I study the large-scale dynamics of the atmosphere and its evolution in a changing climate. I extensively use high-dimensional data analysis, mathematical modeling, numerical computing, and ideas from machine learning in my research.
I am a postdoctoral researcher at Stanford University and I focus on using modern advances in machine learning to improve the representation of atmospheric-climate variability in global climate models. I work with Aditi Sheshadri.
Prior to this, I worked at the Ludwig-Maximilian University of Munich (LMU) where I investigated gravity wave forcing in the midlatitude stratosphere, with Thomas Birner in collaboration with the German Aerospace Center (DLR).
I received my Ph.D. from the Courant Institute of Mathematical Sciences, New York University, where, motivated by uncertainties in the Antarctic ozone hole recovery projections, I assessed stratosphere-troposphere coupling & tracer transport in idealized climate models. I was advised by Edwin P. Gerber.
Email: I can be reached at : ag4680 |at| stanford |dot| edu, or ag4680 [at] nyu [dot] edu
Figure: True vs. predicted gravity wave momentum fluxes around Drake Passage during late July 2015 using a global AI flux predictor we developed by finetuning Prithvi WxC - a large AI foundation model for weather and climate. Climate models typically fail to resolve atmospheric variability at scales shorter than ~a hundred kilometers. ML can help bridge this prevailing gray-zone gap by learning the "missing physics" from high-resolution models (in this case ERA5 reanalysis) and then representing it in coarser-climate models. Our ML predictor skillfully captures the lateral propagation of GW packets over the Southern Ocean, which can be identified as flux blobs moving in the zonal direction. Foundation models are highly versatile and can be used to accomplish a range of weather and climate sub-tasks. Read more about Prithvi WxC here.
Research
Please refer to my CV and my Google Scholar profile for details. Please contact me for access to any of my publications behind a paywall.
2024
A computationally tractable method to compute small-scale momentum fluxes using Helmholtz decomposition.
Using existing Python spherical harmonics packages out-of-the-box one can compute small-scale momentum fluxes from a kilometer-scale global climate model, using Helmhotlz, in around 600 calendar days (on a typical supercomputer with limited node allocation). We optimized the workflow and reduced it to just 15! The fluxes are now publicly available for anyone to use.
[paper] [dataset]
Our ICML'24 paper on machine learning representation of mesoscale gravity waves.
Large AI models are demonstrably good at simulating the large-scale atmospheric evolution but struggle with small-scales. In this paper, we show how deep learning models can be effectively used to generate flux predictors for small-scale processes using a hierarchy of modeling architectures: from single-column ANNs to regionally nonlocal ANNs to global Attention U-Net CNNs. The proof-of-concept indicates that data-driven parameterization can be effectively used to represent mesoscale dynamical processes in climate models.
[paper] [blog]
Does horizontal propagation of gravity waves matter for midltitude stratospheric circulation? Yes, here's a quantitative evaluation.
The reason for horizontal propagation of mesoscale atmospheric waves is well-known, but its true impact on long timescales has not been well-quantified. This study quantifies the contribution of horizontal propagation (due to wave refraction, Doppler shifting etc.) on the midlatitude stratospheric circulation.
[paper] [slides]
2023
Estimating gravity wave momentum fluxes in the Southern Hemisphere
by blending LIDAR observations, renanalyis, 1 km global models, and coarse climate models [paper] [poster] [blog]
Where does ML fit in the existing climate model hierarchy?
The last decade has seen a meteoric rise in the application of machine learning principles in climate science. This paper reviews how ML/AI has transformed multiple aspects of climate modeling, and discusses future avenues for climate model development to encompass climate risk evaluation. We reason why ML adds an independent dimension to the existing climate model hierarchy.
[paper]
2022
Deep tropics isolated from wave mixing
[paper]
Wave-enhanced Tracer Dispersion
[paper]
2021
Importance of gravity waves for springtime stratospheric circulation
[paper] [slides]
Understanding trace gas transport differences among climate models
[paper] [slides]
2020
How robust is trace gas transport representation in state-of-the-art climate models?
[paper] [slides]
News & Miscellaneous Links
-
(September 2024) Excited to share the release of NASA and IBM's Prithvi WxC - an AI foundation model for weather and climate application trained on NASA MERRA2 - is now available on Hugging Face [paper] [blog1] [blog2] [media]. We leveraged it to develop a data-driven climate model parameterization for atmospheric gravity waves. Interestingly, not only was training the finetuned model cheaper, but using the encoder-decoder from Prithvi made the model more accurate and more generalizable, leading us to believe that large AI models can be effectively used to improve process representation (precipitation, clouds, radiation, etc.) in global climate models.
-
(May 2024) Prithvi-Weather-Climate: the efforts to build the first multi-modal open foundation model, jointly with NASA, IBM, ORNL, and CSU, over the past several months, have shaped up well. Glad to be a part of this amazing team. Click here to learn more about Prithvi-WxC, and what makes it special.
-
(Jul 2023) In June 2023, I delivered a set of four lectures on "Machine Learning Methods in Atmosphere, Ocean, and Climate Science" at an atmosphere, ocean, and climate science workshop organised at the International Centre for Theoretical Science, TIFR, Bengaluru. These lectures were aimed at introducing undergraduate, graduate and postgraduate researchers to machine learning (ML) methods. If you have wanted to explore the possibilities of ML in your research and would like to start somewhere, I would like to recommend these lectures specially prepared for this. The first lecture discusses the basics of ML, the second and third lectures are hands-on Python tutorials to code neural networks using PyTorch, and the fourth lecture discusses three novel use cases of ML in climate science: data-driven physical parameterizations, equation discovery, and weather forecasting.
[Lecture 1]: Machine Learning Fundamentals
[Lecture 2]: Implementing Artificial Neural Networks in PyTorch | [Jupyter Notebook 1]
[Lecture 3]: Implementing Convolutional Neural Networks in PyTorch | [Jupyter Notebook 2] | [Jupyter Notebooks as HTML]
[Lecture 4]: Machine Learning Applications in Climate Research
-
(Jan 2023) Humanity has a plastic problem. Conventional plastic manufacturing is still cheaper, but the greener alternatives, bioplastics in particular, are catching up fast. Here's a video that highlights some really innovative initiatives to manufacture bioplastics from Avocados, Sugarcanes, Mushrooms, and even Algae! Interestingly, the Sugarcane based bioplastic company featured in the video is based quite close to my hometown of Ghaziabad, India.
- NCAR Command Language (NCL) pressure interpolation script. Update : NCL is being pivoted in favor of Python!
- Scientific Writing : A Pulitzer prizewinner novelist’s tips on how to write a great science paper.
- Peer-Review : A nice article discussing a three-step process to efficiently review a scientific paper.
- SSW animations : Interesting (read cool!) Potential Vorticity (PV) evolution animations for past stratospheric sudden warming (SSW) events.
- N2 climatology : Zonal mean Brunt-Vaisala frequency in the southern hemisphere for June, July and August 2018 computed from ERA5 reanalysis. I wanted to check my gravity wave potential energy computations but couldn't find a reliable source online to compare the southern hemisphere climatology for N2. In case you find yourself in a similar situation, your search ends here.
- Linear Algebra Writtens' Workshop: Resources for Fall 2016 Linear Algebra Writtens' Workshop. Click here to view the Math Wiki and solutions to past years' problems.
