Time Domain Transformations to Improve Hydrologic Model Consistency: Parameterization in Flow-Corrected Time

Abstract:

Streamflow modeling is highly complex. Beyond the identification and mapping of dominant runoff processes to mathematical models, additional challenges are posed by the switching of dominant streamflow generation mechanisms temporally and dynamic catchment responses to precipitation inputs based on antecedent conditions. As a result, model calibration is required to obtain parameter values that produce acceptable simulations of the streamflow hydrograph. Typical calibration approaches assign equal weight to all observations to determine the best fit over the simulation period. However, the objective function can be biased toward (i.e., implicitly weight) certain parts of the hydrograph (e.g., high streamflows). Data transformations (e.g., logarithmic or square root) scale the magnitude of the observations and are commonly used in the calibration process to reduce implicit weighting or better represent assumptions about the model residuals. Here, we consider a time domain data transformation rather than the more common data domain approaches. Flow-corrected time was previously employed in the transit time modeling literature. Conceptually, it stretches time during high streamflow and compresses time during low streamflow periods. Therefore, streamflow is dynamically weighted in the time domain, with greater weight assigned to periods with larger hydrologic flux. Here, we explore the utility of the flow-corrected time transformation in improving model performance of the Catchment Connectivity Model. Model process fidelity was assessed directly using shallow groundwater connectivity data collected at Tenderfoot Creek Experimental Forest. Our analysis highlights the impact of data transformations on model consistency and parameter sensitivity.