If the Glasgow COP26 summit was the global platform to discuss and accelerate efforts to combat the dangers of climate change, the recently concluded NeurIPS (Neural Information Processing Systems) 2021 - one of the eminent gatherings of industry and academia - presented ML solutions to support efforts. 

Founded in June 2019, Climate Change AI (CCAI) is an organisation with volunteers from industry and academia that believes machine learning can play an impactful role in tackling climate change. The team working at the intersection of climate change and machine learning presented three different papers at the 35th edition of NeurIPS and provided the details in a Tweet, shown below. 

Machine learning is an effective tool to reduce and respond to climate change; rather than considering it as a silver bullet, we should look at how best it can help to supplement our efforts. Keeping this in mind, let’s try to understand the significance of the solutions mentioned above. 

Paper 1: A GNN-RNN Approach for Harnessing Geospatial and Temporal Information: Application to Crop Yield Prediction 

About: A team of researchers from Cornell University presented a machine learning method that embeds geographical knowledge in crop yield prediction and predicts crop yields at the county level nationwide.   

Significance: Crop production is highly susceptible to changes in climatic conditions like temperature, soil moisture, precipitation, and a variety of other factors. Many recent studies have emphasised the importance of adapting agricultural methods in light of climate change; to do so, it is vital to forecast how climate change will affect crop productivity. Food security, supply stability, seed breeding, and economic planning can all benefit from crop output prediction. 

The team deployed a GNN-RNN framework to take both – geospatial and temporal knowledge to predict crop yield. As opposed to the earlier approaches, they demonstrated that adding information about a county's geospatial neighbourhood and recent history can significantly improve the prediction accuracy of deep learning methods. 

Paper 2: Predicting Atlantic Multidecadal Variability 

About: Researchers from the Massachusetts Institute of Technology and Woods Hole Oceanographic Institution presented models to predict Atlantic Multidecadal Variability (AMV). Simply put, it measures the fluctuations of sea-surface temperatures (SST) with a typical cycle between 60 and 70 years. 

Significance: Atlantic Ocean, the second largest ocean with Europe and Africa to the East and Americas to the west. This work tests multiple machine learning models to predict sea surface temperature, salinity, and sea level pressure in the North Atlantic region. It strongly impacts the local climate over North America and Europe. Take, for example, the positive phase of AMV was associated with warm summers in northern Europe and hot, dry summers in southern Europe, and increased hurricane activity. These impacts in themselves highlight the importance of predicting extreme AMV states. 

Paper 3: DeepQuake: Artificial Intelligence for Earthquake Forecasting Using Fine-Grained Climate Data 

About: A researcher from The Nueva School has proposed DeepQuake, a hybrid physics and deep learning model for fine-grained earthquake forecasting using time-series data of the horizontal displacement of earth’s surface measured from continuously operating Global Positioning System (cGPS) data. 

Significance: Real-time earthquake forecasting can ensure the safety and security of humans. As the paper points, every year, around 20,000 earthquakes occur worldwide. In the last 20 years, earthquakes have caused around 750,000 deaths and have displaced over 125 million people.  

Seasonal precipitation, temperature gradients, and air pressure are all climate variables that can produce deformations in the Earth's crust for a short period of time which can lead to earthquakes. Hence, predicting the depth, magnitude and location of an earthquake get vital. To that end, the deep learning-based neural network model – DeepQuake establishes the relation between climate change and earthquakes to give accurate outputs. 

Conclusion 

Climate change is a pressing issue with significant ramifications for societal well-being, especially for the world's poorest people. In collaboration with key stakeholders, addressing climate change requires quick, sustained, equitable, and scientifically based mitigation and adaptation initiatives. Emerging technologies such as machine learning is a strong tool with a wide range of applications in many technological and societal settings, and they should be used in accordance with its strengths, limitations and climate change objectives. 

Source: Climate Change AI

Image from Unsplash