Analysis of parameter uncertainty in SWAT model using a Bayesian Box–Cox transformation three-level factorial analysis method: a case of Naryn River Basin

Abstract

Abstract Hydrological models are often plagued by substantial uncertainties in model parameters when analyzing water balance, predicting long-time streamflow, and investigating climate-change impact in watershed management. In this study, a Bayesian Box–Cox transformation three-level factorial analysis (BBC-TFA) method is developed for revealing the influence of parameter uncertainty on the runoff in the Naryn River Basin. BBC-TFA cannot only quantify the uncertainty through Bayesian inference but also investigate the individual and interactive effects of multiple parameters on model output. Main findings disclose that: (i) the contribution rate of runoff potential parameter during the non-melting period reaches 88.22%, indicating a flood risk in the rainy season; (ii) the contribution rate of snow temperature lag factor is the highest during the snow-melting period and the entire year (respectively occupying 76.69 and 53.70%), indicating that the glacier melting exists in the Naryn River Basin throughout the year; (iii) the Box–Cox transformation can successfully remove residual variance and enhance the correlation between input and output variables. These findings serve to revealing the presence of glacial resources in the study basin and the significant runoff during the rainy season. Policymakers can consider water storage during the rainy season while developing glacier resources to alleviate water scarcity.