Understanding the role of climate and humans in long-term ecosystem dynamics in South Asia through AI detection of paleo-proxies such as fossil pollen, microcharcoal and phytoliths
Dr. Meghna Agarwala (Department of Environmental Studies) is leading an innovative approach to understanding long-term ecosystem dynamics in South Asia by developing AI-driven methods for automated identification and quantification of paleo-ecological proxies that have traditionally required labor-intensive manual analysis.
The research framework addresses the critical need for higher taxonomic resolution in paleo-proxy identification, particularly for South Asian taxa that differ significantly from those studied in the Global North. Current limitations include misclassification issues such as the apparent presence of cereal pollen at 17,000 years ago in Central India—over 10,000 years before expected agricultural emergence—highlighting the necessity for more precise identification methods to distinguish between wild and cultivated varieties.
The methodology employs a three-pronged approach combining library creation, innovative image acquisition, and AI-driven detection systems. The team is developing a comprehensive database of South Asian taxa through systematic collection from diverse ecosystems including grasslands, wetlands, and indigenous agricultural varieties. Their innovative image acquisition system utilizes custom robotics developed with the Digital Makerspace, featuring automated stage movement, multi-focal imaging, and a novel “press” mechanism that reorients pollen grains to capture three-dimensional morphological features essential for accurate identification.
The AI-ML detection component will process images across multiple orientations and focal planes, employing image segmentation to differentiate objects from background materials and classification algorithms based on diagnostic properties including size, roundness, pore characteristics, and orientation-specific features. This automated approach promises to dramatically accelerate data generation from sediment cores spanning 6,000 to 19,000 years, enabling high-resolution temporal analysis of ecosystem dynamics.
The project’s broader significance extends to understanding savanna maintenance mechanisms, and human adaptation strategies during prehistoric climate events, ultimately informing contemporary ecosystem restoration policies through empirically grounded historical perspectives on landscape dynamics.
