Landsat Gap Filling with TERAS

PANCROMA currently provides three methods for gap filling a Landsat ETM+ SLC-Off image. The Transfer method conducts a global histogram match between the Adjust and Reference images and then transfers the missing pixels of pixels from the Adjust image to the Reference image. The Hayes method uses an approximation algorithm to reconstruct each missing pixel in the Reference using a comparison between a sliding window sample of Adjust and Reference image pixels. The (newest) TERAS method is another sliding window algorithm that uses a local optimization algorithm that can produce excellent color tone matches between the base image and the filled gaps. This tutorial describes the basic two-file method. PANCROMA also offers a six-file automatic processing method that automatically co-registers and gap fills two Landsat data sets. (See the PANCROMA Instruction Manual for more information).

The Landsat gap filling procedure requires a Reference image (the one you are trying to correct, i.e. with gaps) and at least one Adjust image (the one you will extract data from to fill the gaps in the Reference image. The Adjust image is usually gap-free, but it does not have to be. Because the gaps in two images may not coincide exactly, at least some useful data may be transferred from the Adjust image to the Reference image despite possible gaps in both. You will need three Reference bands and three Adjust bands (blue green and red) for each image.

The TERAS algorithm is a modified sliding window, local optimization method. Because the mapping computation is very localized (for example within a frame of size 20 pixels) with respect to the size of the image, the match is only influenced by the DNs in the immediate vicinity of the center target DN. This can result in a high fidelity match between the filled DNs and the DNs adjacent to the gap. A square processing frame is established so that the edges of the frame are twice the user-specified searchRadius in length and width. The frame is positioned over the first target Digital Number (DN). Computations are conducted to map the gap DNs to the reference image DNs, using only common valid DNs within the frame. After computing the target DN, the Reference is modified, the frame is displaced and the process is repeated.

You start the TERAS filling procedure by opening first the Reference (grayscale) image (the one with the gaps), followed by the Adjust image (in that order). When the images are loaded, select 'Gap Fill' | 'Gap Fill TERAS Method'. A Gap Fill Data Form will appear. You can select the algorithm parameters: Search Extents, the Gap Threshold and the Cloud Saturation Cutoff. The default values are 12, 1 and 100, respectively. The Search Extents defines the size of the sliding window within which PANCROMA will look for similar pixels. PANCROMA will search the target +/- the Search Extents, so the size of the window is twice the Search Extents value. The Gap Threshold determines which pixels in the Reference image will be computed. Any pixel with a brightness value less than the Gap Threshold will be considered a missing pixel and PANCROMA will attempt to fill it. The Cloud Saturation Cutoff value is designed to minimize cloud pixel "blooming". This problem is caused when the Adjust (no gap) image has clouds that the Reference (gap) image does not. These highly saturated pixels can distort the matching computation and can propagate across the entire sliding window frame. In order to avoid this, the algorithm will ignore any pixel with a DN above the Cloud Saturation Cutoff value in its computations. Note that this will mitigate the phenomenon but may not completely eliminate it.

The images below show subsets taken from a Reference gap image, and the gap filled image. The image match very good for this example. The RGB color composites are made from the gap and gap filled images and are shown in the third and fourth images below.

A possible issue with the TERAS algorithm is the tendency for clouds in the Adjust image to bias the grayscale tone matching algorithm and to distort the gap filled image. A section from such a problem area is shown below. The setting of the Cloud Saturation Cutoff will determine the extent of this problem to some degree. The optimal setting can be determining by opening the Adjust image and interrogating the cloud areas to determine the best Cloud Saturation Cutoff setting.

The PANCROMA TERAS algorithm provides a method for gap filling Landsat SLC-OFF images with reasonably fast processing times. The fidelity of the match between the Reference and Adjust images can be excellent, as can be seen in the images above. See www.PANCROMA.com for more information on this algorithm and the PANCROMA Satellite Image Processing application.