Spatial Prediction of Soil Salinity Using Kriging with Measurement Errors and Probabilistic Soft Data

Abstract

In this study it is shown how kriging with measurement errors (KME) is useful as opposed to more conventional kriging methods. The goal of the study was to properly account for field measured soil electrical conductivity (EC) as soft data for the spatial prediction of soil salinity. Samplings were done in autumn 2009 (first dataset), spring and autumn 2010 (second and third datasets) around Uromieh Lake, northwest of Iran. The salinity was measured both in the field and laboratory for the first and second datasets. The first dataset was used for error measurements from which an error variance can be estimated. The measured errors were then used for characterizing probabilistic type soft data using the second dataset. The KME with only soft data (SKME), KME with both soft and hard data (HSKME) and ordinary kriging methods were compared. Validation criteria, mean error (ME) and mean squared error (MSE) were used for comparing the methods. Finally, the SKME method was applied as a way of improving the salinity prediction for the third dataset where only field measured soil salinity data were available. Comparing different kriging methods, Ordinary Kriging (OK) showed the best results among the comparing methods with ME and MSE equal to −0.12 and 0.55 respectively. SKME with ME equal to −0.13 was slightly different from OK and SKME with ME equal to −0.24 resulted in more bias predictions among others. KME method has shown to be useful for soil salinity monitoring and can effectively reduce sampling time.

Keywords:

• electrical conductivity
• kriging
• probabilistic soft data
• soil salinity
• Urmia lake

Notes

N: number of samples; Min: minimum value; Max: maximum value; NA: not available.

r: correlation coefficient; ME: mean error; MSE: mean square error; OK: ordinary kriging; HSKME: hard-soft kriging with measurement error; SKME: soft kriging with measurement error.