Recently, drought occurrences have imposed irreparable damages on human societies and natural ecosystems in Iran. Due to significant socio-economic damages, droughts have received considerable attention from the researchers. One of the most common indices which investigate occurrence probability of drought is low flow index. In this study regional low flow frequency analysis is applied for a reliable estimation of low flow quantiles in ungauged sites. To perform regional analysis of low flow, 26 stations were selected and homogeneity of these stations was checked. Homogeneity measures were employed on the 7-day annual minimum series as low flow index. Results showed that the study area is definitely heterogeneous (H 1 =13.68). To form homogenous regions, Fuzzy Cluster Analysis (FCA) and multivariate techniques were used according to physiographic characteristics of subbasins. Although the results of homogeneity test based on FCA method showed better results than multivariate techniques, but homogeneity test was not acceptable for resulted clusters. Therefore combination of 8 at site statistics and 5 at site characteristics of the selected stations were used together for homogenous cluster formation. Applying hierarchical cluster separated basin into the eastern and western parts. Removing of Gilvan, Yengikand and Firooz-abad stations from western part, homogeneity test revealed that two separated parts are homogenous. In continue, Pearson III and Generalized logistic distributions were selected as the robust regional distributions based on Z Dist criteria. Then at-site low flow statistics as dependent variables and physiographic and land use characteristics as independent variables were used to establish regional low flow models. According to these models, drainage area, average basins slope and land use characteristics were identified as important factors in low flows features as hydrological drought index. On the other hand, uncertainty analysis of two types of regression models was performed using asymptotic approximation and Bayesian analysis. According to uncertainty analysis 95% confidence intervals are remarkably reduced using the Bayesian approach. Finally, results showed that 2-parameters distributions are fitted for upstream regions with small sample sizes. Keywords: Hydrological drought, Regional frequency analysis, L-moments, multivariate techniques.
