Abstract: Climate change and forest vegetation affect the distribution of PM_2.5 concentrations, and PM_2.5 as an important air pollutant can also affect forest vegetation growth directly or indirectly. Currently, the technique of inverting daytime PM_2.5 based on optical aerosol thickness（AOD） data is relatively mature, and as a complement to daytime PM_2.5, nighttime PM_2.5 is of great significance for the all-day PM_2.5 monitoring. Based on the radiation transmission theory, the machine learning estimation model of nighttime PM_2.5concentration in the three northeastern provinces was established with nighttime light brightness, enhanced vegetation index and seven meteorological factors（2 m dewpoint temperature, 2 m temperature, u component of wind speed, v component of wind speed, atmospheric surface pressure, evaporation,precipitation） as input variables, and nighttime PM_2.5 concentration as response variable, aiming to provide a reference for monitoring nighttime PM_2.5 concentration in the three northeastern provinces. The results show that the model constructed based on the integration tree has high estimation accuracy, with a goodness of fit（R2） of 0. 68, a mean absolute error（MAE） of 7. 05 μg/m~3, and a root mean square error（RMSE） of 11. 62 μg/m~3. In addition, the model is found to have certain spatial and temporal sensitivity by analyzing the errors between the estimated and true PM_2.5 values at each monitoring station in the three northeastern provinces. It can provide a reference for the forest vegetation conservation work by timely and accurately controlling the distribution of nighttime PM_2.5 concentration.