SEMIVARIOGRAM MODELLING USING PARAMERE ESTIMATION ORDINARY LEAST SQUARE (STUDY CASE: BUILDING LAND COVER DATA IN EAST JAVA PROVINCE)

• Main Author: Nurkholik Adi Sunaji, Rahmat
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/79808
• Institution: Institut Teknologi Bandung

Semivariogram is one of statistics method that useful to modelize spatial correlation between pairs of locations. East Java Province is second most popolus province in Indonesia after West Java Province. Geographically, East Java province consists of East Java mainland and Madura archipelago. East Java’s economic growth is 3.57% as of 2021, with total of Gross Regional Domestic Product (GRDP) of 2,454.5 trillion Rupiah. Land cover for building data in East Java Province can be spatially modeled using a semivariogram model. This research aims to build an isotropic semivariogram model based on land cover data in East Java Province, then carry out Ordinary Kriging interpolation to predict variable values at unobserved points. There are 9 variables in this data, including coordinates, physical factor data, and socio-economic factor data. From this data, samples were selected randomly and then calculated its experimental isotropic semivariogram values. After that, the calculation results were adjusted to the theoretical semivariogram model using the Ordinary Least Square (OLS) parameter estimation method. The theoretical semivariogram results are then used to predict variable values at unobserved points using the interpolation Ordinary Kriging. The results of this kriging can later be used to estimate the value of land cover for building data variables in East Java Province. From the research results, the best semivariogram model used to model elevation, slope, density and distance to main roads in East Java is the Gauss model. This model provides value Sum of Squared Error (SSE) smaller compared to the exponential and spherical models. Next, the model results were re-examined using the Ordinary Kriging method to measure the values of 4 variables at 22 points. The kriging estimation results were then compared with the actual data and it was found that 61 of the 88 kriging data results were over estimated. This illustrates that most of the model calculation results produce predictions that are greater than the actual value.