Peruvian fishermen have long observed a link between unusually warm ocean waves, or the “El Niño phenomena,” and droughts and floods on land.
But for decades, climate scientists have struggled to understand the intricate El Niño phenomena itself and to make precise hydrologic projections about how El Niño may affect local weather patterns.
According to Auroop Ganguly, co-director of Northeastern’s Global Resilience Center, identifying the connection was thought to require creating an immensely sophisticated physics model that included difficult-to-measure flows between the ocean and atmosphere and between the atmosphere and land.
Ganguly and three co-authors demonstrated in a recent paper published that using machine learning to analyze existing data can produce explainable insights about the effects of El Niño on the world’s major river systems, the Ganges, Congo, and Amazon, and ultimately, regional weather patterns.
Treasure troves of data
According to Ganguly, climate scientists have gathered a considerable quantity of information about weather patterns, ocean temperatures from all around the world, flood levels, droughts, and other climatic phenomena. This information is based on both observation and simulations.
He claims that the vast amount of data has not yet been completely utilized and that it has occasionally been kept in different climate science silos.
Nevertheless, according to Ganguly, recent advances in machine or deep learning make it possible to leverage powerful computers to harness the enormous amounts of data and create prediction algorithms.
He claims that this approach to deep learning is novel. They can infer that this is how the correlation between sea surface temperature and river flow occurs. And the past teaches us that. Prior to now, that was not a possibility.
According to Ganguly, who is also the lead for Climate-AI at the Institute for Experiential AI, deep learning can identify what is essentially a long-distance relationship between sea surface temperatures in the eastern Pacific, where El Niño or La Nina intermittently occurs, and what that means for river flows around the world.
Deep learning to address society’s major issues
He claims that rather than using simple indices of this complicated phenomenon, the power of these approaches is in the ability to extract this information from the immense ocean—pun intended!—of data.
Co-author Yumin Liu of Amazon, who also worked on the study as a Northeastern Ph.D. student says, Think about being able to assess what is happening by correlating ocean temperatures to cloud development to precipitation influencing rivers, in different parts of the world.
Many now understand that collaborating across the machine learning and climate communities can be advantageous to both.
Stakeholders could better plan for floods, droughts, and other climatic disasters that have an impact on people’s lives, homes, businesses, transportation, and food production with the assistance of the upcoming information.
According to co-author Jennifer Dy, a professor of electrical and computer engineering at Northeastern University, developing and applying machine learning technologies to address societal great issues is an important necessity of our times.
Co-author Kate Duffy, a Ph.D. student at Northeastern University’s Sustainability and Data Science Lab, says, This study is an interesting display of how data sciences, particularly deep learning and complicated network structures, can fill gaps in our predictive understanding about hydroclimates.
Developing predictive models
Although it is relatively simple to measure temperature, enormous volumes of data are required to monitor temperatures across the vast majority of the planet’s oceans.
According to NASA, measuring precipitation is more challenging because precipitation systems “may be relatively random and vary extremely rapidly. The deep learning approach enables researchers to make use of both sorts of data to create a potentially predictive model, they claim.
The study, according to Duffy, demonstrates how enhancements to earth systems models can also improve so-called couplers, which connect substantial model components like ocean, atmosphere, and land models and enable greater feedback and information flow. One illustration of such a coupler is the Energy Exascale Earth System Model.
According to what the paper suggests, it may be crucial to investigate whether gaps in the science of coupling can be filled by creating so-called hybrid physics-AI approaches, in which, for instance, numerical models and partial differential equation-based systems can be at least partially connected via machine learning.
Strengthening information flows
The Institute for Experiential AI at Northeastern University will work on creating “generalizable and reliable solutions” by fusing customized machine learning with global climate models, according to Dy, head of experiential AI postdoc education.
According to Ganguly, a distinguished professor at Northeastern College of Engineering, being able to forecast information about river flows using maps of sea surface temperature may seem like a specialized answer.
Nonetheless, it enables for amazing chances to arise, he claims.
The study, according to Duffy, who recently left her position as a NASA scientist in order to launch her own firm in AI-based satellite remote sensing with a NASA SBIR grant, demonstrates how data-driven methodologies can enable enhanced climate-informed water resource estimates.
According to Ganguly and Dy, the exciting possibility is that this allows for the creation of hybrid-AI systems for a more efficient coupling of model elements within earth systems and global climate models.
