2020-09-22, 09:15–09:35, Room 1
Google Earth Engine is a powerful tool for earth science, but its main entrypoint is through a web-based JavaScript editor. Here we’ll talk about when and how to use Google Earth Engine in Python. The focus will be about getting data and connecting with popular frameworks such as Tensorflow for analysis. This talk is aimed towards geospatial data scientists trained in Python who are interested in using Google Earth Engine for their workflow and want to be immediately productive in using the tool.
With the boom of data science comes a renewed interest in geospatial, one of the most concrete and visual solutions for different problems faced by organizations of different sizes and motivations. This is seen in increasing number of product offerings for geospatial products, along with full-blown conferences dedicated to spatial data science. With new technologies and frameworks now capable of challenging tasks such as processing and storing petabytes of data and increased availability of satellite imagery, there is an even greater need for the skills to analyze and extract insights from this potentially useful data. Currently, Python enjoys a continued rise in popularity among data scientists for its quick to start, modular and generally applicable nature. This talk aims to make data scientists interested in including satellite imagery analysis in their workflow, to be immediately productive in using GEE.
The talk will start with a case for Google Earth Engine in Python in the context of renewed interest in geospatial with popularity of Python. This part will also include an introduction to GEE, details on how it was developed, the community that uses it, the team that maintains it, its capabilities such as its dataset inventory, and limits of use.
The main content of the talk will be focused on onboarding the user on interacting with the platform, starting with an intro to the web based code editor. We follow it up with a quick look at the functionalities by reading the documentation for javascript and Python and we proceed with learning the commands using a Google Colab environment for development. A large part of this section will be dedicated to detailed steps on an existing workflow[1] for land use classification via Tensorflow. We walk through different steps such as how to get a license, checking out which satellite images are available, correcting satellite images with high cloud cover, the limitations for pixel wise classification, and exporting tfrecords of images to connect with Tensorflow.
Finally, the last part of the talk will focus on other analysis compatible with the tool, different ways to contribute, and avenues to seek help for very specific errors. The last part will also include how our company uses GEE in our work.