Abstract: The Robinia pseudoacacia in Mizhi County, located in a semi-arid region, and Changwu County, situated in a semi-humid area, within Shaanxi Province were selected as the subjects of study. By measuring the midday leaf water potential of R. pseudoacacia on the ground, multi-spectral images of R. pseudoacacia were obtained synchronously and its band reflectance was extracted to construct the vegetation index. Firstly, the difference of midday leaf water potential between the 2 places was analyzed. Secondly, the correlation between spectral index and ground measured data is analyzed. Finally, combine variables in 3 ways: 5-band combination variable, 5-band + 10 vegetation index combination variable and full subset screening variable. Random forest, support vector machine and radial basis neural network models were constructed respectively. The results showed that the correlation between vegetation index and midday leaf water potential was good, in which the ratio index of red to green was good and improved enhanced vegetation index had the highest correlation with the midday leaf water potential, and the correlation coefficients were 0.39 and −0.51, respectively. Normalized vegetation index, non-linear index, red-green ratio index and improved enhanced vegetation index were the optimal combinations of all the selected variables. Random forest method is the optimal model construction method, and the R² and RMSE of random forest model validation set are 0.76 and 0.27 MPa.