VARIABILITY OF VARIOGRAMS AND SPATIAL ESTIMATES DUE TO SOIL SAMPLING - A CASE-STUDY

Measuring of electromagnetic soil conductivity (EMC) is used to follow the temporal evolution of soil salinity, because it is a rapid technique with a portable instrument. A test site of 1.2 km by 2.4 km is selected as a reference. The suitability of sampling schemes for monitoring soil salinity during the first years of irrigation is studied. An initial sampling consists of 17 rows and 33 columns of observation points 75 m apart, i.e. 561 data points, regularly spaced on 288 ha; it allows to determine the ''actual variogram'' of EMC on this site. This structure is complex, as is often the case in nature, with alternating strongly and weakly salted areas, due to small creeks across the site. The sample size required to get accurate estimates of soil salinity and to follow temporal changes is assessed by subsampling the data set. Simulation by repeated sub-sampling is preferred because of the complexity of the data structure. Five series of 20 sub-samples are randomly taken from the initial sample, with size 50, 75, 100, 150 and 200 data points, respectively. For each sample size, the analysis consists in computing the mean squared error of the 20 sub-sample variograms relative to the ''actual sample variogram''', and similarly for the fitted models. Confidence limits for the theoretical variogram were directly estimated from the 20 fitted models. Finally, the effect of this uncertainty is studied by comparing kriged estimates with observed values. The consistency of both experimental and fitted variograms increases with sample size. In this case, a choice of 150 data points appears to be consistent. Despite a large variability in experimental variograms, the fitted models and the kriged estimates are not so different as could be expected.