Skip to content

Cloud Detection

Cloud serve as one of the larhest hindrances in optical remote sensing, making it near impossible to look at ground pixels value and depending on opacity often skew the final image values. Not only do clouds by themselves pose a challenge but they are accompanied by issues such as cloud shadow and sometimes haze along with differetiating between cloud and snow in some areas. Our next problem would be a sugegstion to tackle this problem using multiple approaches such as index based as well as machine learning algorithm to look more closely at performance of different algorithms. The idea would be to look for images with clouds near 60-70% of the entire scene and to allow them to think of machine learning algorithms and/or other approaches to detect cloudy versus cloud free area. How do you score per pixel based on what is cloud versus a clear pixel? Last but not least how do you differentiate snow versus cloud in certain areas.

Data Source

We are providing you with open California data for an area with varying stakcs of cloud cover from 20-80% cloud cover if possible.

//Date Range 2016-01-01 to 2018-04-08
var cloud_aoi=ee.FeatureCollection('ft:1MENjn5Pakbgf6iXsxppS0W0pUtnbJu6nJLWJoK1B') //AOI boundary for Area with cloud coverage
var cloud_img=ee.ImageCollection('projects/sat-io/Planet/cloud_ps') //Image Collection for Area with cloud coverage with PlanetScope 4Band analytic imagery

Simple

Suggested Methods

This is an open ended problem, you can try simple indexing approach and classification methods to machine learning algorithm looking at morphology instead of just the spectral information. The output will probably consist of a binary that could be used as a mask to mask out cloudy areas over the image and works consistenly on multiple images and not just a single image.

Otsu Method