中国1km逐日气象六要素重构产品（1961-2021）

高分辨率且高精度的气象产品对于深入理解气候变化机制及其在水文、水资源管理和生态环境等相关领域的影响至关重要。本研究提出了一种分层递进式重构框架（HPRF），用于生成高分辨率气象产品。该框架采用多层感知器（MLP）深度学习算法，通过多阶段过程逐步映射中国大陆六个气象要素——平均气温、最高气温、最低气温、大气压力、相对湿度和日照时长——的空间分布，解析这些要素与地形因素（包括纬度、经度和海拔）之间的复杂非线性关系以及要素之间的相互作用。深度学习分层渐进框架（DL-HPRF）能够生成任意所需空间分辨率的逐日气象产品。本研究中，生成了1公里分辨率的产品，并利用118个地面观测站的数据对其质量进行了评估。结果表明，所有重构的气象产品均具有较高的准确性。平均气温、最高气温和最低气温的误差较小（RMSE中位数分别为1.03℃、1.19℃、1.34℃；ME中位数分别为-0.09℃、-0.10℃、-0.08℃），且与地面数据高度一致（CC中位数分别为1.00、0.99、0.99）。大气压力的误差也较小（RMSE中位数为2.48 hPa；ME中位数为-0.02 hPa），一致性较高（CC中位数为0.98）。尽管相对湿度的准确性相较于以上5种产品较低（RMSE中位数为6.02%；ME中位数为-0.5%；CC中位数为0.90），但其精度依然达到了优异标准。日照时长的精度较高（RMSE中位数为1.48小时；ME中位数为0.05小时；CC中位数为0.93），整体产品质量表现优异。进一步对比分析显示，使用地面数据评估网格化气象产品在复杂地形中的准确性存在局限性，本研究的模型在高海拔地区的表现实际上优于地面数据与网格产品评估指标之间的结果，展现了更高的准确性。

High-resolution and high-precision meteorological products are critical for in-depth understanding of climate change mechanisms and their impacts in fields such as hydrology, water resource management, and ecological environment. This study proposes a Hierarchical Progressive Reconstruction Framework (HPRF) for generating high-resolution meteorological products. The framework adopts the Multi-Layer Perceptron (MLP) deep learning algorithm to gradually map the spatial distributions of six meteorological variables over mainland China—including average air temperature, maximum air temperature, minimum air temperature, atmospheric pressure, relative humidity, and duration of sunshine—via a multi-stage process, and analyzes the complex nonlinear relationships between these variables and topographic factors (latitude, longitude, and elevation) as well as the interactions among the variables themselves. The Deep Learning-based Hierarchical Progressive Reconstruction Framework (DL-HPRF) can generate daily meteorological products with any required spatial resolution. In this study, products with a 1-kilometer resolution were generated, and their quality was evaluated using data from 118 ground observation stations. The results show that all reconstructed meteorological products have high accuracy. The errors of average air temperature, maximum air temperature, and minimum air temperature are small (median RMSE: 1.03 ℃, 1.19 ℃, 1.34 ℃ respectively; median ME: -0.09 ℃, -0.10 ℃, -0.08 ℃ respectively), and they have high consistency with ground observation data (median CC: 1.00, 0.99, 0.99 respectively). The errors of atmospheric pressure are also small (median RMSE: 2.48 hPa; median ME: -0.02 hPa), with high consistency (median CC: 0.98). Although the accuracy of relative humidity is lower than that of the above five products (median RMSE: 6.02%; median ME: -0.5%; median CC: 0.90), its precision still meets the excellent standard. The duration of sunshine has high precision (median RMSE: 1.48 hours; median ME: 0.05 hours; median CC: 0.93), and the overall product quality performs excellently. Further comparative analysis reveals that using ground data to evaluate the accuracy of gridded meteorological products has limitations in complex terrain, and the model of this study actually performs better in high-altitude areas than the results derived from the comparison between ground observation data and gridded product evaluation metrics, demonstrating higher accuracy.

搜集汇总

背景与挑战

背景概述

该数据集是中国地区1961年至2021年逐日气象六要素重构产品，空间分辨率为1公里，包含平均气温、最高气温、最低气温、大气压力、相对湿度和日照时长六个要素，采用深度学习分层递进框架生成，数据准确性高，如平均气温RMSE中位数为1.03℃，适用于气候变化、水文和生态环境研究。数据以开放获取方式共享，格式为栅格文件，便于GIS工具读取。