Spectral preprocessing methods combined with data downscaling techniques improved the prediction accuracy of soil structure indicators

Rapid and accurate estimation of soil structure is significant for soil quality assessment and farm management. The Generalized Soil Structure Index (GSSI) and Soil Three-phase Structure Distance (STPSD) are two important indices for soil structure analysis. However, the inefficiency of traditional measurements has limited research progress. This study focuses on hyperspectral technology, aiming to explore efficient measurement methods for soil structure indicators and provide preliminary technical references for their quantitative inversion. The raw spectral reflectance was pre-processed in various ways, and the spectral characteristics of the soil structure indices were analyzed. The hyperspectral data were downscaled using spectral resolution resampling and the successive projection algorithm (SPA), and then the estimation models of the soil structure indices were developed using partial least squares (PLS). The correlation between spectral reflectance and soil structure indicators was improved by the first derivative (1ST), improving by 5.941% for GSSI and 3.134% for STPSD. The method of spectral resolution resampling technique combined with SPA not only greatly reduces the data dimensions, but also better realizes the preliminary estimation of soil structure indicators. Compared with the constructed models, the DET-15nm-SPA model yielded the best estimation of GSSI (R 2 =0.651, RMSE=10.464, rRMSE=13.053%), and the 1ST-15nm-SPA model provided the best estimation of STPSD (R 2 =0.568, RMSE= 4.646, rRMSE=32.966%).The spectral preprocessing technique combined with spectral resolution resampling greatly reduces the spectral data dimension while enhancing spectral features. It contributes to the construction of less complex and more stable monitoring models for soil structure indices.