蒸散发（ET）作为连接陆地与大气水循环的关键环节，在全球水循环中具有重要作用。本研究通过对比现场观测数据，评估了两种实际蒸散发数据集（GLEAM和ERA5_Land）的精度。为提高数据准确性，我们引入地表温度和净辐射作为协变量进行数据融合，并提出一种基于深度学习的多源ET融合模型，该模型通过挖掘时空依赖关系整合相应数据，其核心由图卷积神经网络（GCN）和门控循环单元（GRU）构成。实验结果表明：\n
（1）本研究提出的GCN-GRU融合模型在精度上显著优于单一数据源；\n
（2）模型测试显示其均方根误差（RMSE）低于1.25毫米/天，平均绝对误差（MAE）低于1.1毫米/天，相对偏差（RB）小于22%，相关系数（CC）达0.83；\n
（3）该模型同时提升了青海省区域原始GLEAM数据和ERA5_Land再分析数据的ET空间精度，其中均方根误差分别降低65%和54%，平均绝对误差分别降低67%和53%；\n
（4）最终利用GCN-GRU模型融合生成2012-2016年逐日ET数据集，该数据集在青海省区域具有更高空间分辨率（0.01°）和更优的数据精度。
", "ds_source": "本研究驱动模型所使用的数据包括地面气象站点观测数据、再分析气象数据和遥感数据。地面气象站点观测数据来源于国家气象信息中心（https://data.cma.cn）， 时间分辨率为1天。同时采用了欧洲气象预报中心生成的第五代大气再分析数据集ERA5_land。针对气象站点观测数据和ERA5_land数据，将小时尺度的总蒸散发量和净辐射数据通过处理转化为日尺度数据。这些数据集原始空间分辨率为0.1°×0.1°，通过最近邻插值法降尺度至0.01°×0.01°。
\n遥感数据来源于NASA的MODIS产品和GLEAM（阿姆斯特丹陆地蒸散发模型）数据集。其中GLEAM数据（版本3.7a）基于卫星与再分析数据的融合，提供时间分辨率为1天、空间分辨率为0.25°×0.25°的全球ET数据。特别地，我们将各数据源（GLEAM和ERA5_Land）的原始数据降尺度至与融合模型相匹配的0.01°分辨率。降尺度处理采用最近邻插值法——该方法通过将目标像元值赋值为最邻近已知像元值，是一种简洁高效的空间插值方法。
\n对于MODIS产品，我们选用MOD11A1数据集。该数据集提供分辨率为0.01°×0.01°的全球日尺度地表温度数据，原始数据采用正弦曲线投影的HDF格式文件。我们使用NASA提供的MRT（Modis Reprojection Tools）软件对原始数据进行批量拼接和投影转换，最终生成采用GCS_WGS_1984坐标系的TIF格式文件，并采用双线性插值法进行重采样。GCS_WGS_1984是基于世界大地测量系统1984版（WGS84）的地理坐标系，采用经纬度表示地球位置，是目前广泛使用的标准地理坐标系。
", "ds_process_way": "本研究采用图卷积网络（GCN）捕捉多源数据间的空间依赖性。通过将地理相邻区域的关系构建为图结构，GCN能够聚合节点特征及其邻近节点的特征。同时，利用门控循环单元（GRU）提取单源数据内部的时间依赖性并捕获时序模式。通过整合GCN与GRU，我们实现了多源数据时空依赖关系的联合建模，最终生成具有时空表征能力的特征向量。这些特征向量将被输入融合模型，以提升蒸散发（ET）估算精度。
", "ds_quality": "通过采用GCN-GRU深度学习方法，成功构建了1990-2023年青海省区域1公里分辨率的蒸散发（ET）产品。该产品融合了ERA5_Land和GLEAM网格气象数据集及相关协变量数据。通过青海省区域内不同数据源（包括融合ET数据、ERA5_Land ET和GLEAM ET）与站点观测值的均方根误差（RMSE）对比。模型生成的融合数据显著降低了与站点观测的RMSE：相较于ERA5_Land数据，融合数据的RMSE降低了1.43毫米/天（降幅53%）；相较于GLEAM数据则降低了2.3毫米/天（降幅65%）。融合数据相较于ERA5_Land数据的平均绝对误差（MAE）降低1.14毫米/天（降幅54%），相较于GLEAM数据则降低2毫米/天（降幅67%）。这些结果表明，基于GCN-GRU方法的融合模型能获得更精确的ET估算结果，在青海省区域显著优于单一数据源（ERA5_Land和GLEAM）的表现。
", "ds_acq_start_time": "1990-01-01 00:00:00", "ds_acq_end_time": "2023-12-31 00:00:00", "ds_acq_place": "青海省", "ds_acq_lon_east": 103.4, "ds_acq_lat_south": 31.366666666666667, "ds_acq_lon_west": 89.5, "ds_acq_lat_north": 39.86666666666667, "ds_acq_alt_low": 1600.0, "ds_acq_alt_high": 6500.0, "ds_share_type": "apply-access", "ds_total_size": 369973405, "ds_files_count": 0, "ds_format": "*.nc", "ds_space_res": "1000", "ds_time_res": "日", "ds_coordinate": "无", "ds_projection": "", "ds_thumbnail": "d624ed8f-6c77-4b4d-9eb5-37ffce8fcb6a.png", "ds_thumb_from": 0, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "", "ds_from_station": "", "organization_id": "5b99d600-008a-4069-8fc3-7adb9c3f2f8b", "ds_serv_man": "权晨", "ds_serv_phone": "18397119373", "ds_serv_mail": "quanchen007@sina.com", "doi_value": "", "subject_codes": [ "170.15" ], "quality_level": 0, "publish_time": "2026-06-10 10:03:12", "last_updated": "2026-06-10 10:03:12", "protected": false, "protected_to": "2027-08-20 00:00:00", "lang": "zh", "cstr": "", "i18n": { "en": { "title": "Daily Evapotranspiration Fusion Data Set in Qinghai Province Based on GCN_GRU Model (1990-2023)", "ds_format": "*.nc", "ds_source": "The data used for driving the model in this study include surface meteorological station observation data, reanalysis meteorological data and remote sensing data. Observation data from surface meteorological