Calculating uncertainties on predictions of palaeoprecipitation from the magnetic properties of soils

Quantitative predictions of past climate states based on calibrated proxy data are key to the reconstruction of palaeoenvironments and are essential for climate model validation. Magnetic climofunctions have been used to make predictions concerning past climates based on soil magnetic mineral assemblages. For example, detailed time series of Quaternary mean annual precipitation and palaeoprecipitation gradients across wide geographic regions have been predicted from the rock magnetic properties of Chinese loess and palaeosol units. Quantitative prediction requires full assessment of the uncertainties associated with predictions. However, little attention has been given to this important aspect of climofunction prediction. We present an analysis of an ensemble of published rock magnetic climofunctions and estimate the uncertainty of the associated predictions. We find that existing climofunctions have associated uncertainties that are so large that their subsequent predictions are effectively invalid. Thus, palaeoprecipitation reconstructions must be treated with extreme caution. In the future climofunctions that are constrained geologically through the inclusion of theoretical models of soil development may provide predictions with lower uncertainties. ► We place empirical magnetic climofunctions within a statistical framework. ► The uncertainties associated with palaeoprecipitation predications are assessed. ► Discrimination intervals show that climofunction prediction uncertainties are large.