We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A high-precision prediction method for coarse grids based on deep learning and the Weather Research and Forecasting model.
- Authors
He, Junyi; Liu, Xinyu; Wang, Hanqing; Zhu, Dongnan; Liu, Zhenming
- Abstract
The Weather Research and Forecasting (WRF) model improves the accuracy of climate prediction and obtains meteorological parameters for fine grids; however, fine-grid climate predictions for different time periods and regions often consumes a great amount of computational resources. In this letter, the Multi Residual Attention Generative Adversarial Network (MRA-GAN) is proposed based on the generative adversarial network; the technique is applied to restore a simulated image from a coarse-grid WRF model to a simulated image from a fine-grid WRF model. The fine-grid image generated by MRA-GAN is very similar to the original fine-grid image. When compared with Super-Resolution Wasserstein Generative Adversarial Network with Gradient Penalty (SRWGAN-GP), the average of peak signal-to-noise was improved by approximately 0.54dB, and the average structural similarity index was improved by approximately 3%. After comparing the downscaling results of temperature, wind speed, and relative humidity of various models with the original data, the results show that MRA-GAN has the highest correlation, the lowest data dispersion, and the smallest data error. The trained network model was able to efficiently transform the coarse-grid meteorological fields from different time periods and different simulation variables into fine-grid meteorological fields, and it greatly reduced the computational workload.
- Subjects
METEOROLOGICAL research; WEATHER forecasting; DEEP learning; GENERATIVE adversarial networks; WIND speed; HUMIDITY
- Publication
Theoretical & Applied Climatology, 2024, Vol 155, Issue 1, p117
- ISSN
0177-798X
- Publication type
Article
- DOI
10.1007/s00704-023-04592-0