A multi-temporal cloud method for detecting cloud and shadow cloud in Landsat 5/8 images on Google Earth Engine

Abstract

Satellite images provide a source of data given its spatial and temporal coverage with different level of
detail, and are often used in monitoring climate changes, disasters, water management in river basins
and wetlands. However, cloud and cloud shadow cover on most satellite images, requiring a special
handling in order to improve the accuracy and to have more representative results. In this paper, a
multi-temporal cloud removal method (consisting of three main steps: (i) radiometric correction, (ii)
cloud and cloud shadow detection, and (iii) image reconstruction) was implemented on the Google
Earth Engine (GEE) cloud computing platform using Java Script language, and then it was applied to
remove cloud and cloud shadow in 35 Landsat 5/8 images (in which the area of cloud and cloud shadow
is less than 10% of total image area) that were collected from 1984 to 2018. The results showed that the
multi-temporal cloud removal method was implemented and applied suceessfully to all selected Landsat
5/8 images of the domain of interest. A great improvement of computing capacity in comparison with the
use of traditional softwares such as ENVI and a great potential for processing satellite images with
regional large-scale mapping were also observed and presented. Futhermore, the use of GEE platform
allows for using all available satellite images to investigate spatial distribution and temporal variation
of the relevant field of interest.