Early warning of an incoming disaster — a monster storm, extreme rainfall, or fast-moving wildfire — can save lives and livelihoods by giving people more time to prepare. But for much of the world, early warning remains a distant dream, even as climate change-worsened extremes increasingly ravage the planet. Cost-benefit analyses have repeatedly shown that money invested in mitigating risk from extreme weather can yield big payoffs in reducing damage. According to the UN, just 24 hours of advance notice can reduce damage by 30 per cent, and $800 million invested in early warning systems (EWS) in developing countries could prevent losses of up to $16 billion. Yet currently, thirty per cent of the world’s population still lacks access to EWS. In Africa, that percentage doubles. To close the early warning gap, the UN has launched a goal of ‘Early Warning for All’ by 2027.

Role of AI in Disasters

The backbone of any effective EWS is accurate weather forecasts. Artificial intelligence can provide faster and, in many instances, more finely tuned forecasts than conventional approaches. In just the last few years, artificial intelligence has become a powerful tool to fast-track forecasts and is rapidly changing the landscape for early warnings.

Consider California’s AI-powered wildfire detection system implemented by Cal Fire, the state’s main firefighting agency. Using a network of over one thousand high-definition cameras, the system detects and warns firefighters of emerging threats at the earliest signs of smoke. During piloting, California officials reported that the software was able to alert firefighters accurately even before dispatch centres received emergency calls around 40 per cent of the time.

Google is similarly co-developing FireSat with Muon Space, the Earth Fire Alliance, and others. The initiative will utilize a wide array of satellites to monitor the Earth’s surface and update imagery for authorities every 20 minutes, far more rapidly than satellite imagery currently available to officials. The AI is trained on infrared sensors that identify heat signatures and feed the detection models, allowing for early detection of fires as small as five-square metres, enabling effective responses.

Collaborative Efforts to Clear the Hurdles

In some cases, the challenge for weather forecasting is providing more local, granular predictions that conventional weather models miss. A good example is the collaboration among Oxford University, the World Food Programme, and the IGAD Climate Prediction and Applications Centre. This team has trained AI models to refine conventional low-resolution forecasts, zooming-in to create highly specific rainfall and flooding forecasts across East Africa. Notably, these models can be run on a laptop, instead of the expensive supercomputers required for conventional forecasts.

To be sure, barriers to effective early warning systems still exist. They include — among other things — a paucity of trained meteorologists in the developing world, a lack of historical data to support AI, funding challenges, limited infrastructure, and energy use during AI training. But given the payoff of early warning in reducing loss of both life and economic damage, AI has already proven to be one of the most promising ways to close the early warning gap. #

Alice C Hill is David M Rubenstein Senior Fellow for Energy and the Environment Council on Foreign Relations; and Colin McCormick is Chief Innovation Officer at Carbon Direct.