降水是青海省水资源的主要来源之一，对区域生态环境和社会经济发展具有重要影响。利用深度学习技术，设计了基于3D卷积神经网络和卷积长短期记忆网络3D CNN_ConvLSTM的多源降水数据融合模型，构建了青海省范围内1990年-2020年降水融合数据集，时间分辨率为逐日、空间分辨率为0.01°×0.01°，旨在解决青海省不同时空尺度、不同来源降水数据不确定性、稀疏性、多样性以及数据质量等问题。通过数据融合模型衍生长时间序列高时空精度的青海省降水融合数据集，提升降水数据的精确性和可靠性，为青海省及三江源区生态环境效应变化分析和数字生态保护提供高质量数据支撑。", "ds_source": "
ERA5是由欧洲中期天气预报中心（ECMWF）提供的最新气候再分析数据集，该数据是基于全球观测数据，通过数值天气预报模式和数据同化系统生成。本研究采用ERA5的0.25°空间分辨率小时降水数据。\nGsMap是由日本宇宙航空研究开发机构（JAXA）开发的全球卫星降水数据产品，基于多颗卫星的被动微波观测数据，结合地面雷达和其他辅助数据生成。本研究采用的是0.1°空间分辨率、1小时时间分辨率的GSMaP-Gauge （version 8）。 \n
GLDAS是由美国国家航空航天局（NASA）和美国国家海洋和大气管理局（NOAA）联合开发的全球陆地数据同化系统。GLDAS通过融合地面观测、卫星遥感和数值模型输出，生成高分辨率的陆地表面数据产品。本研究采用了空间分辨率为0.25°、逐三小时的GLDAS（Version 2.1）。\n
IMERG是由全球降水测量任务（GPM）提供的卫星降水数据产品，旨在通过整合多颗卫星的观测数据，生成全球降水估算。本研究使用的是第6版的IMERG最终运行产品，时空分辨率为0.1h和0.1°。 \n
CMORPH是由美国国家海洋和大气管理局（NOAA）开发的基于卫星观测的降水估算产品，利用红外和被动微波卫星观测数据，并通过形态学算法生成高时空分辨率的降水估算。本研究使用的是空间分辨率为8km，时间分辨率为0.5h的CMORPH。\n
本研究使用的实测数据来自青海省气象科学研究所。我们选取了研究区域内经过质量控制的48个气象站点的小时降水数据。", "ds_process_way": "
数据融合处理流程包括：\n
（1）空间降尺度与对齐——系统比较双线性、最邻近与立方卷积三种插值后，选择最邻近插值作为生产级降尺度方法以降低融合前误差传播；\n
（2） 特征标准化——使用归一化，使其标准差为1，均值为0，以避免数据集中的大幅波动导致模型不稳定，使多源特征在同一量纲内进入模型学习；\n
（3）深度融合建模——采用自注意力 + 3D CNN +（双向）ConvLSTM联合结构，同时捕捉降水的全局空间相关性与时间依赖性；\n
（4）正则化与建模细节——在全连接映射阶段引入 Dropout 抑制过拟合、稳定训练，形成从数据预处理到模型推理的端到端出数流程。
", "ds_quality": "将ERA5、IMERG和GLDAS三种公开的青海省范围内降水产品数据和选取青海省内的10个国家级气象观测站的降水量数据转换为日累积降水量，并使用最邻近插值方法将三种降水产品的空间分辨率降尺度至0.01°×0.01°。然后根据青海省内10个国家级气象观测站的经纬度，从降水产品中提取降水特征数据用于多源降水数据融合模型输入。将提取完毕的特征数据集输入3D CNN_ConvLSTM模型，得到降水融合预测值。最后用青海省内10个国家级气象观测站观测数据从相对误差（RB）、平均绝对误差（MAE）、均方根误差（RMSE）、关键成功指数（CSI）、监测率（POD）和误报率（FAR）指标方面，综合评估模型预测结果的准确性。\n
青海省信息中心利用青海省内10个国家级气象观测站观测数据从相对误差（RB）、平均绝对误差（MAE）、均方根误差（RMSE）、关键成功指数（CSI）、监测率（POD）和误报率（FAR）指标方面综合评估模型预测结果的准确性。评估结果表明，相对误差均值为3.17%，均方根误差均值为0.609，平均绝对误差均值为0.275，监测率均值为0.923，关键成功指数均值为0.621，误报率均值为0.343。", "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": null, "ds_acq_alt_high": null, "ds_share_type": "apply-access", "ds_total_size": 2406531452, "ds_files_count": 0, "ds_format": "*.nc", "ds_space_res": "0.01°", "ds_time_res": "日", "ds_coordinate": "无", "ds_projection": "", "ds_thumbnail": "d11e3dcc-99cc-4a30-9752-2f963229d7a8.jpeg", "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": "张小丹", "ds_serv_phone": "13897247931", "ds_serv_mail": "xdzhang@qhu.edu.cn", "doi_value": "", "subject_codes": [ "170.45" ], "quality_level": 0, "publish_time": "2026-06-10 10:03:15", "last_updated": "2026-06-10 10:03:15", "protected": false, "protected_to": "2027-08-20 00:00:00", "lang": "zh", "cstr": "", "i18n": { "en": { "title": "A fused dataset of daily 1km precipitation in Qinghai Province based on 3D CNN_ConvLSTM model (1990-2023)", "ds_format": "*.nc", "ds_source": "
ERA5 is the latest climate reanalysis dataset provided by the European Center for Medium-Range Weather Forecasts (ECMWF). The data is based on global observation data and generated through a numerical weather prediction model and data assimilation system. This study uses 0.25° spatial resolution hourly precipitation data from ERA5.\r\nGsMap is a global satellite precipitation data product developed by the Japan Aerospace Exploration Agency (JAXA). It is generated based on passive microwave observation data from multiple satellites, combined with ground-based radar and other auxiliary data. This study used a GSMaP-Gauge (version 8) with a spatial resolution of 0.1° and a temporal resolution of 1 hour. \r\n
GLDAS is a global terrestrial data assimilation system jointly developed by the National Aeronautics and Space Administration (NASA) and the National Oceanic and Atmospheric Administration (NOAA). GLDAS generates high-resolution land surface data products by integrating ground observations, satellite remote sensing and numerical model outputs. This study used GLDAS (Version 2.1) with a spatial resolution of 0.25° and three-hour.\r\n
IMERG is a satellite precipitation data product provided by the Global Precipitation Measurement Mission (GPM) that aims to generate global precipitation estimates by integrating observation data from multiple satellites. This study uses version 6 of IMERG's final operating product with spatiotemporal resolutions of 0.1h and 0.1°. \r\n
CMORPH is a satellite-based precipitation estimation product developed by the National Oceanic and Atmospheric Administration (NOAA). It uses infrared and passive microwave satellite observation data and generates precipitation estimates with high temporal and spatial resolution through morphological algorithms. This study used CMORPH with a spatial resolution of 8km and a temporal resolution of 0.5 h.\r\n
The measured data used in this study came from the Institute of Meteorological Science of Qinghai Province. We selected hourly precipitation data from 48 quality controlled meteorological stations in the study area.", "ds_quality": "
