An Algorithm for Precipitation Correction in Flood Season Based on Dendritic Neural Network

Area of Science:

• Climate Science
• Meteorology
• Artificial Intelligence

Background:

• The National Climate Center utilizes the Climate-Weather Research and Forecasting (CWRF) model for dynamic downscaling and summer precipitation prediction.
• Current CWRF model predictions exhibit deviations, limiting accurate forecasting capabilities.

Purpose of the Study:

• To evaluate and compare the effectiveness of dendritic neural networks (DD) and artificial neural networks (ANNs) in correcting summer precipitation forecasts generated by the CWRF model.
• To identify the optimal machine learning algorithm for improving the accuracy of regional climate model precipitation predictions.

Main Methods:

• Simulated summer precipitation forecast data from the CWRF model (1996-2019) were analyzed.
• Dendritic neural network (DD) and artificial neural network (ANN) algorithms were employed for comparative analysis of precipitation correction techniques.
• Correction models were established by integrating CWRF simulated data with actual ground station precipitation observations.
• Model performance was assessed using evaluation indices such as anomaly correlation coefficient (ACC), temporal correlation coefficient (TCC), mean square error (MSE), and trend anomaly (Ps) test score.

Main Results:

• The dendritic neural network (DD) algorithm demonstrated a superior correction effect compared to the baseline CWRF historical data.
• Key performance metrics improved significantly: anomaly correlation coefficient (ACC) and temporal correlation coefficient (TCC) increased by 0.1.
• Mean square error (MSE) decreased by approximately 26%, indicating a substantial reduction in prediction error.
• The overall trend anomaly (Ps) test score also showed improvement, signifying enhanced prediction of precipitation trends.

Conclusions:

• Machine learning algorithms, particularly the dendritic neural network, can effectively correct summer precipitation predictions from the CWRF regional climate model.
• The application of these algorithms leads to a notable improvement in the accuracy of weather forecasts.
• This study highlights the potential of integrating advanced AI techniques with regional climate models for more reliable climate predictions.