TY - GEN
T1 - High-Resolution Near-Surface Soil Water Content Estimation, Using Remote Sensing Data and Deep Computer Vision Methods
AU - Rafiei, M
AU - Asif, MR
AU - Nørremark, M
AU - Sørensen, CAG
PY - 2024
Y1 - 2024
N2 - This paper proposes a novel methodology for high spatial resolution near-surface daily Soil Water Content (SWC) estimation using remote sensing data and deep computer vision techniques. Current SWC estimation methods have accuracy, spatial resolution, and globalization limitations. To overcome these limitations, the proposed approach integrates diverse data sources, including satellite images (land elevation, SAR data, and a vegetation index), time-series data (precipitation, temperature, global radiation, and wind), and soil and land cover properties. To do this, we use a hybrid deep model consisting of a U-Net for image processing, a Temporal Convolutional Network (TCN) for time-series features, and a Feed-Forward Network (FNN) for constant properties. This solution makes it possible to capture spatial variability and temporal dynamics simultaneously. We used the HOBE dataset as SWC ground truth to conduct initial experiments that showed promising results in accurately estimating SWC across different soil depths. Further refinements are needed for a potential solution for various hydrology, environmental, and agricultural applications.
AB - This paper proposes a novel methodology for high spatial resolution near-surface daily Soil Water Content (SWC) estimation using remote sensing data and deep computer vision techniques. Current SWC estimation methods have accuracy, spatial resolution, and globalization limitations. To overcome these limitations, the proposed approach integrates diverse data sources, including satellite images (land elevation, SAR data, and a vegetation index), time-series data (precipitation, temperature, global radiation, and wind), and soil and land cover properties. To do this, we use a hybrid deep model consisting of a U-Net for image processing, a Temporal Convolutional Network (TCN) for time-series features, and a Feed-Forward Network (FNN) for constant properties. This solution makes it possible to capture spatial variability and temporal dynamics simultaneously. We used the HOBE dataset as SWC ground truth to conduct initial experiments that showed promising results in accurately estimating SWC across different soil depths. Further refinements are needed for a potential solution for various hydrology, environmental, and agricultural applications.
UR - http://www.scopus.com/inward/record.url?scp=85214783379&partnerID=8YFLogxK
U2 - 10.3997/2214-4609.202420111
DO - 10.3997/2214-4609.202420111
M3 - Article in proceedings
T3 - EAGE Conference Proceedings
SP - 1
EP - 5
BT - NSG 2024 30th European Meeting of Environmental and Engineering Geophysics
PB - European Association of Geoscientists and Engineers
ER -