| dc.description.abstract | The Bay of Bengal, the world's largest bay, along with the Andaman Sea, a peripheral sea situated in the southeastern part of the bay, is crucial to the economic and maritime security of India. Understanding the dynamics and uncertainties of the Bay of Bengal is particularly important, given India's increased investment in the Deep Ocean Mission and Blue Economy. This thesis examines the variability in the features of the Bay of Bengal in the past and future using reanalysis and climate projections. In two parts, this thesis makes the following specific contributions. In the first part, a new statistical analysis is completed on the reanalysis data to uncover the relationship of salinity dynamics in the Andaman Sea with the Southwest Monsoon Current and a data-assimilative forecast system is developed for synoptic forecasts of the Andaman Sea. In the second part, a novel deep learning model is developed to correct the future projections of the climate models in the Bay of Bengal and uncover new dynamical insights.
To transition from analysis to forecasting, we developed a data assimilation system that integrates satellite observations with numerical models, producing data-driven synoptic forecasts for the Andaman Sea. This research marks the first application of a time-evolving, high-resolution data assimilation ocean model in the Andaman Sea, offering an accurate assessment of the ocean's state in this area. First, we performed a high-resolution data-driven numerical Regional Ocean Modeling System (ROMS) control run. Second, a data assimilative ROMS simulation is performed with MODIS SST observations and compared to the control run. The initial and boundary conditions are taken from the Nucleus for European Modeling of the Ocean (NEMO) reanalysis model for the control and data assimilation runs. The incremental strong-constraint 4-dimensional variational data assimilation scheme (IS4D-Var) was used. The data assimilation run significantly reduces the difference between the Andaman Sea simulations and the in situ observations from a moored OMNI buoy (BD12) over the control run. Significantly, the data assimilation run reduces the surface temperature error by 0.5°C compared to the control run. This data assimilation setup is crucial for improving our understanding of the synoptic dynamics of the Andaman Sea.
In the second part of the thesis, we study the variability of the features of the Bay of Bengal in the future. Climate change impacts the ocean state, including temperature, salinity, and sea level, affecting monsoons and ocean productivity. Future projections based on shared socioeconomic pathways (SSP) from the Coupled Model Intercomparison Project (CMIP) are widely used to understand the effects of climate change. However, CMIP models exhibit considerable errors compared to observations in the Bay of Bengal for the time period when both projections and observations are available. In this region, there is a 1.5$^{\circ}$C root mean square error (RMSE) in the sea surface temperature of CMIP6 compared to the Ocean Reanalysis System (ORAS5). We introduce data-driven deep learning models to correct for this error in three important variables: sea surface temperature, sea surface salinity, and dynamic sea level. The deep neural model for each variable is trained using pairs of monthly CNRM-CM6 projections and the corresponding month's ORAS5 as input and output. This model is trained and validated with historical data (1950-2014) and future projection data (2015-2020) and tested with future projections from 2021 to 2024. Ablation studies are conducted to identify the best neural architecture. The final developed model has a UNet architecture and uses a climatology-removed CMIP6 projection as input and predicts the climatology-removed corrected fields. The trained model is then used to correct the future projections from 2025 to 2100. Our new deep learning-based CMIP6 correction approach has 15% lower RMSE compared to the traditional statistical correction method called the Equidistant Cumulative Distribution Function (EDCDF). Further goodness of correction metrics, such as pattern correlation coefficient and image-based similarity metrics, all indicate the superiority of our correction model.
To uncover the dynamical implication of the corrected projections, a detailed analysis of the monthly, seasonal mean, and variability of the projections is performed. Compared to the raw projections, our corrected projections indicate increased warming in the bay, altered patterns of salinity, and dynamic sea level that affect monsoons and vital oceanographic features of the BoB. The new findings from the corrected projections are as follows. In winter, the north-south temperature gradient weakens more than the raw projections, potentially influencing the northeast monsoon dynamics. The intensified warming in the corrected projections during the pre-monsoon season in the central bay has implications for cyclogenesis and may potentially delay or weaken the monsoon onset. The corrected SST shows a stronger coastal warm front and a changed thermal structure in central and southern bays during the monsoon, potentially impacting monsoon rainfall and SMC propagation. The corrected salinity shows a stronger variability in the propagation of SMC in the BoB, potentially impacting the influx of high salinity to the region and subsequently affecting productivity in the region. The increased post-monsoon bay warming in the corrected projections may aid cyclogenesis and affect the EICC and its eddies. In general, the impact of climate change on the mean and variability of the characteristics in the BoB, revealed by our new corrected model, would be substantially different from the projections of the uncorrected climate models. | en_US |
