MODELING OF ANISOTROPIC SEMIVARIOGRAM USING ITERATIVE REWEIGHTED LEAST SQUARES AS A PARAMETER ESTIMATION METHOD (CASE STUDY: HEAVY METAL CONTENT OF FE, MN, AND PB IN WELL WATER IN BANDUNG REGENCY)

• Main Author: Mirratinia
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84070
• Institution: Institut Teknologi Bandung

Well water is one of the clean water sources used by residents of Bandung Regency in their daily lives. The heavy metals content in well water that is consumed can affect their health if it exceeds the maximum levels stated in the Regulation of the Health Ministry of Indonesia 2023 number 2. Water flows from one place to another indicating that the heavy metals content in it could have a spatial relation which can be influenced by distance and direction. The analysis that can be used to see the relation between locations is an anisotropic semivariogram. The data studied is heavy metal content data at 181 locations spread across 7 districts. This study aims to build the best anisotropic semivariogram model using Iterative Reweighted Least Squares as the parameter estimation method which is a weighted numerical solution for data with outliers. The experimental semivariogram was calculated using the Cressie-Hawkins approach and the distance lag classified by the Sturges and Scott rules. There was a significant decrease in the value of the Sturges rule which can reduce the modeling accuracy, so only the result of the Scott rule is modeled with the theoretical semivariogram of the Exponential and Gauss models. The best anisotropic model for the Fe and Pb metal variables was the geometric Gauss, whereas for the Mn metal variable was the zonal Gauss with each Sum of Squared Errors being smaller than the Exponential model. The best model was applied to interpolate values at three unobserved district points using Ordinary Kriging because the average was unknown. The interpolated values of the three variables were within the range of the actual data and did not differ significantly from the surrounding observed values.