A probabilistic framework encasing a model for aerial transport of pollution

Project description

Risk analysis after a chemical or radiological incident in the Netherlands is supported by an atmospheric dispersion model, NPK-PUFF. This model currently gives a deterministic forecast of the risks in the atmosphere and on the ground, caused by the incident.

The deterministic nature of the simulation has a number of drawbacks, all relating to  uncertainties about circumstances. For example, in the case of a large chemical fire, not only uncertainties about how much is released, of what substances, how high in the air, chemically stable or active, etcetera, but also uncertainties about weather, about the time of extinguishing, and about changes in release (e.g. smoldering, explosions) during the fire.

StochasticPuff is the result of the project ‘Quantification of uncertainty in radiological risk assessement after radiological and nuclear accidents’, funded by RIVM, the National Institute for Public Health and the Environment. This is a follow-up project of the PhD research of Paul Hiemstra. His proof-of-concept of MC-simulation and data assimilation for uncertainty assesment in atmospheric dispersal modelling is the basis for our implementation of StochasticPuff for MC simulation and Particle Filtering at RIVM in operational settings.

The StochasticPuff framework uses the PCRaster Modelling Framework to assign uncertainties to input variables for NPK-PUFF. After the simulation, results are summarized for visualisation in Aguila. Aguila is capable of showing maps of estimated values and uncertainties, timeseries, probability charts, maps of estimated percentiles and  confidence intervals. We demoed StochasticPuff as part of a lecture at NMDC about NPK_PUFF, and the treatment of uncertainties and the communication of uncertain results were welcomed as a ‘must-have’ property. We now also employ StochasticPuff as a tool to investigate the data and model requirements for decision support at different stages after a chemical release. The picture below reflects the dispersion of a hypothetical ‘unit release’ under different scenarios of data availability and uncertainty, visualised in Aguila.

StochasticPuff screenshot


Arien Lam, Derek Karssenberg, Kor de Jong, Paul Hiemstra, Arjan van Dijk (RIVM).


Hiemstra, P.H., (2011) Ensemble modeling and statistical mapping of airborne radioactivity, PhD thesis, Utrecht Studies in Earth Sciences 001, ISBN  978-90-6266-278-4. PDF