Search

Soil scientists require cost-effective methods to make accurate regional predictions of soil organic carbon (SOC) content. We assess the suitability of airborne radiometric data and digital elevation data as covariates to improve the precision of predictions of SOC from an intensive survey in Northern Ireland. Radiometric data (K band) and, to a lesser extent, altitude are shown to increase the precision of SOC predictions when they are included in linear mixed models of SOC variation. However the statistical distribution of SOC in Northern Ireland is bimodal and therefore unsuitable for geostatistical analysis unless the two peaks can be accounted for by the fixed effects in the linear mixed models. The upper peak in the distribution is due to areas of peat soils. This problem may be partly countered if soil maps are used to classify areas of Northern Ireland according to their expected SOC content and then different models are fitted to each of these classes. Here we divide the soil in Northern Ireland into three classes, namely mineral, organo-mineral and peat. This leads to a further increase in the precision of SOC predictions and the median square error is 2.2 %2. However a substantial number of our observations appear to be mis-classified and therefore the mean squared error in the predictions is larger (30.6 %2) since it is dominated by large errors due to mis-classification. Further improvement in SOC prediction may therefore be possible if better delineation between areas of large SOC (peat) and small SOC (non-peat) could be achieved.