| Abstract [eng] |
Master thesis “Baltic Sea Data of Depth and Salinity with Non- Stationary Average Modeling and Prediction” the aim is to choose the best prediction method of depth and salinity by using scientific literature for Baltic Sea region: 15 – 25 degree of east longitude and 54 – 66 degree of north latitude. Master’s thesis consists of three parts: spatial data, review of modeling and prediction methods, prediction methods analyze and choice of the best depth and salinity prediction at before mention Baltic Sea region. At first work part is generalize the modeling and prediction methods of spatial data. Was noticed spatial data modeling stages: spatial data analyze, mean’s and semivariogram’s models. Last stage is used for geostatistical data. There are widely written about interpolation methods of kriging and spline. Also in this part is attention is put on non-stationary mean’s models as Trend Surface and Regression. The main aim of this part is to do analysis of literature. In the second part is made analyze of mean’s models usage at prediction models. There are written about stationary mean’s model usage for simple and ordinary krigings and B-spline, also about non-stationary mean’s models fit for universal, regression, median krigings, and universal cokriging and for linear, quadratic and thin plate splines. There was observed, that different from kriging method) at spline method variables didn’t belong from covariate and degree of trend. The main idea of this part is to make methodological review of prediction methods. In the third, practical part is trying to show which interpolation method: universal, regression krigings, universal cokriging and thin plate spline the best fits for depth and salinity prediction. The prediction is made by R system by using for gstat, fields and spatial package. In the conclusions are generalize the main stages of spatial data modeling, the difference of kriging’s and spline’s methods and they practical usage at R system’s packages. Also is present the best prediction method for before Baltic’s Sea region: universal cokriging the best fits for initial depth data, for the depth data without divided points best fits universal kriging and for salinity data – thin plate spline. |
