Kernel Density Estimation for Seismic Hazard Mapping in Indonesia: Influence of Kernel Function, Bandwidth Size, and Grid Resolution

Identifying earthquake-prone areas is critical for disaster mitigation to reduce casualties and economic losses. This study applies Kernel Density Estimation (KDE) to analyze and rank earthquake-prone regions in the context of seismic hazard mapping, focusing on variations in kernel functions, bandwidth sizes, and grid resolutions. The Indonesian Earthquake Catalog (1964 - 2023) is used as a case study. The results indicate that different kernel functions have unique strengths. The Epanechnikov kernel provides an even density distribution, particularly in low and medium categories, while the Gaussian kernel captures high concentrations effectively, especially in high and extreme categories. The Biweight kernel performs well in medium and high categories but less effectively identifies extreme density concentrations. Grid resolution also significantly impacts results; smaller grids 0.25o 0.25o reveal detailed density patterns but may overemphasize localized concentrations, whereas larger grids 5o 5o are suited for macro-scale analyses but can obscure finer variations. Bandwidth size selection significantly affects density estimates. Smaller bandwidths (0.1) spread density widely, resulting in many grids in the low category but fewer in the medium, high, and extreme categories. Medium bandwidths (0.3) increase the proportion of medium and high categories, while larger bandwidths (0.5) produce the highest proportion of grids in the medium category, though extreme values remain limited. These variations demonstrate how bandwidth choices influence the balance between localized detail and broader distribution patterns. KDE effectively identifies earthquake-prone areas with varying densities and cluster significance, providing essential insights for disaster mitigation and spatial planning. The Gaussian kernel, 0.5 bandwidth, and 1o 1o grid combination yields the most significant results in mapping earthquake risk. However, the study has some limitations, including sensitivity to dataset completeness, parameter selection, and the smoothing effect of KDE that may underrepresent low-frequency events, particularly in sparse-data regions, which may introduce uncertainties in density estimations. Future studies could explore adaptive bandwidth selection and refined spatial resolution to enhance the robustness of KDE-based earthquake hazard mapping. HIGHLIGHTSThis research investigates using Kernel Density Estimation (KDE) to map earthquake-prone areas in Indonesia. The primary findings include identifying and rating seismic risk zones using KDE, a statistical method for spatially representing earthquake density. KDE allows for a more detailed visualization of earthquake-prone zones by analyzing differences in kernel functions, bandwidth size, and grid resolution. This method permits seismic risk assessment using historical earthquake data, providing valuable insights for disaster mitigation and spatial design. GRAPHICAL ABSTRACT