Abstract: To explore the feasibility of estimating SPAD （Soil and plant analyzer development） of spring wheat （Triticum aestivum L.） using the three dimensional spectral index （TBI）, this study focused on spring wheat at the field scale, collected canopy hyperspectral data at the heading stage, and calculated the three dimensional spectral index for any band combination to construct a SPAD estimation model for spring wheat based on the optimal three dimensional spectral index. The results showed that the three-dimensional spectral indexes TBI-1_{（849, 850, 850 nm）}, TBI-2_{（849, 850 997 nm）}, TBI-3_{（850, 849, 850 nm）}, and TBI-4_{（849, 849, 850 nm）} were significantly correlated with SPAD at the range of 400~1 300 nm （P<0.01）; a spring wheat SPAD estimation model was established using three machine learning algorithms： artificial neural network （ANN）, k-nearest neighbor （KNN）, and support vector machine （SVM）. By comparing the estimation results of the models, it was found that the model constructed using KNN algorithm had better estimation performance （R²=0.79, RMSE=2.68, RPD=2.25） than ANN and SVR algorithms.

Key words: spring wheat （Triticum aestivum L.）, SPAD, machine learning, three-dimensional spectral index, hyperspectral