Analysis of class surface temperature
We computed the mean, min, max and standard deviation of surface temperature pixels that fell within each cell of the STURLA grid using zonal statistics (Table 1) and joined these results with the STURLA class variable. Averaging was necessary because Landsat 7 thermal bands are resampled to 30 meters for distribution (Roy et al., 2010) while the STURLA grid is 120 m. Thus, we averaged sixteen 30 m pixels that fell within each 120 m cell. Similar to Hamstead et al. (2016) and Larondelle et al. (2014) we focused the class temperature analysis on the most frequently occurring classes, which cumulatively comprise 90% of the city’s land area. As done with comparison of STURLA classes between cities, permutational t-tests with Bonferroni correction were employed to test significance. Likewise, the null hypothesis of the permutational t-test is that ST does not differ between the STURLA classes.