ISOTROPIC SEMIVARIOGRAM MODELING WITH KERNEL APPROACH (CASE STUDY: KALIMANTAN PEATLAND DATA)

• Main Author: Dhiya Rachmy Ariffia, Andi
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81906
• Institution: Institut Teknologi Bandung

Indonesia has the largest peatland in the tropical zone in the world with an area of 21 million hectares. As much as 32% (5.76 million hectares) of the peatland is on the island of Kalimantan. Peat is a type of soil formed from piles of plant residues that are undergoing and/or have undergone a decomposition process. Peat contains a lot of carbon elements. Therefore, degraded peatlands become vulnerable to fires. The level of availability of combustible peat fuel is affected by the moisture content of the peat soil. Peat moisture is affected by the groundwater level (TMA) of peat. Peatland drought due to low TMA triggers fires. Peat data including TMA variables were spatially analyzed using geostatistics. An isotropic semivariogram is a statistic used to model the spatial relationship of data between locations that only depends on the distance of the location pair. The purpose of this study is to determine the experimental semivariogram of peat TMA variables with the Matheron and Cressie-Hawkins approach and determine the appropriate isotropic semivariogram model. The semivariogram model uses a nonparametric approach with a kernel approach using the Nadaraya-Watson estimator, with Gaussian and Epanechnikov kernel types. Nonparametric statistics are used because the semivariogram points are few and their distribution is unknown. The optimal kernel bandwidth value is selected by the leave-one-out cross validation method. Modification of many distance lags is done by dividing Scott's class into 2 classes of equal width. The best model for Matheron's semivariograms was a modified Gaussian kernel with a bandwidth of 0.47 (MSE: 1.9145), while for Cressie-Hawkins it was a modified Gaussian kernel with a bandwidth of 1.3 (MSE: 0.002421). In general, Gaussian kernels are better because they produce smooth curves. Both models were used to predict the TMA values in three unobserved locations by the ordinary kriging interpolation method because the mean was unknown.