The three published precipitation product data in Qinghai Province, ERA5, IMERG and GLDAS, and the precipitation data from 10 national meteorological observation stations in Qinghai Province were converted into daily cumulative precipitation, and the spatial resolution of the three precipitation products was downscaled to 0.01°×0.01° using the nearest neighbor interpolation method. Then, according to the latitude and longitude of 10 national meteorological observation stations in Qinghai Province, precipitation characteristic data are extracted from precipitation products for input of multi-source precipitation data fusion model. Input the extracted feature data set into the 3D CNN_ConvLSTM model to obtain the precipitation fusion prediction value. Finally, observation data from 10 national-level meteorological observation stations in Qinghai Province are used to comprehensively evaluate the accuracy of the model prediction results in terms of relative error (RB), mean absolute error (MAE), root mean square error (RMSE), key success index (CSI), monitoring rate (POD) and false alarm rate (FAR) indicators.\r\n
The Qinghai Province Information Center uses observation data from 10 national-level meteorological observation stations in Qinghai Province to comprehensively evaluate the accuracy of the model prediction results in terms of relative error (RB), mean absolute error (MAE), root mean square error (RMSE), key success index (CSI), monitoring rate (POD) and false alarm rate (FAR) indicators. The evaluation results show that the average relative error is 3.17%, the average root-mean-square error is 0.609, the average absolute error is 0.275, the average monitoring rate is 0.923, the average key success index is 0.621, and the average false alarm rate is 0.343.", "ds_ref_way": "", "ds_abstract": "
Precipitation is one of the main sources of water resources in Qinghai Province and has an important impact on regional ecological environment and social and economic development. Using deep learning technology, a multi-source precipitation data fusion model based on a 3D convolutional neural network and a convolutional long-term memory network 3D CNN_ConvLSTM was designed, and a precipitation fusion data set from 1990 to 2020 within Qinghai Province was constructed. The temporal resolution is daily and the spatial resolution is 0.01°×0.01°. It aims to solve the problems of uncertainty, sparsity, diversity and data quality of precipitation data from different temporal and spatial scales and sources in Qinghai Province. The data fusion model is used to derive a long-time series and high spatio-temporal precision precipitation fusion data set in Qinghai Province to improve the accuracy and reliability of precipitation data and provide high-quality data support for the analysis of changes in ecological and environmental effects and digital ecological protection in Qinghai Province and the Sanjiang Source Area.", "ds_time_res": "", "ds_acq_place": "Qinghai Province", "ds_space_res": "", "ds_projection": "", "ds_process_way": "
The data fusion processing process includes:\r\n
(1) Spatial downscaling and alignment-After systematically comparing the three interpolation types of bilinear, nearest neighbor and cubic convolution, the nearest neighbor interpolation is selected as the production-level downscaling method to reduce error propagation before fusion;\r\n
(2) Feature standardization-Use normalization so that its standard deviation is 1 and the mean value is 0 to avoid large fluctuations in the data set causing model instability, and allow multi-source features to enter model learning within the same dimension;\r\n
(3) Deep fusion modeling-adopts the joint structure of self-attention + 3D CNN +(bidirectional) ConvLSTM to simultaneously capture the global spatial correlation and time dependence of precipitation;\r\n
(4) Regularization and modeling details-Dropout is introduced in the fully connected mapping stage to suppress overfitting and stabilize training, forming an end-to-end output process from data preprocessing to model reasoning.
", "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": [ "长时序", "深度学习", "降水" ], "ds_subject_tags": [ "地理学" ], "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": "张小丹", "email": "xdzhang@qhu.edu.cn", "work_for": "青海大学", "country": "中国" }, { "true_name": "游少杰", "email": "3118935574@qq.com", "work_for": "青海大学", "country": "中国" }, { "true_name": "黄远琛", "email": "ys240854040438@qhu.edu.cn", "work_for": "青海大学", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "张小丹", "email": "xdzhang@qhu.edu.cn", "work_for": "青海大学", "country": "中国" } ], "ds_managers": [ { "true_name": "张小丹", "email": "xdzhang@qhu.edu.cn", "work_for": "青海大学", "country": "中国" } ], "category": "气象" }