Order and spatial dependence in risk analysis


Modelling order and dependence

In hazard vulnerabiltiy analysis, fragility curves based on the lognormal cumulative distribution are often used to model the relation between hazard intenisty and damage due to their empirical performance and theoretical properties. In this two-part project, we show how accounting for the ordering and spatial dependence in damage data improves damage models and regional portfolio loss estimation.


We compare the conventional approach of fitting such fragility curves per damage grade via separate probit regressions against fitting them simultaneously using an ordinal model.


We model the spatially correlated differences in modelled and measured hazard intensity and the residual spatial correlation through a spatial ordinal model.


Risk Analysis publication

Comparing the ordinal and  nominal approaches to fragility curve fitting.

GitHub code repository

Accompanying code for the analysis conducted in the Risk Analysis publication.

EOS Annual Meeting 2022 poster

Presented to members of the Earth Observatory of Singapore (EOS). 

GYSS 2022 presentation

Short talk on modelling spatial dependence, its applications in public health and disaster risk analysis, and the pitfalls of neglecting it in model calibration. Presented at the 10th Annivesary Edition of the Global Young Scientists Summit in January 2022. 

To come:
– Paper on the spatial ordinal model and its implications for fragility curve fitting as well as regional portfolio estimation. 
– R package containing functions and demos for fitting ordinal fragility curves and spatial ordinal models.