In Short : AI tools can enhance weather forecasting accuracy by analyzing vast datasets, identifying patterns, and improving prediction models. Machine learning algorithms can process real-time data, including satellite imagery and atmospheric conditions, providing more precise forecasts. This increased accuracy aids in better understanding and preparing for extreme weather events, contributing to climate change mitigation and adaptation efforts.
In Detail : Watsonx.ai, an AI tool jointly built by IBM and NASA, might make it possible for us to anticipate hurricanes, droughts, and other catastrophic weather occurrences with greater accuracy. How does it work?
At the recent COP28, NASA and IBM announced that an Artificial Intelligence (AI) tool called watsonx.ai would be available on the open-source AI platform Hugging Space. Watsonx.ai will help users monitor the Earth from space, measuring environmental changes that have already happened while also making predictions about the future.
Utilising NASA’s trove of data and IBM’s AI technology, the model can help scientists estimate the past and future extent of wildfires, floods, and urban heat maps.
Here is a look at what the technology does, and how similar applications of AI help mitigate climate change.
How AI has helped in weather forecasting
A key factor in understanding and combating climate change rests in our ability to predict weather patterns. In recent decades, weather prediction has improved rapidly with today’s six-day forecast as accurate as a five-day forecast 10 years ago. Hurricane tracks can be predicted with more accuracy three days in advance than they could 24 hours in advance 40 years ago.
This achievement is due to improvements in atmosphere and ocean technology, and in parallel, the progress made in high-computing power. Weather models today base their predictions on massive computing simulations that run on interpreted data. However, two key challenges remain.
First, access to that data is hard to come by. Second, analysing the same is an ever-harder task. Estimates from NASA suggest that by 2024, scientists will have 250,000 terabytes of climate data sets to work with. Climate data sets are massive and take significant time to collect, analyse and subsequently utilise to make informed decisions.
With advancements in technology, particularly the use of AI, these data sets become easier to interpret. In the last year, the European Center for Medium-Range Weather Forecasting started using deep-learning models known as AI emulators to generate forecasts based on historical weather patterns. According to IBM, while the laws of physics are not encoded into AI emulators, they can be inferred from the data, meaning that a forecast can be generated by a desktop computer in minutes instead of the hours typically taken by current systems.
So how have NASA and IBM built upon this?
Like Microsoft’s Bing, OpenAI’s ChatGPT, and other chatbots, watsonx.ai is also built on a foundation model — it’s trained on a broad set of uncategorised data allowing the model to apply information about one situation to another. In the case of watsonx.ai, NASA provides the datasets (in terms of satellite images instead of words,) and IBM created the foundation model to interpret them.
In order to train the model to comprehend visual sequences that unfold over time, scientists filled in blank areas in each image and asked the model to piece it back together. It became increasingly adept at figuring out how the photos connected to one another as it reassembled additional images. The model was then adjusted for certain tasks like segmenting and categorising photos.
In beta tests across the last year, the model has demonstrated a 15 per cent improvement in mapping flood and burn scars over the continental United States, using half as much labelled data compared to existing techniques.
“We believe that foundation models have the potential to change the way observational data is analysed and help us to better understand our planet,” said Kevin Murphy, Chief Science Data Officer, at NASA. “And by open sourcing the model and making it available to the world, we hope to multiply its impact.”
The model is also designed to be extremely simple to use. A user would merely need to select a location and a date, and the model will highlight changes in floodwater, reforestation efforts and other relevant factors.
What will be its impact?
According to IBM, this approach has the potential to minimise the amount of data cleaning and labelling needed to train a typical deep-learning model, and it could speed up geographical analysis by a factor of three to four. Information from the visualisations may be used to lessen the effects of flooding, develop infrastructure, assist in disaster response, and safeguard the environment.
When this type of generative AI is used in weather forecasting in the future, it may be possible to anticipate hurricanes, droughts, and other catastrophic weather occurrences with greater accuracy. This may clarify for us the precise ways in which alterations in the environment, such as the melting of ice in the poles, may affect our daily existence.
The technology could also apply to businesses, helping disaster response teams to prepare for fires impacting residential housing or helping supply chain logistics companies better understand macro weather patterns.
According to Juan Bernabe-Moreno, Director of IBM research for Ireland and the UK, in theory, this system could even be used to plan where to travel or buy a house. “There are many ideas about what you can do – the use of the application is really up to the people,” Bernabe-Moreno told BBC Science Focus. “But instead of having to be a big tech to create this application, making it open-source means putting it in the hands of the community.”
AI and Climate change
AI is already significantly impacting climate change strategies. According to the Boston Consulting Group (BCG) AI survey report, 87 per cent of private and public sector CEOs believe that AI is an essential tool in the fight against climate change.
In the transportation industry, AI-enabled vehicles have the potential to minimise energy use by mapping and identifying the most efficient routes. In agriculture, 40 per cent of freshwater usage is wasted on average but with AI technology, farmers can optimise crop irrigation, reducing water wastage and leading to more productive harvests. In India, AI-equipped peanut farmers have already witnessed a 30 per cent increase in yield.
AI may also be used to assess emissions at the macro and micro levels, cut emissions and the impacts of greenhouse gases, and remove already-existing emissions from the environment. According to BCG’s experience, AI may be utilised to help cut greenhouse gas emissions by five to 10 per cent of an organisation’s carbon footprint.