Explicit Haze & Cloud Removal for Global Land Cover Classification

Abstract

Haze and clouds in Earth's atmosphere obstruct a seamless monitoring of our planet via optical satellites. Prior work shows that models can learn to adapt and perform remote sensing downstream tasks even in the presence of such sensor noise. So what are the auxiliary benefits of incorporating an explicit cloud removal task, and what is its relation to other tasks in the remote sensing pipeline? We address these questions and show that explicit cloud removal makes models for land cover classification furthermore robust to haze and clouds. Finally, we explore the relation to a self-supervised pre-text task (including abundant cloudy data) and demonstrate how to further ease the need for costly annotations on the land cover classification task.