Using Machine Learning for Enhanced Seasonal Water Level Predictions in the Great Lakes

The Great Lakes of North America form one of the largest freshwater systems on Earth, exhibiting significant seasonal fluctuations in lake levels, with variations of up to 0.5 meters. These fluctuations present substantial challenges for coastal resilience, flood risk management, and navigation planning. Accurate seasonal water level predictions using mechanistic models have been difficult due to the complexity of the physical mechanisms and coupled processes within the Great Lakes hydroclimate. Recently, machine learning has shown great potential in geoscience, demonstrating its ability to identify patterns in complex multi-physical datasets. We have developed a novel machine learning approach, the dual-transformer model, using the CFSR (Climate Forecast System Reanalysis) dataset, the model has shown promise for estimating historical seasonal water levels in the Great Lakes. However, the model's dependence on historical data limits its ability to provide forward-looking seasonal predictions. To address this, one potential approach is to transform the CFSv2 (Climate Forecast System Version 2) dataset into a format with bias characteristics comparable to CFSR. This transformation would allow the dual-transformer model to perform well without relying on CFSR. Our study examines a suite of machine learning approaches to achieve this transformation, including MLP (multi-layer perceptron), CGAN (conditional generative adversarial network), and Conditional Diffusion Model. By integrating these transformation models with the dual-transformer framework, we aim to enable predictions with a similar level of accuracy as those made with the original CFSR dataset. Our results demonstrate that this integrated approach successfully aligns the datasets, offering great potential for improving seasonal water level forecasts for the Great Lakes. This advancement holds significant value for the community by enhancing the ability to manage water resources proactively, reducing economic losses associated with flooding, and supporting navigation planning, thereby contributing to the resilience and sustainability of the Great Lakes region.