Abstract: Based on the three-source merged grid real-time precipitation data, the verification and evaluation were performed for the area rainfall forecast effect of SWCWARMS, ECMWF, GRAPES_-Meso and intelligent grid forecast of Sichuan province for the upper reaches of Minjiang River during April 2019 to December 2021 by using several evaluation indicators. And on this basis, the area rainfall ensemble forecast research was performed by using regression ensemble method, TS ensemble method and Nash ensemble method. The results showed that the effect of intelligent grid forecast and ECMWF of area rainfall in the upper reaches of Minjiang River was better. In the area rainfall classification verification, the TS scores of the four single-model forecasts gradually decreased with the increase of magnitude, while the empty forecast rate and missing forecast rate gradually increased. ECMWF had the best effect in light rain forecast, and the intelligent grid forecast was superior to other models in moderate rain forecast. The SWCWARMS had better effect in heavy rain forecast. The area rainfall ensemble forecast could improve the prediction effect well. In the three ensemble methods, regression ensemble method had the smallest error, the highest AS score and efficiency coefficient. In the area rainfall classification forecast, three-model multiple regression integration could be used for light rain forecast, four-model TS ensemble could be used for moderate rain forecast, and three-model TS ensemble could be considered for heavy rain forecast.