Probabilistic Quantitative Precipitation Forecasting Using Machine Learning Methods and Probable Maximum Precipitation

Abstract

Precise prediction of rainfall, especially in dry and semidry regions, can play a significant role in effective
management of water resources. The recent challenging problems in management of water resources and
dealing with climatic risks is finding an accurate forecasting method. Although extensive research has
been conducted in employing computational intelligence methods to predict precipitation and runoff, the
probable maximum precipitation (PMP) method is one of the more practical tools used to design water
structures and predict maximum floods in Iran. In this study the efficiency of this method compared to
using artificial neural network methods has been challenged. The results show while an artificial neural
network method will produce an accurate response for predicting precipitation processes, yielding a very
high correlation coefficient of about 97%, a synoptic method will estimate maximum values far better
than does the neural network approach. Concurrent use of these two methods will result in the most
satisfactory investigation and comparison as well as the most useful modeling.