The role of prior model calibration on predictions with ensemble Kalman filter

This paper, based on a real world case study (Limmat aquifer, Switzerland), compares inverse groundwater flow models calibrated with specified numbers of monitoring head locations. These models are updated in real time with the ensemble Kalman filter (EnKF) and the prediction improvement is assessed in relation to the amount of monitoring locations used for calibration and updating. The prediction errors of the models calibrated in transient state are smaller if the amount of monitoring locations used for the calibration is larger. For highly dynamic groundwater flow systems a transient calibration is recommended as a model calibrated in steady state can lead to worse results than a noncalibrated model with a well-chosen uniform conductivity. The model predictions can be improved further with the assimilation of new measurement data from on-line sensors with the EnKF. Within all the studied models the reduction of 1-day hydraulic head prediction error (in terms of mean absolute error [MAE]) with EnKF lies between 31% (assimilation of head data from 5 locations) and 72% (assimilation of head data from 85 locations). The largest prediction improvements are expected for models that were calibrated with only a limited amount of historical information. It is worthwhile to update the model even with few monitoring locations as it seems that the error reduction with EnKF decreases exponentially with the amount of monitoring locations used. These results prove the feasibility of data assimilation with EnKF also for a real world case and show that improved predictions of groundwater levels can be obtained.     

Determination of specific yield for the Biscayne Aquifer with a canal-drawdown test

Data from a large-scale canal-drawdown test were used to estimate the specific yield (sy) of the Biscayne Aquifer, an unconfined limestone aquifer in southeast Florida. The drawdown test involved dropping the water level in a canal by about 30 cm and monitoring the response of hydraulic head in the surrounding aquifer. Specific yield was determined by analyzing data from the unsteady portion of the drawdown test using an analytical stream-aquifer interaction model (Zlotnik and Huang 1999). Specific yield values computed from drawdown at individual piezometers ranged from 0.050 to 0.57, most likely indicating heterogeneity of specific yield within the aquifer (small-scale variation in hydraulic conductivity may also have contributed to the differences in sy among piezometers). A value of 0.15 (our best estimate) was computed based on all drawdown data from all piezometers. We incorporated our best estimate of specific yield into a large-scale two-dimensional numerical MODFLOW-based ground water flow model and made predictions of head during a 183-day period at four wells located 337 to 2546 m from the canal. We found good agreement between observed and predicted heads, indicating our estimate of specific yield is representative of the large portion of the Biscayne Aquifer studied here. This work represents a practical and novel approach to the determination of a key hydrogeological parameter (the storage parameter needed for simulation and calculation of transient unconfined ground water flow), at a large spatial scale (a common scale for water resource modeling), for a highly transmissive limestone aquifer (in which execution of a traditional pump test would be impractical and would likely yield ambiguous results). Accurate estimates of specific yield and other hydrogeological parameters are critical for management of water supply, Everglades environmental restoration, flood control, and other issues related to the ground water hydrology of the Biscayne Aquifer.     

SUR LA PUISSANCE DES CRUES AU JAPON A MADAGASCAR ET A LA REUNION

Abstract

Hydrological modelling has faced the problem of ungauged basins for many years: how does one estimate hydrological characteristics for a river for which there are no data? Whatever the kind of model, it needs at least hydroclimatic input data and discharge data for calibration. However, the Yates model does not need any discharge data for calibration: it is a pre-calibrated model from a vegetation—climate classification map. In the specific context of West and Central Africa, where data are often of poor quality and very scarce, it is interesting to compare the performance of such a model with those of calibrated models, and with observed data. For this study, a platform including different semi-global rainfall—runoff models which allow the estimation of monthly runoff at a spatial resolution of 0.5° × 0.5° was used. The performance of the Yates model is very close to those of calibrated models, so that one can say that this simple model, based simply on a vegetation—climate classification, can be a very useful prediction tool in regions of scarce and unreliable data, such as those of interest to the International Association of Hydrological Sciences (IAHS) initiative on prediction in ungauged basins (PUB). Therefore, this model was applied to a period covering the last 30 years, and to a data set covering the first decades of the 21st century, from a climatic scenario of doubling the CO₂ concentration in the atmosphere. The results show that, in West Africa, where drought conditions have now prevailed for 35 years, water resources should still be decreasing in the future, following the general decreasing trend of rainfall projected by the climatic scenarios.     

