Talking About the Weather
Through its ownership of cloud-computing services, Big Tech is setting the terms for how we understand the climate crisis—and how we respond.
One thing we do: talk about the weather.[1] People in the UK apparently—though not surprisingly—devote six months over the span of our lives to talking about whether it’s hot or raining or snowing. Assuming you live to 80, that’s 56.6 hours per month—rather a lot of chitchat. Talking about the weather is an effective icebreaker; when there is nothing left to say, or when we still don’t know what to say, this trusty cliché provides easy fodder for small talk. We can even politicize it: if in summer it’s very hot, and we recall that it didn’t used to be quite so sweltering, we can mention climate change. We all talk about the weather—but we don’t all do so in the same way.
Take, for instance, the National Oceanic and Atmospheric Administration (NOAA), an agency under the United States Department of Commerce. Its functions include weather forecasting, monitoring atmospheric conditions, exploring the depths of the ocean and protecting endangered marine species. One of its sub-agencies is the National Weather Service, which provides Americans with their daily weather updates.
Among the projects NOAA is currently testing is the FourCastNet Global Forecast System, a weather forecasting model created by scientists from NVIDIA, Lawrence Berkeley National Laboratory, the University of Michigan and Rice University. This is how a post on Amazon’s cloud website, Amazon Web Services (AWS), describes the model:
The FourCastNet Global Forecast System (FourCastNetGFS) is an experimental system set up by the National Centers for Environmental Prediction (NCEP) to produce medium-range global forecasts. The model runs on a 0.25-degree latitude–longitude grid (about 28 km) and 13 pressure levels. It produces forecasts four times a day, at 00Z, 06Z, 12Z, and 18Z cycles. Major atmospheric and surface fields—including temperature, wind components, geopotential height, relative humidity, 2-meter temperature, and 10-meter winds—are available. The products are 6-hourly forecasts up to 10 days ahead. The data format is GRIB2.
The FourCastNetGFS system is an experimental weather forecast model built upon the pre-trained Nvidia’s FourCastNet Machine Learning Weather Prediction(MLWP) model version 2. The FourCastNet (Bonev et al., 2023) was developed by Nvidia using Adaptive Fourier Neural Operators. It uses a Fourier transform-based token-mixing scheme with the vision transformer architecture. This model is pre-trained with ECMWF ERA5 reanalysis data.
And so on. So much for small talk. FourCastNetGFS is an initiative listed under the Amazon Sustainable Data Initiative (ASDI), hosted by AWS. It is a repository of climatological databases that provides open and fast access, as well as storage opportunities. The description goes on:
The Amazon Sustainability Data Initiative (ASDI) seeks to accelerate sustainability research and innovation by minimizing the cost and time required to acquire and analyze large sustainability datasets. These datasets are publicly available to anyone. In addition, ASDI provides cloud grants to those interested in exploring the use of AWS’ technology and scalable infrastructure to solve big, long-term sustainability challenges with this data.
In simpler terms, what this means is that datasets like those compiled by NOAA are stored on Amazon’s cloud-computing network. As these datasets can only be used inside the cloud, and not downloaded and taken offline, users, in turn, must enmesh themselves in Amazon’s infrastructure. To promote the use of these datasets through AWS, Amazon offers AWS promotional credits that grant free use of the model for some time, or until the consumption of processing power reaches a certain level. Once the credit is exhausted, whatever modelling or further analysis is done with ASDI data comes with a price tag.
Google follows a similar pattern with its “solution” for the climate transition. In 2010, Google developed an AI platform to access and analyse data from Google Earth, a sort of Google Maps for the Earth and space, called Google Earth Engine. In the years since, thanks to its 90,000 users who have used the tool mainly for research and educational purposes, Google Earth Engine’s services have vastly improved.
These users create specific tools (or applications) for global conservation and restoration which then feed back into the platform’s machine learning models. One such tool is Restor, developed by Crowther Lab in Zurich. For a specific area, it provides data on local biodiversity, the current and potential amount of solid carbon stored in the soil, land cover, soil pH and annual rainfall. According to Google, these data allow users to assess the restoration potential of any area of the Earth.
Google Earth Engine was originally free; however, it became a paid service on Google Cloud for governments and private companies in 2022. Among those using the service is the UN Food & Agricultural Organization, which uses the engine to create measurement, reporting and verification tools. Google is now adding new paid functionalities like generative design features for Google Earth for buildings. Another example of a green-tech service is Google’s Tree Canopy, a tool based on Google Earth’s proprietary data. It is already used by 350 cities to map tree canopies to optimise planting.
Tapestry, “an Alphabet-incubated moonshot” in tech speak, is another example of the tools developed by Google for addressing ecological issues. Tapestry is aimed at developing a single virtualized view of the global electricity system. Because the grid was not planned as a unified project, and many now-essential technologies did not exist when it was built, many of the often public actors that run, plan and maintain grids lack detailed virtual mapping tools. By rendering the grid visible, Google claims Tapestry will be able to offer AI tools for predictions and simulations, from milliseconds to decades into the future. Yet, while Tapestry makes the grid visible, only Google can see the global grid; governments and operators must pay to get access their corresponding bit of it.
What initiatives like ASDI, Google Earth Engine and Tapestry demonstrate, beyond the familiar tactic of profit-seeking by creating structural dependencies, is a reorganization of the epistemology of climate discourse. In other words, by controlling who has access to data, and how their access is mediated—with which tools and indicators, defined by Big Tech—these companies create and restrict the language that can be used to describe and explain a phenomenon, and the scales of knowledge available to different actors. By limiting the type and scale of data they can access, tech companies force users—from governments and UN agencies to researchers and everyday people—into discussing climate and environment from within siloed viewpoints, while only companies like Google and Amazon keep the whole, systemic picture, the one that is needed to make anything like an informed policy decision. By defining the models and arranging the data that is increasingly used for forecasting weather—and the progress of the climate crisis—they silently influence how we speak about the weather and our planetary future.