Catchment responses to plausible parameters and input data under equifinality in distributed rainfall‐runoff modeling

Abstract

Abstract:In a hydrological simulation using a complex model, equifinality is likely to occur in many different ways, due to various sources associated with the model structure, parameter as well as data uncertainties involved in modeling process. This study aims to demonstrate the equifinality problem in streamflow prediction with a distributed rainfall-runoff model and also investigate the effect of parameter and input uncertainties on an internal catchment response in time and space under the equifinality condition. For this objective, several experiments were carried out step by step as follows. First, we estimated plausible parameter sets, which lead to similarly good objective function values, using the shuffled complex evolution metropolis (SCEM-UA) algorithm and also generated plausible input scenarios with different spatial patterns from radar rainfall field, which showed minimum runoff errors. Then, we adopted a computational tracer method based on a time-space accounting scheme, combined with a distributed model for analyzing how a catchment behaves under a given equifinality condition by parameter and input uncertainties. It was found that the global responses of a catchment to the predefined equifinality factors were very similar; all simulated hydrographs were almost identical in spite of the different model parameter values and different spatial patterns of rainfall data. On the other hand, the internal catchment responses, tracked by the computational tracer method, were sensitive to plausible scenarios. The results of the internal catchment behavior show that the distributed rainfall-runoff model has multiple alternative flow pathways, providing identical hydrographs in time and space when the rainfall over the catchment transforms into runoff. Our approach can be used to understand catchment responses to various uncertainty sources by visualizing the spatiotemporal runoff generation process at the watershed scale.

Improving runoff prediction using agronomical information in a cropped, loess covered catchment

“…Considering the complexity of the interactions among physical processes, performing the validation at the outlet can mask important spatial variation in the catchment (Lee, Tachikawa, Sayama, & Takara, ). Therefore, predicted spatial runoff patterns and connectivity were compared with information collected with internal detectors (P1 and P2, Figure ) and from direct observations on runoff traces.…”

“…Considering the complexity of the interactions among physical processes, performing the validation at the outlet can mask important FIGURE 2 Rainfall intensity (mm mn-1), observed discharge (L s-1) together with predicted discharge according to the calibration strategies: event per event and temporal constraint strategy (EES and TCS) for the nine runoff events within the catchment (measured outflow for May 21 was hydraulically influenced by a sluice gate) spatial variation in the catchment (Lee, Tachikawa, Sayama, & Takara, 2012). Therefore, predicted spatial runoff patterns and connectivity were compared with information collected with internal detectors (P1 and P2, Figure 1) and from direct observations on runoff traces.…”

Comparison of the rise of water level in the typical catchments, Three Gorges Reservoir area

“…However, in some studies and water management practices, a linear relationship is used to convert hydrologic indicators among catchments under different areas (Smajstrla 1990) to obtain data for ungauged catchments. In particular, distributed rainfall-runoff models have been widely used (Lee et al 2012) with hydrologic response units (HRUs). Therefore, it is necessary to ask what is the threshold of catchment size in which the hydrologic processes of a catchment have only qualitative changes in a certain area and with a certain method?…”

Overcoming equifinality: time-varying analysis of sensitivity and identifiability of SWAT runoff and sediment parameters in an arid and semiarid watershed

“…The SWAT model has a complex structure and many parameters, and most of these parameters need to be calibrated with the help of measured data (Li et al 2021). Theoretically, parameter calibration mainly depends on the objective function, constraints, and measured data, but there is likely to be "equifinality for different parameters" due to uncertainty (Lee et al 2012). To avoid equifinality and overparameterization of the distributed hydrological modeling, accurately identifying sensitive parameters is necessary to ensure the reliability of the SWAT model application (Zadeh et al 2015).…”