Assessment of Increasing Temporal Frequency for Landsat-Based Time Series Land Surface Temperature Product

Thermal radiance measurements are atmospherically compensated, converted to Landsat surface temperatures, and delivered as part of the USGS Landsat analysis ready data (ARD). Reliable monitoring of urban heat island (UHI) based on Landsat ARD-based time series is often difficult because surface temperature variations occur rapidly compared to the temporal frequency of cloud-free satellite observations. Landsat ARD thermal data gaps (or missing data) can limit the ability to monitor annual, seasonal, and monthly variations in the surface thermal condition, and result in different capabilities for seasonal and monthly modeling that uses the temperature data to fit time series for historical trend analysis with background climate variations. A spatiotemporal gap filling model, as a feasible and low-cost solution for producing Landsat time-series surface temperature (ST) with both high spatial and temporal resolution, has been developed to produce increased temporal frequency Landsat ST (based on Landsat acquisition dates). This gap-filled Landsat ST, therefore, can be used to improve monitoring of annual, seasonal, or even monthly landscape thermal conditions. Accuracy assessment and uncertainty analysis of gap-filled Landsat ST is crucial for users to select appropriate products for their own applications. Input data and models are the two major sources of uncertainty in our gap filling procedures, and understanding these uncertainties is important in supporting informed users. We compared Landsat ARD gap-filled ST data to different sources of validation data, including NOAA Global Historical Climate Network station observation data, Daymet data, and Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature (LST). We compared these sources as independent datasets with Landsat ARD gap-filled ST in different locations and times. Our case study conducted in Atlanta, GA; Phoenix, AZ; and Sioux Falls, SD during selected years (1991, 2000, and 2016) demonstrated that the Landsat ARD gap-filled products can better differentiate the performances of the spatiotemporal gap filling model with improved training data strategy.