Effect of uncertainty of discharge data on uncertainty of discharge simulation for the Lake Michigan Diversion, northeastern Illinois and northwestern Indiana

First posted November 10, 2022 For additional information, contact: Director, Central Midwest Water Science Center U.S. Geological Survey405 North Goodwin Urbana, IL 61801Contact Pubs Warehouse Simulation models of watershed hydrology (also referred to as “rainfall-runoff models”) are calibrated the best available streamflow data, which typically published discharge time series at outlet watershed. Even after calibration, model generally cannot replicate discharges because simplifications physical system embedded in structure and uncertainties input data estimated parameters, which, although optimized for given calibration remain uncertain. The errors caused by forcing such precipitation other climatological used calibration. In numerical algorithms often assumed be without error, but they themselves uncertain, having been computed using ratings, fitted uncertain measurements.In this study, uncertainty daily how is transmitted parameter values Hydrological Program–FORTRAN (HSPF) rainfall-runoff simulated both prediction locations were investigated Lake Michigan diversion northeastern Illinois northwestern Indiana. HSPF study Army Corps Engineers part quantifying water from State Illinois. jointly two watersheds area; resulting considered base study. Seven gaged area testing predictive simulations. A Bayesian rating curve estimation (BaRatin) approach, BaRatin stage-period-discharge (SPD) method, was estimate watersheds. To characterize effect on values, parameters recalibrated 17 nonrandomly selected pairs SPD analysis. provide an indicator compare uncertainty, 1,000 sets also randomly generated covariance matrix model. random then simulations seven Selected summary statistics—the period-of-study (POS, years 1997 2015) mean discharge, flow-duration (FDC) quantiles, year discharges—are variability between discharge.A normalized index (VN) a measure flow statistics arising sources When least 1, large enough explain median error offsetting likely. appreciably less than clearly insufficient values. At results simulation together, VN ranged 0.2 0.8 POS 0.3 0.6 set FDC 0.1 discharges. These indicate that substantial remains unexplained. though highly constrained it applied simultaneously subject regularization adjustment their initial constraints avoid overfitting thus increase likelihood would give accurate not created transfer shown balancing Additional remaining include structural meteorological degree unable adjust account these errors. watersheds, corresponding almost always substantially lower those This result expected have including error.The work described report provides preliminary estimates limited range predicted uncertainty. Future beneficial obtain better statistical characterization address (model)