Hawaii's flash-flood downpours are intensifying inside a long-term drying trend

The two trends point opposite ways at the same gauges. That divergence is the result, not a wrinkle in it. Rain common enough to recur every couple of years is getting lighter ^[5], while the one-to-three-hour burst in the 20-to-100-year class is getting heavier ^[6]. A distribution whose middle sinks while its tail lifts is changing shape ^[4], and no single statewide adjective for Hawaii's rainfall survives that. Yinphan Tsang, the principal investigator, gave the methodological version: understanding storms and floods requires short-duration, high-resolution data rather than daily or monthly averages ^[11].

The rarest numbers deserve the most caution. A 20-to-100-year event is a return level, not a tally. The record runs 1966 through 2020, which is 55 calendar years ^[1], so a 100-year quantile is extrapolated roughly 1.8 times beyond the longest interval anyone actually observed ^[2]. That is how engineering hydrology normally produces such figures, and it is also where the fitted model does more work than the gauges. My read: treat the direction of the one-to-three-hour signal as the finding, and the specific 100-year magnitudes as provisional. The paper, compiled primarily by Maxime Gayte with co-authors including the state climatologist Pao-Shin Chu, is peer reviewed and carries a DOI ^[14].

Density matters as much as duration. There are 117 gauges across five islands, about 23 apiece ^[3][2], set against terrain where the authors report intensity rising in one neighborhood and falling a few miles away ^[8]. So the finding is statewide, and its resolution is coarser than a watershed. A county can conclude that its short-duration extremes are trending up and still not know which gulch.

This record stops short of telling you what the water then does. Gauges record depth; whether a stronger two-hour burst overtops a culvert depends on slope and imperviousness, neither of which sits in a rain record ^[5]. The study also stops short of a mechanism. That storms of six hours and longer are weakening faster than one-to-three-hour storms ^[7] is a description of the record, not an explanation of it, and the honest version of the causal question is still open. What the work does hand emergency managers is a calendar, including a February-to-March peak at the tail of the wet season that the authors themselves call unexpected ^[10].