Predictive modeling of daily precipitation occurrence using weather data of prior days in various climates

Precipitation is one of the most important climatic parameters in water resources management. Accurate forecasting of the precipitation occurrence can effectively help to flood risk management, agriculture and irrigation planning, and tourism calendar. In this research, daily precipitation occurrence in Isfahan, Rasht and Tabriz cities (in Iran) with different climates was predicted and evaluated using the meteorological data of prior days. The data set of 2000–2019 (20-years) was applied and the prediction was done using the well-known intelligent methods. Results showed that in Isfahan, the shorter the time lag was (1–3 days), the more effective the parameter “RH” on predicting the daily precipitation was. In the longer the time lag (4–5 days), the more effective was the parameter “T”. In Isfahan, there would be precipitation in four days after the days with T_max−4 less than 12.93 ºC. Otherwise there would be no precipitation. The decision tree had a precision of about 80.4%. Rasht city, with a very humid climate, has an inverse prediction pattern compared to Isfahan (with arid climate). In Rasht, the shorter the time lag (1–3 days), the more effective was the parameter “T” while, the longer the time lag was (4–5 days), the more effective the parameter “RH” on forecasting the daily precipitation was. In Rasht, precipitation would occur in five days after the days with \({T}_{mean-5}\le 24.87\) $T_{mean-5}\le 24.87$ ºC. Also, five days after the days with  \({RH}_{min-5}\le 49.67\) ${\mathit{RH}}_{min-5}\le 49.67$ % precipitation would occur, if \({T}_{mean-5}>24.87\) $T_{mean-5}>24.87$ ºC. Otherwise there would be no precipitation. The precision of the decision tree was 78.9%. In Tabriz, with semi-arid climate, “Sun” (CCI = 65.8%, correctly classified instances) and “Cloud” (CCI = 68.7%) are the most appropriate parameters to forecast the daily precipitation occurrence for the scenarios of the previous 1 and 2 days, respectively. In Tabriz, the longer the time lag (3–5 days), the more effective the parameter “T” on forecasting the daily precipitation was. In Tabriz, precipitation would occur in five days after the days with \({T}_{max-5}\le 25.2\) $T_{max-5}\le 25.2$ ºC. If \({T}_{max-5}>25.2\) $T_{max-5}>25.2$ ºC, then there would be no precipitation in the fifth day. The accuracy of this decision tree was about 62.1%.