O. Rakovec

Multiscale and Multivariate Evaluation of Water Fluxes and States over European River Basins

AbstractAccurately predicting regional-scale water fluxes and states remains a challenging task in contemporary hydrology. Coping with this grand challenge requires, among other things, a model that makes reliable predictions across scales, locations, and variables other than those used for parameter estimation. In this study, the mesoscale hydrologic model (mHM) parameterized with the multiscale regionalization technique is comprehensively tested across 400 European river basins. The model fluxes and states, constrained using the observed streamflow, are evaluated against gridded evapotranspiration, soil moisture, and total water storage anomalies, as well as local-scale eddy covariance observations. This multiscale verification is carried out in a seamless manner at the native resolutions of available datasets, varying from 0.5 to 100 km. Results of cross-validation tests show that mHM is able to capture the streamflow dynamics adequately well across a wide range of climate and physiographical character...

Improving the realism of hydrologic model functioning through multivariate parameter estimation

Increased availability and quality of near real-time observations provide the opportunity to improve understanding of predictive skills of hydrologic models. Recent studies have shown the limited capability of river discharge data alone to adequately constrain different components of distributed model parameterizations. In this study, the GRACE satellite-based total water storage (TWS) anomaly is used to complement the discharge data with the aim to improve the fidelity of mesoscale hydrologic model (mHM) through multivariate parameter estimation. The study is conducted on 83 European basins covering a wide range of hydroclimatic regimes. The model parameterization complemented with the TWS anomalies leads to statistically significant improvements in (1) discharge simulations during low-flow period, and (2) evapotranspiration estimates which are evaluated against independent data (FLUXNET). Overall, there is no significant deterioration in model performance for the discharge simulations when complemented by information from the TWS anomalies. However, considerable changes in the partitioning of precipitation into runoff components are noticed by in-/exclusion of TWS during the parameter estimation. Introducing monthly averaged TWS data only improves the dynamics of streamflow on monthly or longer time scales, which mostly addresses the dynamical behavior of the base flow reservoir. A cross-evaluation test carried out to assess the transferability of the calibrated parameters to other locations further confirms the benefit of complementary TWS data. In particular, the evapotranspiration estimates show more robust performance when TWS data are incorporated during the parameter estimation, in comparison with the benchmark model constrained against discharge only. This study highlights the value for incorporating multiple data sources during parameter estimation to improve the overall realism of hydrologic models and their applications over large domains.

Computationally inexpensive identification of noninformative model parameters by sequential screening

Environmental models tend to require increasing computational time and resources as physical process descriptions are improved or new descriptions are incorporated. Many-query applications such as sensitivity analysis or model calibration usually require a large number of model evaluations leading to high computational demand. This often limits the feasibility of rigorous analyses. Here we present a fully automated sequential screening method that selects only informative parameters for a given model output. The method requires a number of model evaluations that is approximately 10 times the number of model parameters. It was tested using the mesoscale hydrologic model mHM in three hydrologically unique European river catchments. It identified around 20 informative parameters out of 52, with different informative parameters in each catchment. The screening method was evaluated with subsequent analyses using all 52 as well as only the informative parameters. Subsequent Sobols global sensitivity analysis led to almost identical results yet required 40% fewer model evaluations after screening. mHM was calibrated with all and with only informative parameters in the three catchments. Model performances for daily discharge were equally high in both cases with Nash-Sutcliffe efficiencies above 0.82. Calibration using only the informative parameters needed just one third of the number of model evaluations. The universality of the sequential screening method was demonstrated using several general test functions from the literature. We therefore recommend the use of the computationally inexpensive sequential screening method prior to rigorous analyses on complex environmental models.

Toward seamless large domain parameter estimation for hydrologic models

Estimating spatially distributed parameters remains one of the biggest challenges for large domain hydrologic modeling. Many large domain modeling efforts rely on spatially inconsistent parameter fields, e.g., patchwork patterns resulting from individual basin calibrations, parameter fields generated through default transfer functions that relate geophysical attributes to model parameters, or spatially constant, default parameter values. This paper provides an initial assessment of a multi-scale parameter regionalization (MPR) method over large geographical domains to derive seamless parameters in a spatially consistent manner. MPR applies transfer functions at the native scale of the geophysical data, and then scales these model parameters to the desired model resolution. We developed a stand-alone framework called MPR-flex for multi-model use and applied MPR-flex to the Variable Infiltration Capacity model to produce hydrologic simulations over the contiguous USA (CONUS). We first independently calibrate 531 basins across the CONUS to obtain a performance benchmark for each basin. To derive the CONUS parameter fields, we perform a joint MPR calibration using all but the poorest behaved basins to obtain a single set of transfer function parameters that are applied to the entire CONUS. Results show that the CONUS-wide calibration has similar performance compared to previous simulations using a patchwork quilt of partially calibrated parameter sets, but without the spatial discontinuities in parameters that characterize some previous CONUS-domain model simulations. Several avenues to improve CONUS-wide calibration remain, including selection of calibration basins, objective function formulation, as well as MPR-flex improvements including transfer function formations and scaling operator optimization.

