关键词: 土壤水同位素, 提取方式, 土壤质地, 植物水源划分, 平衡时间

Abstract: To select the appropriate pre-treatment method and thus to improve the predictiveaccuracy of leaf SPAD value, the leaf spectra were processed with normalized correction (NC), multiple scatter correction (MSC), baseline correction (BC) and their combinations. Successive projection algorithm (SPA) and stepwise multiple linear regression (SMLR) were used for establishing the predictive model and determining the optimal spectral pre-treatments. The results showed that increased N application rate could improve the SPAD values. The second top leaf and the first top leaf had the highest value of near-infrared spectral reflectance at early growth stage and after the flowering stage, respectively. The pre-treatment of NC method had the highest-predictive model performance with R²=0.770 andRMSE=1.483, whereas  the MSC pre-treatment resulted in poorer model performance. The validated model with the pre-treatment of BC+NC combination had the best prediction (R²=0.755, RMSE=1.540). Pre-treatment obviously increased the prediction; however, improper pre-treatment would reduce the prediction ability. Our results indicated the single pre-treatments or the combinations did not always improve the model performance.

Key words: equilibrium time, soil texture., soil water isotope, extraction method, plant water source segmentation