Comparative study of flash flood in ungauged watershed with special emphasizing on rough set theory for handling the missing hydrological values

Prediction of the flash floods in ungauged or poorly gauging watershed is one of the challenging tasks in the field of hydrology and needs implication of advanced techniques to obtain the reliable results. In this study, an innovative artificial intelligence-based rough set theory (RST) was used to retrieve missing hydro-meteorological data which were utilized to build a forecast model to predict the flood event in an ungauged watershed in Pakistan (Thor Nullah). The RST-based forecast model was calibrated for 1986 to 2004 and tested for 2008 to 2016. The result showed that 9 out of 10 forecasting objects were predicted precisely. Basin data model technique along with rainfall–runoff (R.F-R.O) model and RST forecasting model was used to estimate the peak discharge of flood event occurred in 2015. The modeled peak discharge (1152 m3 s−1) was compared with the field observation-based highest flood marks (HFMs—1189 m3 s−1), which showed slight discrimination due to indeterminate model calibration sparse rain gauge density. Moreover, flood inundation map showed high flood risk to the 80% localities with a flood depth of 0.1–1.67 m in locality. Overall, this study suggested a reliable use of RST for data mining and flood modeling; however, the absence of adequate flow data at study site limits the reliability of R.F-R.O model calibration. Moreover, based on the array of flood hazard simulation studies, provision of channelization and cross-drainage works is suggested to protect the catchment against floods and debris brought down through catchment.