**1. Introduction**

Since summer precipitation is of grea<sup>t</sup> concern regarding meteorological disasters, a lot of research work has been done on the influence of climate system subsystem changes and their interactions on summer precipitation. Some progress has been made in the study of the characteristics, causes and prediction methods of summer precipitation in Chongqing and its surrounding areas. Li [1–3] analyzed the characteristics of summer precipitation, drought and flood in the eastern part of southwest China, and pointed out that it had obvious inter-annual and inter-decadal changes. Zhou et al. [4] studied the basic climatic characteristics of summer precipitation in the three gorges reservoir area, and the results showed that the summer precipitation in the three gorges reservoir area had a good consistency, the frequency of drought years was significantly higher than that of flood years, and the summer precipitation in the three gorges reservoir area had an obvious inter-decadal variation. Ma Zhenfeng [5] analyzed the main physical factors a ffecting summer precipitation in southwest China, such as plateau factors, westerly belt system, subtropical high and other factors, and established a summer precipitation prediction model with a certain physical basis on this basis, which achieved good results in precipitation prediction in flood season in recent years. Zhang Qiang et al. [6] analyzed the correlation between Sea Surface Temperature (SST) index and drought and flood disaster in the upper reaches of the Yangtze river, showing that the occurrence of El Nino event increases the probability of drought in the upper reaches of the Yangtze river, while the occurrence of La Nina event increases the probability of waterlogging in the upper reaches of the Yangtze river. Liu De et al. [7] analyzed the characteristics of Eurasian circulation in summer rainfall in Chongqing, and established a conceptual model for forecasting summer precipitation in Chongqing by using circulation index in key areas in early winter.

In recent years, artificial intelligence technology has also begun to be applied in the field of atmospheric science such as severe convection weather forecast and climate prediction. There are many applications of machine learning in severe convection weather prediction. In 2017, Shenzhen Meteorological Bureau and Alibaba jointly organized the CIKM data science competition themed "smart city, smart country", and made climate precipitation forecast with radar images. Xiu Yuanyuan et al. [8] used machine learning supervised learning model support vector machine SVM to identify and forecast severe convection weather. Sun Quande et al. [9] showed the potential of machine learning methods in improving local accurate weather prediction. Li Wenjuan et al. [10] founded that the physical significance of factors selected by the random forest algorithm was relatively clear. In the field of climate, researchers have used artificial intelligence systems to help them rank climate models over the past few years [11] to detect hurricanes and other extreme weather events in real and simulated climate data, and thus find new climate models. The above study is based on considering the influence of a single system or physical factor on precipitation in and around Chongqing. The e ffects of anomalies of multiple systems or physical factors on precipitation in Chongqing are considered. In fact, due to the non-linear and chaotic nature of the climate system, the factors that affect the precipitation prediction constitute the comprehensive e ffects of many sea temperatures (ENSO, Kuroshio, etc.), plateau snow, land surface temperature, volcanic activity, astronomical factors, monsoon, subtropical high, high resistance and plateau topography. We aim to analyze the synergistic effect of the factors that lead to the precipitation change through sorting, statistics, analysis and processing of big data, machine learning, etc., and distinguish which factors are excellent forecasting factors, and the weight of these excellent factors in di fferent regions, that is, how much forecasting information these factors can provide. If these issues are resolved, precipitation prediction will become possible and credible.

As an excellent representative of the machine learning algorithm, the decision tree model adopts recursive segmentation technology to continuously divide the data space into di fferent subsets so as to detect the potential structure, important patterns and relationships of data [12]. Compared with traditional parametric statistical methods, the decision tree model does not need to make assumptions about the relationship between independent variables and dependent variables in advance, and it can effectively overcome the multi-collinearity of independent variables. However, the results of a single decision tree are unstable and prone to overfitting. The random forest model builds decision trees by randomly extracting some samples from the original samples through Bootstrap sampling technology, and combines multiple decision trees to e ffectively avoid overfitting [13]. At present, decision trees and random forest algorithms are more and more widely used in meteorology. Shi Dawei et al. [14] used the decision tree algorithm to establish a more accurate classification and prediction model for road icing disaster. Shi Yimin et al. [15] studied the classification and prediction model of regional summer precipitation days based on the data mining Classification and Regression Tree (CART) algorithm. Qin Pengcheng et al. [16] Hubei rapeseed yield limiting factor analysis based on a decision tree and random forest model also achieved good application.

Based on the actual forecast business, this paper adopts the decision tree classification method to establish the precipitation prediction model with multi-factor collaborative influence for the average summer precipitation in Chongqing. Based on decision tree modeling, random forest is used to conduct an integrated prediction test, and evaluate its prediction e ffects.

#### **2. Materials and Methods**
