Sampling Distribution of Nitrates in Irrigated Fields1
- J. O. Reuss,
- P. N. Soltanpour and
- A. E. Ludwick2
The object of this study was to determine the components of variability encountered in sampling farmers' fields for soil nitrates. This information is required for formulating recomendations for sampling plan and intensity. Twenty-four irrigated farm fields in northeastern Colorado and Western Nebraska were sampled on 61 ✕ 61 m grids. One soil core was taken in 30-cm increments from each grid to a depth of 120 cm. Duplicate cores were obtained from each third grid to allow determination of within grid variability. The laboratory-induced variability was also estimated. The results indicated that within and among grid standard deviations increased as the means increased. The standard errors of laboratory determinations (subsampling plus analytical errors) also increased as the means increased. A logarithmic transformation produced histograms that approximated a normal distribution better than the untransformed values. Therefore, the geometric mean represented the nitrate content of a sampled field better than an arithmetic mean. Regression analyses showed that geometric mean could be estimated by multiplying the arithmetic mean of nitrates by 0.91. The nitrate values changed significantly in parallel and perpendicular directions with respect to the crop rows in a nonpredictive manner, indicating that a systematic sampling plan will lead to a more representative sample than a random sampling plan. By separation of a spatial component of variance it was calculated that for the average field sampled to a depth of 120 cm, 7% of the area had nitrate levels below 67% of the geometric mean while on 9% of the area the nitrates exceeded 150% of the geometric mean. In the most variable 10% of the fields 20% or more of the area had nitrate levels below 67% of the geometric mean while on 21% of the area the nitrate exceeded 150% of this value. The laboratory standard deviation was consistently 5% of the mean, indicating a lognormal distribution. With this laboratory precision doubling laboratory variability will not significantly increase the confidence interval of the geometric mean at a low level of sampling intensity, but it will have much larger effect as sampling intensity increases. The number of cores required for a 90% confidence interval of about ± 15% and ± 26% of the geometric mean was calculated to be 82 and 20 per field, respectively.Please view the pdf by using the Full Text (PDF) link under 'View' to the left.
Copyright © . .