Abstract:Utilizing the real-time grid dataset CLDAS spanning from 2017 to 2020 and an optimal set of variables from the ECMWF IFS model, we developed a temperature prediction model based on Spatiotemporal Stacked ResNet (Res-STS). To assess the efficacy of the Res-STS model, we devised two additional modelling schemes: the first, denoted as the LSTM scheme, replaced the spatiotemporal stacking technology with long short-term memory (LSTM) networks on the basis of the Res-STS model, while the second, denoted as the T2m scheme, replaced the optimal factor set with a single meteorological element, specifically 2 m temperature. Consequently, we generated three sets of hourly temperature and daily maximum/minimum temperature forecasts. A comparative analysis was then conducted, pitting the predictions from the three schemes against those from the IFS model to evaluate the performance of the Res-STS model. The findings showed a weak unimodal pattern in the Mean Absolute Error (MAE) of hourly forecasts across all models, with errors peaking in the afternoon when the daily maximum temperature was reached. All schemes revealed superior and more consistent performance in the minimum temperature prediction compared to their maximum temperature counterparts. Notable errors were observed in high-altitude areas. The Res-STS scheme, using the Res-STS model with spatiotemporal stacking technology, outperformed the IFS model, especially in forecasting the maximum temperature. This improvement extended throughout the entire province, particularly in high-altitude mountainous regions. Comparing with the alternative schemes, the better performance of the Res-STS scheme underscored the greater contribution of the spatiotemporal stacking technology compared to the optimal factor set. Particularly in high-altitude regions, the Res-STS scheme demonstrated superior performance over the T2m scheme, showing its capability to reduce forecast errors caused by complex terrain, where the optimal factor set played a crucial role.

TempCast: A Multi-modal Transformer for Short-Term Temperature Forecasting

TFEformer: Temporal Feature Enhanced Transformer for Multivariate Time Series Forecasting

MTTF: a Multimodal Transformer for Temperature Forecasting

Hidformer: Hierarchical Dual-Tower Transformer Using Multi-Scale Mergence for Long-Term Time Series Forecasting

TENT: Tensorized Encoder Transformer for Temperature Forecasting

NTDformer: A Multi-Scale Forecasting Model for Non-Stationary Multilevel Time Series

FATE: Focal-modulated Attention Encoder for Temperature Prediction

Spatio-Temporal Transformer Network for Weather Forecasting

WeatherFormer: Empowering Global Numerical Weather Forecasting with Space-Time Transformer

Long-Term Prediction of Sea Surface Temperature by Temporal Embedding Transformer With Attention Distilling and Partial Stacked Connection

Efficient Subseasonal Weather Forecast using Teleconnection-informed Transformers

FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

Precipitation Nowcasting Using Diffusion Transformer with Causal Attention

TERMCast: Temporal Relation Modeling for Effective Urban Flow Forecasting

Learning Robust Precipitation Forecaster by Temporal Frame Interpolation

Stecformer: Spatio-temporal Encoding Cascaded Transformer for Multivariate Long-term Time Series Forecasting

W-MAE: Pre-trained weather model with masked autoencoder for multi-variable weather forecasting

Improved temperature prediction using deep residual networks in Hunan Province, China

A Lightweight Transformer-Based Spatiotemporal Analysis Prediction Algorithm for High-Dimensional Meteorological Data

MSLKSTNet: Multi-Scale Large Kernel Spatiotemporal Prediction Neural Network for Air Temperature Prediction

Deep Learning for Weather Forecasting: A CNN-LSTM Hybrid Model for Predicting Historical Temperature Data