基于深度学习方法改进CMIP6模式对中国东部沿海复合极端风雨事件的模拟能力 

展开
  • 1. 成都信息工程大学大气科学学院/高原大气与环境四川省重点实验室/成都平原城市气象与环境四川省野外科学观测研究站/四川省气象灾害预测预警工程实验室,四川 成都 610225
    2. 广东海洋大学海洋与气象学院,广东 湛江 524088

网络出版日期: 2025-04-11

基金资助

国家自然科学基金项目(42375047U2442210);四川省自然科学基金项目(2024NSFSC0064);高原与盆地暴雨旱涝灾害四川省重点实验室开放研究基金项目(SZKT202304

Improving the Capability of CMIP6 Simulations for Compound Extreme Wind and Precipitation Events in the Eastern Coastal Region of China Using Deep Learning Methods

Expand
  • 1. School of Atmospheric SciencesPlateau Atmosphere and Environment Key Laboratory of Sichuan ProvinceChengdu Plain Urban Me-teorology and Environment Observation and Research Station of Sichuan ProvinceSichuan Meteorological Disaster Prediction and Early Warning Engineering LaboratoryChengdu University of Information TechnologyChengdu 610225SichuanChina
    2. College of Ocean and MeteorologyGuangdong Ocean UniversityZhanjiang 524088GuangdongChina

Online published: 2025-04-11

摘要

相对于单一极端天气气候事件,由极端强风和极端降水造成的复合极端风雨事件(Compoundwind and precipitation extremesCWPE)对沿海地区的经济和人民生活造成巨大的影响。本文利用第六次国际耦合模式比较计划(Coupled Model Intercomparison Project Phase 6CMIP6)中 17 个模式模拟1961-2000 年中国东部沿海地区 CWPE 作为训练集,采用多层感知机神经网络建立深度学习(DeepLearningDL)模型。通过构建适用于 CWPE损失函数的方式对模型进行优化,发展适用于降低 CMIP6模式对CWPE模拟偏差和不确定性的DL模型。研究结果表明,大多数CMIP6模式对中国东部沿海地区CWPE有较好的模拟能力,其中多模式集合平均值(Multi-Model Ensemble MeanMME-Mean)和多模式集合中位数(Multi-Model Ensemble MedianMME-Median)的结果相对于单一模式的评估表现更好。以均方误差(Mean Squared ErrorMSE)函数作为损失函数构建的 DL 模型,在泰勒技巧评分(Taylor SkillScoreTS)和均方根误差(Root Mean Squared ErrorRMSE)评分上的表现不如多模式集合统计结果。在MSE损失函数基础上加入气候评估指标中的标准差之比(Ratio of the Standard DeviationRSD)和构建的低估值约束函数,可以有效提高DL模型在TSRMSE评分上的表现。因此,将由MSERSD以及低估值约束函数构建的加权损失函数训练的DL模型定义为DL-MRM,仅以MSE作为损失函数训练的DL模型定义为 DL-MSE。对比分析 2001-2014 年间两个 DL 模型对中国东部沿海 CWPE 模拟表现以及 DLMRM相对于多模式集合方法的表现得出:(1)两个DL模型模拟结果均表现低估,但DL-MRM的偏差相比于 DL-MSE 更低且更接近观测,其中在研究区域内 DL-MRM 的相对偏差低于 DL-MSE 的面积约为63%,且相对偏差平均降低了约20%;(2DL-MRM相较于MME-MeanMME-Median,整体偏差较低,其模拟结果更接近观测,在研究区域内 DL-MRM 的相对偏差较低的面积占比分别为 67%62%,且相对偏差分别平均降低约 10%20%。总体而言,通过融合 RSD和低估值约束函数构建加权损失函数的方式对模型进行优化,建立了适用于提高CMIP6模式模拟CWPE能力的DL模型,并表明结合深度学习方法相对于传统多模式集合方法能更有效地降低CMIP6模式模拟CWPE的偏差。

本文引用格式

谭淇昌, 张 宇, 葛 非, 蒋毅飞, 邬钰嫣, 王康宁 . 基于深度学习方法改进CMIP6模式对中国东部沿海复合极端风雨事件的模拟能力 [J]. 高原气象, 0 : 1 . DOI: 10.7522/j.issn.1000-0534.2025.00029

Abstract

Relative to individual extreme weather and climate eventsCompound Wind and Precipitation ExtremesCWPE),which result from extreme winds and extreme precipitationhave a profound impact on the economy and daily life in coastal areas. In this studywe utilized the simulations from 17 models within the Coupled Model Intercomparison Project Phase 6CMIP6for the period 1961-2000 of CWPE in the eastern coastal region of China as a training setand established a Deep LearningDLmodel employing a multi-layer perceptron neural network. By constructing a loss function suitable for CWPE and optimizing the model accordinglywe have developed a DL model aimed at reducing the simulation bias and uncertainty of CMIP6 models for CWPE. The research results indicate that the majority of CMIP6 models possess a relatively good simulation capability for CWPE in the eastern coastal region of Chinawith the Multi-Model Ensemble MeanMME-Meanand the Multi-Model Ensemble MedianMME-Mediandemonstrating better performance in assessments com‐ pared to individual models. The DL model constructed with the Mean Squared ErrorMSEfunction as the loss function performs worse in terms of Taylor Skill ScoreTSand Root Mean Squared ErrorRMSEcompared to the statistical results of the Multi-Model Ensemble. Incorporating the Ratio of the Standard DeviationRSDfrom climate evaluation metrics and an underestimation constraint function into the MSE loss function can significantly enhance the performance of the DL model in terms of TS and RMSE. Thereforethe DL model trained with a weighted loss function constructed from MSERSDand underestimation constraint function is defined as DL-MRMwhile the DL model trained solely with MSE as the loss function is defined as DL-MSE. By com‐ paring and analyzing the performance of the two DL models in simulating CWPE over the eastern coastal region of China from 2001 to 2014as well as the performance of DL-MRM relative to multi-model ensemble methodswe conclude:(1Both DL models exhibit underestimation in their simulation resultsbut the bias of the DLMRM is lower than that of the DL-MSEbeing closer to the observations. Specificallyin the study areathe relative bias of the DL-MRM is lower than that of the DL-MSE by about 63%and the average relative bias is reduced by approximately 20%.2The DL-MRM has a lower overall bias compared to the MME-Mean and MME-Medianwith simulation results that are closer to the observations. In the study areathe DL-MRM has a lower relative bias in 67% and 62% of the area compared to the MME-Mean and MME-Medianrespectivelyand the average relative bias is reduced by approximately 10% and 20%respectively. Overallby integrating the RSD and underestimation constraint functions to construct a weighted loss function for model optimizationa DL model suitable for improving the simulation of CWPE by CMIP6 models was established. This indicates that the combination of deep learning methods can more effectively reduce the biases in CMIP6 model simulations of CWPE compared to traditional multi-model ensemble methods.

文章导航

/