Flow depletion in a small stream caused by ground water abstraction from wells

The interaction between a gaining stream and a water-table aquifer is studied at an outwash plain. The aquifer is hydraulically well connected to the stream. Pumping tests were carried out in 1997 and 1998 in two wells 60 m from the stream, screening different depths of the aquifer. Drawdown was measured on both sides of the stream. Hydraulic head, drawdown, and stream depletion data were analyzed using numerical flow models. Similar models were fitted to each of two different data sets: Model A was fitted to steady-state hydraulic head and streamflow gain data not influenced by pumping; and model B was fitted to drawdown data measured during the 1998 pumping test. Each calibrated model closely fits its calibration data; however, predictions were biased if model A was used to predict the calibration data of model B, and vice versa. To further test the models, they were used to predict streamflow depletion during the two pumping tests as well as the drawdown during the 1997 test. Neither of these data were used for calibration. Model A predicted the measured depletions fairly accurately during both tests, whereas the predicted drawdowns in 1997 were significantly larger than actually measured. Contrary to this, the 1997 drawdowns predicted by model B were nearly unbiased; the predicted depletions deviate significantly from the measured depletions in 1997, but they compare well with the observations in 1998. Thus, although field work and analyses were extensive and done carefully to develop a ground water flow model that could predict both drawdown and streamflow depletion, the model predictions are biased. Analyses indicate that the deviations between model and data may be because of error in the models' representations of either the release of water from storage or of the hydrology in the riparian zone.     

From models to performance assessment: the conceptualization problem

Today, models are ubiquitous tools for ground water analyses. The intent of this paper is to explore philosophically the role of the conceptual model in analysis. Selection of the appropriate conceptual model is an a priori decision by the analyst. Calibration is an integral part of the modeling process. Unfortunately a wrong or incomplete conceptual model can often be adequately calibrated; good calibration of a model does not ensure a correct conceptual model. Petroleum engineers have another term for calibration; they refer to it as history matching. A caveat to the idea of history matching is that we can make a prediction with some confidence equal to the period of the history match. In other words, if we have matched a 10-year history, we can predict for 10 years with reasonable confidence; beyond 10 years the confidence in the prediction diminishes rapidly. The same rule of thumb applies to ground water model analyses. Nuclear waste disposal poses a difficult problem because the time horizon, 1000 years or longer, is well beyond the possibility of the history match (or period of calibration) in the traditional analysis. Nonetheless, numerical models appear to be the tool of choice for analyzing the safety of waste facilities. Models have a well-recognized inherent uncertainty. Performance assessment, the technique for assessing the safety of nuclear waste facilities, involves an ensemble of cascading models. Performance assessment with its ensemble of models multiplies the inherent uncertainty of the single model. The closer we can approach the idea of a long history with which to match the models, even models of nuclear waste facilities, the more confidence we will have in the analysis (and the models, including performance assessment). This thesis argues for prolonged periods of observation (perhaps as long as 300 to 1000 years) before a nuclear waste facility is finally closed.     

Simulation of submarine ground water discharge to a marine estuary: Biscayne Bay, Florida

Variable density ground water flow models are rarely used to estimate submarine ground water discharge because of limitations in computer speed, data availability, and availability of a simulation tool that can minimize numerical dispersion. This paper presents an application of the SEAWAT code, which is a combined version of MODFLOW and MT3D, to estimate rates of submarine ground water discharge to a coastal marine estuary. Discharge rates were estimated for Biscayne Bay, Florida, for the period from January 1989 to September 1998 using a three-dimensional, variable density ground water flow and transport model. Hydrologic stresses in the 10-layer model include recharge, evapotranspiration, ground water withdrawals from municipal wellfields, interactions with surface water (canals in urban areas and wetlands in the Everglades), boundary fluxes, and submarine ground water discharge to Biscayne Bay. The model was calibrated by matching ground water levels in monitoring wells, baseflow to canals, and the position of the 1995 salt water intrusion line. Results suggest that fresh submarine ground water discharge to Biscayne Bay may have exceeded surface water discharge during the 1989, 1990, and 1991 dry seasons, but the average discharge for the entire simulation period was only approximately 10% of the surface water discharge to the bay. Results from the model also suggest that tidal canals intercept fresh ground water that might otherwise have discharged directly to Biscayne Bay. This application demonstrates that regional scale variable density models are potentially useful tools for estimating rates of submarine ground water discharge.     

