基于特征提取的海面气象水文要素时间序列数据重构研究

doi:10.11978/2026095

热带海洋学报

基于特征提取的海面气象水文要素时间序列数据重构研究

马长兰¹，丘仲锋¹，唐榕¹，王文硕¹，毛科峰²，聂娟¹

1. 南京信息工程大学, 江苏南京 210044
2. 国防科技大学前沿交叉学科学院, 江苏南京 210014

收稿日期:2026-06-22 修回日期:2026-08-19 接受日期:2026-08-27
通讯作者: 丘仲锋
基金资助:
国家卫星气象中心项目(FY-3(03)-AS-11.10-ZT，FY-3(03)-AS-11.12-ZT)

Feature extraction-based reconstruction of time series data for seasurface meteorological and hydrological variables

MA Changlan^¹, QIU Zhongfeng^¹, TANG Rong^¹, WANG Wenshuo^¹, MAO Kefeng^², NIE Juan^¹

1. Nanjing University of Information Science & Technology, Nanjing 210044, China;
2. College of Advanced Interdisciplinary Studies, National University of Defense Technology, Nanjing 210014, China

Received:2026-06-22 Revised:2026-08-19 Accepted:2026-08-27
Supported by:
Project of National Satellite Meteorological Center[FY-3(03)-AS-11.10-ZT, FY-3(03)-AS-11.12-ZT]

摘要/Abstract

摘要： 本研究围绕浮标观测数据，开展高精度时间序列重构研究：首先，评估再分析资料用于气象水文要素重构的适用性，结果显示，受空间分辨率不足、资料来源差异等因素影响，再分析资料与浮标实测资料存在系统性偏差，难以直接满足高精度浮标气象水文要素的重构需求。针对此问题，本研究分别构建了两类重构模型：一是基于小波分解的多时间尺度赋权重构模型(wavelet decomposition based multi time scale weighted reconstruction model ,WD—MTSWR)，二是基于随机森林的时间降尺度重构模型(random forest based temporal downscaling reconstruction, RF—TDR)，针对气温、纬向风分量、经向风分量、海平面气压和海表温度5个核心气象水文要素开展高频重构。其中，WD—MTSWR模型用于修正再分析数据与浮标观测数据之间的偏差，提升小时尺度要素的重构精度；RF—TDR模型用于开展时间降尺度处理，学习整时观测与10min高频观测数据之间的映射关系。验证结果表明：WD—MTSWR模型的重构精度表现优异，各要素重构结果的相关系数可达0.75~1.00，均方根误差(root mean square error, RMSE)和平均绝对误差(mean absolute error, MAE)分别为0.24~1.89、0.20~1.40，重构精度显著高于再分析数据和基于原始序列的重构模型。RF-TDR模型能够有效提升时间降尺度处理的精度，与传统线性插值方法相比，降尺度结果的相关系数提升0.00~0.03，RMSE和MAE分别降低0.02~0.20、0.00~0.10。本研究构建的WD-MTSWR模型和RF-TDR模型，能够有效提升浮标缺失气象水文要素的重构精度，可为海洋环境监测提供高质量的高时间分辨率数据支撑。interpolation, Lerp）方法相比，降尺度结果的相关系数r提升0.00~0.03，RMSE和MAE分别降低0.02~0.20、0.00~0.10。本研究构建的WD-MTSWR模型和RF-TDR模型，能够有效提升浮标缺失气象水文要素的重构精度，可为海洋环境监测提供高质量的高时间分辨率数据支撑。

关键词: 锚系浮标, 小波分解, 多层感知机模型, 时间降尺度, 时间序列重构

Abstract: Reconstructing sea-surface meteorological and hydrological data offers an effective solution to issues such as missing data, data anomalies, and discontinuous sequences in buoy observations. This approach significantly enhances the completeness and accuracy of observational data, thereby providing reliable data support for marine environmental analysis.This study is centered on the high - precision reconstruction of missing buoy meteorological and hydrological data. Initially, we evaluated the suitability of reanalysis data for reconstructing buoy - measured meteorological and hydrological elements. The findings reveal that, owing to factors like inadequate spatial resolution and disparities in data sources, systematic deviations exist between reanalysis data and buoy - measured data. Consequently, it is challenging to directly meet the requirements for high-precision reconstruction of these elements using reanalysis data.To overcome this challenge, we developed two types of reconstruction models: a wavelet-decomposition-based multi-time- cale weighted reconstruction model (WD-MTSWR) and a random - forest - based temporal downscaling reconstruction model (RF-TDR). These models are designed to conduct high - frequency reconstruction of five core meteorological and hydrological elements: air temperature (T), zonal wind component (U), meridional wind component (V), sea - level pressure (MSL), and sea-surface temperature (SST). The WD-MTSWR model rectifies the discrepancies between reanalysis data and buoy observation data, thereby improving the reconstruction accuracy of hourly - scale elements. The RF-TDR model performs temporal downscaling by learning the variation patterns between hourly observations and 10 - minute high - frequency observations.Validation results demonstrate that the WD - MTSWR model exhibits outstanding reconstruction accuracy. The correlation coefficients (r) for each reconstructed element range from 0.75 to 1.00, while the root mean square error (RMSE) and mean absolute error (MAE) range from 0.24 to 1.89 and 0.20 to 1.40, respectively. This reconstruction accuracy is significantly higher than that of the original reanalysis data and traditional reconstruction models based on the original sequences.The RF-TDR model effectively enhances the accuracy of temporal downscaling. Compared with traditional linear interpolation methods, the correlation coefficient (r) of the downscaling results increases by 0.00 to 0.03, while the RMSE and MAE decrease by 0.02 to 0.20 and 0.00 to 0.10, respectively.The WD - MTSWR and RF-TDR models developed in this study can effectively improve the reconstruction accuracy of missing meteorological and hydrological elements from buoys, offering high-quality, high-temporal-resolution data support for marine environmental monitoring.

Key words: Moored buoys, wavelet decomposition, multilayer perceptron model, time downscaling, time series reconstruction

马长兰丘仲锋唐榕王文硕毛科峰聂娟. 基于特征提取的海面气象水文要素时间序列数据重构研究[J]. 热带海洋学报, 0, (): 1-.

MA Changlan QIU Zhongfeng TANG Rong WANG Wenshuo MAO Kefeng NIE Juan. Feature extraction-based reconstruction of time series data for seasurface meteorological and hydrological variables[J]. Journal of Tropical Oceanography, 0, (): 1-.