrogen concentration classes. This suggested using different thresholds for each nitrogen concentration class. Moreover, by applying the calculated thresholds we showed that outliers accounted for approximately 30% of the data in almost all carbon and nitrogen concentration classes for soil samples and less than 20% in almost all carbon and nitrogen concentration classes for litter samples.