MULTIVARIATE SIMULATION AND ASSESSMENT OF THREE DIMENSIONAL SPATIAL PATTERNS OF COASTAL SOIL SALINITY USING ANCILLARY VARIABLES

Rapid and reliable spatial estimates of field soil salinity aid farmers and researchers in understanding the development of salinization and identifying areas in the field where special management practices are required to control its speed. Apparent soil electrical conductivity, measured by electromagnetic induction instruments, has been widely used as an auxiliary variable to estimate spatial distribution of field soil salinity owing to the high-efficient and non-invasive nature. Methods used for this purpose are mostly a series of kriging-based algorithms. This study adopted stochastic simulation algorithms, sequential Gaussian simulation (SGS) and sequential Gaussian co-simulation (SGCS), to generate equiprobable realizations of soil salinity patterns, to evaluate the uncertainty associated with those patterns, and to determine which algorithm is more reliable for the local and spatial uncertainty assessment. Results showed that experimental semivariograms were generally isotropic at all directions with the identical nugget, sill and range parameters despite the limited data in X and Z directions. Nugget contributions comprised from 19.5 to 40.1% of model sill values at the horizontal orientation, which negligibly ranged from 2.1 to 6.0% at the vertical orientation. The spatial patterns of soil salinity generated by SGS and SGCS algorithms showed consistency with the actual values. It was found that the distribution of soil salinity was generally normal and there existed the probability of secondary salinization across the spatial domain. The estimation precision of SGS algorithm was superior to that of SGCS algorithm. In addition, accuracy plots and standard deviation maps both indicated that SGS approach performed better in terms of modelling local uncertainty and propagating the spatial uncertainty. Thus, the SGS method was suitable for the spatial estimation of the soil property, distribution type and data arrangement in our study. We concluded that the SGS algorithm with data conversion (converting EM measurements at unsampled sites to soil salinity values) prior to simulation modeled the uncertainty better than the SGCS algorithm (soil salinity as primary variate and EM measurements as covariate)