Impacts of assimilating observations from connected vehicles into a numerical weather prediction model
Numerical Weather Prediction (NWP) is a method of weather forecasting using equations to describe atmospheric behavior. However, it requires high-density surface observations to function best. A paper from researchers at the National Center for Atmospheric Research in Boulder, Colorado proposes the use of Connected Vehicle (CV) data to fill gaps in weather observation networks.
The researchers performed a pilot study, using a simulated vehicle probe dataset to compare with the Weather Research and Forecasting (WRF) model for selected case studies. Simulated data was necessary as the current scale of CV deployment would be insufficient to improve weather calculations. CV market penetration rates of 5 and 30 percent were both tested, and compared to a baseline model that did not use any CV-gathered data.
To verify the accuracy of the results, the modeled outputs were compared to rader-derived observations.
In general, the WRF model was improved by the additional CV-derived data.
The paper noted that in most instances, the bias-corrected root mean squared error (BCRMSE), a measure that represents the standard deviation of errors and is appropriate for determining the spread of error, was improved by between 1 and 3 percent across the different outputs studied. These improvements tended to be flat over time, meaning there was not a strong dependence on lead hour.
- CV-augmented data resulted in performance improvements of up to 8 percent compared to the baseline when used to predict temperature, though this was not sustained over the whole period of the model. In general, the augmented data had the largest positive impact on temperature and windspeed predictions.
- Data from the 5 percent CV penetration simulation was found to have a negative impact on the accuracy of dewpoint temperature predictions.