A spatiotemporal gap-filling method for building a seamless MODIS land surface temperature dataset

Land surface temperature (LST) is one of the most important variables for studying land surface processes. Remote sensing based LST products (e.g., MOD11A1 and MYD11A1) can provide high spatial and temporal distribution information of LST. However, due to factors such as cloud and shadows, there are missing values in LST products. Filling missing values is highly needed to create seamless LST datasets. There are still issues such as spatial continuity and computing efficiency in current gap-filling techniques. In this study, we proposed a spatiotemporal gap-filling method for developing a seamless LST dataset using MODIS products. First, we classified the global tiles into nine regions and divided each region into small blocks (10 by 10 pixels). Second, we derived the overall mean of each pixel by implementing a temporal fitting. Third, we calculated residuals and interpolated missing residuals using an inverse distance weighting method. Finally, we calculated missing LSTs by combining means and interpolated residuals. The cross-validation indicates that the average root mean squared error (RMSE) for mid-daytime (1:30pm) and mid-nighttime (1:30am) LST is 1.88K and 1.33K, respectively. There are no obvious boundary effects since the spatial interpolation of block average residuals in regional scale can avoid sudden change caused by large area lack of values, especially near the boundary of tiles. Moreover, this method is computationally efficient, by implementing a parallel scheme. The seamless daily (mid-daytime and mid-nighttime) LST product at a 1 km spatial resolution is of great use in global studies of urban, climate, and ecosystems.