IIT Mandi develops AI-powered model to predict wildlife population shifts

IIT Mandi develops AI-powered model to predict wildlife population shifts

New mathematical framework forecasts predator-prey dynamics up to five years, offering fresh insights into ecosystem conservation

Munish Sood
MANDI:
In a significant scientific development, researchers at the Indian Institute of Technology (IIT) Mandi have developed an artificial intelligence (AI)-based mathematical model capable of predicting changes in wildlife populations, potentially strengthening scientific understanding of ecosystem dynamics and wildlife conservation.

The innovative framework uses decades of historical wildlife data to forecast fluctuations in predator and prey populations, even when available information is limited, irregular or uncertain. It can predict population trends up to five years ahead, offering researchers a new tool to understand how changes in one species can influence the survival and growth of another.

The research focuses on the wolf and moose populations of Isle Royale National Park, a 544-square-kilometre island in Lake Superior, Michigan, USA. Wolves are the primary natural predators of moose in the park, with moose accounting for nearly 90 per cent of the wolves’ winter diet. Their populations have been monitored annually since 1959, providing one of the world’s longest-running records of predator-prey interactions.

Historical records reveal considerable fluctuations in both populations. Wolf numbers, for instance, declined sharply in the early 1980s following a disease outbreak, while the moose population witnessed a major decline in the mid-1990s. Such changes highlight how disruptions in predator-prey relationships can influence the balance and health of an entire ecosystem.

The research has been carried out by Dr Nitu Kumari, Professor at IIT Mandi’s School of Mathematical and Statistical Sciences, along with her PhD scholar Anurag Singh.

The researchers developed a self-adaptive modelling framework that analyses historical population data to estimate moose reproduction rates, natural changes in wolf populations and the impact of predation on prey numbers. The model combines physics-informed neural networks with transfer learning, enabling it to extract meaningful patterns from complex ecological data.

While testing two different approaches to understanding how wolves hunt moose, the researchers found that the model based on the relationship between predator and prey population sizes performed better than the alternative approach. It also successfully identified periods of unusual population decline in historical records.

Notably, for the 2020–2024 period, the model predicted an increase in wolf numbers alongside a decline in the moose population, broadly corresponding with the trends subsequently observed in the national park.

According to Prof. Kumari, to the best of her knowledge, this is the first study to employ a self-adaptive mathematical modelling framework integrating physics-informed neural networks and transfer learning to investigate fluctuations in predator-prey populations.

The findings underline the growing potential of AI-assisted mathematical modelling in ecological research, particularly in situations where long-term environmental data is incomplete or uncertain. Such predictive tools could help scientists better understand wildlife population dynamics and assess possible changes in fragile ecosystems.

The research has also received international attention, with Prof. Kumari recently appearing in an interview with US-based news outlet The Primary. An article on the research is expected to be published online by the publication shortly.

MUNISH SOOD

MUNISH SOOD

Leave a Reply

Your email address will not be published. Required fields are marked *