Extreme heat seasons stretched in opposite directions across half the world's land

The sharpest number in this paper sits inside one city and points two ways at once. Phoenix's dry-heat extremes now have a median date 10 days later in the year than in the 1980s, while its humid-heat extremes have a median 5.5 days earlier ^[10]. The first measure is a plain thermometer reading. The second is wet-bulb globe temperature, which folds humidity and solar radiation in with air temperature to approximate the heat load a body has to shed ^[4]. Which of the two a jurisdiction writes its alert trigger on decides which end of its calendar it prepares for.

Extremes here are defined relatively, as days in the hottest 5% of the daily record, counted once in the baseline decade and again in the recent one ^[3]. That is a deliberately low-tech definition, and it is what gives the two-month windows on either side of each region's historical heat season something stable to be measured against ^[6]. The team also ran the analysis on two datasets, NASA's and the European Centre for Medium-Range Weather Forecasts', which is the control that matters most when the underlying product is a reanalysis rather than a station record, and the pattern largely held ^[15]. Averages had already been mapped: midlatitude summer conditions have lengthened by roughly six days per decade since 1990, but individual extreme events need not move in step with the seasonal mean ^[19].

Phoenix's count grew by 155 days between the two decades, an increase of 85% ^[21]. The more useful split is where they fell. None of the 1980s extremes landed after the heat season had ended; 6% of the 2015 to 2024 ones did ^[9], which works out to about 20 days across that decade, roughly two a year ^[22]. Small in count, and the acclimation argument is what makes it worth attention: the same temperature is harder on a body that has not had weeks to adjust, and harder again after months of enduring it, which is the sense in which a summer cooling-center schedule is mistimed rather than merely undersized ^[18]. That acclimation argument sits outside the study's method, which only counts days ^[3]. The 2024 tail was visible regardless, with 113 consecutive days above 100 F and then, from late September into mid-October, 21 straight days tying or breaking daily records ^[11].

The county mortality figures invite a causal read they cannot carry. July held 64% of Maricopa County's 645 heat-related deaths in 2023 and 46% of 608 in 2024 ^[13][12], so deaths outside July rose from roughly 232 to roughly 328 even as the annual total fell ^[23]. Two years is a distribution, not a trend, and one hot July can move that split by itself. The paper's data are heat days, counted independent of mortality ^[3].

Ivanovich and her coauthors, Benjamin Cook at GISS and Sonali Shukla McDermid at New York University, went in expecting warming to make dangerous thresholds uniformly easier to cross all year, and that is not what came out ^[2][16]. Ivanovich puts the result no higher than a compelling first line of evidence, on the grounds that extreme heat is rare by definition and out-of-season extreme heat rarer still ^[17]. So the work maps which direction each region's heat season is drifting, leaving the calendar dates for switching a heat plan on and off as the next open question.