The global warming potential (GWP) of a number of compounds used by U.S. industry is not known or poorly known. Working with the EPA, a set of nearly 1,200 compounds that are known or reasonable suspected to have a significant GWP has been developed and quantum chemistry calculations have been made on these compounds. Using the data from these calculations, the radiative forcing (RF) can be reasonably well established, and indeed these predictions compare favorably with other models. The atmospheric lifetime of these compounds, predicted based on the rate of reaction with OH, is much more difficult to predict. Using artificial neural network (ANN) software, we have developed a technique for predicting rates of chemical reaction with OH based on quantum chemical calculation of the reactant species alone. The ANN has been trained on more than 100 chemical reactions and used to predict OH reaction rates for more than 800 compounds on the EPA list. The fit produced by the ANN is of very good quality, and the results agree reasonably well with current models. We believe that the predictive power of the ANN for new compounds will exceed that of currently available models. Currently, we are working to extend the types of reaction on which the ANN is trained, and increase the robustness of the ANN.
- Use ANN methods to fit RF and OH rate constant data k(OH).
- Be able to predict RF, k(OH), and GWP with reasonable accuracy using readily-available information.
- Create a screening tool based on the ANNs.
- Improve the performance of rate constant prediction.
- Compare rate constant predictions to other models.
- Incorporate transition state theory estimates for atmospherically important reactions.
- Created predictive model for estimating RF values from simple, 2-d structural information.
- Created ANN model for estimating OH reaction rate constants.
Lead Organizational Unit: