Because extreme precipitation events have the potential to cause severe damage and economic loss, many studies have searched for trends in heavy precipitation across the United States. However, these previous analyses have either failed to include common regional information that could reduce uncertainty and improve trend detection or have only focused on small and homogeneous regions.
To better identify significant spatial trends—especially across larger areas—Sun and Lall have developed a new statistical “clustering” model that analyzes multiple sources of data to identify statistically homogeneous areas, group these data together, and share chosen model parameters within each group. Because this new approach greatly reduces estimation uncertainties, it improves the detection of statistically significant trends at regional and local scales.
The team used this model to search for trends in the maximum daily precipitation across the United States each year from 1941 to 2010. After analyzing data collected from 90 HadEX2 weather observation stations, the researchers identified statistically significant trends of increasing precipitation in the Midwest, the Northeast, and the northern reaches of the Southeast. Although these trends are consistent with the results of a number of other studies, the new model is much better at discerning them, the authors report. In this analysis, all 14 stations in the Midwest displayed a significant statistical trend, compared to only 4 stations when a previous single-site analysis is used.
This new approach results in predictions that are more precise, including for extreme events. The authors argue that the reduction in model uncertainty is especially important for engineering design standards and for improving the seasonal forecasting of rare events, including those related to the El Niño–Southern Oscillation. (Geophysical Research Letters, doi:10.1002/2015GL066483, 2015)
—Terri Cook, Freelance Writer
Citation: Cook, T. (2016), Improving the identification of extreme precipitation trends in the U.S., Eos, 97, doi:10.1029/2016EO050215. Published on 14 April 2016.