Assessing effects of model complexity and structure on predictions of hydrological responses using serial and parallel model design

Abstract

By utilizing functional relationships based on observations at plot or field scales, water quality models first compute surface runoff and then use it as the primary governing variable to estimate sediment and nutrient transport. When these models are applied at watershed scales, this serial model structure, coupling a surface runoff sub-model with a water quality sub-model, may be inappropriate because dominant hydrological processes differ among scales. A parallel modeling approach is proposed to evaluate how best to combine dominant hydrological processes for predicting water quality at watershed scales. In the parallel scheme, dominant variables of water quality models are identified based entirely on their statistical significance using time series analysis. Four surface runoff models of different model complexity were assessed using both the serial and parallel approaches to quantify the uncertainty on forcing variables used to predict water quality. The eight alternative model structures were tested against a 25-year high-resolution data set of streamflow, suspended sediment discharge, and phosphorous discharge at weekly time steps. Models using the parallel approach consistently performed better than serial-based models, by having less error in predictions of watershed scale streamflow, sediment and phosphorus, which suggests model structures of water quantity and quality models at watershed scales should be reformulated by incorporating the dominant variables. The implication is that hydrological models should be constructed in a way that avoids stacking one sub-model with one set of scale assumptions onto the front end of another submodel with a different set of scale assumptions. K E Y W O R D S model complexity and structure, time series model, serial and parallel modeling approaches, Watershed scale hydrological responses

A comprehensive intercomparison study between a lumped and a fully distributed hydrological model across a set of 50 catchments in the United Kingdom

“…Furthermore, the median value of the NSEs was 0.88 and 0.87 for GR4J and mHM, respectively. Despite the high and strongly correlated scores of GR4J and mHM, it must be mentioned that both the models might be transporting water through different hydrological processes (Chien & Mackay, 2020). To confirm the veracity of different hydrological states and fluxes as produced by the different models, we would need more measured data especially in the subsurface region.…”

Comparison of conceptual rainfall–runoff models in semi‐arid watersheds of eastern Algeria

“…Indeed, over‐parameterisation increases the complexity of the models, which can prevent them from reaching their potential performance levels, as shown by Perrin, Michel, and Andréassian (2001) and Willems, Mora, Vansteenkiste, Taye, and van Steenbergen (2014), and generates the equifinality of the parameters, which is an important issue in hydrological modelling, as highlighted by Beven and Binley (1992). Therefore, to choose the most efficient model, considering the effect of the number of parameters, we applied the models' selection criteria that differentiate models based on the performance of each model and the number of parameters it contains (Asl‐Rousta, Mousavi, & Ehtiat, 2017; Chien & Mackay, 2019; Symonds & Moussalli, 2011).…”

Increasing Hydroperiod in a Karst-depression Wetland Based on 165 Years of Simulated Daily Water Levels

“…Such differences-especially varying degrees of model complexity-can have major implications for model transferability either in space (i.e., extrapolation to new wetlands in other geographic locations) or in time (i.e., model forecasting to predict future hydrologic conditions or hindcasting to simulate wetland hydrology over the decades prior to the beginning of hydrologic observations; Wenger and Olden 2012;Clark et al 2017;Lute and Luce 2017). Substantial evidence indicates that model transferability benefits from parsimony, meaning that simpler models (i.e., with fewer parameters and storage compartments) tend to perform as well or better than more complex models when validated using independent data sets (Beven 1989;Jakeman and Hornberger 1993;Leplastrier et al 2002;Perrin et al 2003;Chien and Mackay 2014;Lute and Luce 2017;Chien and Mackay 2020). Whereas more complex models may better approximate localized physical processes and generally enable closer fit in calibration, favorable calibration statistics for highly complex models may represent a form of overfitting (i.e., "overparameterization"; Beven 1989;Perrin et al 2001) causing such models to perform poorly when extrapolated to novel contexts (Wenger and Olden 2012;Lute and Luce 2017;Chien and Mackay 2020).…”

“…Substantial evidence indicates that model transferability benefits from parsimony, meaning that simpler models (i.e., with fewer parameters and storage compartments) tend to perform as well or better than more complex models when validated using independent data sets (Beven 1989;Jakeman and Hornberger 1993;Leplastrier et al 2002;Perrin et al 2003;Chien and Mackay 2014;Lute and Luce 2017;Chien and Mackay 2020). Whereas more complex models may better approximate localized physical processes and generally enable closer fit in calibration, favorable calibration statistics for highly complex models may represent a form of overfitting (i.e., "overparameterization"; Beven 1989;Perrin et al 2001) causing such models to perform poorly when extrapolated to novel contexts (Wenger and Olden 2012;Lute and Luce 2017;Chien and Mackay 2020). Hydroclimatic non-stationarity (e.g., from climate change) implies that the hydroclimatic conditions of the model calibration and validation time periods (generally the recent past) may be substantially different from those of the more distant past (hindcasting) or future (forecasting), creating potential for transferability problems (Vaze et al 2010;Coron et al 2012;Coron et al 2014;Magand et al 2015;Duethmann et al 2020).…”

“…Because of data limitations, hydrologic models for depression wetlands are commonly calibrated and validated using less than a decade of hydrologic observations (e.g., Sun et al 2006;Barton et al 2008;Naughton et al 2012;Lee et al 2015;Chandler et al 2016;Qi et al 2019). Even in the more data-rich field of streamflow modeling, 25-year datasets are considered "long term" and relatively rare (Chien and Mackay 2014;Chien and Mackay 2020). In this context, calibration and validation datasets typically represent only a subset of the long-term hydroclimatic variability and thus may poorly represent the hydroclimatic extremes of droughts and pluvial periods (Perrin et al 2001;Kling et al 2015).…”