Seasonal Co-Variation in Surface Properties and the Urban Heat Island in Boston

Abstract:

Understanding the drivers behind the urban heat island (UHI) effect – the phenomenon of elevated temperatures in urban areas – is an important goal in urban climatology, particularly in the context of an increasingly urbanized and warming planet. Remote sensing offers a useful source of information for UHI studies by providing spatially explicit measures of both temperature and surface properties over time. However, key questions remain, particularly regarding what controls spatio-temporal dynamics in the UHI in and around cities. The objective of this study is to characterize seasonality in the daytime and nighttime UHI over Boston for the period 2001-2010, paying special attention to the roles of (1) green leaf phenology and (2) urban land use and shading in urban canyons as explanatory variables. We use 1 km 8-day land surface temperature (LST) data from MODIS to characterize temperature variability. Initial results are consistent with previously described UHI characteristics, with the highest daytime urban-rural temperature difference occurring during the summer. However, seasonal hysteresis for Boston is apparent in an enhanced UHI signature in the spring versus the fall, even when rural temperatures are equivalent during the two time periods. To characterize how surface cover variations control surface temperatures over the course of the year, we use spectral mixture analysis (SMA) applied to 30 m multi-temporal Landsat data. SMA is particularly well suited for studies of spatially heterogeneous urban areas because unlike classification methods or traditional vegetation indices, SMA takes explicit advantage of sub-pixel compositional variability. Preliminary results for 2010 suggest that spatio-temporal patterns in surface properties, and by extension land surface temperatures, in and around Boston are well explained as combinations of (a) green vegetation, (b) substrate/soil, (c) urban impervious, and (d) shade derived from SMA of multi-temporal Landsat data.