Accurate and timely precipitation estimates are very important for preventing the occurrence of disasters related to atmospheric hazards. Sudden changes in weather of the local scope cause challenges in accurate and short-term forecasts, especially in developing countries that often lack ground and radar equipment to accurately measure precipitation. In order to reconstruct rainfall data in areas without meteorological stations, various traditional methods including spatial interpolation and statistical regression are used. These methods use ground data, especially station precipitation data, to estimate precipitation. For areas that do not have rain gauge stations or the dispersion of stations is high, as well as in the hot months of the year when the rains are more convective and spot, rainfall estimation is associated with a high error only with ground data. New methods for rainfall estimation are the use of satellite and ground data with the help of machine learning models, including neural networks and deep learning. Unlike station data, satellite data has the advantage of global coverage, high and regular spatial and temporal resolutions. In this research, the brightness temperature of the second generation Meteoest 8 and 9 satellites in two infrared channels with wavelengths of 11 and 6.7 micrometers, as well as the geographical characteristics of weather stations, including station altitude, longitude and latitude, are considered as input variables, while precipitation Recorded by the stations has been used as an output variable. The station precipitation data is collected daily for a four-year period from 2019 to 2022, especially for the months of June, July, and August. In this research, the data of 115 meteorological stations were used, which located in the range of longitude 48.1 to 60 degrees east and latitude 28 to 37.2 degrees north, which almost covers the area of the central draining basin of Iran, also it is located at heights of 482 to 2875 meters above sea level. Recently, machine learning technologies such as artificial neural network, deep learning and adaptive neural fuzzy inference system have been widely used for rainfall estimation. One of the advantages of using machine learning is its relatively few and fast calculations. Unlike numerical atmospheric models, data-based models face fewer assumptions and limitations in modeling. To develop and build neural network models, 75% of the data was allocated for model training, 15% for validation and 10% for the final test of the model. In the testing phase, the correlation coefficient between the station daily rainfall and the rainfall estimated from the models is 0.65 for the perceptron neural network, 0.55 for the generalized regression neural network model and 0.47 for the adaptive neural fuzzy inference system model. The comparisons show that all three models have the lowest estimation error in rainfall close to 3 mm, compared to the daily rainfall of the stations. But due to the non-normal distribution of the positive skewness of the stations, the estimation of the models has the biggest error in high rainfalls.