Urban Temperature Bias as Determined by Polar Orbiting Satellite Data

A method of determining urban temperature bias from remotely sensed data is developed and successfully tested in this study. First, atmospheric sounding products from NOAA's polar orbiting satellites were used to derive predictive equations of shelter-level maximum and minimum temperatures. Sounding data from both winter (January) and summer (July) months were combined with surface data from over 5300 cooperative weather stations in the continental United States to develop multiple linear regression equations. Predictive equations were then used to estimate rural ("background") temperatures, unaffected by urbanization. Clear and partly cloudy sounding retrievals proved superior to cloudy retrievals. Validation tests showed the models' abilities to predict rural temperatures in different months and in specific climatic regions. Using these equations, estimates of urban temperature bias for 37 cities in the United States were made. These estimates compared favorably to ground truth data. Largest differences between observed and predicted bias were found at coastal cities, and those at higher elevations in the western United States. Mean differences between observed and predicted bias for groups of cities were not significantly different, making the potential application of this technique to corrections of urban bias in large datasets very plausible. Other products obtained from polar orbiting satellites, including normalized difference vegetation index (NDVI) values, were also found to be useful descriptors of urban temperature bias. NDVI urban minus rural values were highly correlated to daily and monthly minimum temperature bias at most of the cities studied.