Reconstruction of Missing Rainfall in Small Mountainous Watersheds Based on Multi-Source Data Fusion and Deep Learning

ZHANG Minxiong; ZHAN Changxun; ZHENG Wenshen; YUAN Xuehua; WANG Hongna; ZHANG Ting

doi:null

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

Reconstruction of Missing Rainfall in Small Mountainous Watersheds Based on Multi-Source Data Fusion and Deep Learning

更新时间：2026-06-22

- Reconstruction of Missing Rainfall in Small Mountainous Watersheds Based on Multi-Source Data Fusion and Deep Learning
- PEARL RIVER Pages: 1-12(2026)
- 作者机构：
  
  1.福州市闽江下游防洪工程建设有限公司，福建福州 350007
  2.福州大学土木工程学院，福建福州 350116
  3.福州市水利水电勘测设计有限公司，福建福州 350007
  4.福建省溪源水库管理处，福建福州 350001
- 作者简介：
- 基金信息：
- DOI：
  CLC： TV672+.2
- Received：11 April 2026，
  
  Revised：2026-05-31，
  
  Accepted：02 June 2026，
  
  Online First：22 June 2026，
- 稿件说明：
移动端阅览
张闽雄，詹昌洵，郑文燊等.基于多源数据融合与深度学习的山区小流域缺失雨量重构研究［J］.人民珠江，DOI：10.3969/j.issn.1001-9235.XXXX.XX.001.

ZHANG Minxiong,ZHAN Changxun,ZHENG Wenshen,et al.Reconstruction of Missing Rainfall in Small Mountainous Watersheds Based on Multi-Source Data Fusion and Deep Learning[J].PEARL RIVER,
张闽雄，詹昌洵，郑文燊等.基于多源数据融合与深度学习的山区小流域缺失雨量重构研究［J］.人民珠江，DOI：10.3969/j.issn.1001-9235.XXXX.XX.001. DOI：

ZHANG Minxiong,ZHAN Changxun,ZHENG Wenshen,et al.Reconstruction of Missing Rainfall in Small Mountainous Watersheds Based on Multi-Source Data Fusion and Deep Learning[J].PEARL RIVER, DOI：10.3969/j.issn.1001-9235...001.

摘要

为解决山区小流域因雨量站布设稀疏而难以准确获取降水过程的问题，本文提出一种基于多源数据融合与深度学习的雨量重构方法，通过克里金插值法（Kriging）提升卫星降水产品（IMERG）的空间分辨率，并融合周边雨量站逐时观测数据，构建具有时空关联特征的雨量序列，进而利用长短期记忆网络（LSTM）学习雨量时序演化规律，实现山区小流域缺失雨量重构。结果表明，Kriging-LSTM 模型重构雨量与实测值高度一致，纳什效率系数（NSE）达到0.92，峰值误差（PE）为0.1，均方根误差（RMSE）为1.59 mm/h；雨量站配置数量的增加能改善模型对无站点区域缺失雨量的重构性能，但其精度提升呈明显的边际递减特征；当雨量站所处的空间位置能够反映山区小流域的降水变化时，通过少数站点即可对该区域雨量场分布进行准确估计，研究成果可为山区小流域缺失雨量重构及雨量站网优化提供参考。

Abstract

In small mountainous watersheds

complex topography

short flow concentration times

and highly localized storm development often cause strong spatial heterogeneity in rainfall processes. However

rain-gauge networks in these regions are usually sparse

making it difficult to obtain reliable hourly rainfall series for ungauged or poorly gauged areas. To address this challenge

this study proposes a rainfall reconstruction method that integrates multi-source data fusion with deep learning. The method first spatially downscales the IMERG satellite precipitation product using Kriging interpolation: the original satellite precipitation grid values are taken as spatial samples

and the spatial autocorrelation embedded in the satellite precipitation field is exploited to generate higher-resolution precipitation data for the study area. The downscaled satellite precipitation data are then combined with hourly observations from surrounding rain gauges to construct input sequences that incorporate both spatial distribution information and temporal variation characteristics. A Long Short-Term Memory (LSTM) network is subsequently established to learn the nonlinear temporal evolution of rainfall and to reconstruct missing rainfall processes in small mountainous watersheds. The proposed Kriging-LSTM model is evaluated using observed rainfall data from a representative rain gauge as the validation benchmark. The results show that the reconstructed rainfall process agrees well with the observed process in terms of occurrence time

peak intensity

and overall temporal pattern. The Nash–Sutcliffe efficiency (NSE) reaches 0.92

the peak error (PE) is 0.1

and the root mean square error (RMSE) is 1.59 mm/h

indicating that the model can effectively reproduce hourly rainfall processes under sparse-monitoring conditions. Compared with using satellite precipitation data alone

the integration of rain-gauge observations improves the correction of local rainfall variations