Improving the calibration strategy of the physically-based model WaSiM-ETH using critical events

Abstract

The use of a physically-based hydrological model for streamflow forecasting is limited by the complexity in the model structure and the data requirements for model calibration. The calibration of such models is a difficult task, and running a complex model for a single simulation can take up to several days, depending on the simulation period and model complexity. The information contained in a time series is not uniformly distributed. Therefore, if we can find the critical events that are important for identification of model parameters, we can facilitate the calibration process. The aim of this study is to test the applicability of the Identification of Critical Events (ICE) algorithm for physically-based models and to test whether ICE algorithm-based calibration depends on any optimization algorithm. The ICE algorithm, which uses the data depth function, was used herein to identify the critical events from a time series. Low depth in multivariate data is an unusual combination and this concept was used to identify the critical events on which the model was then calibrated. The concept is demonstrated by applying the physically-based hydrological model WaSiM-ETH on the Rems catchment, Germany. The model was calibrated on the whole available data, and on critical events selected by the ICE algorithm. In both calibration cases, three different optimization algorithms, shuffled complex evolution (SCE-UA), parameter estimation (PEST) and robust parameter estimation (ROPE), were used. It was found that, for all the optimization algorithms, calibration using only critical events gave very similar performance to that using the whole time series. Hence, the ICE algorithm-based calibration is suitable for physically-based models; it does not depend much on the kind of optimization algorithm. These findings may be useful for calibrating physically-based models on much fewer data.

Editor D. Koutsoyiannis; Associate editor A. Montanari

Citation Singh, S.K., Liang, J.Y., and Bárdossy, A., 2012. Improving calibration strategy of physically-based model WaSiM-ETH using critical events. Hydrological Sciences Journal, 57 (8), 1487–1505.     

Assessment of a seasonal calibration technique using multiple objectives in rainfall–runoff analysis

The hydrological response characteristics for the catchments in the Republic of Korea are related to a strong seasonality in the rainfall and streamflow distributions with distinct wet and dry seasons. This study aims to improve a model's ability to predict streamflows by minimizing information loss from the available data during the calibration processes. This study assesses calibration techniques incorporating a multi‐objective approach and seasonal calibration. The lumped conceptual rainfall–runoff model IHACRES was applied to selected catchments in Korea. The model was calibrated based on three different methods: the classical approach using a single performance statistic (the single‐objective method), the multi‐objective approach (the multi‐objective method (I)) and the combined approach incorporating multi‐objective and seasonal calibrations (the multi‐objective method (II)). In the multi‐objective approach, the ‘best fit’ models in the calibration period were selected by considering the trade‐offs among multiple statistics. During seasonal calibration, the calibration period was divided into four seasons to investigate whether these calibrated models can improve the model performance with regards to seasonal climate, rainfall and streamflow distributions. The adequacy of the three different calibration methods was assessed through comparison of the variability of model performance in high and low flows and water balance for the entire period and for each seasonal period. The multi‐objective methods yielded more accurate and consistent predictions for high and low flows and water balance simultaneously, compared to the single‐objective method. In particular, the multi‐objective method (II) produces the best modelling capacity to capture the non‐stationary nature of the hydrological response under different climate conditions. The pattern of improvement with the multi‐objective method (II) was generally consistent through the seasons, with the exception of the winter period in the regions partially affected by snow. This exception is due to a potential limitation of the IHACRES model in reflecting the impact of snow on the catchment hydrology. Copyright © 2013 John Wiley & Sons, Ltd.     

