The most important parameter in the kernel density estimator is the bandwidth or spread or window width. The bandwidth of the kernel density estimator, which follows the power law, is determined using the nearest neighborhood technique for the earthquake catalog which is divided into bins. For reliable hazard estimates, the magnitude bins used in developing the power law and estimating the spatial activity rate density function should be the same. It is important that consistency be maintained between the earthquake epicenters used in determining the bandwidth and the epicenters to which the bandwidth is applied subsequently. In this paper, the effect of epicenter data inconsistency on hazard estimates for various return periods for Chennai is evaluated. Two methods of binning are used, one in which the epicenters used in deriving the bandwidth is in line with the epicenters used in arriving at the spatial activity rate and the other where the epicenters used in deriving the bandwidth are just grouped by dividing the catalogue into equal bins. Seismic hazard estimations are compared using these two approaches of forming the magnitude bins for Chennai, Tamil Nadu, India. The peak ground acceleration (PGA) values obtained from Binning Methods 1 and 2 for 475 years return period are 0.0955g and 0.0802g respectively. The difference in PGA and peak spectral acceleration (PSA) from the two binning methods ranges from 20 to 10% with respect to Binning Method 1 for the return periods of 72 to 2475 years.
Some Improvement on Convergence Rates of Kernel Density Estimator