while the Kriging-based spatial downscaling provides richer spatial information for the subsequent deep learning reconstruction. Further analysis of different rain-gauge configurations reveals that increasing the number of stations improves the reconstruction accuracy for ungauged areas

but the accuracy gain exhibits a clear diminishing marginal effect. This implies that the representativeness of station locations is no less important than the total number of stations. When the selected rain gauges are able to capture the main spatial variability of precipitation across the watershed

a limited number of stations can provide sufficient information for reliably estimating the regional rainfall field. This study demonstrates that combining satellite precipitation data

ground rain-gauge observations

spatial interpolation

and LSTM modeling is a feasible approach for reconstructing missing rainfall processes in small mountainous watersheds. The method can offer technical support for rainfall data infilling

hydrological simulation

flood forecasting

and rain-gauge network optimization in areas with limited monitoring conditions. The findings also suggest that future station-network planning should consider both gauge density and spatial representativeness

especially in watersheds with complex terrain and uneven rainfall distribution.

关键词

Keywords

references

袁泽，周秋文，杨航，等 . 贵州榕江县“2025·6”暴雨洪水特征分析［J］. 人民珠江， 2026 ， 47 （ 1 ）： 73 - 82 .

Yuan Ze ， Zhou Qiuwen ， Yang Hang ， et al . Analysis of Characteristics of Heavy Rain and Flood in Rongjiang County， Guizhou Provincein June 2025 ［J］. Pearl River ， 2026 ， 47 （ 1 ）： 73 - 82 . （in Chinese）

Hu Jinlong ， Miao Chiyuan ， Su Jiajia ， et al . An upgraded high-precision gridded precipitation dataset for the Chinese mainland considering spatial autocorrelation and covariates ［J］. Earth System Science Data ， 2025 ， 17 （ 8 ）： 3987 - 4004

李新，金凌，陈元芳 . 日降水场分解的典型解集模型及应用［J］. 水力发电学报， 2022 ， 41 （ 8 ）： 63 - 76 .

Li Xin ， Jin Ling ， Chen Yuanfang . Typical disaggregation models and their application indisaggregating daily precipitation field ［J］. Journal of Hydroelectric Engineering ， 2022 ， 41 （ 8 ）： 63 - 76 . （in Chinese）

陈华，盛晟，夏润亮，等 . 基于矩阵分解的降水时空插值方法［J］. 河海大学学报（自然科学版）， 2021 ， 49 （ 1 ）： 35 - 41 .

Chen Hua ， Sheng Sheng ， Xia Runliang ， et al . Spatiotemporal interpolation method of rainfall based on matrix decomposition ［J］. Journal of Hohai University （Natural Sciences）， 2021 ， 49 （ 1 ）： 35 - 41 . （in Chinese）

Zhang Tuantuan ， Liang Zhongmin ， Wang Haijun ， et al . Merging Multisatellite Precipitation Products Using Stacking Method and the Censored-Shifted Gamma Ensemble Model Output Statistics in China's Beimiaoji Basin ［J］. Journal of Hydrology ， 2023 ， 618 . DOI： 10.1016/j.jhydrol.2023.129263 http://dx.doi.org/10.1016/j.jhydrol.2023.129263 .

郎立晨，唐诚，高星，等 . 复杂地形下降水的高空间分辨率插值方法研究［J］. 地理学报， 2023 ， 78 （ 01 ）： 101 - 120 .

Lang Lichen ， Tang Cheng ， Gao Xing ， et al . Spatial interpolation of high-resolution daily precipitationover complex terrains ［J］. Acta Geographica Sinica ， 2023 ， 78 （ 1 ）： 101 - 120 . （in Chinese）

Zhao X ， Pan S ， Sun Z ， et al . Advances of satellite remote sensing technology in earthquake prediction ［J］. Natural Hazards Review ， 2021 ， 22 （ 1 ）. DOI： 10.1061/（ASCE）NH.1527-6996.00004 http://dx.doi.org/10.1061/（ASCE）NH.1527-6996.00004 .

陈双，陈涛，符娇兰，等 . 卫星反演降水在华北极端降水监测中的适用性评估［J］. 大气科学， 2025 ， 49 （ 3 ）： 873 - 891 .

Chen Shuang ， Chen Tao ， Fu Jiaolan ， et al . Evaluation of the Applicability of Satellite-Derived Precipitationin Monitoring Extreme Precipitation in North China ［J］. Chinese Journal of Atmospheric Sciences ， 2025 ， 49 （ 3 ）： 873 - 891 . （in Chinese）

王忠静，石羽佳，张腾 . TRMM遥感降水低估还是高估中国大陆地区的降水？［J］. 地球科学进展， 2021 ， 36 （ 6 ）： 604 - 615 .