Comparison of two rainfall–runoff models: effects of conceptualization on water budget components

Rainfall–runoff models with different conceptual structures for the hydrological processes can be calibrated to effectively reproduce the hydrographs of the total runoff, while resulting in water budget components that are essentially different. This finding poses an open question on the reliability of rainfall–runoff models in reproducing hydrological components other than those used for calibration. In an effort to address this question, we use data from the Glafkos catchment in western Greece to calibrate and compare the ENNS model, a research-oriented lumped model developed for the river Enns in Austria developed for the river Enns in Austria, with the operational MIKE SHE model. Model performance is assessed in the light of the conceptual/structural differences of the modelled hydrological processes, using indices calculated independently for each year, rather than for the whole calibration period, since the former are stricter. We show that even small differences in the representation of hydrological processes may impact considerably on the water budget components that are not measured (i.e. not used for model calibration). From all water budget components, direct runoff exhibits the highest sensitivity to structural differences and related model parameters.

EDITOR M.C. Acreman

ASSOCIATE EDITOR S. Huang     

On the value of isotope-enabled hydrological model calibration

ABSTRACT

Calibration of hydrological models is challenging in high-latitude regions where hydrometric data are minimal. Process-based models are needed to predict future changes in water supply, yet often with high amounts of uncertainty, in part, from poor calibrations. We demonstrate the utility of stable isotopes (¹⁸O, ²H) as data employed for improving the amount and type of information available for model calibration using the isoWATFLOOD^TM model. We show that additional information added to calibration does not hurt model performance and can improve simulation of water volume. Isotope-enabled calibration improves long-term validation over traditional flow-only calibrated models and offers additional feedback on internal flowpaths and hydrological storages that can be useful for informing internal water distribution and model parameterization. The inclusion of isotope data in model calibration reduces the number of realistic parameter combinations, resulting in more constrained model parameter ranges and improved long-term simulation of large-scale water balance.     

Performance of temperature and radiation index models for point-scale snow water equivalent (SWE) simulations in the Moroccan High Atlas Mountains

ABSTRACT

Four models of increasing complexity were tested and compared to simulate snow water equivalent at the local scale in the Moroccan High Atlas range. A classical temperature index model (TI) and three enhanced temperature index models that respectively include the potential clear-sky direct radiation (HTI), the incoming solar radiation (ETI-A) and net solar radiation (ETI-B), were subjected to annual and multi-annual calibration and cross-validated over the period 2003–2010. When calibrated yearly, the ETI models could be better transferred to other years, whereas all models, including the simple TI model, showed good transferability when calibrated over a longer period that includes inter-annual climate variability. No strong and recurrent relationships emerged between yearly calibrated model parameters and annual climate conditions. However, strong parameter compensation was observed for the enhanced models, which can be explained partly by the collinearity of air temperature and solar radiation causing equifinality of model parameters.     

Streamflow forecasting in remote areas

Forecasts of runoff volumes are required in order to maximize the utility of water-supply sources. In remote areas where hydrologic and land-use data are sparse, forecast models are needed; such models should be conceptually rational so they can be transferred to remote watersheds where data are sparse. A series of models were calibrated for a large watershed in India. A spatially-distributed seasonally-varying model having a structure similar to the rational method was found to provide precise, unbiased estimates of 10-day streamflow volumes. The model was tested on a watershed that was not used for calibration, with the results indicating a high correlation between the observed and measured streamflow. Thus, the model should provide good estimates of streamflow volumes on other ungaged watersheds.     

Assessing the impact of distributed snow water equivalent calibration and assimilation of Copernicus snow water equivalent on modelled snow and streamflow performance

