Cloud Removal using Machine Learning for BRDF/Albedo Retrievals in the Arctic

• Main Author: Hanasoge Nataraja, V.
• Format: Conference Object
• Language: English
• Published: 2023
• Subjects: Arctic; albedo; Sea ice
• Online Access: https://gfzpublic.gfz-potsdam.de/pubman/item/item_5018824

Since the late 1990s, NASA’s Earth Observing System constellation of satellites has provided continuous, long-term observations of atmospheric and surface processes on Earth including surface albedo and Bidirectional Reflectance Distribution Function (BRDF) that are generated as an operational product using cloud-cleared, multi-angle surface reflectances over the course of several days. The BRDF is central to imagery-based cloud and aerosol retrievals, while the surface albedo is a fundamental Earth energy budget parameter. Yet, this product is currently unavailable at higher latitudes where (1) the low contrast between clouds and sea ice/snow poses a challenge for cloud clearing, and (2) drifting ice floes are not accounted for, resulting in a significant gap in our understanding of the Arctic radiation budget. To address this gap, we propose the development of a BRDF/albedo product for moving sea ice floes and snow called the Sea Ice Floe and Snow Albedo Tracker (SIF-SAT). By leveraging multi-overpass, multi-angular satellite data, SIF-SAT will retrieve BRDF and albedo under low contrast and moving surface conditions. We combine existing cloud masks with machine learning (ML) models to produce cloud-cleared scenes in the Arctic. These scenes are then fed to a segmentation algorithm to identify individual sea ice floes and their reflectances are tracked over time to obtain BRDF and albedo. This presentation will primarily focus on the cloud-clearing model which has implications for radiation science in polar regions. SIF-SAT will enhance our capabilities in the Arctic and enable more accurate estimates of the cloud-radiative effect and ice-albedo feedback.