Model Report: Typhoon Impact Model
This brief report provides a peer review of the trigger mechanism – developed by the Netherlands Red Cross 510 team – used in the anticipatory action programme responding to typhoons in the Philippines. The trigger utilizes typhoon track and rainfall forecasts to predict the percentage of houses severely damaged per municipality; this model predicts for the municipalities of Bicol and Eastern Visayas. The technical review analyzes the: 1) intended use (recommendation: make a clear statement about the model combinations for probability and threshold), 2) model development (recommendation: more information about input sources and how often data is updated), 3) model evaluation (recommendation: specify acceptable variance and how robust model performance is over different typhoons), and 4) operational readiness (recommendation: for every model version, make a model release). The ethical review addressed 1) inaccuracy (recommendation: report latest error scores, contextualize the quality of error scores, and include rationale for calculation), 2) false negatives (recommendation: make the possibility of a false negative clear), 3) missing attributes (recommendation: clarify importance and consistency of input variables), 4) lead times (recommendation: more clarity on potential operational impacts of long lead times), 5) decision support (recommendation: clarity about weightage of model output on trigger activation and details on operational readiness in case of model failure), 6) statistical bias (recommendation: clear documentation of the different datasets, their rationale, and parameters), 7) lack of transparency (recommendation: better documentation of the modeling process), and 8) systemic bias (recommendation: clarity on parameters like social vulnerability and clarity in weights on variables when ranking).
August 31, 2022
PDF, 238.60 KB
OCHA / Centre for Humdata /510 Initiative
Briefing Sheet / Fact Sheet, Report
Cyclone / Typhoon / Hurricane