Development of a methodology to fill gaps in MODIS LST data for Antarctica

• Main Author: Alasawedah, Mohammad Hussein Mohammad
• Other Authors: Meyer, Hanna, Valdes, Maite Lezama, Guerrero, Ignacio
• Format: Master Thesis
• Language: English
• Published: 2021
• Subjects: Convolutional Neural Networks; Extreme learning machine; Gam; Antarc*; Antarctica
• Online Access: http://hdl.handle.net/10362/113760

Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies Land Surface Temperature (LST) is an essential parameter for analyzing many environmental questions. Lack of high spatio-temporal resolution of LST data in Antarctica limits the understanding of climatological, ecological processes. The MODIS LST product is a promising source that provides daily LST data at 1 km spatial resolution, but MODIS LST data have gaps due to cloud cover. This research developed a method to fill those gaps with user-defined options to balance processing time and accuracy of MODIS LST data. The presented method combined temporal and spatial interpolation, using the nearest MODIS Aqua/Terra scene for temporal interpolation, Generalized Additive Model (GAM) using 3-dimensional spatial trend surface, elevation, and aspect as covariates. The moving window size controls the number of filled pixels and the prediction accuracy in the temporal interpolation. A large moving window filled more pixels with less accuracy but improved the overall accuracy of the method. The developed method's performance validated and compared to Local Weighted Regression (LWR) using 14 images and Thin Plate Spline (TPS) interpolation by filling different sizes of artificial gaps 3%, 10%, and 25% of valid pixels. The developed method performed better with a low percentage of cloud cover by RMSE ranged between 0.72 to 1.70 but tended to have a higher RMSE with a high percentage of cloud cover.