River floods are the most destructive flood type, hence the need for accurate riverine flood mapping (RFM). Previous studies often chose methods and criteria subjectively or after extensive testing, an approach that is time-consuming and resource-intensive, and sometimes yielded conflicting results. The present study uses accuracy metrics from past RFM studies worldwide to statistically evaluate the influence of five variables on accuracy (i.e., applied method, topographic type, selected criteria in studies that applied multi-criteria methods (MCM), area extent, and reference dataset size). Across studies, overall method rankings show remote sensing (RS) as the most accurate approach, followed by machine-learning and deep-learning (MLDL) and physically based (Ph-B) methods, then statistical approaches, with subjective methods lowest. MLDL and statistical methods perform best in mountainous regions, whereas RS and Ph-B are more appropriate for lowlands. Ph-B accuracy is uniquely sensitive to area extent, performing highest in small areas. Distance from the river (DFR), elevation, and slope are the most influential criteria, and prioritizing these three significantly improves accuracy. Based on these findings, this review proposes general guidelines for RFM, supporting objective, context-appropriate method and criterion selection, and addressing key challenges. • State-of-the-art review identifying key gaps in the riverine flood mapping literature (subjectivity in method and criteria selection, and sometimes conflicting results). • A meta-analysis of riverine flood mapping accuracy across method type, topography, area extent, reference dataset size, and selected criteria. • Synthesized guidelines for riverine flood mapping, emphasizing the most relevant methods and criteria with respect to study area characteristics.
Moumen et al. (Sun,) studied this question.