For years the EROS Data Center processed and distributed Landsat terrain corrected satellite imagery providing users the option of using the Nearest Neighbor resampling algorithm.

Since the advent of the "Free" Landsat imagery in the recent past, all Landsat imagery, past and present, that is being distributed through the EROS Data Center, is now being resampled during terrain correction using only the Cubic Convolution resampling algorithm. 

The Cubic Convolution algorithm is based on computing a (distance-related) weighted average of the 16 nearest pixels and is known to both smooth and possibly distort the image pixel values as this averaging operation is performed.  A bimodal distribution of two distinctly different values tends towards a normally distributed distribution centered about the average of the two distinctly different values after just one pass of the Cubic Convolution algorithm.  Spectral and statistical distortion of the original data can be shown to occur through the use of this algorithm.

The Nearest Neighbor algorithm transfers the values of the nearest pixel to the destination pixel and performs no smoothing or averaging as it transfers the values.  It is known to preserve the original values as it is applied.  There is a possible spatial distortion of as much as 1/2 pixel width (maximum 21.2m along the diagonal of the pixel) through the application of this algorithm.

Do we want our satellite image pixel values possibly moved spatially by as much as 1/2 pixel width or spectrally and statistically distorted, but maintained in their same position?

For all the mapping applications and classification processes GRS uses, the statistical and spectral distortion of the Cubic Convolution algorithm introduces numerous errors into the final map data sets we produce based on Landsat imagery.

These averaged Cubic Convoluted image data may comprise ANY studies, analyses, or mapping efforts performed using this L1T image product.

Many new advanced products are now being developed and released, all based on the spectrally and statistically distorted Cubic Convoluted imagery.

Third party contractors are not available to perform this service and produce an image product comparable to the previously produced EROS Data Center Nearest Neighbor resampled imagery, in particular when EROS Data Center will only provide the Level 0 raw image data.

The failure to produce Nearest Neighbor imagery may mean that one of the Primary Goals of the Landsat Mission, which is to provide continuity over time, is no longer being met.

The purpose of this petition is to seek the REINSTATEMENT by the EROS Data Center of the Nearest Neighbor resampling algorithm as an optional means of resampling during the development of the Landsat Level 1 Terrain Corrected imagery (L1T).  If this cannot be accomplished, then at a minimum the EROS Data Center should at least make available the Level 1 R (L1R) radiometrically corrected imagery so that Landsat users can terrain correct the radiometrically corrected image data ourselves using the Nearest Neighbor resampling algorithm.