Abstract: During the growth process of tobacco, accurate and rapid acquisition of soil alkaline nitrogen (SAN) distribution in tobacco fields can provide an important basis for the reasonable application of nitrogen fertilizers. Firstly, four tobacco fields which are located in Baokang and Xuanen counties in Hubei Province, China, were selected in this study. Unmanned aerial vehicle (UAV) multispectral remote sensing images and SAN data from sampling points were obtained. Secondly, because the sensor contains two red edge bands, improvements have been made by introducing the red edge band to traditional spectral indices. Thirdly, four machine learning methods, including random forest (RF), support vector machine (SVM), back propagation neural networks (BPNN) and extreme gradient boosting (XGBoost), were used to construct models for estimating SAN content in tobacco fields. The results showed that: (1) The gray correlation shows that improved spectral indices are higher than traditional spectral indices in terms of correlation with SAN. (2) The combination of the improved spectral indices and RF algorithm is the best method to construct the prediction model. The coefficient of determination (R²) and root mean square error (RMSE) for the highest modeling and validation sets are 0.833, 0.784 and 0.846, 0.951, respectively. (3) The RF algorithm was validated on 30% of the overall sample, and the R² and RMSE were 0.814 and 0.906, respectively, indicating that the stability of the RF algorithm was consistent with its performance when modeled alone. Overall, the prognosis is accurate. This study provides a new method for prediction of SAN content in tobacco fields of the Wuling mountain area.