Data assimilation (DA) plays a pivotal role in diverse applications, ranging from climate predictions and weather forecasts to trajectory planning for autonomous vehicles. A prime example is the widely used ensemble Kalman filter (EnKF), which relies on linear updates to minimize variance among the ensemble of forecast states. Recent advancements have seen the emergence of deep learning approaches in this domain, primarily within a supervised learning framework. However, the adaptability of such models to untrained scenarios remains a challenge. In this study, we introduce a novel DA strategy that utilizes reinforcement learning (RL) to apply state corrections using full or partial observations of the state variables. Our investigation focuses on demonstrating this approach to the chaotic Lorenz ’63 system, where the agent’s objective is to minimize the root-mean-squared error between the observations and corresponding forecast states. Consequently, the agent develops a correction strategy, enhancing model forecasts based on available system state observations. Our strategy employs a stochastic action policy, enabling a Monte Carlo-based DA framework that relies on randomly sampling the policy to generate an ensemble of assimilated realizations. Results demonstrate that the developed RL algorithm performs favorably when compared to the EnKF. Additionally, we illustrate the agent’s capability to assimilate non-Gaussian data, addressing a significant limitation of the EnKF.

We investigate the impact of ocean data assimilation using the Ensemble Adjustment Kalman Filter (EAKF) from the Data Assimilation Research Testbed (DART) on the oceanic and atmospheric states of the Red Sea. Our study extends the ocean data assimilation experiment performed by Sanikommu et al. (2020) by utilizing the SKRIPS model coupling the MITgcm ocean model and the Weather Research and Forecasting (WRF) atmosphere model. Using a 50-member ensemble, we assimilate satellite-derived sea surface temperature and height and in-situ temperature and salinity profiles every three days for one year, starting January 01 2011. Atmospheric data are not assimilated in the experiments. To improve the ensemble realism, perturbations are added to the WRF model using several physics options and the stochastic kinetic energy backscatter (SKEB) scheme. Compared with the control experiments using uncoupled MITgcm with ECMWF ensemble forcing, the EAKF ensemble mean oceanic states from the coupled model are better or insignificantly worse (root-mean-square-errors are 30% to -2% smaller), especially when the atmospheric model uncertainties are accounted for with stochastic perturbations. We hypothesize that the ensemble spreads of the air–sea fluxes are better represented in the downscaled WRF ensembles when uncertainties are well accounted for, leading to improved representation of the ensemble oceanic states in EAKF. Although the feedback from ocean to atmosphere is included in this two-way regional coupled configuration, we find no significant effect of ocean data assimilation on the latent heat flux and 10-m wind speed, suggesting the improved skill is from downscaling the ensemble atmospheric forcings.

Ensemble Kalman Filters (EnKFs), which assimilate observations based on statistics derived from samples of ocean states called ensemble, have become the norm for ocean data assimilation (DA) and forecasting. These schemes are commonly implemented with inflation and localization techniques to increase their ensemble spread and to filter out spurious long-range correlations resulting from the limited-size ensembles imposed by computational burden constraints. Such ad hoc methods were found not necessary in ensemble DA experiments with simplified ocean/atmospheric models and large ensembles. Here, we conduct a series of 1-year-long ensemble experiments with a fully realistic EnKF-DA system in the Red Sea using tens-to-thousands of ensemble members. The system assimilates satellite and in-situ observations and accounts for model uncertainties by integrating a 4km-resolution ocean model with ECMWF atmospheric ensemble fields, perturbed internal physics and initial conditions for forecasting. Our results indicate that accounting for model uncertainties is more beneficial than simply increasing the ensemble size, with the improvements due to large ensemble leveling off at about 250 members. Besides, and in contrast to what is commonly observed with simplified models, the investigated ensemble DA system still required localization even when implemented with thousands of members. These findings are explained by (i) amplified spurious long-range correlations produced by the low-rank nature of the ECMWF atmospheric forcing ensemble, and (ii) non-Gaussianity generated by the perturbed internal physical parameterization schemes. Large ensemble forcing fields and non-Gaussian DA methods might be needed to take full benefits from large ensembles in ocean DA.

