Detecting salinity in early stages using electromagnetic survey and multivariate geostatistical techniques, A Case study of Nong Suang district, Nakhon Ratchasima, Thailand

Detecting salinity in early stages using electromagnetic survey and multivariate geostatistical techniques. This research is aimed at increasing the accuracy of detecting soil salinity in early stages of the process. This phenomenon develops at depth; optical remote sensing (e.g. satellite images) cannot be useful in detecting subsoil variability. Therefore Electromagnetic (EM) survey, using EM-38 sensor were used to image the subsoil variability, and answer three questions: (1) How much is EM-38 sensor reliable to predict salinity profile? (2) Is the technique used to predict salinity profile robust for soil texture variation? (3) Which technique is more accurate to predict unsampled location univarite or multivariate geostatistics, using EM data as a co-variable? The research was conducted in Nong Suang district, Nakhon Ratchasima, Thailand in an area 7400 ha. The result of Aerial photo interpretation (API) was digitised after orthorectifying the interpreted aerial photos to prepare a base map. In forty-two locations ECe at 0-30, 30-60,and 60-90 cm depths, and the EM38 horizontal (H) and vertical (V) mode were collected for general salinity survey purpose. Two experimental plots; three-hectare each, representing homogenous sandy and clayey soil selected to conduct EM survey, with grid 2x5m. Both EM-38modes were collected. Seventy-five soil samples at the previous three depth were selected from EM survey grid with spacing 20x20 m to measure soil salinity (ECe), and to calibrate the geophysical data. The significance of stratifying homogenous saline soils using API was tested with (ANOVA) test, In sand and clay plots regression analysis was performed on 51 randomly selected points, while remaining points were used to calculate RMSE for validation. Using the same 51-point Ordinary (OK) and Regression kriging (RK) with EM data as a co-variable was applied to predict un-sampled locations in the sand plot only. Using ANOVA test (API) was successful to stratify the study area. EM data explained of ECe variation in top, middle and bottom layer 53%, 64%, and 46% (In sand plot), 32%, 42%, and 37% (in clay plot). Validation RMSE were 0.44, 0.69,and 0.65(ms/cm-1) for the same sequence of layers in the sand plot, while for clay set the values increased to 0.6, 1.6,and 1.8.RK-RMSE are 0.53 and 0.69 (ms/cm-1) for first two layers, while OK-RMSE are 0.77 and 0.83. Final conclusions are: (1) The maximum salinity variation that EM data can explain in the study area is 64% (2) H-mode is sensitive to salinity variation of the first 60 cm of the soil profile. (3) EM survey technique is sensitive to soil texture variation so texture stratification should be applied before conduct the survey. (4) RK is better than OK in predicting un-sampled locations, and a comparison with nonspatial regression is advisable before applying the multivariate technique.