A Multi-Gage Calibration Approach for Modeling a Semiarid Santa Cruz watershed in Arizona-Mexico Border

Flow at the outlet does not sufficiently represent underlying processes for an entire watershed in semiarid climates, where the flow from much of the catchment may not reach the outlet most of the time. For modeling these watersheds, calibration to discharge is required not only at the watershed outlet but also in various locations within the watershed to best capture real watershed conditions. The objective of this study is to develop a reliable model for the Santa Cruz watershed to analyze the impact of land use and climate change on hydrology and water quality in this semiarid basin. The semiarid Santa Cruz watershed (Lat:34^oN and Long:-111^oW) is 9,000 km² and located in southern Arizona, United States (85%) and northern Sonora, Mexico (15%). In this study, the Soil and Water Assessment Tool (SWAT) was used to calibrate the flow at seven monitoring stations to increase the reliability of flow and therefore water quality predictions for the entire basin. The model performance was evaluated based on observed hydrograph, Mass balance error (MBE) coefficient of determination (R²) and Nash-Sutcliffe efficiency (NSE). We began the calibration by identifying the most sensitive parameters using the Latin hypercube global sensitivity approach, identifying that the baseflow recession coefficient, the effective hydraulic conductivity in main channel (transmission loss), the curve number and the soil evaporation compensation factor as the most sensitive parameters. Next, we manually calibrated the model at the USGS gage near the outlet by modifying these parameters. Although the model was validated well at the outlet, the model was not able to validate the flow at other USGS gages inside the watershed. Thus we calibrated the model in sequential order from upstream to downstream gages by further adjusting these parameters. A significant amount of water is lost through transmission loss in our modeling of the Santa Cruz. In SWAT, by default the loss from the streambed is set to zero and so needs to be calibrated to simulate transmission losses. The model calibrated well for all 7 stream gages based on model evaluation parameters. The model was validated for 5 stations but suffered from uncertainty of precipitation data for the other 2 stations, the model predicted high flow when there was low/no flow in the observed data and vice versa. Sufficient precipitation gages are required to reduce the effect of uncertainty in precipitation data and to improve model calibration and validation in semiarid watersheds. While the model calibrated only at the outlet was not able to validate the flow at other monitoring gages inside the watershed, further fine calibration using data from those gaging stations much improved the validation. This study thus emphasizes the importance of multi-gage calibration.