Spatial prediction of soil electrical conductivity using soil axillary data, soft data derived from general linear model and error measurement

Abstract

Indirect measurement of soil electrical conductivity (EC) has become a major data source in spatial/temporal monitoring of soil salinity. However, in many cases, the weak correlation between direct and indirect measurement of EC has reduced the accuracy and performance of the predicted maps. The objective of this research was to estimate soil EC based on a general linear model via using several soil properties. Through calibration equations, the error involved in such model-based data was calculated and employed in mapping soil EC using kriging with measurement errors (KME) method. The results were then compared with those of ordinary kriging (OK) and co-kriging (CK). Soil samples were taken from the depth of 0-20 cm in 78 points with spatial intervals of 500 m from an area of 40 km2, and they were analyzed for their electrical conductivity (EC) and certain other soil properties. Measured soil EC data (hard data) and auxiliary soil data were further used to develop the semi-variance and cross-semi-variance functions; moreover, soil salinity prediction was done on a grid of 100 m with OK and CK methods. Afterwards, the most optimal EC estimation model was developed using auxiliary soil data and GLM. As predicted values always involve uncertainty, the error involved with the predicted values was calculated and then the calibration equations were adjusted. Lastly, soil salinity was predicted using KME method. Results showed that the OK method had the lowest MSE and RMSE values, 0.65 and 0.8 dS m-1, respectively. Furthermore, among the auxiliary data, pH and silt content resulted in some of the best cross-semi-variance functions, among which, silt had a better performance regarding the spatial prediction of soil EC. The GLM model developed with the calculated error and KME resulted in predictions close to those of OK method (with MSE and RMSE of 0.74 and 0.86 dS m-1, respectively). KME method provided the possibility of merging error resulting from the use of soft data, derived from prediction equations; therefore, it successfully improved the spatial prediction of soil electrical conductivity