Posted February 13, 2026 · Submitted by Priya Nair, Indian Institute of Science
Dynamical downscaling delivers detailed regional climate information but at prohibitive computational expense. We develop a deep-learning emulator, trained on a limited ensemble of high-resolution regional simulations, that reproduces precipitation and temperature fields at a small fraction of the cost. Validation against withheld scenarios confirms that the emulator preserves extreme-event statistics, enabling the large ensembles required for robust climate-risk assessment.
Priya Nair, Henrik Vasquez (2026). Deep Learning Emulators for Rapid Regional Climate Downscaling. TerraNova preprint.
Priya Nair, Sofia Reyes
Thawing permafrost represents one of the largest and least constrained feedbacks in the global carbon cycle. We combine controlled incubation experiments with in-situ flux measurements along a natural thaw gradient to partition carbon relea…
Read preprint →Priya Nair, Amara Okeke
Reconstructing the natural range of monsoon variability is essential for contextualising contemporary rainfall trends. We present a precisely uranium-thorium-dated stalagmite oxygen-isotope record resolving monsoon intensity over the past e…
Read preprint →