High-Quality Cloud Masking of Landsat 8 Imagery Using Convolutional Neural Networks

M Joseph Hughes,Robert Kennedy

doi:10.3390/rs11212591

Abstract

The Landsat record represents an amazing resource for discovering land-cover changes and monitoring the Earth’s surface. However, making the most use of the available data, especially for automated applications ingesting thousands of images without human intervention, requires a robust screening of cloud and cloud-shadow, which contaminate clear views of the land surface. We constructed a deep convolutional neural network (CNN) model to semantically segment Landsat 8 images into regions labeled clear-sky, clouds, cloud-shadow, water, and snow/ice. For training, we constructed a global, hand-labeled dataset of Landsat 8 imagery; this labor-intensive process resulted in the uniquely high-quality dataset needed for the creation of a high-quality model. The CNN model achieves results on par with the ability of human interpreters, with a total accuracy of 97.1%, omitting only 3.5% of cloud pixels and 4.8% of cloud shadow pixels, which is seven to eight times fewer missed pixels than the masks distributed with the imagery. By harnessing the power of advanced tensor processing units, the classification of full images is I/O bound, making this approach a feasible method to generate masks for the entire Landsat 8 archive.

Highlights

The sensors aboard Landsat 8 have been collecting high-quality imagery of the Earth since 2013
Landsat 8 imagery is accompanied by a bitwise quality mask (BQA) that encodes information about each pixel’s quality and includes masks for clouds, cloud-shadows, and snow/ice
Humans can identify clouds and cloud shadows in most single-date Landsat imagery when given spatial context. This insight led to the SPARCS algorithm (Spatial Procedures for Automated Removal of Cloud and Shadow), which was developed in 2014 for the Landsat 4 and 5 Thematic Mapper [9], and here we extend it to Landsat 8

Summary

Introduction

The sensors aboard Landsat 8 have been collecting high-quality imagery of the Earth since 2013. Several deep learning approaches have been developed for Landsat imagery [25,26,27], in part due to free access to the high-quality training and evaluation data for these sensors that was used to validate the CFMask algorithm, and which is freely available from the United State Geological Survey. These data includes the Landsat 7 and Landsat 8 Biome Cloud Cover Assessment Validation Data (Biome) as well as the dataset developed for training and evaluating the algorithm described in this paper (SPARCS, or Spatial Procedures for Automated Removal of Cloud and Shadow) [10]. This dataset (Dataset S1) is available at http://emapr.ceoas.oregonstate.edu/sparcs/

Neural Network Architechture

Processing

Evaluation

Performance of CNN SPARCS

Human Interpreter Consistency

Conclusions