stations come from the National Meteorological Information Center (https://data.cma.cn) with a time resolution of 1 day. At the same time, the fifth-generation atmospheric reanalysis data set ERA5_land generated by the European Center for Meteorological Forecasting is used. For meteorological station observation data and ERA5_land data, hourly scale total evapotranspiration and net radiation data are processed and transformed into daily scale data. The original spatial resolution of these datasets was 0.1°×0.1°, which was downscaled to 0.01°×0.01° using nearest neighbor interpolation.
\r\nRemote sensing data are derived from NASA's MODIS product and the GLEAM (Amsterdam Land Evapotranspiration Model) dataset. Among them, GLEAM data (version 3.7a) is based on the fusion of satellites and reanalysis data and provides global ET data with a temporal resolution of 1 day and a spatial resolution of 0.25°×0.25°. In particular, we downscaled raw data from each data source (GLEAM and ERA5_Land) to a 0.01° resolution that matched the fusion model. The down-scaling process uses the nearest neighbor interpolation method-this method is a concise and efficient spatial interpolation method by assigning the value of the target pixel to the value of the nearest known pixel.
\r\nFor MODIS products, we chose the MOD11A1 dataset. This dataset provides global diurnal surface temperature data with a resolution of 0.01°×0.01°, and the raw data is in a sinusoidal projection HDF format file. We used the MRT (Modis Reprojection Tools) software provided by NASA to perform batch splicing and projection conversion on the raw data, and finally generated a TIF format file in the GCS_WGS_1984 coordinate system, and resampled using bilinear interpolation method. GCS_WGS_1984 is a geographical coordinate system based on the 1984 version of the World Geodetic System (WGS84). It uses latitude and longitude to express the earth's position. It is a standard geographical coordinate system widely used at present.
", "ds_quality": "By using the GCN-GRU deep learning method, we successfully constructed evapotranspiration (ET) products with a resolution of 1 kilometer in Qinghai Province from 2012 to 2016. This product integrates ERA5_Land and GLEAM grid meteorological datasets and related covariate data. Compare the root mean square error (RMSE) of different data sources (including fused ET data, ERA5_Land ET, and GLEAM ET) and station observations within the Qinghai Province region. The fusion data generated by the model significantly reduced the RMSE with station observations: compared to ERA5_Land data, the RMSE of the fusion data decreased by 1.43 millimeters per day (a decrease of 53%); Compared to GLEAM data, it decreased by 2.3 millimeters per day (a decrease of 65%). The average absolute error (MAE) of the fused data is reduced by 1.14 millimeters per day (a decrease of 54%) compared to ERA5_Land data, and by 2 millimeters per day (a decrease of 67%) compared to GLEAM data. These results indicate that the fusion model based on GCN-GRU method can obtain more accurate ET estimation results, which is significantly better than the performance of single data sources (ERA5_Land and GLEAM) in the Qinghai Province region.
