Long-term Changes in Warm Season Convective Storm Frequency Over the Northeastern United States
Warm season (April-September) composites were calculated for all quantities to understand spatial variations in changes in convective precipitation over the NEUS. Convective frequency is greatest across inland regions, especially southern Pennsylvania, and shows a sharp decrease along the immediate coast. The annual number of convective storm days has shown a significant (p < 0.01) decrease in southwestern Pennsylvania from 1979-2010, while the interannual trend in convective frequency becomes increasingly more positive towards the coast, where there has been a modest increase. Total warm season convective precipitation has increased significantly over the Adirondack Mountains and eastern New England.
The decrease in convective activity west of the Appalachians was accompanied by a substantial decrease in convective available potential energy (CAPE) and low-level equivalent potential temperature, while the coastal increase in convective frequency was associated with a significant increase in warm season CAPE, largely the result of increased in low-level moisture. The lack of a statistically significant increase in storms over NYC/Long Island despite environmental ingredients that would strongly favor more storms suggests that there are fewer convective systems moving towards the coast from west of the Appalachians.
Using linear discriminant analysis (LDA) to create a single threshold line that attempts to predict convective storm days based on reanalysis parameters over a large region, as other studies have done, failed to capture significant spatial differences in convective frequency over the NEUS in this study. However, using separate thresholds for smaller (1-degree lat/lon) domains better captured the coastal minimum in convective frequency and decreased false alarm rates, illustrating that localization is an important consideration for utilizing this LDA approach in future climate studies of convection that use climate model output.