20230143145. Increasing Accuracy and Resolution of Weather Forecasts Using Deep Generative Models simplified abstract (ClimateAI, Inc.)
Increasing Accuracy and Resolution of Weather Forecasts Using Deep Generative Models
Organization Name
Inventor(s)
Ilan Shaun Posel Price of Oxford (GB)
Increasing Accuracy and Resolution of Weather Forecasts Using Deep Generative Models - A simplified explanation of the abstract
This abstract first appeared for US patent application 20230143145 titled 'Increasing Accuracy and Resolution of Weather Forecasts Using Deep Generative Models
Simplified Explanation
The patent application describes the use of a conditional generative adversarial network (GAN) to improve global ensemble weather or climate forecasts. The network consists of a generator deep neural network (G-DNN) with a corrector DNN (C-DNN) and a super-resolver DNN (SR-DNN). The C-DNN corrects coarse meteorological forecasts, considering other relevant meteorological fields, while the SR-DNN downscales the corrected output to a higher resolution. The GAN is trained in three stages: C-DNN training, SR-DNN training, and overall GAN training, each with separate loss functions. The technology outperforms interpolation methods and approaches the performance of regional high-resolution forecast models.
- Conditional generative adversarial network (GAN) used to correct and downscale global ensemble weather or climate forecasts
- Generator deep neural network (G-DNN) consists of a corrector DNN (C-DNN) and a super-resolver DNN (SR-DNN)
- C-DNN bias-corrects coarse meteorological forecasts, considering other relevant meteorological fields
- SR-DNN downscales the corrected output to a higher target spatial resolution
- GAN trained in three stages: C-DNN training, SR-DNN training, and overall GAN training, each with separate loss functions
- Outperforms interpolation methods and approaches the performance of regional high-resolution forecast models
Potential Applications
- Improving global ensemble weather or climate forecasts
- Enhancing the accuracy and resolution of meteorological predictions
Problems Solved
- Correcting and downscaling coarse meteorological forecasts
- Improving the resolution and accuracy of global ensemble weather or climate forecasts
Benefits
- Significantly outperforms interpolation methods
- Approaches the performance of regional high-resolution forecast models
- High-resolution predictions can be generated in seconds on a single machine
Original Abstract Submitted
embodiments of the present invention provide the use of a conditional generative adversarial network (gan) to simultaneously correct and downscale (super-resolve) global ensemble weather or climate forecasts. specifically, a generator deep neural network (g-dnn) in the cgan comprises a corrector dnn (c-dnn) followed by a super-resolver dnn (sr-dnn). the c-dnn bias-corrects coarse, global meteorological forecasts, taking into account other relevant contextual meteorological fields. the sr-dnn downscales bias-corrected c-dnn output into g-dnn output at a higher target spatial resolution. the gan is trained in three stages: c-dnn training, sr-dnn training, and overall gan training, each using separate loss functions. embodiments of the present invention significantly outperform an interpolation baseline, and approach the performance of operational regional high-resolution forecast models across an array of established probabilistic metrics. crucially, embodiments of the present invention, once trained, produce high-resolution predictions in seconds on a single machine.
