Transferability of Conceptual Hydrological Models Across Temporal Resolutions: Approach and Application

Abstract. Calibration of terrestrial ecosystem models is important but challenging. Bayesian inference implemented by Markov chain Monte Carlo (MCMC) sampling provides a comprehensive framework to estimate model parameters and associated uncertainties using their posterior distributions. The effectiveness and efficiency of the method strongly depend on the MCMC algorithm used. In this work, a differential evolution adaptive Metropolis (DREAM) algorithm is used to estimate posterior distributions of 21 parameters for the data assimilation linked ecosystem carbon (DALEC) model using 14 years of daily net ecosystem exchange data collected at the Harvard Forest Environmental Measurement Site eddy-flux tower. The calibration of DREAM results in a better model fit and predictive performance compared to the popular adaptive Metropolis (AM) scheme. Moreover, DREAM indicates that two parameters controlling autumn phenology have multiple modes in their posterior distributions while AM only identifies one mode. The application suggests that DREAM is very suitable to calibrate complex terrestrial ecosystem models, where the uncertain parameter size is usually large and existence of local optima is always a concern. In addition, this effort justifies the assumptions of the error model used in Bayesian calibration according to the residual analysis. The result indicates that a heteroscedastic, correlated, Gaussian error model is appropriate for the problem, and the consequent constructed likelihood function can alleviate the underestimation of parameter uncertainty that is usually caused by using uncorrelated error models.

Lake monitoring is essential for sustaining aquatic ecosystems, and accurate estimation/prediction of water quality parameters is crucial to this effort. Despite its importance, the performance of predictive models built on varying temporal resolutions remains underexplored systematically. This study used daily and 4 h high temporal resolution (HTR) datasets to assess the performance of multiple machine learning models—namely, Support Vector Regression (SVR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM) networks—under consistent data scales. The results indicate that dissolved oxygen (DO) exhibits pronounced sensitivity to temporal resolution, while total nitrogen (TN), total phosphorus (TP), and ammonia nitrogen (NH3-N) show distinct, parameter-specific response patterns that align with the temporal characteristics of their underlying biogeochemical processes. This research helps to deepen the understanding of how temporal data resolution influences model performance in water quality prediction, offering valuable insights for selecting optimal data resolutions and modeling techniques to enhance lake monitoring and protection strategies.

The current study evaluates the impacts of various sources of uncertainty involved in hydrologic modeling on parameter behavior and regionalization utilizing different Bayesian likelihood functions and the Differential Evolution Adaptive Metropolis (DREAM) algorithm. The developed likelihood functions differ in their underlying assumptions and treatment of error sources. We apply the developed method to a snow accumulation and ablation model (National Weather Service SNOW17) and generate parameter ensembles to predict snow water equivalent (SWE). Observational data include precipitation and air temperature forcing along with SWE measurements from 24 sites with diverse hydroclimatic characteristics. A multiple linear regression model is used to construct regionalization relationships between model parameters and site characteristics. Results indicate that model structural uncertainty has the largest influence on SNOW17 parameter behavior. Precipitation uncertainty is the second largest source of uncertainty, showing greater impact at wetter sites. Measurement uncertainty in SWE tends to have little impact on the final model parameters and resulting SWE predictions. Considering all sources of uncertainty, parameters related to air temperature and snowfall fraction exhibit the strongest correlations to site characteristics. Parameters related to the length of the melting period also show high correlation to site characteristics. Finally, model structural uncertainty and precipitation uncertainty dramatically alter parameter regionalization relationships in comparison to cases where only uncertainty in model parameters or output measurements is considered. Our results demonstrate that accurate treatment of forcing, parameter, model structural, and calibration data errors is critical for deriving robust regionalization relationships.

Performance evaluation of hydrological models using ensemble of General Circulation Models in the northeastern China

In the distributed hydrological simulations for complex mountain areas, large amounts of meteorological input parameters with high spatial and temporal resolutions are necessary. However, the extreme scarcity and uneven distribution of the traditional meteorological observation stations in cold and arid regions of Northwest China makes it very difficult in meeting the requirements of hydrological simulations. Alternatively, regional climate models (RCMs), which can provide a variety of distributed meteorological data with high temporal and spatial resolution, have become an effective solution to improve hydrological simulation accuracy and to further study water resource responses to human activities and global climate change. In this study, abundant and evenly distributed virtual weather stations in the upper reaches of the Heihe River Basin (HRB) of Northwest China were built for the optimization of the input data, and thus a regional integrated environmental model system (RIEMS) based on RCM and a distributed hydrological model of soil and water assessment tool (SWAT) were integrated as a coupled climate–hydrological RIEMS-SWAT model, which was applied to simulate monthly runoff from 1995 to 2010 in the region. Results show that the simulated and observed values are close; Nash–Sutcliffe efficiency is higher than 0.65; determination coefficient (R2) values are higher than 0.70; percent bias is controlled within ±20%; and root-mean-square-error-observation standard deviation ratio is less than 0.65. These results indicate that the coupled model can present basin hydrological processes properly, and provide scientific support for prediction and management of basin water resources.

