Generating synthetic training data that can improve the accuracy of machine learning models – You have probably read about fake images and videos being generated by machine learning (ML) models. While this application might sound more like a fun exercise or, in some cases, malicious activity, synthetic (aka fake) data can help improve the accuracy of ML models. For example, research has shown that Generative Adversarial Networks (GANs) can generate synthetic data to augment real medical image training data that improves liver lesion classification and medical diagnosis.

As part of a project to tackle the scarcity of training data for agricultural monitoring applications, we are using GANs to generate synthetic Sentinel-2 satellite imagery. The results reveal that our GAN model can generate realistic imagery that can be used in classification models. Check out isthisplacereal.com and see how many of the images you can correctly identify as real or synthetic.

Tackling data challenges and incorporating physics into machine learning models will help unlock the potential of artificial intelligence to answer Earth science questions.

The Earth sciences present uniquely challenging problems, from detecting and predicting changes in Earth’s ecosystems in response to climate change to understanding interactions among the ocean, atmosphere, and land in the climate system. Helping address these problems, however, is a wealth of data sets—containing atmospheric, environmental, oceanographic, and other information—that are mostly open and publicly available. This fortuitous combination of pressing challenges and plentiful data is leading to the increased use of data-driven approaches, including machine learning (ML) models, to solve Earth science problems.

Results show models based on Generative Adversarial Networks perform better than Convolutional Neural Networks in classifying land cover classes outside of the training dataset.

In many supervised machine learning (ML) applications that use Earth observations (EO), we rely on ground reference data to generate training and validation data. These reference data are the building block of those applications and require geographical diversity if one aims to deploy the models across various geographies. Ground reference data collection, however, is an extensive process and extremely scarce in remote areas that would most benefit from the use of EO.