Detection and attribution of past changes in cyclone activity are hampered by biased cyclone records due to changes in observational capabilities. Here, we relate a homogeneous record of Atlantic tropical cyclone activity based on storm surge statistics from tide gauges to changes in global temperature patterns. We examine 10 competing hypotheses using nonstationary generalized extreme value analysis with different predictors (North Atlantic Oscillation, Southern Oscillation, Pacific Decadal Oscillation, Sahel rainfall, Quasi-Biennial Oscillation, radiative forcing, Main Development Region temperatures and its anomaly, global temperatures, and gridded temperatures). We find that gridded temperatures, Main Development Region, and global average temperature explain the observations best. The most extreme events are especially sensitive to temperature changes, and we estimate a doubling of Katrina magnitude events associated with the warming over the 20th century. The increased risk depends on the spatial distribution of the temperature rise with highest sensitivity from tropical Atlantic, Central America, and the Indian Ocean. Statistically downscaling 21st century warming patterns from six climate models results in a twofold to sevenfold increase in the frequency of Katrina magnitude events for a 1 °C rise in global temperature (using BNU-ESM, BCC-CSM-1.1, CanESM2, HadGEM2-ES, INM-CM4, and NorESM1-M).