Reconstruction of MASTER Images Using DINEOF and 3D Inpaint

One of the fundamental limitations associated with airborne or satellite remote sensing is the presence of unusable data in the form of cloudy pixels. The presence of whitecaps in ocean color imagery also represents unusable data at high spatial resolution, and error for imagery that doesn't spatially resolve the whitecaps. This study compares the efficiency of two data reconstruction methods, the Data INterpolating Empirical Orthogonal Functions (DINEOF) and 3D Inpaint programs, applied to clouds/whitecaps in MASTER images. MASTER is the MODIS/ASTER airborne simulator, a spectrometer that captures high resolution imagery of the Earth's surface. The DINEOF program was initially developed to utilize time series images to develop patterns that it uses to fill in the gaps in satellite data under clouds. This project evaluated whether the mathematical techniques of DINEOF could be utilized to fill in otherwise unusable data (clouds or white caps) using discrete bands from a MASTER image. The 3D Inpaint program, which is MATLAB code that replaces the missing data by interpolating the non-missing elements, was evaluated compared to DINEOF. For the purpose of this study three MASTER images were selected from the DC-8 Airborne Science Laboratory flight during June 18, 2013 over the Santa Barbara Channel in California. These images contain clouds and whitecaps, factors that disturb or influence both quality and quantity of light received by the instrument. The images were reconstructed using both programs to assess which makes the most accurate reconstruction. With respect to data missing due to the presence of clouds, this study concludes that the mathematical technique employed in DINEOF makes a reasonable reconstruction of MASTER clouded images, reducing the size of the clouds. With respect to data which is unusable due to the presence of whitecaps, 3D Inpaint works better to reconstruct images which have a large number of whitecaps and small percentage of clouds. The results were validated using statistical methods and histograms. For a better understanding of these programs more studies have to be developed with a large quantity of images, but these results suggest that it is possible to “de-noise” the images (remove whitecaps) and reduce or eliminate small cloud-gaps, increasing the retrieval of information from the MASTER airborne imagery.