In this study, a combination of Matlab and ArcGIS was used. The in-situ data was available as a post-processed product to include spatial processing and interpolation. The Landsat 8 level 1 imagery was downloaded from the United States Geological Survey (USGS) Earth Explorer website. Landsat data was processed to LST using an algorithm in Matlab from previous research. Clouds were identified using visual inspection and removed. Regression analysis was used to determine the best equation that would model the relation between in-situ data from the LST.
The results of this study show an R² value of 0.67. Since the data has a range [72, 96] degrees Fahrenheit, the difference between the data has a maximum absolute value of 7.77° F. The regression algorithm for this set of data proves accurate around the traverse routes, and inaccurate around untraversed areas with considerable amounts of vegetation. The difference in temperatures may be from the direct observation of the surface from satellite data and the estimated temperature from the machine learning algorithm applied to the in-situ data over heavily vegetated and water land covers. More in-situ studies will need to be conducted to prove that regression techniques can be used to convert from LST to air temperature.
This study is of interest because heat waves are increasing in severity, duration, and periodicity. City planners must increase extreme heat resiliency to mitigate the effects of the UHI and having an UHI anomaly map could provide a vital first step in combatting the rising number of heat injuries. If a city is unable to acquire the resources to conduct in-situ measurements throughout the city, then an alternate approach could be utilized.