Inversion of soil moisture in bare soil based on multi-spectral remote sensing

Abstract

The monitoring of soil water content in the dumping area of mining area is of great significance to the development of mineral resources, ecological restoration and early warning of drought in China.In this paper, four different depths(1cm, 3 cm, 5 cm, 10 cm) of soil samples were continuously monitored from 10:00 to 14:00 using Spequoia multispectral camera and ECH2O soil moisture sensor.Soil spectral reflectance of four bands(550 nm, 660 nm, 735 nm, 790 nm) and soil water content data were collected.The soil water content inversion model with single-band or multi-band spectral reflectance combination as the inversion factor was established by using the methods of partial least squares regression, ridge regression and back propagation(BP) neural network.The results show that the inversion accuracy of the two regression methods is low, and the highest R2 is only 0.606.The G-R-N-BP model with the combination of green(550 nm), red edge(735 nm) and near-infrared(790 nm) as the inversion factor has the best inversion effect, corresponding to the soil water content inversion model of 1 cm, 3 cm, 5 cm and 10 cm depth.The coefficient of determination(R2) is 0.866, 0.8, 0.975, 0.911, and the root mean square error(RMSE) is 0.333, 0.361, 0.103, 0.315.In our research, a variety of methods are used to invert the soil water content at different depths.The optimal inversion model is G-R-N-BP model.The best inversion depth is 5 cm.The method has high precision and can further develop the surface moisture monitoring of the mining area.

Keywords

• multispectrum
• soil water content
• bp neural network
• mining area