Predictive soil mapping with limited sample data

Existing predictive soil mapping (PSM) methods often require soil sample data to be sufficient to represent soil-environment relationships throughout the study area. However, in many parts of the world with only a limited quantity of soil sample data to represent the study area, this is still an issue for PSM application. This paper presents a method, named individual predictive soil mapping' (iPSM), which can make use of limited soil sample data for PSM. With the assumption that similar environmental conditions have similar soils, iPSM uses the soil-environment relationship at each individual soil sample location to predict soil properties at unvisited locations and estimate prediction uncertainty. Specifically, the environmental similarities of an unvisited location to a set of soil sample locations are used in a weighted average method to integrate the soil-environment relationships at sample locations for prediction and uncertainty estimation. As a case study, iPSM was applied to map soil organic matter (SOM) content (%) in the topsoil layer using two sets of soil samples. Compared with multiple linear regression (MLR), iPSM produced a more accurate SOM map (root mean squared error (RMSE) 1.43, mean absolute error (MAE) 1.16) than MLR (RMSE 8.54, MAE 7.34) the ability of the sample set to represent the study area is limited and achieved a comparable accuracy (RMSE 1.10, MAE 0.69) with MLR (RMSE 1.01, MAE 0.73) when the sample set could represent the study area better. In addition, the prediction uncertainty estimated by iPSM was positively related to prediction residuals in both scenarios. This study demonstrates that iPSM is an effective alternative when existing soil samples are limited in their ability to represent the study area and the prediction uncertainty in iPSM can be used as an indicator of its prediction accuracy.