Prediction of Surface Soil Moisture Content using Multispectral Remote Sensing and Machine Learning

Information on near-surface soil moisture content (SMC) is very important for various applications such as irrigation scheduling, precision farming, watershed management, climate change analysis, drought prediction, meteorological investigations etc. Soil moisture information acquired from remotely sensed satellite data has been widely used in the recent past. However, these remote sensing data's low spatial and temporal resolution is a limitation for agricultural applications. Unmanned aerial vehicles (UAV)-based soil moisture predictions are thriving, but the studies are limited with fewer ground truth data. This study aims to predict the surface soil moisture content using UAV-based multispectral data and machine learning techniques. The UAV-based multispectral data are acquired from an altitude of 40 m. Surface soil samples were collected at an interval of two days to estimate gravimetric soil moisture content. Four machine-learning algorithms (Linear Regression, SVR, RFR, KNN) were used to develop the relationship between near-surface SMC and multispectral data. At high surface SMC, the soil has low spectral reflectance as compared to low surface SMC. The linear regression algorithm performed best, with R² as 0.89 among the other ML algorithms. Also, blue band reflectance was correlated well with the surface SMC as compared to green, red, NIR and red-edge bands. The findings indicated that UAV-based high-resolution multispectral image analytics could accurately predict the surface SMC. The developed approach of estimation of near SMC may be helpful for farmers and irrigation planners to schedule irrigation and crop management accordingly.

Keywords:  Surface soil moisture content; Remote sensing; UAV; Multispectral imageries; Machine learning