Abstract:Heavy precipitation from tropical cyclones (TCs) may result in disasters, such as floods and landslides, leading to substantial economic damage and loss of life. Prediction of TC precipitation based on ensemble post-processing procedures using machine learning (ML) approaches has received considerable attention for its flexibility in modeling and its computational power in managing complex models. However, when applying ML techniques to TC precipitation for a specific area, the available observation data are typically insufficient for comprehensive training, validation, and testing of the ML model, primarily due to the rapid movement of TCs. We propose to use the convolutional neural network (CNN) as a deep ML model to leverage the spatial information of precipitation. The proposed model has three distinct features that differentiate it from traditional CNNs applied in meteorology. First, it utilizes data augmentation to alleviate challenges posed by the small sample size. Second, it contains geographical and dynamic variables to account for area-specific features and the relative distance between the study area and the moving TC. Third, it applies unequal weights to accommodate the temporal structure in the training data when calculating the objective function. The proposed CNN-all model is then illustrated with the TC Soudelor's impact on Taiwan. Soudelor was the strongest TC of the 2015 Pacific typhoon season. The results show that the inclusion of augmented data and dynamic variables improves the prediction of heavy precipitation. The proposed CNN-all outperforms traditional CNN models, based on the continuous probability skill score (CRPSS), probability plots, and reliability diagram. The proposed model has the potential to be utilized in a wide range of meteorological studies.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the data insufficiency in tropical cyclone (TC) precipitation forecasting. Specifically, due to the fast - moving characteristics of tropical cyclones, the amount of data available for training machine - learning models is limited, thus affecting the training, validation, and testing effects of the models. To solve this problem, the author proposes an integrated post - processing method based on convolutional neural networks (CNN) and introduces data augmentation techniques to expand the training data set. In addition, the model also combines geometric and dynamic features to improve prediction accuracy. ### Main Problem Summary: 1. **Insufficient Data**: The rapid movement of tropical cyclones makes the amount of observational data available for specific areas very limited, which restricts the effective training of machine - learning models. 2. **Complexity of Spatio - Temporal Structure**: The movement and precipitation distribution of tropical cyclones have complex spatio - temporal structures, and traditional machine - learning methods are difficult to fully capture these features. 3. **Limitations of Existing Methods**: Existing statistical post - processing methods (such as EMOS and BMA) are limited in the number of parameters by the amount of data and cannot fully utilize a large number of input variables. ### Solutions: 1. **Convolutional Neural Network (CNN)**: Use CNN to extract spatial information and improve the prediction ability of tropical cyclone precipitation. 2. **Data Augmentation**: Increase the amount of training data through methods such as linear combination and noise injection to alleviate the small - sample problem. 3. **Geometric and Dynamic Features**: Introduce geographical information (such as longitude, latitude, and altitude) and dynamic features (such as the distance from the tropical cyclone center) to better describe the characteristics of specific areas. 4. **Unbalanced Weight Loss Function**: When calculating the loss function, assign different weights to data at different time points to consider the time structure of the data. ### Experimental Verification: The paper conducts a case study on the impact of Typhoon Soudelor on Taiwan, demonstrating the superior performance of the proposed CNN - all model in predicting heavy precipitation. The experimental results show that adding augmented data and dynamic features can significantly improve the prediction effect and perform well in evaluations such as the Continuous Ranked Probability Skill Score (CRPSS), probability maps, and reliability diagrams. Through the above methods, this research aims to provide a more effective tropical cyclone precipitation forecasting tool, especially suitable for data - scarce situations.

A Convolutional Neural Network-based Ensemble Post-processing with Data Augmentation for Tropical Cyclone Precipitation Forecasts

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

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

A Data-Driven Approach for Regional-Scale Fine-Resolution Disaster Impact Prediction under Tropical Cyclones

Post-processing of numerical precipitation forecast based on spatial-temporal deep learning model

Enhanced Tropical Cyclone Inflow Flood Forecasts by Using Deep Learning and Spatial&#8209;temporal Information

A Neural Network Framework for Fine-Grained Tropical Cyclone Intensity Prediction

A multimodal deep learning approach for hurricane tack forecast based on encoder-decoder framework

Global Tropical Cyclone Precipitation Estimation Via a Multitask Convolutional Neural Network Based on HURSAT-B1 Data

TCP-Diffusion: A Multi-modal Diffusion Model for Global Tropical Cyclone Precipitation Forecasting with Change Awareness

A hybrid CNN-LSTM model for typhoon formation forecasting

Tropical Cyclone Track Prediction with an Encoding-to-Forecasting Deep Learning Model

A Novel Data-Driven Tropical Cyclone Track Prediction Model Based on CNN and GRU with Multi-Dimensional Feature Selection

Short-Term Intensity Prediction of Tropical Cyclones Based on Multi-Source Data Fusion with Adaptive Weight Learning

Short-term prediction of the intensity and track of tropical cyclone via ConvLSTM model

Near Real-Time Hurricane Rainfall Forecasting Using Convolutional Neural Network Models with Integrated Multi-satellitE Retrievals for GPM (IMERG) Product

Malonyl Coenzyme A Decarboxylase Deficiency: Early Dietary Restriction and Time Course of Cardiomyopathy

A CNN-Based Hybrid Model for Tropical Cyclone Intensity Estimation in Meteorological Industry

Deep Learning Approaches in Predicting Tropical Cyclone Tracks: an Analysis Focused on the Northwest Pacific Region

Tropical Cyclone Intensity Prediction Using Deep Convolutional Neural Network

Tropical Cyclone Intensity Estimation Using Multidimensional Convolutional Neural Network From Multichannel Satellite Imagery