Utilizing AI for drought prediction

Rising temperatures and intensifying drought continue to worsen with the global climate crisis. According to the World Health Organization, an estimated 55 million people worldwide are affected by drought each year - a number expected to grow as climate change becomes more extreme. 

Through the power of artificial intelligence , Andrew Watford, a fourth-year Faculty of Science student at the University of Waterloo, is addressing this challenge by developin g more a ccurate and interpretable tools for forecasting drought.

As part of his co-op term in the Mathematical Physics program and his stellar promise as a researcher in the field , Watford was afforded the opportunity to contribute to a peer-reviewed published study on the AI to analyze vegetation health and forecast drought patterns in Kenya. The paper compares the performance of a mechanistic model to two physics-informed machine learning approaches. 

Watford’s role under the supervision of Dr s. Chris Bauch (Faculty of Mathematics) and Madhur Anand (University of Guelph) involved writing code to predict the ormalized d’ifference egetation i ndex (NDVI) in drought-prone regions of Kenya. Through further refinement of these models, the research aims to enhance machine learning methods to improve drought prediction , which could lead to the develop ment of early warning systems and mitigation strategies.

"Our goal was to bring together mathematics and machine learning to develop new methodologies and push the field forward to predict drought," Watford says. "We are still far off from predicting drought five years in the future with certainty, but it’s a step towards trying to find the best way to do that." 

The ability to predict droughts earlier offers immense benefits, including enabling local governments to implement effective water management strategies, allowing farmers to select drought-resistant crops, and significantly enhancing natural disaster preparedness that could save lives. 

In a time where climate change and natural disasters are becoming more prevalent, incorporating machine learning models to help mitigate these threats becomes increasingly important. Home to the largest co-op program at a research-intensive university , with more than 70 per cent of students gaining up to two years of employment experience during their studies , W atford credits the U niversity of Waterloo for being able to apply his learning to this real-world problem.

"The research doesn’t end with being able to predict drought," says. "It is an evolving tool that will help people and save lives."  

A co-op stream of study is offered in most Science programs at the University of Waterloo. Learn more about the programs and opportunities available on the Science website  or Waterloo  co-operative education website  for

Dynamical systems - inspired machine learning methods for drought prediction was published in Ecological Informatics, volume 84, in December 2024.