Abstract
Accurate spatio-temporal modeling of regionalized data is indispensable to environmental problems such as water and air pollutions. This modeling can contribute to specify influence factors and clarify mechanism of the worsening of environments. Geostatistics is adopted to construct a distribution model by considering spatio-temporal correlation structures included in the regionalized data. Data of sea surface temperature (SST) and concentration data of four nutritive salts, NO2-N, NO3-N, NH4-N, and PO4-P, in the Ariake Sea, southwest Japan, were chosen for a case study of geostatistics to environmental problem in closed seas. These data were sampled monthly between 1975 and 2000 at the 43 stations over the Ariake Sea. The first examination is proper modeling of spatio-temporal semivariogram of univariate on which estimation accuracy of kriging strongly depends: piecewise polynomial model using smoothed cubic spline was demonstrated to be superior to product-sum model, because it could follow complicated changes in experimental semivariogram. The difference in spatio-temporal correlations between the SST and nutritive salts was clarified by the semivariogram models. Estimation accuracy of ordinary kriging was checked through the cross-validation for all the sample data, which proved high estimation accuracy for the SST and acceptable accuracy for the nutritive salts with considering only the space direction. Temporal changes of the SST and four nutritive salt concentrations in summer and winter were depicted with clarifying the characteristics of fluctuation patterns. One possible influence factor for sea environmental change is decline of nitrification activity in the northern part of Ariake Sea in winter, because the rations of NO2-N/NH4-N and NO3-N/NH4-N have seemed to decrease from 1997.
