Zero-Shot Learning for Salinity Forecasting

Salinity forecasting is vital for ensuring sustainable aquatic ecosystems, optimizing agricultural practices, and safeguarding water resources from the adverse impacts of salinization. In time series forecasting, limited data often poses significant challenges. This research specifically addresses this constraint in the context of salinity forecasting. Central to our approach is the use of Zero-Shot Learning, a self-supervised technique that allows models to make predictions in scenarios they have never seen during training. Following this, we employ Transfer Learning, fine-tuning the model using the weights obtained from Zero-Shot Learning to enhance its forecasting accuracy further. Code is available at: https://github.com/neko941/Self-Supervised-Learning-for-Salinity-Forecasting .