Research & Development
From the Navier–Stokes equations to generative neural networks — coupled dynamical models, data assimilation, and physics-ML, developed into operational products for climate-risk decisions.
Research → Operations
Development & products
Weather2Grid: Probabilistic Weather-to-Power-Outage Risk
An open risk system that turns official NHC, numerical, and AI forecasts into calibrated county-level power-outage distributions for emergency managers on a 24–96 hour horizon. A common gust interface lets HRRR, GEFS, ECMWF/AIFS, and WeatherNext feed one two-stage vulnerability model trained on EAGLE-I outage data, covering tropical cyclones, derechos, and high-wind events alike. Outputs give P50/P90 severity, exceedance probabilities, cost–loss thresholds, and a meteorological-versus-impact uncertainty split, running now as a live experimental forecast with leakage-proof replay verification.
SURMA-Flow & BRISHTI-05: Generative High-Resolution Rainfall Analysis
SURMA-Flow (Score-guided Unified Rainfall Modeling and Assimilation with Rectified Flow) learns fine-scale precipitation structure from CPC, ERA5, and CHIRPS, then assimilates BMD rain-gauge and GPM IMERG satellite observations at analysis time through score guidance — producing a stochastic ensemble rather than a deterministic interpolation. Its first public product, BRISHTI-05, is a 30-member daily precipitation reanalysis over Bangladesh on a 0.05° (~5 km) grid, validated on rotated spatial holdout folds of withheld BMD stations with correlated-footprint treatment of satellite error.
Dynamical-AI Kilometer-Scale Downscaling & Reanalysis
Hybrid WRF + diffusion/flow-matching pipelines that downscale MERRA-2/MERRA-21C global reanalysis to 3-km and site-specific wind, solar, moisture, and flood products — supporting NASA fire modeling and a 2025-selected NASA project on downscaled-projection uncertainty.
Global Probabilistic S2S Flow-Matching Emulator
Dynamical-model-informed stochastic flow matching for global subseasonal forecasting: generates 1,000+ physically realistic ensemble members in seconds for tail-probability and extreme-event analysis, without running a numerical model at inference.
AirCast-SR: Atmospheric Super-Resolution
Co-developed a foundation model for kilometer-scale atmospheric super-resolution via latent consistency diffusion — a standalone learned pathway to high-resolution fields when full regional dynamical downscaling is impractical.
Multi-Hazard Climate & Catastrophe-Risk Engine
Probabilistic hazard information for tropical cyclones, floods, extreme precipitation, heat, and renewable-energy resources: GEOS/WRF/CMIP ensembles fused with generative scenarios, river-flow and inundation mapping, and geospatial exposure layers for government and commercial insurance and energy partners.
Science
Research highlights
Global Kilometer-Scale GEOS/ECCO Nature Run
Co-developed and executed a global coupled atmosphere–ocean simulation at kilometer scale (C1440-LLC2160) — a digital-twin "truth" dataset for satellite observing-system simulation experiments, pre-launch sensor evaluation, and extreme-event reconstruction, produced on petabyte-scale HPC.
GEOS-MITgcm NWP-to-Decadal Prediction System
Co-led development of NASA's modular coupled prediction system spanning weather to decadal horizons: Fortran/MPI coupling architecture, dynamically balanced initialization, adjoint data assimilation, and legacy-kernel optimization for 30% higher HPC throughput.
Green's-Function, Adjoint & Bayesian Parameter Optimization
Two physically interpretable strategies for tuning Earth-system models: Green's-function response maps with Gaussian-process multi-fidelity search, and ocean-adjoint gradients transferred into coupled GEOS-MITgcm experiments.
NASA GEOS Model Bias Correction for Extreme Monsoon Forecasts
Principal Investigator of a NASA-funded project correcting a persistent systematic error in NASA's GEOS model to improve forecasts of extreme South Asian monsoon precipitation — combining tendency-error diagnosis, explainable AI, and generative bias correction.
Monsoon Extremes & Bay of Bengal Tropical Cyclones
A connected research program on flood-producing monsoon rainfall and the environmental drivers of Bay of Bengal cyclones under climate change — including the finding that climate change quadruples flood-causing extreme monsoon events in Bangladesh and Northeast India.
Publications and preprints
Preprints & under review
Earlier peer-reviewed
Leadership
Funded research
Also: Co-Lead, WCRP CMIP7 Data Analysis Young Scientist Group (2023–present) · Peer reviewer for Nature Climate Change, Journal of Climate, Climate Dynamics, Journal of Hydrometeorology, and Weather, Climate, and Society.
Origins
Doctoral research
My Ph.D. (George Mason University, 2020, advised by Dr. Natalie Burls) investigated what controls the strength, seasonality, and future of subtropical anticyclones — the great high-pressure systems over the subtropical oceans that shape rainfall, drought, and marine climate in both hemispheres. Using CESM experiments, CMIP5/6 ensembles, and dynamical diagnostics, I showed how tropical diabatic heating, direct CO₂ radiative forcing, and indirect SST warming each imprint on the anticyclones and the storm tracks — mechanisms that now inform my model bias-correction work at NASA. This program also contributed to the Cape Town "Day Zero" drought study and emergent-constraint assessments of CMIP6 circulation projections.