Author(s): Takuya Matsumoto; Yusuke Hiraga; Shunsuke Aita
Linked Author(s):
Keywords: Soil moisture machine learning UAV viticulture
Abstract: This study utilized UAV multispectral imagery and machine learning algorithms to estimate time series and spatial distribution of soil moisture content at high resolution. This study focused on the vineyard with steep slope in Miyagi, Japan (Ryomi Vineyard & Winery). The soil moisture content measured at 55 sites using TDR (Time Domain Reflectometry) method were augmented to 5,500 based on UAV imagery. Based on the large number of samples, the random forest algorithm was applied to estimate the spatial distribution of soil moisture content with high resolution, resulting in the high accuracy of estimation (R2=0.96). We also developed a random forest model to estimate the daily variation of soil moisture content based on the observed temperature, rainfall, and irradiance. Although the model was able to roughly capture the daily variation of soil moisture content, the large variation was not hindcasted well, which needs to be addressed in future studies.
Year: 2025