Evaluation of ground-motion models for the Himalayan region

Uncertainties associated with empirical ground-motion models (GMMs) are a crucial component of any probabilistic seismic hazard analysis (PSHA) study. Conceptually, such uncertainties are divided into two components, that is, aleatory uncertainty and epistemic uncertainty. The former represents inherent randomness in ground-shaking, while the latter refers to that arising from limitations in knowledge and incomplete data. In regions like the Himalayas, where data availability is limited, constraining epistemic uncertainty (as well as the aleatory uncertainty) becomes a major challenge. This becomes an issue for the scenarios that dominate hazards at major urban centres across the Indo-Gangetic plains of northern India. In this study, we investigated these components of uncertainties using a recently compiled strong motion data set provided by PESMOS. The dataset involves around 600 records from approximately 140 earthquakes along the Himalayan arc. For our evaluation, we selected 13 GMMs developed for the Himalayan region. The evaluation of GMMs involves three major steps: 1) Model visualization in terms of median comparison (between-model uncertainty), 2) Residual analysis, and 3) Evaluation with NGA West2 models. Our analysis indicates that the existing Himalayan GMMs, when compared with the most recent compiled dataset, exhibit a significant bias both with magnitude and distance ranges. Importantly, there are strong differences in the median magnitude and distance scaling of the median models, which indicates large epistemic uncertainty. Additionally, the model bias is seen increasing towards longer periods. Overall, our analysis demonstrates a large aleatory and epistemic uncertainty associated with the Indian GMMs. Additionally, for completeness, we evaluated global models such as NGA-West2 (Next Generation Attenuation) for the Himalayan region. This analysis indicated that such global models might not be directly applied to the Himalayan conditions without adjustments and that they do not capture the range of epistemic uncertainty that is implied by Indian GMMs.