Many earthquakes in Indonesia have caused a large number of fatalities. Disaster risk-reduction of fatalities requires a representative fatality model derived from fatality data caused by historical earthquakes in Indonesia. We develop an empirical fatality model for Indonesia by relating macroseismic intensity to fatality rate using compiled subdistrict level fatality rate data and numerically simulated ground shaking intensity for four recent damaging events. The fatality rate data are compiled by collecting population and fatality statistics of the regions impacted by the selected events. The ground shaking intensity is numerically estimated by incorporating a finite fault model of each event and local site conditions approximated by topographically-based site amplifications. The macroseismic intensity distribution of each event is generated by using ShakeMap software with a selected pair of ground motion predictive equation (GMPE) and ground motion to intensity conversion equation (GMICE). The developed fatality model is a Bayesian generalized linear model where the fatality rate is assumed to follow a mixture of a Bernoulli and a gamma distribution. The probability of zero fatality rate and the mean non-zero fatality rate is linked to a linear function of shaking intensity by the logit and the log link functions, respectively. We estimate posterior distribution of the parameters of the model based on the Hamilton Monte Carlo algorithm. For validation of the developed model we calculate fatalities of the past events from the EXPO-CAT catalog and compare the estimates with the EXPO-CAT fatality records. While the developed fatality model can provide an estimate of the range of fatalities for future events it needs on-going refinement by incorporation of additional fatality rate data from past and future events.
