Application of error analysis to a Marsh Hydrology Model

Abstract

The contribution of errors on individual parameters to uncertainties in model predictions is studied in a hydrologic model of a marsh. Daily values of water level and storage are calculated in the model as a function of precipitation, storm runoff from the urban watershed, physical properties of soil, groundwater movement, and daily values of evapotranspiration. Errors associated with each parameter were propagated through the model calculations by Monte Carlo simulation. The distribution of each hydrologic parameter, considered as a random variable, was specified by mean, variance, maximum, and minimum values. Random numbers drawn from these distributions were used to simulate 244 days of water level and storage in the marsh. Errors for the marsh model are bounded, returning toward zero rather than growing monotonically through time. Although the simulations show considerable error for some days, the overall behavior of the model is robust. The seasonal contribution of specific parameters has important implications for model calibration and validation. Our analysis indicates that Monte Carlo analysis using actual estimates of parameter error leads to different conclusions than sensitivity analysis.