| 摘要: |
| 【目的】为研究本地化的智能网格客观预报算法，提升数值模式气温和降水预报在贵州省黔西南州地区的预报准确率。【方法】基于消除偏差集合平均和加权消除偏差2种多模式集成方法，采用动态误差权重系数、高低温单独建模和降水分级统计的方式建模，对欧洲中心（ECMWF）、德国（GERMANY）和日本（JAPAN）数值模式在黔西南州及周边地区的气温、降水预报进行站点和格点订正，并使用多种评估检验方法对比分析订正前后的预报误差、准确率和预报技巧。【结果】（1）基于消除偏差法的订正预报有效地减小了数值模式的预报误差，2 m气温和12 h累积降水预报误差分别减小至2.5 ℃和2.5 mm以下。（2）基于动态误差的加权消除偏差订正预报，明显地提升了模式的气温和降水预报准确率，通过高低温单独建模和降水分级统计方式，进一步改善了高低温和分级降水预报效果。其中，格点和站点的2 m气温预报准确率，比评分最优的模式ECMWF提升了41.48%和12.47%，小雨站点预报准确率比评分最优的模式JAPAN提升了23.9%。最低气温和小雨预报准确率超过了本地预报员历史预报，分别提升了1.44%和27.8%。【结论】通过高低温单独建模和降水分级统计方式构建多模式集成订正预报模型，有效地减小了模式的2 m气温和12 h累积降水的预报误差，提升了预报准确率。尤其是2 m气温格点预报和降水站点订正预报，订正效果明显。其中，小雨预报明显优于本地预报员的历史预报，对于实际预报业务有较大参考价值。 |
| 关键词: 多模式集成；消除偏差集合平均；加权消除偏差集合平均；高低温单独建模；动态误差权重 |
| DOI： |
| 投稿时间：2024-05-08修订日期：2024-10-12 |
| 基金项目:国家自然科学基金项目(42265001)“复杂地形下贵州多模式气温预报订正技术研究”；黔西南州气象局科研项目(2019-15)：黔西南州地质灾害气象风险监测预警系统研发；黔西南州科技计划项目(2023-4-88)：基于人工智能技术的黔西南地区高精度短临降水预报模型研究。 |
|
| Research on Temperature and Precipitation Correction Algorithms Bansed on Multi-Model Ensemble Methods for Qianxina Prefecture of Guizhou Province |
| KONG Dexuan,WANGY Yao,TANG Yuanzhi,ZHOU Zecheng,YANG Chunyan,LI Gang |
| (Guizhou Mountainous Meteorological Science Research Institute;Guizhou Meteorological Observatory;Jiangxi Meteorological Observatory;Meteorological Bureau of Qianxinan Buyei and Miao Autonomous Prefecture) |
| Abstract: |
| This study investigates localized objective forecasting algorithms for intelligent grid systems aimed at improving the accuracy of temperature and precipitation forecasts from numerical models in the Qianxinan region of Guizhou Province. We employed two multi-model ensemble methods—bias correction ensemble averaging and weighted bias correction—utilizing dynamic error weighting, separate modeling for high and low temperatures, and precipitation classification statistics. Corrections were applied to temperature and precipitation forecasts from numerical models (ECMWF, GERMANY, and JAPAN) at both station and grid levels. A comparative analysis of forecast errors, accuracy, and skill was conducted before and after the corrections using various evaluation metrics.The results indicate that (1) the bias correction forecasts significantly reduced numerical model forecast errors, with 2 m temperature and 12-hour accumulated precipitation errors decreasing to below 2.5 °C and 2.5 mm, respectively; (2) the weighted bias correction based on dynamic errors notably enhanced the accuracy of temperature and precipitation forecasts. The separate modeling for high and low temperatures, along with precipitation classification, further improved forecasting performance. Specifically, the accuracy of 2 m temperature forecasts increased by 41.48% and 12.47% at grid and station levels, respectively, compared to the best-performing model, ECMWF. Additionally, the accuracy of light rain forecasts at stations improved by 23.9% compared to the top model, JAPAN. Moreover, the accuracy for minimum temperature and light rain forecasts exceeded local forecasters" historical records, with improvements of 1.44% and 27.8%, respectively.The multi-model ensemble bias correction forecasting model, utilizing separate modeling for high and low temperatures and graded precipitation statistics, effectively reduced forecast errors for 2 m temperature and 12-hour accumulated precipitation, significantly enhancing forecast accuracy. Notably, the improvements in 2 m temperature grid forecasts and precipitation station corrections demonstrate considerable effectiveness, with light rain forecasts outperforming historical predictions from local forecasters, providing substantial reference value for operational forecasting. |
| Key words: Multi-model ensemble; bias correction ensemble averaging; weighted bias correction ensemble; separate modeling for high and low temperatures; dynamic error weighting. |
