Effects of salt content and particle size on spectral reflectance and model accuracy: Estimating soil salt content in arid, saline-alkali lands

Mingyue Sun,Hongguang Liu,Pengfei Li,Ping Gong,Xuyong Yu,Fuhai Ye,Yong Guo,Zhenkun Wu

DOI: https://doi.org/10.1016/j.microc.2024.111666

IF: 5.304

2024-09-22

Microchemical Journal

Abstract:Excessive soil salt content (SSC) leads to soil salinization, which negatively impacts soil structures as well as plant growth. Accurate and rapid monitoring of SSC is essential for the effective management of saline-alkali land resources. This study explores techniques for measuring SSC and spectral reflectance in soil from farmlands in the Manas River Basin. First, SSC was calculated and divided into two sample groups ("High" and "Low"), and then these soil samples were ground into three particle sizes (0.1 mm, 0.5 mm, and 2 mm). The hyperspectral information of the soil samples at each particle size was measured, followed by denoising and three differential transformations of the data. Principal Component Analysis (PCA) was used to identify characteristic spectral bands for sample groups. Estimation models were constructed using Random Forest (RF), Extreme Randomized Trees (ERT), and eXtreme Gradient Boosting techniques (XGBoost). The findings indicate that: 1) SSC and soil particle size are inversely proportional to soil spectral reflectance, with the 0.1 mm-Low-SSC group exhibiting the highest average spectral reflectance. 2) The original spectral reflectance positively correlates with SSC, particularly in visible light and shortwave near-infrared (400–1100 nm) ranges, which are more sensitive to SSC. In addition to this, three first-order differential transformations significantly enhanced this correlation. 3) The predictive accuracy of SSC varies across models and soil conditions. The 0.5 mm-(1/R)′-XGBoost model most effectively estimates SSC in low-salt regions, while the 0.5 mm-(1/R)′-ERT model is optimal for estimating high-salt soil SSC in the study area. This research systematically analyzes SSC predictions and evaluates the fitness between elements in high and low salinity soil samples, three sample particle sizes, three differential transformed hyperspectral data, and three modeling approaches. This study aims to investigate the combined effects of soil salt content and soil particle size on spectral reflectance and modeling accuracy by controlling the variables in the salt prediction process. An experimental scheme is devised to monitor SSCs of different levels based on the prediction performance of various hyperspectral data preprocessing and modeling methods. The proposed scheme offers an effective approach to measuring the salinity of different soil samples for accuracy and efficient SSC monitoring.

chemistry, analytical