Prediction of daily precipitation of Sardasht Station using lazy algorithms and tree models

Due to the heterogeneous distribution of precipitation, predicting its occurrence is one of the primary and basic solutions to prevent possible disasters and damages caused by them. Considering the high amount of precipitation in Sardasht County, the people of this city turning to agriculture in recent years and not using classification models in the studied station, it is necessary to predict the daily precipitation parameter as accurately as possible. On the other hand, although the optimal performance of lazy algorithms and tree models has increased their use for predicting various hydrological phenomena, these algorithms have not been used in Sardasht County. Therefore, in this research, four models Kstar, M5P, learning algorithm with local weighting, and random forest are used to predict the daily precipitation of Sardasht Station. In this study, seven input parameters of average temperature, maximum temperature, average relative humidity, maximum relative humidity, average wind speed, maximum wind speed, and sunshine hours which were the same time as daily rainfall were used for the models. The comparison and evaluation between the input parameters showed that the sunshine hours was one of the most important input parameters, which played a significant role in the prediction accuracy of the used models. The obtained results showed that the M5P tree model had the best performance in the seventh scenario with the highest correlation coefficient (0.734 mm/day) compared to other models. In addition, the seventh scenario showed a high performance compared to the rest of the scenarios. Therefore, it can be said that increasing the input of the models has a direct relationship with their accuracy. In general, it can be said that the M5P tree model is suitable for modeling and forecasting daily rainfall in Sardasht City and it is recommended for future use.