TempEE: Temporal-Spatial Parallel Transformer for Radar Echo Extrapolation Beyond Auto-Regression

by   Shengchao Chen, et al.

The meteorological radar reflectivity data, also known as echo, plays a crucial role in predicting precipitation and enabling accurate and fast forecasting of short-term heavy rainfall without the need for complex Numerical Weather Prediction (NWP) model. Compared to conventional model, Deep Learning (DL)-based radar echo extrapolation algorithms are more effective and efficient. However, the development of highly reliable and generalized algorithms is hindered by three main bottlenecks: cumulative error spreading, imprecise representation of sparse echo distribution, and inaccurate description of non-stationary motion process. To address these issues, this paper presents a novel radar echo extrapolation algorithm that utilizes temporal-spatial correlation features and the Transformer technology. The algorithm extracts features from multi-frame echo images that accurately represent non-stationary motion processes for precipitation prediction. The proposed algorithm uses a novel parallel encoder based on Transformer technology to effectively and automatically extract echoes' temporal-spatial features. Furthermore, a Multi-level Temporal-Spatial attention mechanism is adopted to enhance the ability to perceive global-local information and highlight the task-related feature regions in a lightweight way. The proposed method's effectiveness has been valided on the classic radar echo extrapolation task using the real-world dataset. Numerous experiments have further demonstrated the effectiveness and necessity of various components of the proposed method.


page 1

page 6

page 11


FDNet: A Deep Learning Approach with Two Parallel Cross Encoding Pathways for Precipitation Nowcasting

With the goal of predicting the future rainfall intensity in a local reg...

Multistep Speed Prediction on Traffic Networks: A Graph Convolutional Sequence-to-Sequence Learning Approach with Attention Mechanism

Multistep traffic forecasting on road networks is a crucial task in succ...

DeepReflecs: Deep Learning for Automotive Object Classification with Radar Reflections

This paper presents an novel object type classification method for autom...

Intrinsic Non-stationary Covariance Function for Climate Modeling

Designing a covariance function that represents the underlying correlati...

Deep Temporal Interpolation of Radar-based Precipitation

When providing the boundary conditions for hydrological flood models and...

Convolutional Neural Network for Convective Storm Nowcasting Using 3D Doppler Weather Radar Data

Convective storms are one of the severe weather hazards found during the...

Please sign up or login with your details

Forgot password? Click here to reset