arXiv subjects

Niraj Agarwal

Publications and source records attributed to Niraj Agarwal.

3 recordsLinked to original sources

arXiv2026

Skillful Global Ocean Emulation and the Role of Correlation-Aware Loss

Machine learning emulators have shown extraordinary skill in forecasting atmospheric states, and their application to global ocean dynamics offers similar promise. Here, we adapt the GraphCast architecture into a dedicated ocean-only emulator, driven by prescribed atmospheric conditions, for medium-range predictions. The emulator is trained on NOAA's UFS-Replay dataset. Using a 24 hour time step, single initial condition, and without using autoregressive training, we produce an emulator that provides skillful forecasts for 10-15 day lead times. We further demonstrate the use of Mahalanobis distance as loss that improves the forecast skill compared to the Mean Squared Error loss by explicitly accounting for the correlations between tendencies of the target variables. Using spatial correlation analysis of the forecasted fields, we also show that the proposed correlation-aware loss acts as a statistical-dynamical regularizer for the slow, correlated dynamics of the global oceans, offering a better background forecast for downstream tasks like data assimilation.

physics.ao-ph↗

arXiv2024

Assimilating Observed Surface Pressure into ML Weather Prediction Models

There has been a recent surge in development of accurate machine learning (ML) weather prediction models, but evaluation of these models has mainly been focused on medium-range forecasts, not their performance in cycling data assimilation (DA) systems. Cycling DA provides a statistically optimal estimate of model initial conditions, given observations and previous model forecasts. Here, real surface pressure observations are assimilated into several popular ML models using an ensemble Kalman filter, where accurate ensemble covariance estimation is essential to constrain unobserved state variables from sparse observations. In this cycling DA system, deterministic ML models accumulate small-scale noise until they diverge. Mitigating this noise with a spectral filter can stabilize the system, but with larger errors than traditional models. Perturbation experiments illustrate that these models do not accurately represent short-term error growth, leading to poor estimation of cross-variable covariances.

physics.ao-ph↗

arXiv2021

A comparison of data-driven approaches to build low-dimensional ocean models

We present a comprehensive inter-comparison of linear regression (LR), stochastic, and deep-learning approaches for reduced-order statistical emulation of ocean circulation. The reference dataset is provided by an idealized, eddy-resolving, double-gyre ocean circulation model. Our goal is to conduct a systematic and comprehensive assessment and comparison of skill, cost, and complexity of statistical models from the three methodological classes. The model based on LR is considered as a baseline. Additionally, we investigate its additive white noise augmentation and a multi-level stochastic approach, deep-learning methods, hybrid frameworks (LR plus deep-learning), and simple stochastic extensions of deep-learning and hybrid methods. The assessment metrics considered are: root mean squared error, anomaly cross-correlation, climatology, variance, frequency map, forecast horizon, and computational cost. We found that the multi-level linear stochastic approach performs the best for both short- and long-timescale forecasts. The deep-learning hybrid models augmented by additive state-dependent white noise came second, while their deterministic counterparts failed to reproduce the characteristic frequencies in climate-range forecasts. Pure deep learning implementations performed worse than LR and its noise augmentations. Skills of LR and its white noise extension were similar on short timescales, but the latter performed better on long timescales, while LR-only outputs decay to zero for long simulations. Overall, our analysis promotes multi-level LR stochastic models with memory effects, and hybrid models with linear dynamical core augmented by additive stochastic terms learned via deep learning, as a more practical, accurate, and cost-effective option for ocean emulation than pure deep-learning solutions.

physics.ao-ph↗