This dataset contains digital soil maps (.tiff) of predicted soil contents of oxalate-extractable iron and aluminium at a 25 m spatial resolution across six depth layers (0-5 cm, 5-10 cm, 10-25 cm, 25-60 cm, 60-100 cm and 100-200 cm) for agricultural fields in the Netherlands. For each of these depth layers, there is a map of mean predictions, the 5th, 50th (median) and 95th quantile predictions, as well as the 90% prediction interval (PI90 = 95th - 5th quantile) and prediction interval ratio (PIR = PI90 / median). PI90 and PIR represent absolute and relative uncertainty predictions, respectively. The maps were created using Quantile Regression Forest models, which were calibrated using geo-referenced wet-chemical measurements (n = 12,110) and near-infrared (NIR) estimates (n = 102,393) of oxalate-extractable iron and aluminium and over 150 spatial covariates (spatially explicit environmental variables of soil forming factors). See publication for details, including the assessment of map quality using design-based statistical inference.