The investigation of methods to evaluate the performance of hydrological models has never stopped. This study explores an innovative application, called joint multifractal spectra (JMS) analysis. The JMS analysis, which is demonstrated in three typical basins in Zhejiang Province, China, analyses both flood and low-flow periods. JMS is applied to seasonal-decomposed observed and simulated runoff series, thus helping to analyse the rising and the recession processes of discharge, which are unique features of a hydrograph. This is achieved by using q exponents to magnify or shrink the relative magnitudes of different parts. The joint multifractal spectra revealed many details of modelling, like the evaluation of recession process modelling. Given its versatility, the JMS analysis has great potential for its applications in hydrological modelling.

The assessment of climate change impacts on water resources involves several methodological decisions, including choices of global climate models (GCMs), emission scenarios, downscaling techniques, and hydrologic modeling approaches. Among these, hydrologic model structure selection and parameter calibration are particularly relevant and usually have a strong subjective component. The goal of this research is to improve understanding of the role of these decisions on the assessment of the effects of climate change on hydrologic processes. The study is conducted in three basins located in the Colorado headwaters region, using four different hydrologic model structures [PRMS, VIC, Noah LSM, and Noah LSM with multiparameterization options (Noah-MP)]. To better understand the role of parameter estimation, model performance and projected hydrologic changes (i.e., changes in the hydrology obtained from hydrologic models due to climate change) are compared before and after calibration with the University of Arizona shuffled complex evolution (SCE-UA) algorithm. Hydrologic changes are examined via a climate change scenario where the Community Climate System Model (CCSM) change signal is used to perturb the boundary conditions of the Weather Research and Forecasting (WRF) Model configured at 4-km resolution. Substantial intermodel differences (i.e., discrepancies between hydrologic models) in the portrayal of climate change impacts on water resources are demonstrated. Specifically, intermodel differences are larger than the mean signal from the CCSM–WRF climate scenario examined, even after the calibration process. Importantly, traditional single-objective calibration techniques aimed to reduce errors in runoff simulations do not necessarily improve intermodel agreement (i.e., same outputs from different hydrologic models) in projected changes of some hydrological processes such as evapotranspiration or snowpack.

Understanding hydrological processes at catchment scale through the use of hydrological model parameters is essential for enhancing water resource management. Given the difficulty of using lump parameters to calibrate distributed catchment hydrological models in spatially heterogeneous catchments, a multiple calibration technique was adopted to enhance model calibration in this study. Different calibration techniques were used to calibrate the Soil and Water Assessment Tool (SWAT) model at different locations along the Logone river channel. These were: single-site calibration (SSC); sequential calibration (SC); and simultaneous multi-site calibration (SMSC). Results indicate that it is possible to reveal differences in hydrological behavior between the upstream and downstream parts of the catchment using different parameter values. Using all calibration techniques, model performance indicators were mostly above the minimum threshold of 0.60 and 0.65 for Nash Sutcliff Efficiency (NSE) and coefficient of determination (R2) respectively, at both daily and monthly time-steps. Model uncertainty analysis showed that more than 60% of observed streamflow values were bracketed within the 95% prediction uncertainty (95PPU) band after calibration and validation. Furthermore, results indicated that the SC technique out-performed the other two methods (SSC and SMSC). It was also observed that although the SMSC technique uses streamflow data from all gauging stations during calibration and validation, thereby taking into account the catchment spatial variability, the choice of each calibration method will depend on the application and spatial scale of implementation of the modelling results in the catchment.

Lack of observed data for model calibration hinders the application of hydrological models in many poorly-gauged catchments, particularly in the humid tropical region. Despite much less attention given, it is vital to assess transferability of calibrated parameters in order to apply hydrological models in such catchments to assist their water resources planning and management activities. Thus, this study investigated temporal transferability of a lumped conceptual hydrological model’s (MIKE 11 NAM) calibrated model parameters for rainfall-runoff simulations in two different time periods beyond its calibration period. Study area was selected as Gin catchment located in the humid tropical region. MIKE 11 NAM model was calibrated for the period 1995-1998 [Nash Sutcliffe coefficient (NSE) = 0.73, percent bias (PBIAS) = 3.9%, ratio of the root mean square error to the standard deviation of measured data (RSR) = 0.52] and validated for the period 1999-2002 (NSE=0.66, PBIAS = 8.7%, RSR = 0.59) for the Gin catchment. The temporal transferability of the calibrated model parameters was tested using two scenarios which formulated based on the temporal lag between the calibration period and the transfer period: scenario A having a 4-year time lag and scenario B having a 8-year time lag. Scenario A which evaluated the model performance using 2003-2006 streamflow data indicated only a marginal loss in the model performance in comparison to the calibration. It showed an overall ‘good’ performance (NSE=0.64, PBIAS = 8.6%, RSR = 0.59) including promising capability to reproduce the peak flows (<10th percentile) with Pearson’s correlation coefficient of 0.6. However, scenario B which evaluated the model performance using 2007-2010 streamflow data indicated ‘unsatisfactory’ model performance (NSE=0.42, PBIAS = 13.6%, RSR = 0.76). Therefore, this study suggests that the calibrated parameters of MIKE 11 NAM model can be temporally transferred within a catchment with a 4-year time lag from the calibration period implying the applicability of this modelling framework for rainfall-runoff simulations especially in the catchments where streamflow data is sparse.

