The recent advancements in time series forecasting and flood prediction have seen a significant shift towards leveraging deep learning models that integrate complex temporal and spatial dependencies. Innovations in model architectures, such as the incorporation of wavelet-based tokenization and multi-scale decomposition, have enabled more accurate and efficient forecasting. These models are not only enhancing the accuracy of predictions but also demonstrating superior generalization capabilities across diverse datasets. Additionally, the integration of self-supervised learning and probabilistic density labeling has opened new avenues for improving the accuracy of weather and flood predictions, particularly in the context of extreme weather events. The field is also witnessing a move towards more interpretable models, with methods like Windowed Temporal Saliency Rescaling offering insights into the decision-making processes of complex models. Furthermore, the application of machine learning in urban resilience planning, particularly in flood-prone areas, is providing dynamic and data-driven strategies for enhancing community preparedness and response. Overall, the research is advancing towards more sophisticated, interpretable, and adaptable solutions that address the complexities of time series data and the urgent need for accurate environmental predictions.
Advances in Time Series Forecasting and Flood Prediction
Sources
WinTSR: A Windowed Temporal Saliency Rescaling Method for Interpreting Time Series Deep Learning Models
Short-term Streamflow and Flood Forecasting based on Graph Convolutional Recurrent Neural Network and Residual Error Learning
Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization
KEDformer:Knowledge Extraction Seasonal Trend Decomposition for Long-term Sequence Prediction
Self-Supervised Learning with Probabilistic Density Labeling for Rainfall Probability Estimation
Curse of Attention: A Kernel-Based Perspective for Why Transformers Fail to Generalize on Time Series Forecasting and Beyond
Applying Machine Learning Tools for Urban Resilience Against Floods
APS-LSTM: Exploiting Multi-Periodicity and Diverse Spatial Dependencies for Flood Forecasting
LMS-AutoTSF: Learnable Multi-Scale Decomposition and Integrated Autocorrelation for Time Series Forecasting
Accurate Prediction of Temperature Indicators in Eastern China Using a Multi-Scale CNN-LSTM-Attention model
Towards modeling evolving longitudinal health trajectories with a transformer-based deep learning model
STORM: A Spatio-Temporal Factor Model Based on Dual Vector Quantized Variational Autoencoders for Financial Trading