The Jack Rabbit II (JR II) chlorine field trials in 2015 and 2016 involved nine 5–20 ton releases of pressurized liquefied chlorine from a tank mounted 1 m above a broad flat desert sand surface. A model comparison study was initiated, where 17 widely-used dense-gas dispersion models were run by scientists in seven countries.
Predictions were submitted following specified formats, using specified emissions and meteorology inputs. To compare with the model predictions, sets of observations were defined for the arc-maximum 1–3 s averaged concentrations (arc max C) and for cloud widths and heights (to 20 ppm and 200 ppm contours) at distances from 0.2 to 11.0 km from the release. The initial focus is on the three field trials (1, 6, and 7) that have the highest observed concentrations and that have detailed emissions information. It is found that these models are able to satisfactorily simulate (generally within a factor of two) the observed arc max C’s and their variation with downwind distance at this flat desert site. At each downwind distance, the scatter in the arc max C predictions covers about 1 ½ orders of magnitude, but the observed arc max C is within the range of the predictions. The median of the cloud width predictions is about 50% larger than the observed value for the three trials. The median of the cloud height predictions is within about 10% of the observed value. For both cloud width and/or height, there are a few models with large (factor of 3 or higher) overpredictions. Of the 17 models, when compared to observations, there is a core group of 5 or 6 with consistently (across all three trials and all distances) less mean error and scatter in their predictions of arc max C and cloud width and height. However, as a group, the 17 models are performing adequately (using the “factor of two” rule of thumb). An important caveat is that, at the JR II desert site, chlorine deposition is minimal. At a site with vegetation and/or organic-rich soil, the effects of removal of chlorine by deposition are expected to be significant.
Algorithms were developed to approximate the 3-dimensional internal packing density distributions of fibrous filter media at different levels of detail and complexity. Each algorithm uses certain input information derived from X-ray tomographic data of filter samples.
Algorithm 1 creates a binary media structure consisting of the true (i.e. tomographically determined) void distribution plus regions of uniform packing density. The average packing density of the media is maintained constant. Algorithm 2 creates a model fibrous structure of straight fibres of equal diameter and random length positioned randomly in space, while maintaining the true (i.e. tomographically determined) fibre orientation distribution. The number and length of the fibres on average adds up to the packing density of the filter. The model fibrous structure is recreated by a stationary Poisson process of convex bodies. Algorithm 3 distributes pores of random size and location within a homogeneous matrix, such that the average packing density again coincides with the true (i.e. tomographically determined) packing density. This algorithm is also based on a stationary Poisson process of convex bodies.
The capability of each algorithm to recreate the essential structural features of the media was tested against 'reality' by computing the respective overall pressure drop of the filter as well as the velocity distribution in the filter interior, and comparing with the results obtained for the 'true' packing density distribution of a sample measured by tomography. Compared to the assumption of a completely homogenous filter (which gives roughly 2 times the actual Δp), all algorithms are closer to reality. The binary algorithm deviates in Δp by a factor of 1.8; algorithm 3 comes within a factor of 1.6 of the true Δp. The best approximation is by algorithm 2 which narrows the difference in Δp to a factor of 1.4.