Assessing alternative approaches to predicting soil phosphorus sorption

Twenty-five pasture soils were sampled from high-rainfall zones of southeastern Australia to examine relationships between soil properties, and between soil properties and P buffering capacity (PBC) measures. Correlations between PBC values and soil properties were generally poor, with the exception of oxalate-extractable Al (Al-ox) (r >= 0.97). Predictions of PBC were further improved when clay, as well as Al-ox, was included in a linear regression model (r(2) >= 0.98). When Al-ox and oxalate-extractable Fe were excluded from the modelling exercise, a more complex three-terin linear regression model, including pH(H2O), exchangeable H and cation exchange capacity, adequately fitted both PBC values of the 25 soils examined in this study (r(2) >= 0.76). However, the Al-ox, Al-ox plus clay and the three-term models gave poor predictions of the PBC values when the models were validated using 28 independent soils. These results emphasize the importance of model validation, because predictive models based on soil properties were not robust when tested across a broader range of soil types. In comparison, direct measures of PBC, such as single-point P sorption measures, are more practical and robust methods of estimating PBC for Australian soils.