Wang Zhongjing ， Shi Yujia ， Zhang Teng . Does TRMM Precipitation Underestimate orOverestimate in China's Mainland？［J］. Advances in Earth Science ， 2021 ， 36 （ 6 ）： 604 - 615 . （in Chinese）

欧阳文宇，叶磊，顾学志，等 . 深度学习水文预报研究进展综述Ⅱ——研究进展及展望［J］. 南水北调与水利科技（中英文）， 2022 ， 20 （ 05 ）： 862 - 875 .

Ouyang Wenyu ， Ye Lei ， Gu Xuezhi ， et al . Review of deep learning for hydrological forecasting II：Research progress and prospect ［J］. South-to-North Water Transfers and Water Science & Technology ， 2022 ， 20 （ 5 ）： 862 - 875 . （in Chinese）

LEES T ， BUECHEL M ， ANDERSON B ， et al . Benchmarking Data-Driven Rainfall-Runoff Models in Great Britain： A Comparison of Long Short-Term Memory （LSTM）-Based Models with Four Lumped Conceptual Models ［J］. Hydrology and Earth System Sciences ， 2021 ， 25 （ 10 ）： 5517 - 5534 .

王晓燕，刘刚，刘依嘉，等 . 基于机器学习模型的塔里木河流域卫星降水融合校正［J］. 水土保持学报， 2025 ， 39 （ 06 ）： 419 - 430 .

Wang Xiaoyan ， Liu Gang ， Liu Yijia ， et al . Correction of Satellite Precipitation Product Based on MachineLearning Models in Tarim River Basin ［J］. Journal of Soil and Water Conservation ， 2025 ， 39 （ 6 ）： 419 - 430 . （in Chinese）

闫鑫，陈华，尚志宏，等 . 星地数据融合的日降水产品及其径流模拟研究［J］. 中国农村水利水电， 2023 （ 6 ）： 36 - 46 . doi: 10.12396/znsd.221678 http://dx.doi.org/10.12396/znsd.221678

Yan Xin ， Chen Hua ， Shang Zhihong ， et al . Daily Precipitation Estimates Merging Ground-Based Observations withSatellite-Derived Data and Its Runoff Simulated Performance ［J］. China Rural Water and Hydropower ， 2023 （ 6 ）： 36 - 46 . （in Chinese）

Gao Shiping ， He Dongjie ， Zhang Zhuozhuo ， et al . A novel dynamic interpolation method based0nboth temporal and spatial correlations ［J］. Applied Intelligence ， 2023 ， 53 （ 5 ）： 5100 - 5125 .

纪国良，义崇政，马瑞，等 . 顾及时空变化的流域降雨数据插补技术研究［J］. 水电能源科学， 2025 ， 43 （ 12 ）： 30 - 33，52 .

Ji Guoliang ， Yi Chongzheng ， Ma Rui ， et al . Water Quality Spatial-temporal Analysis of Xinglong River Based on Comprehensive Water Quality Index ［J］. Water Resources and Power ， 2025 ， 43 （ 12 ）： 30 - 33， 52 . （in Chinese）

SINGHAL A ， CHERIYAMPARAMBIL A ， SAMAL N ， et al . Relating forecast and satellite precipitation to generate future skillful ensemble forecasts over the northwest Himalayas at major avalanche and glacier sites ［J］. Journal of Hydrology ， 2023 ， 616 . DOI： 10.1016/j.jhydrol.2022.128795 http://dx.doi.org/10.1016/j.jhydrol.2022.128795 .

石羽佳，王忠静，索滢 . 基于多源数据融合的海河流域降水资源评价［J］. 水科学进展， 2022 ， 33 （ 4 ）： 602 - 613

Shi Yujia ， Wang Zhongjing ， Suo Ying . Evaluation of Haihe River basin precipitation resources based on multisource data fusion ［J］. Advances in Water Science ， 2022 ， 33 （ 4 ）： 602 - 613 . （in Chinese）

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

An Intelligent Method for Detecting Leaks in Dams Based on an Improved YOLOv10 Model

Deformation Prediction of Wuxijiang Buttress Dam Based on Whale Optimization Algorithm and Transformer

Intelligent Recognition of Urban Flood Depth Based on Mask RCNN

Concrete Dam Deformation Prediction Model Based on Attention Mechanism and Deep Learning

Deep Learning-Based Water Level Prediction Model Inspired by Integrator Delay

Related Author

HUANG Xin

YU Zengxin

WANG Liuyi

JU Hairong

LIANG Yuchen

CHAO Yang

QI Huijun

HUANG Lihong

Related Institution

Technology Consulting (Guangzhou) Co., Ltd., Pearl River Water Resources Commission, Ministry of Water Resources

Pearl River Water Conservancy Comprehensive Technology Center, Pearl River Water Resources Commission, Ministry of Water Resources

College of Harbor, Coastal and Offshore Engineering, Hohai University

College of Water Conservancy & Hydropower Engineering, Hohai University

China Huadian Corporation Ltd., Quzhou Wuxijiang Company

AI问答

⁰