| Abstract [eng] |
The aim of this master thesis a stochastic forecast data and deterministic interpolation methods and compare them. Also, to confirm or refute the hypothesis for the Baltic Seas depth arrangement data accuracy between the two methods used is similar. The study analyzed 30 of the Curonian Lagoon and the Baltic Sea depth data obtained from the „Sea Research Center“ in Klaipėda . From this data, interpolation was performed by numerical methods and found the bottom of a digital terrain surface. To find the digital surface used gstat R program, Spatial, Akima, mba, fields, gss packages. Digital Baltic seabed surface is visualised and represented by using the interpolation: universal kriging, median polishing kriging and linear, cubic, cubic Akima and cubic B – spline methods. Interpolation is investigated to ensure the accuracy of the data: the trend type, semivariogram degree, type, and the relationship between the depth and coordinates. Used to assess the accuracy of akaike's information criterion and sum of square of errors methods. The best interpolation method is found using: the cross-validation (comparing mean square error, root mean square of forecast errors and standard deviation values), and numerical methods using ANOVA analysis (F-test comparing p values). After interpolation, and the numerical methods used for comparing the obtained results it was concluded that the Baltic Sea in the eastern part and western part of the Curonian Lagoon area is interpolated best with universal kriging (stating that the data exists in the third degree trend) and a natural cubic spline. The resulting models of the accuracy and surface relief in both cases was almost identical, and therefore confirms the hypothesis that the correct data arrangements, both the accuracy of the methods is similar. |
