Machine learning-enabled weather forecasting is revolutionizing the prediction of radioactive transport and contamination in real-time, offering effective responses to nuclear accidents.
Accurate and rapid predictions of radionuclide transport in the atmosphere are crucial during emergencies. However, real-time forecasting has been a challenge due to the computational intensity of numerical weather simulations required for meteorological predictions.
To address this challenge, researchers have developed statistical emulators for the Weather Research and Forecasting model (WRF). These emulators forecast wind and temperature fields around the source location by utilizing pre-computed WRF simulation results and boundary conditions from global circulation models (GCM).
Two methods have been explored: a deep neural network and a vector autoregressive model with exogenous variables (VARX). These emulators can rapidly predict local-scale wind and temperature fields for several days in advance by ingesting GCM-predicted boundary conditions available online.
The Gaussian plume model is then used to predict the distribution of radioactive material and dose in the environment based on these field inputs. The results demonstrate that both emulators are highly capable of forecasting the wind field in real-time without compromising accuracy.
The autoregressive model has shown strong performance in predicting the temperature field, attributed to the temporal autocorrelations present in the temperature data. This innovative framework provides a powerful tool for informed decision-making during emergencies, enabling meteorology-informed timing of containment venting and real-time guidance for evacuation strategies.
