Effects of single‐ and multi‐site calibration strategies on hydrological model performance and parameter sensitivity of large‐scale semi‐arid and semi‐humid watersheds

Abstract

In large‐scale watersheds, only the calibration of parameters for the outlet may ignore spatial heterogeneity and fail to meet the simulation accuracy of local areas. Based on the runoff observation data of three hydrological stations (Qingyang, Yangjiaping and Zhangjiashan) in the Jinghe River Watershed, single‐ and multi‐site calibration strategies were designed to investigate the performance and applicability of the Soil and Water Assessment Tool (SWAT) in predicting the monthly runoff of nested catchments and evaluate the differences in parameter sensitivity between the upstream and downstream. (i) The two calibration strategies showed little difference in the simulation accuracy of the watershed outlet; the Nash–Sutcliffe efficiency (NSE) and determination coefficient (R2) were approximately 0.8 and 0.6, respectively, and the percentage of bias (PBIAS) was mostly <15%. The single‐site one‐by‐one calibration was significantly better than the multi‐site joint calibration at the interior points of the watershed. Among them, Qingyang decreased most obviously in the joint calibration with Zhangjiashan, with both NSE and R2 < 0.5, whereas Yangjiaping was >0.5 in combination with the other two sites. (ii) The sensitive parameters obtained by the two calibration strategies were different, and different parameters were effective for different‐sized catchments because hydrological processes exhibit astounding variability at all scales. The single‐site one‐by‐one calibration was mainly affected by the hydrological processes of the local catchment, whereas the multi‐site joint calibration was largely constrained by the sensitive parameters of the single‐site calibration. (iii) Parameters at different scales and locations were related to the problem of scaling and regionalization. The multi‐site joint calibration did not significantly improve the model performance of the watershed outlet but increased the capacity for synchronous identification of local sensitive parameters when compared to single‐site calibration. These findings are expected to provide experience for parameter calibration in other similar large‐scale watersheds and to lay a good foundation for the accurate identification of other related hydrological behaviours at watershed scales.

Towards a data-effective calibration of a fully distributed catchment water quality model

“…In contrast, several other studies have found that performance was largely equivalent for multi-site and single-site calibration techniques (e.g., Franco et al, 2020;Lerat et al, 2012;Wu et al, 2022a).…”

SWAT Model Performance Using Spatially Distributed Saturated Hydraulic Conductivity (Ksat) and Varying-Resolution DEMs

“…A standard algorithm in SWAT-CUP called Sequential Uncertainty Fitting 2 (SUFI-2) was applied for multi-site model calibration and validation for monthly time steps [52,53] (Figure 3). SUFI-2 performs sensitivity analysis via global sensitivity analysis to test the sensitivity of model parameters, which is indicated by the t-stat and p-value [28,[54][55][56]. The parameter has greater sensitivity if it has a higher absolute t-stat value [28].…”

Modelling Floodplain Vegetation Response to Climate Change, Using the Soil and Water Assessment Tool (SWAT) Model Simulated LAI, Applying Different GCM’s Future Climate Data and MODIS LAI Data

“…The R 2 value ranges from 0 to 1, where 0 indicates that the regression model explains none of the variance in the dependent variable, meaning it does not fit the data at all. A water balance model is deemed successful when it achieves a higher coefficient of determination R 2 , indicating a more favorable outcome [68]. R 2 is defined as follows [68,69]:…”

“…A water balance model is deemed successful when it achieves a higher coefficient of determination R 2 , indicating a more favorable outcome [68]. R 2 is defined as follows [68,69]:…”