Probabilistic quantile multiple fourier feature network for lake temperature forecasting: incorporating pinball loss for uncertainty estimation

Abstract

Lake temperature forecasting is crucial for understanding and mitigating climate change impacts on aquatic ecosystems. The meteorological time series data and their relationship have a high degree of complexity and uncertainty, making it difficult to predict lake temperatures. In this study, we propose a novel approach, Probabilistic Quantile Multiple Fourier Feature Network (QMFFNet), for accurate lake temperature prediction in Qinghai Lake. Utilizing only time series data, our model offers practical and efficient forecasting without the need for additional variables. Our approach integrates quantile loss instead of L2-Norm, enabling probabilistic temperature forecasts as probability distributions. This unique feature quantifies uncertainty, aiding decision-making and risk assessment. Extensive experiments demonstrate the method’s superiority over conventional models, enhancing predictive accuracy and providing reliable uncertainty estimates. This makes our approach a powerful tool for climate research and ecological management in lake temperature forecasting. Innovations in probabilistic forecasting and uncertainty estimation contribute to better climate impact understanding and adaptation in Qinghai Lake and global aquatic systems.