Automated Detection of Cloud and Cloud Shadow in Single-Date Landsat Imagery Using Neural Networks and Spatial Post-Processing

M Hughes,Daniel Hayes

doi:10.3390/rs6064907

Abstract

The use of Landsat data to answer ecological questions is greatly increased by the effective removal of cloud and cloud shadow from satellite images. We develop a novel algorithm to identify and classify clouds and cloud shadow, SPARCS: Spatial Procedures for Automated Removal of Cloud and Shadow. The method uses a neural network approach to determine cloud, cloud shadow, water, snow/ice and clear sky classification memberships of each pixel in a Landsat scene. It then applies a series of spatial procedures to resolve pixels with ambiguous membership by using information, such as the membership values of neighboring pixels and an estimate of cloud shadow locations from cloud and solar geometry. In a comparison with FMask, a high-quality cloud and cloud shadow classification algorithm currently available, SPARCS performs favorably, with substantially lower omission errors for cloud shadow (8.0% and 3.2%), only slightly higher omission errors for clouds (0.9% and 1.3%, respectively) and fewer errors of commission (2.6% and 0.3%). Additionally, SPARCS provides a measure of uncertainty in its classification that can be exploited by other algorithms that require clear sky pixels. To illustrate this, we present an application that constructs obstruction-free composites of images acquired on different dates in support of a method for vegetation change detection.

Highlights

The Landsat archive provides an unprecedented opportunity to discover how our landscape has changed over the last 30 years
Network size was significant at the 0.05 confidence level; in a post hoc test using Tukey’s honestly significant difference (HSD) criterion [31], statistically significant increases in total accuracy were seen in networks with 30 hidden nodes over those with 10 and 20 hidden nodes, which were similar (Figure 2)
Considering errors of commission, SPARCS performs substantially better by mislabeling 0.5% of clear sky pixels as cloud shadow and 0.2% as clouds, compared to FMask mislabeling 2.4% of clear sky pixels as cloud shadow and 2.8% as clouds

Summary

Introduction

The Landsat archive provides an unprecedented opportunity to discover how our landscape has changed over the last 30 years. Neural networks work by learning h linear combinations of the input data, where h is determined by the operator, and passing each of these through a given non-linear thresholding function These results are temporarily stored as hidden values. The network repeats the process by taking c linear combinations of those hidden values, where c is the number of desired classes, and again passing them through a given non-linear thresholding function. These results are interpreted as the input observation’s membership in each output class and are wholly dependent on the weights of the linear combinations at both stages. A higher number of hidden values (h) allows more complex patterns in the input data to be learned, though it increases the likelihood of learning spurious correlations in the training data and thereby reducing the generality of the classifier [19]

Methods

Results

Conclusion