Staff Research Highlight - Application of Different Weighting Schemes and Stochastic Simulations to Parameterization Processes Considering Observation Error
Lee, E., Lee, H., Park, D., Hwang, H.-T., & Park, C. (2023). Application of Different Weighting Schemes and Stochastic Simulations to Parameterization Processes Considering Observation Error: Implications for Climate Change Impact Analysis of Integrated Watershed Models. In Water (Vol. 15, Issue 10, p. 1880). MDPI AG. https://doi.org/10.3390/w15101880
“The inclusion of different observation types and a detailed characterization of the study site through a highly parameterized problem approach or regularized inversion could be useful to improve the calibration performance and better characterize the spatially varying dynamics of integrated water systems.”
CLICK HERE TO READ THE ARTICLE.
We’re pleased to highlight this publication (co-authored by Aquanty’s senior scientist Hyoun-Tae Hwang) which focuses on improving the calibration of integrated watershed models by addressing uncertainties arising from observation errors. The research explores how different weighting schemes and stochastic simulations can enhance the accuracy of parameter estimation processes, ultimately reducing uncertainty in climate change impact assessments.
“Based on the geological logging database, the subsurface domain was vertically discretized into four divisions (surface soil, alluvial deposits, weathered rock zone, and basement rock), the depths of which were determined using inverse distance weighting interpolation. These vertical geological units were then divided into 10 sublayers to increase the vertical resolution of the model. The total numbers of 3D nodes and elements are 89,530 and 156,420, respectively.”
In this research highlight, the authors developed a three-dimensional integrated model of the Sabgyo watershed in South Korea using HydroGeoSphere (HGS), a powerful tool capable of simulating coupled surface water–groundwater interactions. The study applied the Parameter ESTimation tool (PEST) to calibrate the model, investigating three different weighting schemes that account for variances in observation error, alongside a stochastic simulation approach that treated observation error as a random variable. This innovative calibration framework was used to assess how groundwater and surface water interact under current and future climate conditions.
Figure 1. (a) Location of the study site; (b) monitoring stations and stream network within the watershed; and (c) 3D subsurface model and dominant hydrostratigraphy/soil types of the study site.
The research showed that incorporating observation-error-based weighting into the PEST calibration process improved model performance, particularly in parameter sensitivity and estimation accuracy. Although the differences in surface flow and groundwater level simulations across the three weighting schemes were relatively minor, the estimated parameters—such as hydraulic conductivity and evapotranspiration limits—varied significantly. Notably, the log-transformed weighting approach (Case 3) produced parameter values most consistent with stochastic simulations, indicating a more realistic calibration outcome.
HydroGeoSphere played a central role in the study, enabling the simulation of groundwater and surface water processes at high spatial and temporal resolutions. Its integration with PEST allowed for an efficient calibration process, while also accounting for uncertainties stemming from field data collection and observational limitations. HGS’s ability to model seasonal hydrological cycles and groundwater–surface water exchange fluxes under both current and future climate scenarios was key to understanding how these interactions might evolve.
“The models predicted that groundwater infiltration and seepage would play a significant role in reducing stream discharge during the rainy season and maintaining it during the dry season.”
Using the calibrated model, the authors then explored the predicted hydrological responses and groundwater–surface water interactions under different climate change scenarios. Predictions under RCP2.6 and RCP8.5 climate/emission scenarios suggested notable changes in watershed behavior, with increased evapotranspiration and variable surface discharges across decades. Importantly, groundwater was shown to buffer seasonal hydrological variability, contributing significantly to streamflow during dry periods. These results emphasize the critical role of groundwater in sustaining water systems under changing climatic conditions.
By leveraging HGS’s advanced capabilities and incorporating statistical approaches to deal with observation uncertainty, the study delivers valuable insights for improving hydrological modelling practices. It underscores the need to account for observation error in large-scale watershed models to better inform climate resilience strategies, water resource planning, and environmental decision-making.
Abstract:
We investigated the potential impact of observation error on the calibration performance of an integrated watershed model. A three-dimensional integrated model was constructed using HydroGeoSphere and applied to the Sabgyo watershed in South Korea to assess the groundwater–surface water interaction process. During the model calibration, three different weighting schemes that consider observation error variances were applied to the parameter estimation tool (PEST). The applied weighting schemes were compared with the results from stochastic models, in which observation errors from surface discharges were considered a random variable. Based on the calibrated model, the interactions between groundwater and surface water were predicted under different climate change scenarios (RCP). Comparisons of calibration performance between the different models showed that the observation-error-based weighting schemes contributed to an improvement in the model parameterization. Analysis of the exchange flux between groundwater and surface water highlighted the significance of groundwater in delaying the hydrological response of integrated water systems. Predictions based on different RCP scenarios suggested the increasing role of groundwater in watershed dynamics. We concluded that the comparison of different weighting schemes for the determination of error covariance could contribute to an improved characterization of watershed processes and reduce the model uncertainty arising from observation errors.