The paper addresses the possibilities of spatio-temporal statistical modelling of basic hydrophysical and meteorological parameters of sea surface layer in the south-eastern Baltic Sea, Curonian Lagoon. The aim of the paper is to compare two methods (multivariate linear regression and regression kriging) for the analysis of changes and trends of ice phenomena, their dependence on changes in the air temperature, sea surface temperature and water salinity. The prediction of ice conditions for several locations at different distances from the reference sites shows that spatial information is an extremely important factor in making forecasts. The application of the regression kriging is more efficient than the multivariate linear regression for predicting the ice phenomena in semi-enclosed basins and lagoons.