Decadal Analysis of Mean Land Surface Temperature: Adjusting Landsat Data for Temporal Consistency

Urbanized areas are well-known for being hotter than their surroundings, the Urban Heat Island (UHI) effect. UHI can occur in both the atmosphere and on the surface, with the latter being the main forcing for the former through sensible heat exchange. Land Surface Temperature (LST) can be estimated via remote sensing, offering high spatial resolution but often limited by low temporal resolution and short temporal series. The Landsat satellites (5, 8 and 9) provide the longest consistent LST temporal series (1984–present) with a high spatial resolution (30–120m). However, they have a 16-day revisit time (morning overpass) with data gaps under cloudy conditions, limiting the estimation of reliable monthly and yearly averages. Additionally, Landsat 5 had a non-fixed overpass time, which makes meaningful averages unfeasible. Using the Metropolitan Region of São Paulo, Brazil, as a case study, LST at overpass time was converted to 10:00 AM using an empirically estimated rate of temperature increase under clear-sky conditions. The estimation was made for each pixel using Ordinary Least Squares Method (OLS) and Normalized Difference Vegetation Index (NDVI) and it was compatible with the in situ LST data from the region’s weather station. With that data, decadal averages (1985-1995 to 2015-2025) at 10:00 were estimated using sine/cossine and OLS to account for seasonal discrepancies in sample sizes. Decadal differences were then calculated to estimate the increase/decrease over time. That approach was chosen over linear regression as it is best suited for non-linear changes in LST average, which occur by change in land use. Results show that significant increases (average: +2.5°C/30 years) in LST are mainly associated with urbanization (concentrated in the city’s periphery) and decreases (average: −2.0°C/30 years) are associated with increase in vegetation and/or increase in high-rise buildings (both concentrated in central areas).