A sufficient historical record in grid-based O3 modeling now exists to
permit assessment of its role and limitations in policy analysis. In
this presentation, several topics are examined: past experience in mod
el evaluation and use, the present status of modeling, the value of mo
deling in policy analysis, and key issues and future needs. First, the
role of grid-based O3 modeling in policy analysis was assessed throug
h a case study of nearly 10 years of model applications in the South C
oast (Los Angeles) Air Basin. Changes in quality of performance and in
degree of acceptance of grid-based models (in policy analysis) with t
ime were analyzed and compared. Degree of acceptance appears to depend
on a variety of factors, including level of understanding and familia
rity, perception of need, and relative degree of acceptability, as com
pared with other available models. Improvements in quality of performa
nce with time were limited (and, in any event, knowledge of such chang
es were not available), and thus this factor seems to have had little
or no role in influencing model acceptance. Second, the current state
of modeling, in terms of both science and art, is appraised, consideri
ng the state of knowledge, availability of data bases, adequacy of per
formance evaluation procedures, and quality of predictive performance.
Significant deficiencies have existed in knowledge and treatment of k
ey governing processes in models. Until recently, data bases suitable
for use in evaluating model performance were unavailable. Quality of p
erformance has not been sufficiently good to confer confidence in mode
ls' use in a regulatory environment. Evaluative testing did not provid
e sufficiently for 'stressing' models. For many, but not all of these
issues, attention is now being, or soon will be, given to alleviating
concerns. Third, the role and value of modeling in policy analysis is
discussed. Attention is given to the role of the modeling expert in th
e policy forum, as compared with the model itself, and with the consid
eration of modeling as a longer term process rather than as a direct o
r short term means of generating 'answers'. Finally, key concerns and
future needs are delineated, and some unsolved problems are discussed.