Environmental management of soil phosphorus: Modeling spatial variability in small fields

The mapping of soil P concentration is necessary to assess the risk of P loss in runoff. We modeled the distribution of Mehlich-3 extractable soil P (M3P) in an east-central Pennsylvania 39.5-ha watershed (FD-36) with an average field size of 1.0 ha. Three interpolation models were used: (i) the field classification model-simple field means, (ii) the global model-ordinary kriging across the watershed, and (iii) the within-field model-ordinary kriging within fields with a pooled within-stratum variogram. Soils were sampled on a 30-m grid, resulting in an average of 14 samples per field. Multiple validation runs were used to compare the models. Overall, the mean absolute errors (MAE) of the models were 76, 71, and 66 mg kg(-1) M3P for the field classification, global, and within-field models, respectively. The field classification model performed substantially worse than did the kriging models in five fields; these fields exhibited strong spatial autocorrelation. The within-field model performed substantially better than did the global model in three fields where autocorrelation was confined by the field boundary. However, no differences in P index classification were observed between the three prediction surfaces. The field classification model is simpler and less expensive to implement than the kriging models and should be adequate for applications that are not sensitive to small errors in soil P concentration estimates.