| The purpose of this study is to select suitable sensitive spectral indices and algorithms for predicting the yield of drip-irrigated winter wheat. Through a two-year field trial using New Winter 22 as the experimental material, canopy spectral data were collected during four critical growth stages of drip-irrigated winter wheat: the jointing stage, booting stage, flowering stage, and grain filling stage, while the Leaf Area Index (LAI) was measured using Sunscan. Using Pearson correlation analysis, fifteen spectral indices were screened to identify sensitive spectral indices that characterize yield, and a yield prediction model was constructed combining Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) algorithms. The results showed that when modeling with only the combined spectral indices, the RF model during the flowering stage achieved the highest accuracy (R2=0.70，RMSE=927.86 kg/hm2); integrating the combined spectral indices with LAI resulted in the optimal RF model during the grain filling stage (R2=0.73，RMSE=910.06 kg/hm2); however, when the combined spectral indices and LAI were integrated with sensitive bands, the RF model during the grain filling stage exhibited the highest accuracy (R2=0.76，RMSE=728.47 kg/hm2). The results indicate that the integration of combined spectral indices, LAI, and sensitive bands can most accurately predict the yield of drip-irrigated winter wheat. |