By Zhengkun Jiang.
Abstract:
In case of a nuclear accident, decision makers relied on high resolution and accurate information about the spatial distribution of the radioactivity levels in the surroundings of the accident site. However, the static nuclear monitoring networks, employed in many countries in Europe, were usually too course to reach the required accuracy. Therefore the Dutch authorities considered a strategy in which measurement density is increased in case of emergency by adding measurements from complementary mobile measuring devices. This raised the question where the mobile devices should be placed. This paper proposed a geostatistical methodology to optimize the allocation of the mobile devices, such that the expected weighed sum of false negative and false positive area (i.e., false classification into safe and unsafe zones) is minimized. The radioactivity concentration was modelled as the sum of a deterministic trend and a zero-mean spatially correlated stochastic residual, whereby the deterministic trend was defined as the outcome of a spatially explicit physical atmospheric dispersion model (PUFF). The PUFF model used meteorological data and the characteristics of the radioactive release as input. The residual was characterized by a semivariogram that was estimated from the differences between outputs from various PUFF runs with input settings reflecting the uncertainty in the PUFF inputs (e.g., wind speed, wind direction). Spatial simulated annealing was used to obtain the optimal monitoring design, whereby accessibility of sampling sites (e.g., distance to roads) was also included. The method was computationally demanding but results were promising and the computational speed may be considerably improved to compute the optimal monitoring network in nearly real-time.