The Red Sea is an extremely warm tropical sea that hosts diverse ecosystems; thus, it is important to understand its ecology in the context of global warming. Using a coupled physical–biogeochemical model validated against in situ data, we provide the first report on the diel cycle (i.e., diel variability) in the Red Sea chlorophyll (CHL) concentration, revealing near-sunset CHL maxima at 17h ± 1h local time over the entire basin. This CHL peak time is considerably later than those reported in most other oceans, suggesting low grazing rates in this high-irradiance tropical sea. Model-based analyses reveal that CHL diel cycle is predominantly controlled by irradiance, whereas longer-timescale (e.g., seasonal) CHL variability is regulated by nutrient availability, suggesting a light-limited biological production at diel timescale. The identified CHL diel cycle comprises a fundamental component of the Red Sea ecology and has implications for CHL remote sensing and in situ measurements.

The accuracy of initial conditions is an important driver of the forecast skill of numerical weather prediction models. Increases in the quantity of available measurements, in particular high-resolution remote sensing observational data products from satellites, are valuable inputs for improving those initial condition estimates. However, the data assimilation methods used for integrating observations into forecast models are computationally expensive. This makes incorporating dense observations into operational forecast systems challenging, and it is often prohibitively time-consuming. As a result, large quantities of data are discarded and not used for state initialization. We demonstrate, using the Lorenz-96 system for testing, that a simple machine learning method can be trained to assimilate high-resolution data. Using it to do so improves both initial conditions and forecast accuracy. Compared to using the Ensemble Kalman Filter with high-resolution observations ignored, our augmented method has an average root-mean-squared error reduced by 15%. Ensemble forecasts using initial conditions generated by the augmented method are more accurate and reliable at up to 10 days of forecast lead time.

Waves generated by tropical cyclones can have devastating effects on coastal regions. However, the role of ocean currents in modifying wave amplitudes, wavelengths, and directions is commonly overlooked in wave forecasts, despite the fact that these interactions can lead to extreme wave conditions. Here, we use satellite observations and wave modeling to quantify the effects of ocean currents on the surface waves generated during a tropical cyclone event in the Arabian Sea. As a case study, this paper documents beams of wave heights originating from the eyewall of a tropical cyclone caused by current-induced refraction. Alternating regions of high and low wave heights in the model simulations are consistent with observations and extend for thousands of kilometers all the way to 100 m isobath. Our results highlight the importance of accounting for wave refraction by currents in order to accurately predict the impact of tropical cyclone generated waves on coastal regions.

Recent observations of warming trends in the Red Sea raise more attention to the response of the basin under a warming climate. Using two remotely sensed datasets, the Hadley Centre Sea Ice and Sea Surface Temperature [HadISST] and Extended Reconstructed Sea Surface Temperature [ERSST.v3], we investigate the reported sudden increase in the Red Sea sea surface temperatures (SST) in terms of average and maximum and assess their relation to multi-decadal climate variability. Prior to the analysis, the two datasets are successfully validated with respect to their ability to reproduce the recent observed and reported trends and their spatial features. Analysis of long-term SST variability revealed a sequence of alternating and similar in amplitude positive and negative trends, characterized by a period of nearly 70 years. Similar oscillations have been reported in other basins and have been related to atmospheric disturbances associated with the Atlantic Multidecadal Oscillation (AMO). A point-by-point spectral analysis of SST evolution shows a significant correlation with the basic modes of the AMO that explains a large fraction of its temporal and spatial variability. Projections on the major modes of the spectral analysis suggest a possible decreasing effect on local SST in the near future. Under this assumption, recent projected trends in the Red Sea may be exaggerated, whilst trends that may be related to anthropogenic influence could be masked by the projected negative influence of the AMO in the near future.