400 potty TTI TY TITTY 77 350 -- (a) Station t 2 = sensor 4 300 E - - —- ATMAS FJ caiculation 4 250 E a a e Grab C 200 _ samples 150 — a a e 100 E °F 3 70 1 5 0 4 . ; . tittiritt q a td: 250 mrt ttt T7710 ce & (b) Station 2 Oo & - 4 7 = 200 E _ ~"J 150 — 7 100 F 4 a 50C 0 “T : . 4- ritisyi Vit 0 FAG ° 50 100 Time — 5s Li 150 Comparison be- tween ATMAS calculations and measured gas con- centrations (a) at station 1 and (b) at station 2 in thefirst 5-m?spill experiment. Dots represent the results of mass-spectrometric analysis of grab samples taken by opening evacuated bottles at intervals during the experiment. ' Figure 3 shows the gas concentration data from stations 1 and 2 (the two stations inside the pond) for spill 1. The grab sample results the LNG spill experiments. We used this code to help overcome difficulties encountered in analyzing concentration data that were (dots) agree very well with the Shell significantly affected by changesin gas sensor output (black trace} in the wind field and to evaluate the both cases, indicating that the sen- _impact of these changes on plume movement. sors were working properly. The ATMAScalculation (colored line) agrees well, both in general and in Experimental results The sensorarray for the 5-m° detail, with the sensor outputatstation 2 (Fig. 3b). At station 1 spill experiments was designed to (Fig. 3a) there was.réasonably good evaluate instruments and measurement techniquesfor larger spill exgeneral agreement. However, the pernments, not to reconstruct gasdetailed structure of the experimental data was not reproduced by cloud-concentration contours. ATMAS.This structure is probably However, we were interested in im- due to local wind variations. Duringspill 2 there wasa shift in the wind direction at the far end of the detector array while the vapor cloud was dispersing. Figure 4 consists of three selected markerparticle plots from the ATMAS calculation (at 10, 50, and 90s), showing the kink that this wind shift produced in the developing vapor plume. Although there is some un- certainty as to the exact cause Of the wind shift, we believe that it was due to the presence of a hill immediately downwind from thespill pond, The ATMAScode hassignificant deficiencies when applied to an LNG cloud dispersing in the atmosphere; for example, it completely neglects the effects of density and temperature variations within the cloud. However, it does attempt to simulate the timevarying wind fields encountered in proving our computer modeling and analyzing the fragmentary data obtained to extract whatever new information it might contain on LNG boiloff and dispersion. We were able to estimate the dispersion coefficients through a point-topoint comparison with data from certain stations. We did this in the horizontal plane by repeated calculations, gradually increasing the horizontal dispersion coefficient until the gas just reached those stations that were at the edge of the plume. We foundthat the horizontal dispersion coefficient obtained in this way correspondedto stable conditions, while the atmosphere outside of the cold gas plume was unstable. (Unstable conditions imply a larger dispersion coefficient than do stable conditions.) We determined the vertical dispersion coefficient in much the same way from data taken at two heights at stations 3 and 6. Vertical dispersion corresponded to stable conditions and was nearly identical 32