Urbanization has affected land surface temperature (LST) significantly. The spatial average of LST is widely applied to evaluate urban heat islands (UHI), but a simple comparison of mean temperature is insufficient to evaluate the heterogeneity of LST and the effectiveness of urban heat mitigation strategies. This study took the spatial distribution of LST as the thermal landscape and firstly employed 3D landscape metrics to analyze its spatial heterogeneity. To perform collaborative analysis with UHI, the urban–rural differences of landscape metrics were defined as thermal metric islands (TMI). Significant UHI and TMI were revealed by temperature amplitude, aggregation, and complexity parameters, indicating higher temperature fluctuation and fragmentation in urban regions. Before 2007, the standard deviation of thermal surface, landscape fragmentation, and shape irregularity increased significantly, particularly in the suburbs. Later, benefiting from the Summer Olympic Games, the temperature fluctuation, fragmentation, and shape complexity decreased obviously, and UHI intensity hit the lowest. Both UHI intensities and TMI increased after 2010, and various metrics suggest that maximum temperatures, landscape fragmentation, and shape complexity declined over most of the district, even as the mean temperatures increased. This study concludes that urban temperatures have risen at a higher rate than suburban areas, but the thermal landscape composition and configuration have changed more dramatically in rural areas.