Google’s newest AI model is poised to analyze the Earth and ideally offer support. The goal is to scrutinize in detail the actions we are undertaking on our planet. Significantly, once the model finishes this task, it will also recommend areas where we can introduce initiatives to benefit our world.
AlphaEarth Foundations, a sector of Google’s DeepMind AI model, aims to utilize machine learning alongside the vast data Google has amassed about our planet over the last twenty years to comprehend how certain regions are progressing over time.
The model utilizes a technique referred to as “embeddings” that processes terabytes of satellite data each day, evaluates it, and compresses it to conserve storage. This leads to a model with multiple filters applied to maps, color-coded to reflect material characteristics, types of vegetation, groundwater sources, and human developments like buildings and farms. Google refers to the system as a “virtual satellite,” enabling users to retrieve detailed information about any site on the planet when needed.
Google’s goal is for users of the service to achieve a deeper insight into how particular ecosystems operate, including fluctuations in air quality, sunlight, groundwater, and human construction activities across various terrains. Ultimately, the company hopes the model will assist in addressing inquiries from paying governments and corporations, such as which ecosystems possess greater access to sunlight or groundwater, which could help identify the optimal locations for crop cultivation. It may also aid in pinpointing areas suitable for solar panels or buildings in climate-resilient zones.
The new model has already charted a complex landscape in Antarctica—an area difficult to capture due to irregular satellite imaging—in clear detail. It has also reportedly highlighted variations in agricultural land use in Canada that are invisible to the naked eye.
Google’s new model applies colors to AlphaEarth Foundations’ embedding fields. In Ecuador, the model penetrates persistent cloud cover to detail agricultural plots at various stages of growth.
