Abstract:Deep learning approaches have been widely adopted for precipitation nowcasting in recent years. Previous studies mainly focus on proposing new model architectures to improve pixel-wise metrics. However, they frequently result in blurry predictions which provide limited utility to forecasting operations. In this work, we propose a new Fourier Amplitude and Correlation Loss (FACL) which consists of two novel loss terms: Fourier Amplitude Loss (FAL) and Fourier Correlation Loss (FCL). FAL regularizes the Fourier amplitude of the model prediction and FCL complements the missing phase information. The two loss terms work together to replace the traditional $L_2$ losses such as MSE and weighted MSE for the spatiotemporal prediction problem on signal-based data. Our method is generic, parameter-free and efficient. Extensive experiments using one synthetic dataset and three radar echo datasets demonstrate that our method improves perceptual metrics and meteorology skill scores, with a small trade-off to pixel-wise accuracy and structural similarity. Moreover, to improve the error margin in meteorological skill scores such as Critical Success Index (CSI) and Fractions Skill Score (FSS), we propose and adopt the Regional Histogram Divergence (RHD), a distance metric that considers the patch-wise similarity between signal-based imagery patterns with tolerance to local transforms. Code is available at <a class="link-external link-https" href="https://github.com/argenycw/FACL" rel="external noopener nofollow">this https URL</a>

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the common fuzzy prediction problem in precipitation nowcasting. Specifically, existing deep - learning models usually produce fuzzy prediction results when conducting precipitation nowcasting, and these results have limited practicality for actual forecasting operations. The main cause of this fuzzy prediction is the use of pixel - level loss functions (such as the mean - square error MSE), which leads to uncertainties in image transformation and makes the prediction results fuzzy. To overcome this problem, the author proposes a new loss function - **Fourier Amplitude and Correlation Loss (FACL)**, which consists of two novel loss terms: **Fourier Amplitude Loss (FAL)** and **Fourier Correlation Loss (FCL)**. FAL improves the prediction of high - frequency patterns by regularizing the amplitude component in the Fourier space, while FCL supplements the information of the overall image structure. These two loss terms work together to replace the traditional L2 losses (such as MSE and weighted MSE), thereby improving the prediction accuracy and clarity of signal - data - based in spatio - temporal prediction problems. In addition, in order to further improve the error range of meteorological skill scores (such as the Critical Success Index CSI and the Fractional Skill Score FSS), the author also proposes a new distance metric - **Regional Histogram Divergence (RHD)**, which is used to measure the local similarity between signal - image patterns and has a certain tolerance for local transformations. ### Main contributions 1. **Propose FACL**: Combine FAL and FCL to achieve clear and sharp predictions by adjusting the spatial frequency of the prediction. 2. **Theoretical and experimental proof**: FAL significantly improves image sharpness, and FCL supplements the missing accuracy information. 3. **Apply to generative models**: Apply FACL to generative models, and the results show that it performs better on most metrics. 4. **Propose RHD**: A quantitative measurement method that considers regional similarity and visual similarity and is suitable for evaluating the differences between signal - image patterns. ### Summary By introducing FACL and RHD, this paper provides an efficient and deterministic method that can significantly improve the prediction quality of precipitation nowcasting, especially in terms of image sharpness and meteorological skill scores, and provides more useful prediction results for weather forecasting.

Fourier Amplitude and Correlation Loss: Beyond Using L2 Loss for Skillful Precipitation Nowcasting

Enhanced Rainfall Nowcasting of Tropical Cyclone by an Interpretable Deep Learning Model and Its Application in Real-Time Flood Forecasting

Focal Frame Loss: A Simple but Effective Loss for Precipitation Nowcasting

Enhanced Precipitation Nowcasting via Temporal Correlation Attention Mechanism and Innovative Jump Connection Strategy

SFTformer: A Spatial-Frequency-Temporal Correlation-Decoupling Transformer for Radar Echo Extrapolation

A Spatio-Temporal Graph-Guided Convolutional LSTM for Tropical Cyclones Precipitation Nowcasting.

Effective training strategies for deep-learning-based precipitation nowcasting and estimation

Multi-Task Collaboration Deep Learning Framework for Infrared Precipitation Estimation.

Learning Robust Precipitation Forecaster by Temporal Frame Interpolation

Personalized Federated Learning for improving radar based precipitation nowcasting on heterogeneous areas

Nowcasting Heavy Rainfall With Convolutional Long Short-Term Memory Networks: A Pixelwise Modeling Approach

Deep learning model based on multi-scale feature fusion for precipitation nowcasting

Spacetime Separable Latent Diffusion Model With Intensity Structure Information for Precipitation Nowcasting

Improvements in deep learning-based precipitation nowcasting using major atmospheric factors with radar rain rate

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

SCENT: A new precipitation nowcasting method based on sparse correspondence and deep neural network

A Deep Learning‐Based Methodology for Precipitation Nowcasting with Radar

DiffCast: A Unified Framework via Residual Diffusion for Precipitation Nowcasting

Improving Nowcasting of Convective Development by Incorporating Polarimetric Radar Variables Into a Deep-Learning Model

DYffCast: Regional Precipitation Nowcasting Using IMERG Satellite Data. A case study over South America