Flow prediction in ungauged catchments is a major unresolved challenge in scientific and engineering hydrology. This study attacks the prediction in ungauged catchment problem by exploiting advances in flow index selection and regionalization in Bayesian inference and by developing new statistical tests of model performance in ungauged catchments. First, an extensive set of available flow indices is reduced using principal component (PC) analysis to a compact orthogonal set of “flow index PCs.” These flow index PCs are regionalized under minimal assumptions using random forests regression augmented with a residual error model and used to condition hydrological model parameters using a Bayesian scheme. Second, “adequacy” tests are proposed to evaluate a priori the hydrological and regionalization model performance in the space of flow index PCs. The proposed regionalization approach is applied to 92 northern Spain catchments, with 16 catchments treated as ungauged. It is shown that (1) a small number of PCs capture approximately 87% of variability in the flow indices and (2) adequacy tests with respect to regionalized information are indicative of (but do not guarantee) the ability of a hydrological model to predict flow time series and are hence proposed as a prerequisite for flow prediction in ungauged catchments. The adequacy tests identify the regionalization of flow index PCs as adequate in 12 of 16 catchments but the hydrological model as adequate in only 1 of 16 catchments. Hence, a focus on improving hydrological model structure and input data (the effects of which are not disaggregated in this work) is recommended.

Investigating strategies to improve hydrologic model performance in a changing climate

Multiple spatio-temporal scale runoff forecasting and driving mechanism exploration by K-means optimized XGBoost and SHAP

<p>Daily streamflow data are crucial for various scientific and operational water resources applications, such as climate change impact assessment, flood forecasting, and catchment classification, among others. Streamflow is typically estimated through the implementation of hydrological models, which rely on parameters to represent hypotheses about the dominant processes in a catchment. In most cases, these parameters cannot be measured at the scales relevant for model applications and are therefore estimated through model calibration. Because most streams worldwide remain ungauged, novel parameter regionalisation techniques have been developed to predict daily streamflow over ungauged catchments. These regionalisation techniques transfer calibrated model parameters from gauged to ungauged catchments. To this end, an accurate spatio-temporal representation of crucial meteorological variables such as precipitation is essential, and therefore, most regionalisation studies have been conducted over regions with a dense network of meteorological stations. However, the characterisation of precipitation over data-scarce areas is challenging and might be subject to large uncertainties when only ground-based measurements are used. Despite that few daily regionalisation studies have used gridded precipitation products, there is no precise evaluation on how the selection of a particular precipitation product can affect the performance of the existing regionalisation techniques. Therefore, this work aims to analyse how the choice of gridded daily precipitation products affects the relative performance of three well-known parameter regionalisation techniques (spatial proximity, feature similarity, and parameter regression) over 100 near-natural catchments with diverse hydrological regimes across Chile. For this purpose, we calibrated a conceptual semi-distributed HBV-like hydrological model (TUWmodel) for each catchment, using four precipitation products (CR2MET, RF-MEP, ERA5, and MSWEPv2.8). We assessed the ability of these regionalisation techniques to transfer the parameters of a rainfall-runoff model, implementing a leave-one-out cross-validation procedure for each precipitation product. Despite differences in the spatio-temporal distribution of precipitation, all products provided good performance during calibration (median KGE's > 0.77), two independent verification periods (median KGE's > 0.70 and 0.61, for near normal and dry conditions, respectively), and regionalisation (median KGE's for the best method ranging from 0.56 to 0.63). We show how model calibration can compensate, to some extent, differences between precipitation forcings by adjusting model parameters and thus the water balance components. Feature similarity provided the best results, followed by spatial proximity, while parameter regression resulted in the worst performance, reinforcing the importance of transferring complete model parameter sets to ungauged catchments. Our results suggest that: i) merging precipitation products and ground-based measurements does not necessarily translate into an improved hydrological model performance; ii) a precipitation product that provides the best individual model performance during calibration and verification does not necessarily yield the best performance in terms of parameter regionalisation; iii) the spatial resolution of the precipitation products does not substantially affect the regionalisation performance; and iv) the model parameters and the performance of regionalisation methods are affected by the hydrological regime, with the best results for spatial proximity and feature similarity obtained for rain-dominated catchments with a minor snowmelt component.</p>

Integrated Bayesian Multi-model approach to quantify input, parameter and conceptual model structure uncertainty in groundwater modeling

Evaluating the role of evapotranspiration remote sensing data in improving hydrological modeling predictability