SmaAt-UNet: Precipitation nowcasting using a small attention-UNet architecture

Trebing, Kevin; Stańczyk, Tomasz; Mehrkanoon, Siamak

doi:10.1016/j.patrec.2021.01.036

Cited by 275 publications

(196 citation statements)

References 8 publications

Supporting

Mentioning

195

Contrasting

Order By: Relevance

“…Approaches based on deep learning have been developed that move beyond reliance on the advection equation 5,6,[14][15][16][17][18][19] . By training these models on large corpora of radar observations rather than relying on in-built physical assumptions, deep learning methods aim to better model traditionally difficult non-linear precipitation phenomena, such as convective initiation and heavy precipitation.…”

mentioning

confidence: 99%

Skilful precipitation nowcasting using deep generative models of radar

Ravuri

Lenc

Willson

et al. 2021

Nature

531

302

View full text Add to dashboard Cite

Precipitation nowcasting, the high-resolution forecasting of precipitation up to two hours ahead, supports the real-world socioeconomic needs of many sectors reliant on weather-dependent decision-making1,2. State-of-the-art operational nowcasting methods typically advect precipitation fields with radar-based wind estimates, and struggle to capture important non-linear events such as convective initiations3,4. Recently introduced deep learning methods use radar to directly predict future rain rates, free of physical constraints5,6. While they accurately predict low-intensity rainfall, their operational utility is limited because their lack of constraints produces blurry nowcasts at longer lead times, yielding poor performance on rarer medium-to-heavy rain events. Here we present a deep generative model for the probabilistic nowcasting of precipitation from radar that addresses these challenges. Using statistical, economic and cognitive measures, we show that our method provides improved forecast quality, forecast consistency and forecast value. Our model produces realistic and spatiotemporally consistent predictions over regions up to 1,536 km × 1,280 km and with lead times from 5–90 min ahead. Using a systematic evaluation by more than 50 expert meteorologists, we show that our generative model ranked first for its accuracy and usefulness in 89% of cases against two competitive methods. When verified quantitatively, these nowcasts are skillful without resorting to blurring. We show that generative nowcasting can provide probabilistic predictions that improve forecast value and support operational utility, and at resolutions and lead times where alternative methods struggle.

show abstract

mentioning

confidence: 99%

Skilful precipitation nowcasting using deep generative models of radar

Ravuri

Lenc

Willson

et al. 2021

Nature

531

302

View full text Add to dashboard Cite

show abstract

“…U-NET: U-NET (Ronneberger et al 2015) is a popular convolution based network architecture initially proposed for biomedical imaging segmentation, but recently used in image super-resolution and precipitation downscaling (Agrawal et al 2019;Trebing and Mehrkanoon 2020). Similar to TRU-NET and HCGRU, U-NET features an encoder and a decoder.…”

Section: Baseline Modelsmentioning

confidence: 99%

TRU-NET: a deep learning approach to high resolution prediction of rainfall

et al. 2021

View full text Add to dashboard Cite

Climate models (CM) are used to evaluate the impact of climate change on the risk of floods and heavy precipitation events. However, these numerical simulators produce outputs with low spatial resolution that exhibit difficulties representing precipitation events accurately. This is mainly due to computational limitations on the spatial resolution used when simulating multi-scale weather dynamics in the atmosphere. To improve the prediction of high resolution precipitation we apply a Deep Learning (DL) approach using input data from a reanalysis product, that is comparable to a climate model’s output, but can be directly related to precipitation observations at a given time and location. Further, our input excludes local precipitation, but includes model fields (weather variables) that are more predictable and generalizable than local precipitation. To this end, we present TRU-NET (Temporal Recurrent U-Net), an encoder-decoder model featuring a novel 2D cross attention mechanism between contiguous convolutional-recurrent layers to effectively model multi-scale spatio-temporal weather processes. We also propose a non-stochastic variant of the conditional-continuous (CC) loss function to capture the zero-skewed patterns of rainfall. Experiments show that our models, trained with our CC loss, consistently attain lower RMSE and MAE scores than a DL model prevalent in precipitation downscaling and outperform a state-of-the-art dynamical weather model. Moreover, by evaluating the performance of our model under various data formulation strategies, for the training and test sets, we show that there is enough data for our deep learning approach to output robust, high-quality results across seasons and varying regions.

show abstract

“…U-Net is a U-shaped architecture comprising an encoder and a decoder. The former applies a double convolution and a max-pooling operation for feature extraction, while the decoder comprises the same layers and is used for up-sampling (Dupuy et al, 2020;Trebing et al, 2021). U-Net achieves a pixel-to-pixel mapping process applied between the input and output image.…”

Section: U-net-based Modelsmentioning

confidence: 99%

“…Attention U-Net (Att-UNet) is a U-Net model using attention modules. Attention is a mechanism that suppresses feature responses within an irrelevant background and directs the network to enhance its attention to task-related important features (Bello et al, 2019;Oktay et al, 2018;Schlemper et al, 2018;Trebing et al, 2021). Several attention mechanisms (Ahmed et al, 2017;Luong et al, 2015;Oktay et al, 2018;S.…”

Section: U-net-based Modelsmentioning

confidence: 99%

“…Accordingly, P. R. Larraondo et al (2020) introduced a methodology for optimizing neural network models using a combination of continuous and categorical binary indices in the context of precipitation forecasting. Moreover, Dupuy et al (2020) suggested a U-Net architecture to perform cloud cover forecast post-processing, while Trebing et al (2021) proposed a U-Net-based model to nowcast Netherlands' precipitation.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep learning‐based precipitation bias correction approach for Yin–He global spectral model

Yin

Zhang

2021

Meteorological Applications

View full text Add to dashboard Cite

In this paper, a data-driven bias correction approach based on deep learning is proposed, which is appropriate for the Yin-He global spectral model (YHGSM) re-forecasting. The proposed architecture involves four U-Net-based networks estimating the proper bias correction models for YHGSM re-forecasting that consider as correction factors the geopotential, specific humidity, and vertical velocity on three pressure levels from the YHGSM model. The proposed models are then evaluated for their bias correction capability on the 3-h cumulative precipitation over the region of China between 15 -54.5 N, and 63 -122.5 E. The results revealed that U-Net-based models could reduce the root mean squared error (RMSE) and improve the threat scores (TSs), especially for heavy precipitation and rainstorms.

show abstract

SmaAt-UNet: Precipitation nowcasting using a small attention-UNet architecture

Cited by 275 publications

References 8 publications

Skilful precipitation nowcasting using deep generative models of radar

Skilful precipitation nowcasting using deep generative models of radar

TRU-NET: a deep learning approach to high resolution prediction of rainfall

Deep learning‐based precipitation bias correction approach for Yin–He global spectral model

Contact Info

Product

Resources

About