Statistical estimation of next-day nighttime surface urban heat islands

Estimating future temporal patterns of Surface Urban Heat Islands (SUHIs) on multiple time scales is an ongoing research endeavor. Among these time scales, estimation of next-day SUHIs is of special significance to urban residents, yet we currently lack a simple but efficient approach for making such estimations. In the present study, we propose a statistical strategy for estimating next-day nighttime SUHIs, based on incorporating various SUHI controls into a support vector machine regression (SVR) model. The majority of both the surface controls (including factors related to land cover and solar radiation) and meteorological controls (including temperature fluctuations, relative humidity, accumulated precipitation, wind speed, aerosol optical depth, and soil moisture) that have previously been found to account for daily SUHI variations were used as estimators, and we provide estimations for both the overall SUHI intensity (SUHII) and pixel-by-pixel Gaussian-based LSTs over 59 Chinese megacities. For the overall SUHII, the mean absolute error (MAE) is 0.67 K on average, and the mean absolute percentage error (MAPE) is no more than 25% for more than 90% of the cities. For the pixel-by-pixel LSTs, the associated MAE is less than 2.0 K in most scenarios. In addition, the contribution from each selected estimator to SUHII estimation is assessed comprehensively. Among all the estimators, the contribution from relative humidity is the greatest, followed by rural surface temperature and surface air temperature. Moreover, for nearly 78% of the cities, the estimators related to day-to-day SUHI variations make a larger contribution than those related to intra-annual SUHI variations. We conclude that our simple yet effective statistical approach for estimating next-day SUHIs can potentially help urban residents to better adapt to urban heat stress.