Accuracy of the Copernicus snow water equivalent (SWE) product and the impact of SWE calibration and assimilation on modelled SWE and streamflow was evaluated. Daily snowpack measurements were made at 12 locations from 2016 to 2019 across a 4104 km² mixed-forest basin in the Great Lakes region of central Ontario, Canada. Sub-basin daily SWE calculated from these sites, observed discharge, and lake levels were used to calibrate a hydrologic model developed using the Raven modelling framework. Copernicus SWE was bias corrected during the melt period using mean bias subtraction and was compared to daily basin average SWE calculated from the measured data. Bias corrected Copernicus SWE was assimilated into the models using a range of parameters and the parameterizations from the model calibration. The bias corrected Copernicus product agreed well with measured data and provided a good estimate of mean basin SWE demonstrating that the product shows promise for hydrology applications within the study region. Calibration to spatially distributed SWE substantially improved the basin scale SWE estimate while only slightly degrading the flow simulation demonstrating the value of including SWE in a multi-objective calibration formulation. The particle filter experiments yielded the best SWE estimation but moderately degraded the flow simulation. The particle filter experiments constrained by the calibrated snow parameters produced similar results to the experiments using the upper and lower bounds indicating that, in this study, model calibration prior to assimilation was not valuable. The calibrated models exhibited varying levels of skill in estimating SWE but demonstrated similar streamflow performance. This indicates that basin outlet streamflow can be accurately estimated using a model with a poor representation of distributed SWE. This may be sufficient for applications where estimating flow is the primary water management objective. However, in applications where understanding the physical processes of snow accumulation, melt and streamflow generation are important, such as assessing the impact of climate change on water resources, accurate representations of SWE are required and can be improved via multi-objective calibration or data assimilation, as demonstrated in this study.     

Choosing an arbitrary calibration period for hydrologic models: How much does it influence water balance simulations?

The selection of calibration and validation time periods in hydrologic modelling is often done arbitrarily. Nonstationarity can lead to an optimal parameter set for one period which may not accurately simulate another. However, there is still much to be learned about the responses of hydrologic models to nonstationary conditions. We investigated how the selection of calibration and validation periods can influence water balance simulations. We calibrated Soil and Water Assessment Tool hydrologic models with observed streamflow for three United States watersheds (St. Joseph River of Indiana/Michigan, Escambia River of Florida/Alabama, and Cottonwood Creek of California), using time period splits for calibration/validation. We found that the choice of calibration period (with different patterns of observed streamflow, precipitation, and air temperature) influenced the parameter sets, leading to dissimilar simulations of water balance components. In the Cottonwood Creek watershed, simulations of 50-year mean January streamflow varied by 32%, because of lower winter precipitation and air temperature in earlier calibration periods on calibrated parameters, which impaired the ability for models calibrated to earlier periods to simulate later periods. Peaks of actual evapotranspiration for this watershed also shifted from April to May due to different parameter values depending on the calibration period's winter air temperatures. In the St. Joseph and Escambia River watersheds, adjustments of the runoff curve number parameter could vary by 10.7% and 20.8%, respectively, while 50-year mean monthly surface runoff simulations could vary by 23%–37% and 169%–209%, depending on the observed streamflow and precipitation of the chosen calibration period. It is imperative that calibration and validation time periods are chosen selectively instead of arbitrarily, for instance using change point detection methods, and that the calibration periods are appropriate for the goals of the study, considering possible broad effects of nonstationary time series on water balance simulations. It is also crucial that the hydrologic modelling community improves existing calibration and validation practices to better include nonstationary processes.     

A Compartmental–Spatial System Dynamics Approach to Ground Water Modeling

High-resolution, spatially distributed ground water flow models can prove unsuitable for the rapid, interactive analysis that is increasingly demanded to support a participatory decision environment. To address this shortcoming, we extend the idea of multiple cell (Bear 1979) and compartmental (Campana and Simpson 1984) ground water models developed within the context of spatial system dynamics (Ahmad and Simonovic 2004) for rapid scenario analysis. We term this approach compartmental–spatial system dynamics (CSSD). The goal is to balance spatial aggregation necessary to achieve a real-time integrative and interactive decision environment while maintaining sufficient model complexity to yield a meaningful representation of the regional ground water system. As a test case, a 51-compartment CSSD model was built and calibrated from a 100,000+ cell MODFLOW (McDonald and Harbaugh 1988) model of the Albuquerque Basin in central New Mexico (McAda and Barroll 2002). Seventy-seven percent of historical drawdowns predicted by the MODFLOW model were within 1 m of the corresponding CSSD estimates, and in 80% of the historical model run years the CSSD model estimates of river leakage, reservoir leakage, ground water flow to agricultural drains, and riparian evapotranspiration were within 30% of the corresponding estimates from McAda and Barroll (2002), with improved model agreement during the scenario period. Comparisons of model results demonstrate both advantages and limitations of the CCSD model approach.     

