Abstract: 【Objective】Hyper spectral technology is more and more widely applied in monitoring soil properties. Soil Water Content (SWC) is a key parameter of soil properties. This paper aimed to make clear the spectral response of soil moisture under different soil roughness to predict quantitatively soil water content, and further to provide the basis for rapid monitoring of farmland moisture and other soil properties. 【Method】 Soil samples were sieved through different mesh sizes to classify into different particle sizes, and then set different moisture levels. The spectral reflectance of those samples were compared, and the quantitative prediction models of soil water content were established by Partial Least Squares Regression (PLSR) method. 【Result】The results showed that the spectral reflectance decreased with the increase of soil water content, and the difference became bigger as the increase of wavelength and decrease of soil water content. The absorptions got deeper in 1 400 nm and 1 900 nm with the increase of water content. In those soil samples passed through a sieve with an aperture of 0.15 mm (denoted as D-1), the spectral reflectance increased in 350-1 240 nm and then decreased after 1 240 nm when the water content was more than 40%. Compared with the model constructed from all samples with different sizes, models from the same size were improved in predicting accuracy and stability: the smaller the particle size was, the better the prediction effect and stability of the predicting model were. The RMSE (root mean square error) and R2 of the optimal model were 4.13% and 0.90, respectively. Additionally, normalization of spectral data reduced the influence of noise, and improved the predicting accuracy and stability of the model. 【Conclusion】The spectral generally decreased with soil water content increasing, but soil with small size showed opposite in 350-1 240 nm when the moisture content was greater than 40%. The predicting model for soil water content was improved as size getting smaller and sample number involved getting larger, and the spectral data normalization also improved predicting accuracy and stability of model.

Key words: spectral, soil water content, particle size, model