Recurring Slope Lineae (RSL) are low-albedo features that appear and disappear seasonally on Martian slopes. They provide distinct surface markers that are thought to result from the activity of subsurface water. Reliable characterizations of RSL activity are needed to test this hypothesis. RSL are of high scientific interest, and could inform future mission planning priorities beyond sample return.

Currently, scientists manually trace out RSL features to characterize them in terms of their size, length, and temporal extent. This strategy is not feasible for areas such as Tivat Crater and Valles Marineris, where hundreds of RSL have been spotted. An automated image analysis method could identify RSL in high-resolution orthorectified images of Mars, and quickly generate a catalog of the RSL as well as descriptive features for each one. This new capability would enable the analysis of hundreds of previously collected images. If associated with subsurface water activity, RSL detections could inform estimates of the minimum water volume required to generate the features. However, current methods for change detection are very sensitive to image noise, and especially to seasonal changes in illumination that are an unavoidable element in long-term monitoring. Further, RSL appear more quickly than they fade, so the sensitivity required to detect each type of change varies.

An automated image analysis method was developed that identifies RSL and computes key descriptive features such as size, length, eccentricity, albedo, etc. This approach analyzes orthorectified images and digital terrain model (DTM) files produced by the HiRISE instrument on the Mars Reconnaissance Orbiter (MRO). It employs a data-driven model of pixel intensities, accounting for local terrain, to identify areas that are darker than expected.

The first step is to create a model of expected pixel intensity as a function of topography and solar illumination. Combining the DTM and knowledge about the position of the Sun at the time the image was taken allows one to compute the amount of illumination received at each pixel in the image. A linear fit of the observed pixel values within each image to incident illumination is employed. Areas that are significantly darker than predicted by the model are identified as candidate RSL. The next step applies filters that exclude candidates that do not satisfy known RSL properties (e.g., size and shape attributes such as eccentricity).

Finally, the method compares the RSL candidates in the current image to those identified in the previous image of the same location. Each pixel is classified as one of extant (persistent) RSL, new (lengthening) RSL, faded/prior RSL (indistinguishable from surroundings), or no RSL activity. Once the RSL have been identified, the system generates a characterization of each RSL (e.g., area, length, eccentricity). These values inform image-wide statistics that create a picture of the temporal evolution of RSL activity, and they can be used as inputs to a hydrodynamical model to estimate the amount of subsurface water that would be required to generate the level of activity observed.

Previously, a system was developed to detect RSL in HiRISE images by using a simple image differencing operation that compares the currently observed pixel intensity values to those in a previous image. This approach is very sensitive to changes in lighting that occur seasonally as the Sun moves to the north or south, especially in slope areas where RSL are found. That approach detected new activity, but was far less sensitive to the fading of RSL, which occurs over a longer time span and therefore was often missed in adjacent images.

The new approach does not experience this limitation because it employs a terrain map (DTM) that allows the computation of the degree of illumination each pixel receives, which modulates the expected intensity of that pixel at the time it was observed. Therefore, shadows can be automatically excluded from detection, and areas where the RSL fade (i.e., surface brightens) are easier to detect.

This work was done by Brian D. Bue and Kiri L. Wagstaff of Caltech for NASA’s Jet Propulsion Laboratory. We thank David Stillman of Southwest Research Institute for helping to motivate this work and for his feedback on the results, which has helped them to continually improve. This software is available for license through the Jet Propulsion Laboratory, and you may request a license at: . NPO-49935