The anatomy of uncertainty for soil pH measurements and predictions: Implications for modellers and practitioners

Statistical validation of spatial predictions of soil properties requires assessment of errors against measured values. The objective of this study was to assess the size of errors in the measurement of soil pH from different sources in the United States of America national databases and implications of the size of errors for prediction, validation and management decision making under uncertain conditions. Error sources included measurement methods, laboratory conditions, pedotransfer functions, database manipulations, location accuracy, and spatial and polygon methods of interpolation. The databases consisted of measured soil pH values from the US National Cooperative Soil Survey Characterization Database (NCSS–SCDB) and estimated values from US Soil Survey Geographic (SSURGO) and State Soil Geographic (STATSGO2) databases. The degree of agreement between measurement methods ranged from poor to substantial, with Lin's concordance correlation coefficients (ρc) varying from 0.83 (pH 1:1W against 1:5CₐCₗ₂) to 0.95 (pH 1:1W against pH 1:5W) and root mean square error (RMSE) varying from 0.27 to 0.43. The degree of agreement between pH 1:1W, 1:2CₐCₗ₂ and mid‐infrared spectroscopy (MIR) ranged from poor to moderate. The RMSE for MIR was 0.40 for pH 1:1W and 0.32 for soil pH 1:2CₐCₗ₂. The RMSE for between‐laboratory reproducibility varied from 0.50 (pH 1:1W) to 0.68 (pH 1:2CₐCₗ₂) and was greater than within‐laboratory reproducibility (pH 1:1W, 0.34; pH 1:2 CₐCₗ₂, 0.22) and repeatability (pH 1:1W, 0.19; pH 1:2CₐCₗ₂, 0.04). The RMSE for the relations for profile depth slicing (weighted mean against equal‐area spline) was 0.36. The RMSE for the relation between soil pH 1:1W for the Global Positioning System and Public Land Survey System was 0.57. Predictions based on polygon or spatial interpolation had the largest RMSEs, 0.78 and 0.62, respectively. Soil liming recommendations based on 0.1 pH increments do not reflect error measurements or the uncertainty of spatial prediction. Although it was not possible to establish consistent trends in the size of error (progressively increasing from measurement to aggregation), its assessment can improve modelling and management at various scales. HIGHLIGHTS: We assessed sources of errors and uncertainty for measured and spatial predictions of soil pH. The smallest error was reported for measured pH (0.06). Polygon or spatial interpolation resulted in the largest error (0.68). Differences in error size influenced rates of liming and cost.