", "ds_ref_way": "", "ds_abstract": "Evapotranspiration (ET), as a key link connecting the land and atmosphere water cycle, plays an important role in the global water cycle. This study evaluated the accuracy of two actual evapotranspiration data sets (GLEAM and ERA5_Land) by comparing in-situ observation data. In order to improve data accuracy, we introduced surface temperature and net radiation as covariates for data fusion, and proposed a multi-source ET fusion model based on deep learning. This model integrates corresponding data by mining spatio-temporal dependencies, and its core is graph convolution. Composed of neural network (GCN) and gated loop unit (GRU). Experimental results show that:\r\n
(1) The GCN-GRU fusion model proposed in this study is significantly better than a single data source in terms of accuracy;\r\n
(2) Model tests show that its root-mean-square error (RMSE) is less than 1.25 mm/day, the mean absolute error (MAE) is less than 1.1 mm/day, the relative deviation (RB) is less than 22%, and the correlation coefficient (CC) reaches 0.83;\r\n
(3) The model also improves the ET spatial accuracy of raw GLEAM data and ERA5_Land reanalysis data in Qinghai Province, with the root-mean-square error reduced by 65% and 54% respectively, and the average absolute error reduced by 67% and 53% respectively;\r\n
(4) Finally, the GCN-GRU model fusion was used to generate the daily ET data set from 2012 to 2016, which has higher spatial resolution (0.01°) and better data accuracy in the Qinghai Province region.
", "ds_time_res": "", "ds_acq_place": "Qinghai Province", "ds_space_res": "", "ds_projection": "", "ds_process_way": "This study uses Graph Convolutional Networks (GCN) to capture spatial dependencies between multi-source data. By constructing the relationship between geographically adjacent regions into a graph structure, GCN can aggregate node features and the features of its neighboring nodes. Meanwhile, utilizing Gated Recurrent Units (GRUs) to extract temporal dependencies within single source data and capture temporal patterns. By integrating GCN and GRU, we have achieved joint modeling of spatiotemporal dependencies in multi-source data, ultimately generating feature vectors with spatiotemporal representation capabilities. These feature vectors will be input into the fusion model to improve the accuracy of evapotranspiration (ET) estimation.
", "ds_ref_instruction": "" } }, "submit_center_id": "ncdc", "data_level": 0, "recommendation_value": 0, "license_type": "https://creativecommons.org/licenses/by/4.0/", "doi_reg_from": "reg_local", "cstr_reg_from": "reg_local", "doi_not_reg_reason": null, "cstr_not_reg_reason": null, "is_paper_in_submitting": false, "belong_to_nieer": false, "ds_topic_tags": [ "蒸散发", "GCN_GRU模型" ], "ds_subject_tags": [ "大气科学" ], "ds_class_tags": [], "ds_locus_tags": [ "青海省" ], "ds_time_tags": [ 1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022, 2023 ], "ds_contributors": [ { "true_name": "权晨", "email": "quanchen007@sina.com", "work_for": "青海省气象科学研究所", "country": "中国" }, { "true_name": "张小丹", "email": "xdzhang@qhu.edu.cn", "work_for": "青海大学", "country": "中国" }, { "true_name": "刘畅", "email": "644496605@qq.com", "work_for": "青海大学", "country": "中国" }, { "true_name": "王惠平", "email": "wanghp3176@163.com", "work_for": "青海省气象科学研究所", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "权晨", "email": "quanchen007@sina.com", "work_for": "青海省气象科学研究所", "country": "中国" }, { "true_name": "张小丹", "email": "xdzhang@qhu.edu.cn", "work_for": "青海大学", "country": "中国" }, { "true_name": "刘畅", "email": "644496605@qq.com", "work_for": "青海大学", "country": "中国" }, { "true_name": "王惠平", "email": "wanghp3176@163.com", "work_for": "青海省气象科学研究所", "country": "中国" } ], "ds_managers": [ { "true_name": "权晨", "email": "quanchen007@sina.com", "work_for": "青海省气象科学研究所", "country": "中国" }, { "true_name": "张小丹", "email": "xdzhang@qhu.edu.cn", "work_for": "青海大学", "country": "中国" } ], "category": "气象" }