Assessing gap-filled Landsat land surface temperature time-series data using different observational datasets

Landsat Analysis Ready Data (ARD)-based time-series present challenges in monitoring surface urban heat islands (SUHI) due to rapid changes in land surface temperature (LST) compared to cloud-free satellite observations. This research investigates the use of a spatiotemporal gap-filling model as a feasible and cost-effective solution to produce Landsat time-series LST products with both high spatial resolution and temporal frequency. The study identified and filled Landsat ARD thermal times-series data gaps due to missing data, cloud and shadow effects, and data quality. The accuracy of Landsat gap-filled products was assessed using randomly selected clear observations of Landsat and uncertainty products from the gap-filling model and was evaluated using various existing temperature datasets, including climate data from NOAA Global Historical Climate Network station observations, Daily Surface Weather and Climatological Summaries (DAYMET), and LST including MODIS, VIIRS and ECOSTRESS. The result suggests that the gap-filled Landsat LST has significant correlations with existing datasets including field observation and remote sensing data derived from other sensors that have similar monthly and seasonal variation patterns. The uncertainty maps show spatial distributions of uncertainty for gap-filled pixels that have high or low uncertainties. The Landsat gap-filled time-series datasets can be used to measure annual, seasonal, or even monthly landscape thermal conditions, which are useful for SUHI and relevant research, and to perform multi-decade time-series LST change analysis under climate change conditions.