An empirical comparison of interpolation methods for MODIS 8-day land surface temperature composites across the conterminous Unites States

These results provide strong evidence that spatial and spatiotemporal methods have a greater predictive capability than temporal methods, regardless of the time of day or season. This is true even under extremely high cloud cover (>80%). The Spline method performed best at low cloud cover (<30%) with median absolute errors (MAEs) ranging from 0.2 °C to 0.6 °C. The Weiss method generally performed best at greater cloud cover, with MAEs ranging from 0.3 °C to 1.2 °C. The regression analysis revealed spatial methods tend to perform worse in areas with steeper topographic slopes, temporal methods perform better in warmer climates, and spatiotemporal methods are influenced by both of these factors, to a lesser extent. Assessed covariates, however, explained a low portion of the overall variation in MAEs and did not appear to cause deviations from major interpolation trends at sites with extreme values. While it would be most effective to use the Weiss method for images with medium to high cloud cover, Spline could be applied under all circumstances for simplicity, considering that (i) images with <30% cloud cover represent the vast majority of 8-day LST images requiring interpolation, and (ii) Spline functions are readily available and easy to implement through several software packages. Applying a similar framework to interpolation methods for daily LST products would build on these findings and provide additional information to future researchers.