Homogeneity Testing and Quantile Regression for Regional Flood Estimation in the Narayani Basin, Nepal

Abstract

Regional flood estimation in Nepal is challenged by short, uneven, and spatially sparse hydrometric records, particularly in Himalayan basins where strong physiographic and hydroclimatic contrasts make regional pooling difficult. This study examines the combined use of homogeneity testing and quantile regression for regional flood estimation in the Narayani Basin, Nepal. Annual maximum discharge records from gauged stations, together with selected catchment descriptors, were analyzed within an L-moment framework to evaluate regional homogeneity and identify a pooling group suitable for regional analysis. Quantile regression models were then developed to estimate flood quantiles directly from catchment descriptors and limited annual maximum series. The results show that achieving strict homogeneity in the Narayani Basin is difficult without substantially reducing the available station-year data, reflecting the marked variability of the basin. However, homogeneity screening allowed the identification of a defensible pooling group for regional flood estimation. The quantile regression results indicate that the method makes effective use of limited regional information and provides reasonable flood estimates for low to moderate return periods, with stable performance up to about the 50-year return period. Uncertainty increases for rarer events, and the reliability of regional estimates declines. The study shows that Hosking-Wallis homogeneity assessment combined with quantile regression provides a practical basis for estimating design floods in data-scarce Himalayan basins.