Nishidh Singh
Investigating the effects of temporally induced data distribution shift on deep learning weather prediction models, and developing methods of adaptation to maintain forecast skill under a changing data distribution over time.
MS Thesis Students
Kartik Khurana
Working at the intersection of statistical learning and paleoclimatology, using Gaussian processes to reconstruct past climate from cave-based proxy records. Focused on developing more generalized Gaussian process frameworks that relate paleoclimatic proxies to depth.
Abhishek Menon
Developing a reinforcement-learning method to make large transformer-based weather models (such as Pangu-Weather) cheaper to run, by dynamically skipping parts of the data that carry little useful information for the current weather state — while enforcing physical constraints so these savings never come at the cost of accuracy in high-impact situations like extreme rainfall.
Subhajit Biswas
Implementing and evaluating GenCast, a diffusion-based generative AI weather model, for its ability to represent extreme weather over South Asia — monsoon rainfall, heatwaves, tropical cyclones, and severe convection — and comparing its skill against traditional numerical weather prediction.
Ratul Tarafder
Evaluating how well state-of-the-art AI weather models — used as released, without any extra fine-tuning — detect and track low-pressure systems that form over the Bay of Bengal, and comparing their performance against each other and against ensemble weather forecasts.
Hrithuparna Bharat
Studying how atmospheric variables relate to one another across space, using network-based methods that represent these relationships as connections between locations. Focused on how the structure of these networks changes over time, comparing temperature against other thermodynamic (heat- and energy-related) variables.
Lakshya Chouhan
Building an in-house version of Prithvi-WxC, a large pretrained foundation model for weather and climate, and fine-tuning it on India-specific datasets (rainfall, satellite, radar) to improve regional forecasting skill for tasks like monsoon rainfall prediction and cyclone tracking.