Spatial Transferability Analysis of the Regional Automobile-SpecificHousehold-Level Carbon Dioxide (CO2) Emissions Models

This paper compares the performance of four methods for combining model information developed in one region to a model in another region to improve estimation results. The transfer methods compared are Naive, Joint Context Estimation, Bayesian Updating, and Combined Transfer Estimator. The application is for models developed to estimate household-level carbon dioxide (CO2) emissions from vehicle use developed with data from the 2009 National Household Travel Survey collected by the U.S. Department of Transportation. This is potentially very useful for small regions or communities that would like to quickly/easily estimate household-level CO2 emissions from vehicle use but do not have adequate data for developing their own model. This has the potential to minimize the need for large data collection and/or model estimation efforts. The transfer methods can incorporate model information from other regions to make up for the local travel data shortfall. The results show that automobile-specified CO2 emissions models can be transferred from one geographical region to another region and improve estimation results. The Combined Transfer Estimator method produces superior prediction performance, followed by Bayesian Updating, Joint Context Estimation, and Naive, in that order. The analyst, however, should determine whether the incremental benefits gained from the application of any transfer methods is worth additional computational investment. The results also highlight the potential and significance of regions with smaller sample size to combine travel data/parameters from another region to improve CO2 emissions estimation. These results can assist different agencies to develop a baseline CO2 emissions inventory from vehicle use.