Urban surface and near-surface air temperatures are known to be often higher than their rural counterparts, a phenomenon now labeled as the urban heat island effect. However, whether the elevated urban temperatures are more persistent than rural temperatures at timescales commensurate to heat waves has not been addressed despite its importance for human health. Combining numerical simulations by a global climate model with a surface energy balance theory, it is demonstrated here that urban surface and near-surface air temperatures are significantly more persistent than their rural counterparts in cities dominated by impervious materials with large thermal inertia. Further use of these materials will result in even stronger urban temperature persistence, especially for tropical cities. The present findings help pinpoint mitigation strategies that can simultaneously ameliorate the larger magnitude and stronger persistence of urban temperatures.
Caption: The urban-rural differences of near-surface air temperature and surface temperature persistence timescales and their relations with urban thermal inertia. (A) Urban-rural difference of near-surface air temperature persistence quantified by δΓ/Γ (the urban-rural difference in the persistence timescale normalized by the rural persistence timescale, where δ represents the urban-rural difference, namely, urban minus rural). The stippled regions are those with significant urban-rural differences of lag-1 autocorrelation at the 95% confidence level. (B) Urban-rural difference of surface temperature persistence quantified by δΓ/Γ. The stippled regions are those with significant urban-rural differences of lag-1 autocorrelation at the 95% confidence level. (C) Thermal inertia of urban impervious materials (μ, unit: 104 J m−2 K−1 s−1/2) as used in the climate model. (D) Urban-rural differences of temperature persistence (δΓ/Γ) as a function of the thermal inertia of urban impervious materials (μ, unit: 104 J m−2 K−1 s−1/2) in grid cells with significant urban-rural differences of lag-1 autocorrelation at the 95% confidence level. The large dots/circles in (D) are bin averages, and the shading indicates the SD in each bin.
