Developing a Data-Driven Model to Minimize Adjacency Effects in Landsat-8 Imagery
Christopher Begeman, Goddard Space Flight Center / SSAI, christopher.begeman@ssaihq.com (Presenter)
Nima Pahlevan, SSAI / NASA GSFC, nima.pahlevan@nasa.gov
Atmospheric Correction (AC) is an important tool in achieving accurate aquatic science products from remotely sensed optical measurements. A key aspect of AC, and a largely unaddressed problem, is that of adjacency effects (AE) in nearshore observations of medium to high spatial resolution optical imagery. The purpose of this work is threefold. First, to characterize the AE and its various environmental and physical drivers. Second, to create a machine-learning model that minimizes the corresponding effects. And third, to apply the said model to all water types, including inland lakes and coastal waters, and validate its performance through the AERONET-OC measurements and other in situ methods. For this study, we used Landsat-8/OLI as there is nearly a decade’s worth of imagery to use through the Google Earth Engine environment where a broad range of co-located auxiliary environmental and physical data are readily accessible. To determine the drivers of AE, it was first determined where the AE was located within an image with respect to shoreline distances and other such environmental factors. From there, high-correlation variables were determined to characterize the major drivers behind AE. These variables were used to train a machine-learning model to predict the AE for any water pixel within an image and minimize AE. Initial results of these models indicate a maximum percent difference change of 66% between worst case adjacency affected pixels and its AE-corrected counterpart, suggesting the viability of methodology in addressing the contribution of AE in nearshore satellite measurements.
Poster: Poster_Begeman_1-18_54_35.pptx
Associated Project(s):
Poster Location ID: 1-18
Presentation Type: Poster
Session: Poster Session 1
Session Date: Tue (May 9) 5:00-7:00 PM
CCE Program: OBB