Making the most out of a hydrological model data set: Sensitivity analyses to open the model black‐box

In this work we investigate methods for gaining greater insight from hydrological model runs conducted for uncertainty quantification and model differentiation. We frame the sensitivity analysis questions in terms of the main purposes of sensitivity analysis: parameter prioritization, trend identification and interaction quantification. For parameter prioritization, we consider variance-based sensitivity measures, sensitivity indices based on the L1-norm, the Kuiper metric and the sensitivity indices of the DELSA methods. For trend identification, we investigate insights derived from graphing the one-way ANOVA sensitivity functions, the recently introduced CUSUNORO plots and derivative scatterplots. For interaction quantification, we consider information delivered by variance-based sensitivity indices. We rely on the so-called given-data principle, in which results from a set of model runs are used to perform a defined set of analyses. One avoids using specific designs for each insight, thus controlling the computational burden. The methodology is applied to a hydrological model of a river in Belgium simulated using the well established Framework for Understanding Structural Errors (FUSE) on five alternative configurations. The findings show that the integration of the chosen methods provide insights unavailable in most other analyses.

Anthropogenic warming exacerbates European soil moisture droughts

Anthropogenic warming is anticipated to increase soil moisture drought in the future. However, projections are accompanied by large uncertainty due to varying estimates of future warming. Here, using an ensemble of hydrological and land-surface models, forced with bias-corrected downscaled general circulation model output, we estimate the impacts of 1–3 K global mean temperature increases on soil moisture droughts in Europe. Compared to the 1.5 K Paris target, an increase of 3 K—which represents current projected temperature change—is found to increase drought area by 40% (±24%), affecting up to 42% (±22%) more of the population. Furthermore, an event similar to the 2003 drought is shown to become twice as frequent; thus, due to their increased occurrence, events of this magnitude will no longer be classified as extreme. In the absence of effective mitigation, Europe will therefore face unprecedented increases in soil moisture drought, presenting new challenges for adaptation across the continent.Severe drought plagued Europe in 2003, amplifying heatwave conditions that killed more than 30,000 people. Assuming business as usual, such soil moisture deficits will become twice as frequent in the future and affect up to two-thirds of the European population.

On the choice of calibration metrics for high flow estimation usinghydrologic models

Calibration is an essential step for improving the accuracy of simulations generated using hydrologic models, and a key modeler decision is the selection of the performance metric to be optimized. It has been common to use squared error performance metrics, or normalized variants such as Nash-Sutcliffe Efficiency (NSE), based on the idea that their squared-error nature will emphasize the estimation of high flows. However, we conclude that NSE-based model calibrations actually result in poor reproduction of high flow events, such as the annual peak flows that are used for flood frequency estimation. Using 5 three different types of performance metrics, we calibrate two hydrological models at daily step, the “Variable Infiltration Capacity” model (VIC) and the “mesoscale Hydrologic Model” (mHM) and evaluate their ability to simulate high flow events for 492 basins throughout the contiguous United States. The metrics investigated are (1) NSE, (2) Kling-Gupta Efficiency (KGE) and variants, and (3) Annual Peak Flow Bias (APFB), where the latter is an application-specific metric that focuses on annual peak flows. As expected, the APFB metric produces the best annual peak flow estimates; however, performance on 10 other high flow related metrics is poor. In contrast, the use of NSE results in annual peak flow estimates that are more than 20% worse, primarily due to the tendency of NSE to result in underestimation of observed flow variability. On the other hand, the use of KGE results in annual peak flow estimates that are better than from NSE owing to improved flow time series metrics (mean and variance), with only a slight degradation in performance with respect to other related metrics, particularly when a non-standard weighting of the components of KGE is used. Stochastically generated ensemble simulations based on remaining 15 residuals show ability to improve some of the metrics regardless of the deterministic performances. However, it is emphasized that obtaining the correct fidelity of streamflow dynamics from the deterministically calibrated models is still important as it may improve high flow metrics (for the right reasons). Overall this paper highlights the need for a deeper understanding of performance metric behavior and design in relation to the desired goals of model calibration.

Assimilation of Streamflow Observations

Streamflow is arguably the most important predictor in operational hydrologic forecasting and water resources management. Assimilation of streamflow observations into hydrologic models has received growing attention in recent decades as a cost-effective means to improve prediction accuracy. Whereas the methods used for streamflow data assimilation (DA) originated and were popularized in atmospheric and ocean sciences, the nature of streamflow DA is significantly different from that of atmospheric or oceanic DA. Compared to the atmospheric processes modeled in weather forecasting, the hydrologic processes for surface and groundwater flow operate over a much wider range of time scales. Also, most hydrologic systems are severely under-observed. The purpose of this chapter is to provide a review on streamflow measurements and associated uncertainty and to share the latest advances, experiences gained, and science issues and challenges in streamflow DA. Toward this end, we discuss the following aspects of streamflow observations and assimilation methods: (1) measurement methods and uncertainty of streamflow observations, (2) streamflow assimilation applications, and (3) benefits and challenges streamflow DA with regard to large-scale DA, multi-data assimilation, and dealing with timing errors.