一种融合全球温度直减率网格模型的GPT3改进方法

来源：大地测量与地球动力学杂志2025年第9期北京时间：

作者：牛明昂

单位：1 河南智联时空信息科技有限公司，郑州市经三路15号，450000],figList:[{captionCn:温度直减率年均值的全球分布,magId:a8014741-4fda-4ffd-a255-abb336bde9b2,labelCn:图1,pptUrl:20250902/ddclydqdlx-45-9-888-1.jpg.ppt,labelEn:Fig. 1,figUrl:20250902/ddclydqdlx-45-9-888-1.jpg,titleEn:Fig. 1 Global distribution of annual mean temperature lapse rate,titleCn:图1 温度直减率年均值的全球分布,id:Fig1,captionEn:Global distribution of annual mean temperature lapse rate,thumbnailUrl:20250902/thumbnail/ddclydqdlx-45-9-888-1.jpg},{captionCn:温度直减率在不同温度带中的时间变化特征,magId:fb11148d-8cb2-4dfc-a88f-5a3532c96279,labelCn:图2,pptUrl:20250902/ddclydqdlx-45-9-888-2.jpg.ppt,labelEn:Fig. 2,figUrl:20250902/ddclydqdlx-45-9-888-2.jpg,titleEn:Fig. 2 Temporal variation characteristics of temperature lapse rates across different temperature zones,titleCn:图2 温度直减率在不同温度带中的时间变化特征,id:Fig2,captionEn:Temporal variation characteristics of temperature lapse rates across different temperature zones,thumbnailUrl:20250902/thumbnail/ddclydqdlx-45-9-888-2.jpg},{captionCn:TLRG、UNB3和GPT3模型估计温度直减率的bias和RMSE值的全球分布,magId:1a59f9c7-34f7-4064-b99c-5e36484896f8,labelCn:图3,pptUrl:20250902/ddclydqdlx-45-9-888-3.jpg.ppt,labelEn:Fig. 3,figUrl:20250902/ddclydqdlx-45-9-888-3.jpg,titleEn:Fig. 3 Global distribution of bias and RMSE values for TLRG, UNB3, and GPT3 models in predicting temperature lapse rates,titleCn:图3 TLRG、UNB3和GPT3模型估计温度直减率的bias和RMSE值的全球分布,id:Fig3,captionEn:Global distribution of bias and RMSE values for TLRG, UNB3, and GPT3 models in predicting temperature lapse rates,thumbnailUrl:20250902/thumbnail/ddclydqdlx-45-9-888-3.jpg},{captionCn:TLRG、UNB3和GPT3模型预测温度和气压廓线的bias和RMSE的垂直分布,magId:778c010d-e1e1-4d08-9f39-df802660ad9b,labelCn:图4,pptUrl:20250902/ddclydqdlx-45-9-888-4.jpg.ppt,labelEn:Fig. 4,figUrl:20250902/ddclydqdlx-45-9-888-4.jpg,titleEn:Fig. 4 Vertical distribution of bias and RMSE values for TLRG, UNB3, and GPT3 models in predicting temperature and pressure profiles,titleCn:图4 TLRG、UNB3和GPT3模型预测温度和气压廓线的bias和RMSE的垂直分布,id:Fig4,captionEn:Vertical distribution of bias and RMSE values for TLRG, UNB3, and GPT3 models in predicting temperature and pressure profiles,thumbnailUrl:20250902/thumbnail/ddclydqdlx-45-9-888-4.jpg},{captionCn:IGPT3和GPT3模型预测温度和气压剖面的bias和RMSE的垂直分布,magId:395829f7-897a-4623-b671-3a0506e7b872,labelCn:图5,pptUrl:20250902/ddclydqdlx-45-9-888-5.jpg.ppt,labelEn:Fig. 5,figUrl:20250902/ddclydqdlx-45-9-888-5.jpg,titleEn:Fig. 5 Vertical distribution of bias and RMSE for IGPT3 and GPT3 models in predicting temperature and pressure profiles,titleCn:图5 IGPT3和GPT3模型预测温度和气压剖面的bias和RMSE的垂直分布,id:Fig5,captionEn:Vertical distribution of bias and RMSE for IGPT3 and GPT3 models in predicting temperature and pressure profiles,thumbnailUrl:20250902/thumbnail/ddclydqdlx-45-9-888-5.jpg},{captionCn:IGPT3和GPT3模型预测ZHD和ZTD剖面的bias和RMSE的垂直分布,magId:88f8b2c8-278d-43cf-930d-eebbc88ecc49,labelCn:图6,pptUrl:20250902/ddclydqdlx-45-9-888-6.jpg.ppt,labelEn:Fig. 6,figUrl:20250902/ddclydqdlx-45-9-888-6.jpg,titleEn:Fig. 6 Vertical distribution of bias and RMSE for IGPT3 and GPT3 models in predicting ZHD and ZTD profiles,titleCn:图6 IGPT3和GPT3模型预测ZHD和ZTD剖面的bias和RMSE的垂直分布,id:Fig6,captionEn:Vertical distribution of bias and RMSE for IGPT3 and GPT3 models in predicting ZHD and ZTD profiles,thumbnailUrl:20250902/thumbnail/ddclydqdlx-45-9-888-6.jpg},{captionCn:IGS站点上IGPT3与GPT3预测ZHD和PWV的bias差和RMSE差随高差的分布,magId:73bc1883-5583-43f0-a621-e027fe4c0eef,labelCn:图7,pptUrl:20250902/ddclydqdlx-45-9-888-7.jpg.ppt,labelEn:Fig. 7,figUrl:20250902/ddclydqdlx-45-9-888-7.jpg,titleEn:Fig. 7 Distributions of bias differences and RMSE differences in ZHD and PWV predictions by IGPT3 versus GPT3 with elevation differences at IGS stations,titleCn:图7 IGS站点上IGPT3与GPT3预测ZHD和PWV的bias差和RMSE差随高差的分布,id:Fig7,captionEn:Distributions of bias differences and RMSE differences in ZHD and PWV predictions by IGPT3 versus GPT3 with elevation differences at IGS stations,thumbnailUrl:20250902/thumbnail/ddclydqdlx-45-9-888-7.jpg}

摘要：随着GNSS在民航和无人机平台上的广泛应用，大高差应用场景不断涌现，对对流层延迟垂直修正精度提出了更高的要求。GPT3是目前最先进的ZTD模型，但其对温度和气压的垂直修正存在不足，仅适用于近地表。针对这一问题，提出融合全球温度直减率网格模型和多元大气压高公式的GPT3改进方法，构建IGPT3模型，并对其适应性进行评价。结果表明，与UNB3和GPT3相比，在预测温度直减率、温度廓线、气压廓线3个方面，IGPT3精度分别提高25.0%、18.2%、16.1%和72.2%、50.0%、69.9%。另外，相较于GPT3，IGPT3预测ZHD和ZTD廓线的精度改善量分别达到了29.3 mm和29.0 mm；而且采用IGPT3后，多数IGS站点上的GNSS水汽反演精度得到改善，最大改善量达3.07 mm。

关键词：对流层延迟改正,温度直减率,压高公式,GPT3模型