Earth Lab has focused on a number of different techniques for data analytics including:

• Statistical approaches for model-based prediction and inference
• Machine learning and deep learning for artificial intelligence applications like:
  • Computer Vision
  • Natural Language Processing 

Statistical Approaches

Newer statistical approaches are using machine learning and Bayesian (vs Frequentist) statistics to harness the data revolution so that models can scale and adapt as we make more observations. These methods must leverage spatiotemporal coherence in heterogeneous data, embed domain expertise like scientific knowledge, and incorporate decision-science to contextual and enable inference. 

Models allow us to link observed data back to the hidden rules of nature, and predict what might happen in the future. In the age of big data, we are now ushering in the next phase of scientific research that leverages models to extract insight from the information provided by Earth observation data.

Machine Learning

Machine learning and deep learning have revolutionized artificial intelligence applications like computer vision and natural language processing in the past decade. Computer vision develops models that teach machines to recognize and identify patterns of information from raw data like photons reflected and absorbed by the Earth’s surface. Natural language processing uses similar methods to teach a machine to recognize and identify patterns in written text. These are artificial intelligence applications because they teach machines to mimic the intelligence of the human brain.

This enables use of data science to sort through big data that would otherwise be limited by human capacity for interpretation. 

Computer Vision

Machine learning applications in science are beginning to open up new insights into earth and environmental data, especially in the remote sensing and social sensing domains. High availability of large amounts of remote sensing data and “off-the-shelf” models and code now makes machine learning practical for computer vision tasks interpreting remote sensing data such as image classification, object detection, scene segmentation, data fusion, transfer learning, and time series analysis.

An example of how a deep learning based image segmentation model can be applied to remote sensing imagery.

Natural Language Processing

Another application of machine learning for artificial intelligence in natural language processing. Earth Lab uses social sensing to develop new data sources and methods for curating data from written text. We curate and analyze data at scale related to the societal impact and social disruption of natural disasters. This data is critically important for use in the Earth Sciences, natural hazards research, and in real-time response. As the impact of natural hazard events grows, the ability to pair metrics of social response, societal disruption, and incident management at scale is critical for shaping our understanding of the drivers making a hazard a disaster. The ability to mine relevant social sensing data in real-time has the potential to save lives and support community response. 

Machine learning  techniques are increasingly being used in science to infer hidden dynamics of natural systems from noisy observations. Earth Lab is at the forefront of science-based deep learning in environmental science. While machine learning data science tasks are focused on image classification, sequence prediction, or regression  without reference to any particular scientific model or dynamical system, Earth Lab embeds  science knowledge and domain expertise into machine learning models. Earth Lab draws upon the rich set of methods provided by machine learning techniques to gain new insights from diverse spatiotemporal data streams.