5:15 PM - 7:15 PM
[SGD02-P07] Machine Learning-Based GNSS-R Soil Moisture Estimation Model
Keywords:GNSS-R, CYGNSS, TRITON, Soil Moisture Content, Deep Learning
Due to the difficulty of traditional soil moisture content measurement and the high cost of large-area observations, along with the fact that soil moisture content is highly susceptible to changes in the surrounding environment, it is challenging to obtain accurate and large-scale soil moisture data. With the development of Global Navigation Satellite System-Reflectometry (GNSS-R), which features wide coverage and rapid data transmission, large-scale surface analysis can be conducted. Since its microwave signals are insensitive to atmospheric variations and can penetrate clouds and vegetation of certain thicknesses, coupled with the significant difference in dielectric properties between wet and dry soil, GNSS-R satellites can be used to estimate large-scale soil moisture variations. As a result, GNSS-R has recently emerged as a technology for satellite remote sensing of soil moisture content.
There are many well-known and successful GNSS-R satellites, such as CYGNSS from the United States, the Galileo satellite system from the European Union, and GLONASS from Russia. In 2023, Taiwan also launched its first domestically developed GNSS-R satellite, TRITON. Therefore, in addition to leveraging publicly available data from the U.S. CYGNSS, this study will also incorporate data from Taiwan's TRITON satellite to enhance the diversity of the training dataset.
However, most existing studies examine the relationship between GNSS-R signal-to-noise ratio (SNR) and soil moisture content by linear regression models. To account for variables influencing soil moisture content, this study is expected to incorporate not only GNSS-R data (DDM_SNR, a signal-to-noise ratio) but also factors such as humidity, precipitation, air pressure, temperature, and wind speed, which may affect soil moisture content. Through deep learning, a nonlinear multidimensional relationship analysis model will be developed, with the aim of achieving effective monitoring and prediction of soil moisture distribution over time and space in Taiwan. This approach aims to improve traditional linear regression methods by leveraging multidimensional data to enhance the accuracy of soil moisture content retrieval, ultimately providing a more reliable estimation for environmental monitoring and agricultural applications.
There are many well-known and successful GNSS-R satellites, such as CYGNSS from the United States, the Galileo satellite system from the European Union, and GLONASS from Russia. In 2023, Taiwan also launched its first domestically developed GNSS-R satellite, TRITON. Therefore, in addition to leveraging publicly available data from the U.S. CYGNSS, this study will also incorporate data from Taiwan's TRITON satellite to enhance the diversity of the training dataset.
However, most existing studies examine the relationship between GNSS-R signal-to-noise ratio (SNR) and soil moisture content by linear regression models. To account for variables influencing soil moisture content, this study is expected to incorporate not only GNSS-R data (DDM_SNR, a signal-to-noise ratio) but also factors such as humidity, precipitation, air pressure, temperature, and wind speed, which may affect soil moisture content. Through deep learning, a nonlinear multidimensional relationship analysis model will be developed, with the aim of achieving effective monitoring and prediction of soil moisture distribution over time and space in Taiwan. This approach aims to improve traditional linear regression methods by leveraging multidimensional data to enhance the accuracy of soil moisture content retrieval, ultimately providing a more reliable estimation for environmental monitoring and agricultural applications.