Abstract:The increasing generation of data in different areas of life, such as the environment, highlights the need to explore new techniques for processing and exploiting data for useful purposes. In this context, artificial intelligence techniques, especially through deep learning models, are key tools to be used on the large amount of data that can be obtained, for example, from weather radars. In many cases, the information collected by these radars is not open, or belongs to different institutions, thus needing to deal with the distributed nature of this data. In this work, the applicability of a personalized federated learning architecture, which has been called adapFL, on distributed weather radar images is addressed. To this end, given a single available radar covering 400 km in diameter, the captured images are divided in such a way that they are disjointly distributed into four different federated clients. The results obtained with adapFL are analyzed in each zone, as well as in a central area covering part of the surface of each of the previously distributed areas. The ultimate goal of this work is to study the generalization capability of this type of learning technique for its extrapolation to use cases in which a representative number of radars is available, whose data can not be centralized due to technical, legal or administrative concerns. The results of this preliminary study indicate that the performance obtained in each zone with the adapFL approach allows improving the results of the federated learning approach, the individual deep learning models and the classical Continuity Tracking Radar Echoes by Correlation approach.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to use Personalized Federated Learning (PFL) technology to improve precipitation nowcasting based on radar images in different regions. Specifically, the research aims to address the following challenges: 1. **Heterogeneity of data distribution**: Since the data distribution of weather radar images is highly heterogeneous in different regions, traditional federated learning methods may not converge or generalize effectively. The paper proposes a personalized federated learning method named adapFL to adapt to this heterogeneity. 2. **Data privacy and security**: Radar data usually belong to different institutions and cannot be centrally stored due to technical, legal or administrative reasons. Therefore, a method for collaborative learning without sharing data is required. The PFL method allows all participating parties to jointly train a more powerful model while maintaining data privacy. 3. **Model performance improvement**: The research hopes to verify whether the personalized federated learning method can provide better prediction performance than individual deep - learning models, traditional federated learning methods and the classic Continuity Tracking Radar Echoes by Correlation (COTREC) method. ### Research background and motivation - **Deep - learning techniques in meteorology**: Deep - learning models perform well in processing meteorological radar images, but there are challenges of fuzzy prediction and special loss functions are required for processing. Technologies such as Generative Adversarial Network (GAN) have been successfully applied to solve this problem. - **Federated learning and personalized federated learning**: Federated learning allows models to be trained on distributed data without centralizing data, and is applicable to multiple fields such as medical treatment, intrusion detection, and credit card fraud detection. Personalized federated learning further customizes models to better adapt to the data of each client while maintaining privacy and security. ### Method overview 1. **Data processing**: - Use images captured by X - band meteorological radar, covering an area with a diameter of 400 kilometers. - Divide the image into four quadrants to simulate each radar covering a part of the total area. - The data of each quadrant is used to train different models. 2. **Model architecture**: - **Classic model**: Use COTREC as the baseline model. - **Deep - learning model**: Develop a simple Convolutional Neural Network (CNN) for individual learning and personalized federated learning. - **Federated learning model**: Adopt a horizontal federated learning architecture and aggregate the model weights of each client through weighted averaging. - **Personalized federated learning model**: On the basis of federated learning, each client conducts additional local training to capture the specific characteristics of their respective regions. 3. **Experimental setup**: - Use the AI4EOSC platform for experiments, configured with 16GB RAM, 10GB disk, 8 CPU cores and an NVIDIA Tesla T4 GPU. - The model uses the Adam optimizer and the Mean Squared Error (MSE) loss function, and the monitoring indicators include MSE and Mean Absolute Error (MAE). ### Results and analysis - **Performance comparison**: - Table 4 summarizes the MSE and MAE results on the four - region test sets. - The results show that the adapFL method performs best in most regions, especially in improving the generalization ability and prediction accuracy of the model compared to other methods. ### Conclusion This research shows the potential of the personalized federated learning method in processing distributed radar image data, especially in significantly improving the performance of precipitation nowcasting while maintaining data privacy. Future work can further explore more applications of radar data and the effects on larger - scale and more diverse data sets.

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

People Flow Detection Algorithm Based on a Multiencoder-Classifier Cotraining Architecture for FMCW Radar.

Deep Learning Based Coded Over-the-Air Computation for Personalized Federated Learning

Precipitation Nowcasting Leveraging Spatial Correlation Feature Extraction and Deep Spatio-Temporal Fusion Network

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

When Cars meet Drones: Hyperbolic Federated Learning for Source-Free Domain Adaptation in Adverse Weather

An Long Short-Term Memory Model with Multi-Scale Context Fusion and Attention for Radar Echo Extrapolation

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

A Deep Learning‐Based Methodology for Precipitation Nowcasting with Radar

Enhancing Radar Echo Extrapolation by ConvLSTM2D for Precipitation Nowcasting

CEMA-LSTM: Enhancing Contextual Feature Correlation for Radar Extrapolation Using Fine-Grained Echo Datasets

Deep Precipitation Nowcasting with Dual Regions Displacement Information and Global Spatiotemporal Representations Learning

Spatiotemporal Prediction of Radar Echoes Based on ConvLSTM and Multisource Data

FedRAD: Heterogeneous Federated Learning via Relational Adaptive Distillation

FsrGAN: A Satellite and Radar-Based Fusion Prediction Network for Precipitation Nowcasting

Deep Temporal Interpolation of Radar-Based Precipitation

PrecipLSTM: A Meteorological Spatiotemporal LSTM for Precipitation Nowcasting

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

Machine Learning for Precipitation Nowcasting from Radar Images

Location-Refining Neural Network: A New Deep Learning-Based Framework for Heavy Rainfall Forecast

Spatiotemporal Data Fusion for Precipitation Nowcasting