Research on Soil Moisture Inversion Method for Canal Slope of the Middle Route Project of the South to North Water Transfer Based on GNSS-R and Deep Learning

被引：4

作者：

Hu, Qingfeng ^{[1
]}

Li, Yifan ^{[1
]}

Liu, Wenkai ^{[1
]}

Lu, Weiqiang ^{[1
]}

Hai, Hongxin ^{[1
]}

He, Peipei ^{[1
]}

Liu, Xianlin ^{[1
,2
]}

Ma, Kaifeng ^{[1
]}

Zhu, Dantong ^{[1
]}

Wang, Peng ^{[1
]}

Kou, Yingchao ^{[1
]}

机构：

[1] North China Univ Water Resources & Elect Power, Coll Surveying & Geoinformat, Zhengzhou 450046, Peoples R China

[2] Chinese Acad Engn, Beijing 100088, Peoples R China

来源：

REMOTE SENSING | 2023年 / 15卷 / 17期

基金：

中国国家自然科学基金;

关键词：

data fusion; deep learning; drought detection; GNSS-R; soil moisture; South to North Water Transfer; GPS SIGNALS;

D O I：

10.3390/rs15174340

中图分类号：

X [环境科学、安全科学];

学科分类号：

08 ; 0830 ;

摘要：

The soil moisture from the South-to-North Water Diversion Middle Route Project is assessed in this study. Complex and variable geological conditions complicate the prediction of soil moisture in the study area. To achieve this aim, we carried out research on soil moisture inversion methods for channel slopes in the study area using massive monitoring data from multiple GNSS observatories on channel slopes, incorporating GNSS-R techniques and deep learning algorithms. To address the issue of low accuracy in linear inversion when using a single satellite, this study proposes a multi-satellite and multi-frequency data fusion technique. Furthermore, three soil moisture inversion models, namely, the linear model, BP neural network model, and GA-BP neural network model, are established by incorporating deep learning techniques. In comparison with single-satellite data inversion, with the data fusion technique proposed in this study, the correlation is improved by 12.7%, the root mean square error is reduced by 0.217, the mean square error is decreased by 0.884, and the mean absolute error is decreased by 0.243 with the linear model. With the BP neural network model, the correlation is increased by 15.4%, the root mean square error is decreased by 0.395, the mean square error is decreased by 0.465, and the mean absolute error is reduced by 0.353. Moreover, with the GA-BP neural network model, the correlation is improved by 6.3%, the root mean square error is decreased by 1.207, the mean square error is decreased by 0.196, and the mean absolute error is reduced by 0.155. The results indicate that performing data fusion by using multiple satellites and multi-frequency bands is a feasible approach for improving the accuracy of soil moisture inversion. These research findings provide new technical means for the risk analysis of deformation disasters in the expansive soil channel slopes of the South-to-North Water Diversion Middle Route Project.

引用

页数：34

共 45 条

[1]

Alonso-Arroyo A, 2015, INT GEOSCI REMOTE SE, P3921, DOI 10.1109/IGARSS.2015.7326682

[2] THE LIGHT AIRBORNE REFLECTOMETER FOR GNSS-R OBSERVATIONS (LARGO) INSTRUMENT: INITIAL RESULTS FROM AIRBORNE AND ROVER FIELD CAMPAIGNS [J].