Constraining water table depth simulations in a land surface model using estimated baseflow

Several recent studies have shown the significance of representing groundwater in land surface hydrologic simulations. However, optimal methods for model parameter calibration in order to realistically simulate baseflow and groundwater depth have received little attention. Most studies still use globally constant groundwater parameters due to the lack of available datasets for calibration. Moreover, when models are calibrated, various parameter combinations are found to exhibit equifinality in simulated total runoff due to model parameter interactions. In this study, a simple lumped groundwater model is incorporated into the Community Land Model (CLM), in which the water table is interactively coupled to soil moisture through the groundwater recharge fluxes. The coupled model (CLMGW) is successfully validated in Illinois using a 22-year (1984–2005) monthly observational dataset. Baseflow estimates from the digital recursive filter technique are used to calibrate the CLMGW parameters. The advantage obtained from incorporating baseflow calibration in addition to traditional calibration based on measured streamflow alone is demonstrated by a Monte Carlo-type simulation analysis. Using the optimal parameter sets identified from baseflow calibration, flow partitioning and water table depth simulations using CLMGW are improved, and the equifinality problem is alleviated. For other regions that lack observations of water table depth, the baseflow calibration approach can be used to enhance parameter estimation and constrain water table depth simulations.     

Multiscale characterization of a heterogeneous aquifer using an ASR operation

Heterogeneity in the physical properties of an aquifer can significantly affect the viability of aquifer storage and recovery (ASR) by reducing the recoverable proportion of low-salinity water where the ambient ground water is brackish or saline. This study investigated the relationship between knowledge of heterogeneity and predictions of solute transport and recovery efficiency by combining permeability and ASR-based tracer testing with modeling. Multiscale permeability testing of a sandy limestone aquifer at an ASR trial site showed that small-scale core data give lower-bound estimates of aquifer hydraulic conductivity (K), intermediate-scale downhole flowmeter data offer valuable information on variations in K with depth, and large-scale pumping test data provide an integrated measure of the effective K that is useful to constrain ground water models. Chloride breakthrough and thermal profiling data measured during two cycles of ASR showed that the movement of injected water is predominantly within two stratigraphic layers identified from the flowmeter data. The behavior of the injectant was reasonably well simulated with a four-layer numerical model that required minimal calibration. Verification in the second cycle achieved acceptable results given the model's simplicity. Without accounting for the aquifer's layered structure, high precision could be achieved on either piezometer breakthrough or recovered water quality, but not both. This study demonstrates the merit of an integrated approach to characterizing aquifers targeted for ASR.     

Relating nitrogen sources and aquifer susceptibility to nitrate in shallow ground waters of the United States

Characteristics of nitrogen loading and aquifer susceptibility to contamination were evaluated to determine their influence on contamination of shallow ground water by nitrate. A set of 13 explanatory variables was derived from these characteristics, and variables that have a significant influence were identified using logistic regression (LR). Multivariate LR models based on more than 900 sampled wells predicted the probability of exceeding 4 mg/L of nitrate in ground water. The final LR model consists of the following variables: (1) nitrogen fertilizer loading (p-value = 0.012); (2) percent cropland-pasture (p < 0.001); (3) natural log of population density (p < 0.001); (4) percent well-drained soils (p = 0.002); (5) depth to the seasonally high water table (p = 0.001); and (6) presence or absence of a fracture zone within an aquifer (p = 0.002). Variables 1-3 were compiled within circular, 500 m radius areas surrounding sampled wells, and variables 4-6 were compiled within larger areas representing targeted land use and aquifers of interest. Fitting criteria indicate that the full logistic-regression model is highly significant (p < 0.001), compared with an intercept-only model that contains none of the explanatory variables. A goodness-of-fit test indicates that the model fits the data well, and observed and predicted probabilities of exceeding 4 mg/L nitrate in ground water are strongly correlated (r2 = 0.971). Based on the multivariate LR model, vulnerability of ground water to contamination by nitrate depends not on any single factor but on the combined, simultaneous influence of factors representing nitrogen loading sources and aquifer susceptibility characteristics.