融合地形特征和神经网络的日最高/最低气温预报订正方法研究

展开
  • 1. 气象防灾减灾湖南重点实验室,湖南 长沙 410118
    2. 湖南省气象台,湖南 长沙 410118
    3. 洞庭湖国家气候观象台,湖南 岳阳 414000
    4. 中国气象局高影响天气(专项)重点开放实验室,湖南 长沙 410118
    5. 湘西州气象局,湖南 吉首 416000
    6. 中国气象局武汉暴雨研究所/暴雨监测预警湖北省重点实验室/中国气象局流域强降水重点开放实验室,湖北 武汉 430205

网络出版日期: 2026-01-26

基金资助

湖南省气象局重点项目(XQKJ22A005);湖南省气象台青年创新基金项目(QNJJ202501);武汉市自然科学基金项目
2024020901030454

Research on Daily Maximum and Minimum Temperature Forecasts by Integrating Terrain Features and Neural Networks 

Expand
  • 1. Hunan Key Laboratory of Meteorological Disaster Prevention and ReductionChangsha 410118HunanChina
    2. Hunan Meteorological ObservatoryChangsha 410118HunanChina
    3. Dongting Lake National Climatic ObservatoryChina Meteorological AdministrationYueyang 414000HunanChina
    4. Key Laboratory of High Impact WeatherspecialChina Meteorological AdministrationChangsha 410118HunanChina
    5. Xiangxi Meteorological BureauJiShou 416000HunanChina
    6. China Meteorological Administration Basin Heavy Rainfall Key Laboratory/Hubei Key Laboratory for Heavy Rain Monitoring and Warning ResearchInstitute of Heavy RainChina Meteorological AdministrationWuhan 430205HubeiChina

Online published: 2026-01-26

摘要

日最高、最低气温的预报是天气预报业务的重要组成部分,其精度提升对保障社会经济活动具有重要意义。针对数值模式在复杂地形区域系统性偏差显著的问题,本研究以湖南省为试验区(具有字形三级阶梯地貌,涵盖山地、丘陵、平原等多种下垫面类型),提出一种融合地理特征聚类的深度学习改进方案。基于欧洲中期天气预报中心(European Centre for Medium Range Weather ForecastsECMWF)模式预报场、中国气象局陆面数据同化系统(CMA Land Data Assimilation SystemCLDAS)再分析数据和地理变量,首先构建基准卷积神经网络CNN模型并针对地形特征处理策略设置三类对照实验:方案1K-means聚类地理变量)、方案2(标准化非聚类地理变量)及方案3(无地理变量)。对比实验表明,方案1 24 h 最高/低温度平均绝对误差 MAE 较方案 3 分别降低 4. 7%/9. 4%,较方案 2 预报技巧提升 2. 5%/1. 4%,证实地理特征聚类处理对模型性能具有显著增益。因此,优选方案 1发展了未来 72 h气温预报CNN-TCTerrain Correction)模型。该模型表现出显著预报优势:相较ECMWF产品,最高/低温MAE降幅 达 23. 5%~37. 3%/20. 8%~26. 9%;较 中 央 气 象 台 指 导 预 报 产 品 SCMOC,最 高/低 温 度 误 差 降 低18. 7%~27. 6%/26. 8%~32. 3%,其中 24 h 预报时效下,最高/低温度空间分布 MAE 区间由 1. 2~5. 8 ℃/0. 8~5. 9 ℃ECMWF)降低至0. 9~1. 7 ℃/0. 8~1. 7 ℃,区域稳定性大幅提升。分月检验表明,CNN-TC模型在所有月份均保持最优性能,MAE相对降幅覆盖5. 6%~59. 1%(最高温度)和6. 3%~47. 8%(最低温度)。典型强天气过程检验中,模型成功捕捉202211月寒潮过程的降温特征,较ECWMFSCMOC表现均为最优,显示出优异的极端天气应对能力。本研究证实通过深度学习耦合地形特征聚类,可有效解决数值模式在复杂下垫面区域的系统性偏差问题,为山地气候区精细化气象服务提供了可靠的技术方案。

本文引用格式

卢 姝, 许 霖, 顾 雪, 周 悦, 戴泽军 陶雅琴 . 融合地形特征和神经网络的日最高/最低气温预报订正方法研究[J]. 高原气象, 0 : 1 . DOI: 10.7522/j.issn.1000-0534.2025.00081

Abstract

Accurate forecasting of daily maximumTmaxand minimumTmintemperatures is essential for meteorological operationalas enhanced precision safeguards socioeconomic stability in agriculturetransportation and public health. To address pronounced systematic biases in numerical weather prediction models over complex terrain-particularly regions with heterogeneous underlying surfaces-this study develops an advanced deep learning framework integrating geographic feature clustering. The study focuses on China’s Hunan Provincecharacterized by a distinctive concave-shapedthree-tiered topography encompassing mountainshillsbasinsand plains. A baseline convolutional neural networkCNNwas constructed using ECMWF forecast fieldshigh-resolution CLDASChina Land Data Assimilation Systemreanalysis dataand multi-dimensional geographic variableselevationslopeaspectterrain roughness index. Three comparative experiments rigorously evaluated terrain-processing efficacyMethod 1 K-means clustered geographic variables delineating topographic regimes),Method 2Conventionally standardized non-clustered geographic variables),and Method 3Geographic variable exclusion as terrain-agnostic control. Validation confirmed Method 1's superiorityachieving 24- hour mean absolute errorMAEreductions of 4. 7% for Tmax and 9. 4% for Tmin relative to Method 3while im‐ proving forecast skill by 2. 5%Tmaxand 1. 4%Tmincompared to Method 2. These statistically significant gains validate the benefits of explicit terrain feature clustering. Building on this approachthe CNN-Terrain Cor‐ rectionCNN-TCmodel was developed for 72-hour Tmax/Tmin predictions. CNN-TC framework delivers transformative improvements. Versus ECMWF outputsTmax MAE decreased by 23. 5%~37. 3% and Tmin MAE by 20. 8%~26. 9% across 24~72 hour lead times. Compared to operational SCMOC productserrors reduced 18. 7%~27. 6%Tmaxand 26. 8%~32. 3%Tmin. Criticallythe model compresses spatial error dispersionnarrowing 24-hour MAE ranges from 1. 2~5. 8 ℃/0. 8~5. 9 ℃ECMWFto a stable 0. 9~1. 7 ℃/0. 8~1. 7 ℃ for T max/Tmin-demonstrating breakthrough operational stability. Monthly verification confirmed persistent superioritywith T max MAE reductions of 5. 6%~59. 1% and Tmin improvements of 6. 3% to 47. 8% relative to ECMWF. During the November 2022 cold-air outbreakthe model captured intricate spatiotemporal cooling patternsoutperforming ECMWF and SCMOC with over 30% error reductionunderscoring its capability in extreme weather. This study verifies that deep learning combined with terrain clustering effectively mitigates systematic biases over complex terrain. The CNN-TC framework establishes a robust solution for refined meteorological services in mountainous regions. Cross-regional implementation requires localized hyperparameter optimization and cluster retraining to address spatial heterogeneity in climate regimes and surface properties.

文章导航

/