Deep learning in statistical downscaling for deriving high spatial resolution gridded meteorological data: A systematic review

Sun, Yongjian; Deng, Kefeng; Ren, Kaijun; Liu, Jia; Deng, Chongjiu; Jin, Yongjun

doi:10.1016/j.isprsjprs.2023.12.011

Cited by 16 publications

(3 citation statements)

References 181 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The U-Net regression model consists of two input data streams: normalized high-resolution elevation data (12km) from CCAM and the largescale prognostic predictor variables (1.5°). Our model uses residual convolutional layers (or residual blocks) with batch normalization, which have shown better performance than traditional convolutional layers and help address instability issues in deep-learning models (Rampal, et al, 2024;Sun et al, 2024). Following several residual convolutional blocks and pooling layers, the two input streams are concatenated and mixed to form the latent space of the model.…”

Section: Regression Baselinementioning

confidence: 99%

“…convective high intensity short duration rainfall events) which can have the highest societal impact. While there have been a wide variety of algorithm developments to overcome such issues in regression-based approaches, these issues persist (for recent reviews see Rampal et al, 2024;Sun et al, 2024) .…”

Section: Introductionmentioning

confidence: 99%

“…The effectiveness of GANs for climate downscaling in present-day or future climates has not been well-assessed (Rampal, et al, 2024). Existing research mainly focuses on using traditional error metrics such as root-mean-squared error (Rampal et al, 2024;Sun et al, 2024) instead of climatological metrics. Additionally, GANs are notoriously unstable and challenging to train, where stability is often determined by selecting the correct hyperparameters (Arjovsky et al, 2017;Goodfellow et al, 2014;Gulrajani et al, 2017;Mirza & Osindero, 2014).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

Rampal,

Gibson,

Sherwood

et al. 2024

Preprint

View full text Add to dashboard Cite

Anticipating climate impacts and risks in present or future climates requires predicting the statistics of high-impact weather events at fine-scales. Direct numerical simulations of fine-scale weather are computationally too expensive for many uses. While regression-based (deep-learning or statistical) downscaling of low-resolution climate simulations is several orders of magnitude faster than direct numerical simulations, it suffers from several limitations. These limitations include the tendency to regress to the mean, which produces excessively smooth predictions and underestimates the magnitude of extreme events. Additionally, they also fail to preserve statistical measures that are key for climate research. We use a conditional GAN (c-GAN) architecture to downscale daily precipitation as a Regional Climate Model (RCM) emulator. The c-GAN generates plausible residuals on top of the predictable expectation state produced by a regression-based DL algorithm. The skill of c-GANs is highly sensitive to a hyperparameter known as the weight of the adversarial loss (λadv), and the value of λadv required for accurate results varies with season and performance metric, casting doubt on the robustness of c-GANs as usually implemented. But, by applying a simple intensity constraint to the loss function, it is possible to obtain robust performance results across λadv spanning two orders of magnitude. C-GANs are considerably more skillful in capturing climatological statistics including the distribution and spatial characteristics of extreme events. We expect c-GANs with this modification to be readily transferable to other problems and time periods, making them a useful weather generator for representing extreme event statistics in present and future climates.

show abstract

Section: Regression Baselinementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

Rampal,

Gibson,

Sherwood

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

A Machine Learning-Based Parameterized Tropical Cyclone Precipitation Model

Lu,

Yin,

Chen

et al. 2024

Int J Disaster Risk Sci

View full text Add to dashboard Cite

Current simulation models considerably underestimate local-scale, short-duration extreme precipitation induced by tropical cyclones (TCs). This problem needs to be addressed to establish active response policies for TC-induced disasters. Taking Shanghai, a coastal megacity, as a study area and based on the observations from 192 meteorological stations in the city during 2005–2018, this study optimized the parameterized Tropical Cyclone Precipitation Model (TCPM) initially designed for TCs at the national scale (China) to the local or regional scales by using machine learning (ML) methods, including the random forest (RF), extreme gradient boosting (XGBoost), and ensemble learning (EL) algorithms. The TCPM-ML was applied for multiple temporal scale hazard assessment. The results show that: (1) The TCPM-ML not only improved TCPM performance for simulating hourly extreme precipitations, but also preserved the physical meaning of the results, contrary to ML methods; (2) Machine learning algorithms enhanced the TCPM ability to reproduce observations, although the hourly extreme precipitations remained slightly underestimated; (3) Best performance was obtained with the XGBoost or EL algorithms. Combining the strengths of both XGBoost and RF, the EL algorithm yielded the best overall performance. This study provides essential model support for TC disaster risk assessment and response at the local and regional scales in China.

show abstract

Downscaling sea surface height and currents in coastal regions using convolutional neural network

Yuan,

Jacob,

Chen

et al. 2024

Applied Ocean Research

View full text Add to dashboard Cite

Deep learning in statistical downscaling for deriving high spatial resolution gridded meteorological data: A systematic review

Cited by 16 publications

References 181 publications

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

A Machine Learning-Based Parameterized Tropical Cyclone Precipitation Model

Downscaling sea surface height and currents in coastal regions using convolutional neural network

Contact Info

Product

Resources

About