One chemical element that was missing from the dataset used in this work is aluminium. Strong relationships between soil colour and aluminium content have been found in previous research (e.g. Zhang et al., 1998), and it is felt that the inclusion of this element may have provided higher rates of predictive accuracy for colour in addition to acting as proxy values for elements whose concentration in soil is related to aluminium. However, the dataset that was available does not contain sufficient samples for which aluminium content has been analysed, preventing its inclusion in this work. As aluminium content is a significant factor in the identification and characterisation of podzols, the work carried out here using the data described may be less applicable to podzolic soils. If data for aluminium were present it is assumed that not only would the NN model be more applicable, but also that it would be more accurate for the prediction of other chemical and physical parameters. Aluminium concentration is strongly related to pH, Ca and even texture in some soils, and so the inclusion of this element would possibly increase the robustness of the overall dataset for NN training.
It is also possible however, that the strong relationships between iron, aluminium and organic matter in acid soils (e.g. Reddy et al., 1995) would make the inclusion of aluminium in the model less important, and other parameters already present are so strongly correlated with Al.