Sources Of Uncertainty In Models Research Articles

Effects of spatial data resolution on runoff predictions by the BASINS model

The BASINS model, developed by the United States EPA, is a popular simulation tool for predicting watershed responses, such as runoff, pollution exports, and water quality. It requires large amounts of data to set parameters. Many studies state that model input is a major source of model uncertainty. Thus, improvements to the quality and completeness of the data will improve the certainty of the model. The objective of this study is to discuss the effects of spatial data, including digital elevation models (DEMs) and spatial rainfall records, on predictions of runoff from the BASINS model. The result shows that both DEMs and rainfall data can significantly influence peak flow and runoff volume. Rainfall input has more influence on the curve shape of hydrograph than DEM resolution. DEM resolution can have more impact on peak flow predictions than rainfall input. Because the model uncertainties from DEMs and rainfall records influence each other, the prediction error does not always decrease when DEM resolution increases. The present results show that the BASINS model produces reliable answers in the case area when the grid size is less than 100 m × 100 m and the precipitation records from the Bihu Rainfall Station are correct and complete.

Investigation of the complex dynamics and structure of the 2010 Eyjafjallajökull volcanic ash cloud using multispectral images and numerical simulations

AbstractWe investigated the structure and evolution of the 2010 Eyjafjallajökull volcanic cloud and its dispersal over Iceland and Europe integrating satellite multispectral images and numerical simulations. Data acquired by Medium Resolution Imaging Spectrometer (MERIS) and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) have been analyzed to quantify the cloud extent and composition. The VOL‐CALPUFF dispersal code was applied to reconstruct the transient and 3‐D evolution of the cloud. Source parameters estimated on the base of available a posteriori volcanological data sets have been used. Quantitative comparisons between satellite retrievals and modeling results were performed for two selected instants of time during the first and third eruptive phases on a regional scale. Sensitivity of the model to initial volcanological conditions has been analyzed at continental scale. Several complex non intuitive features of cloud dynamics have been highlighted and strengths and limitations of the adopted methods identified. The main findings are: the level of quantitative agreement between satellite observations and numerical results depends on ash cloud composition (particle sizes and concentration) with better agreement for smaller particles and higher concentrations; the agreement between observations and modeling outcomes also depends on the temporal stability of volcanological conditions and the complexity of the meteorological wind field; the irregular dispersion of ash, as reconstructed from satellite data and numerical modeling, can be well explained by the different response of particle sizes to strong vertical wind‐shear, and by resuspension processes acting at ground level; eruptive source conditions are the main source of uncertainty in modeling, especially during an ongoing crisis and at long‐range scales.

On the need for bias correction in regional climate scenarios to assess climate change impacts on river runoff

Abstract. In climate change impact research, the assessment of future river runoff as well as the catchment-scale water balance is impeded by different sources of modeling uncertainty. Some research has already been done in order to quantify the uncertainty of climate projections originating from the climate models and the downscaling techniques, as well as from the internal variability evaluated from climate model member ensembles. Yet, the use of hydrological models adds another layer of uncertainty. Within the QBic3 project (Québec–Bavarian International Collaboration on Climate Change), the relative contributions to the overall uncertainty from the whole model chain (from global climate models to water management models) are investigated using an ensemble of multiple climate and hydrological models. Although there are many options to downscale global climate projections to the regional scale, recent impact studies tend to use regional climate models (RCMs). One reason for that is that the physical coherence between atmospheric and land-surface variables is preserved. The coherence between temperature and precipitation is of particular interest in hydrology. However, the regional climate model outputs often are biased compared to the observed climatology of a given region. Therefore, biases in those outputs are often corrected to facilitate the reproduction of historic runoff conditions when used in hydrological models, even if those corrections alter the relationship between temperature and precipitation. So, as bias correction may affect the consistency between RCM output variables, the use of correction techniques and even the use of (biased) climate model data itself is sometimes disputed among scientists. For these reasons, the effect of bias correction on simulated runoff regimes and the relative change in selected runoff indicators is explored. If it affects the conclusion of climate change analysis in hydrology, we should consider it as a source of uncertainty. If not, the application of bias correction methods is either unnecessary to obtain the change signal in hydro-climatic projections, or safe to use for the production of present and future river runoff scenarios as it does not alter the change signal. The results of the present paper highlight the analysis of daily runoff simulated with four different hydrological models in two natural-flow catchments, driven by different regional climate models for a reference and a future period. As expected, bias correction of climate model outputs is important for the reproduction of the runoff regime of the past, regardless of the hydrological model used. Then again, its impact on the relative change of flow indicators between reference and future periods is weak for most indicators, with the exception of the timing of the spring flood peak. Still, our results indicate that the impact of bias correction on runoff indicators increases with bias in the climate simulations.

Model Comparison and Uncertainty Quantification for Geologic Carbon Storage: The Sim-SEQ Initiative

Sim-SEQ is a model comparison initiative for geologic carbon storage (GCS). Within this initiative, 15 different modeling teams are developing conceptual models for flow and transport of an injected CO2 plume at the Sim-SEQ study site (the S-3 site) located near Cranfield, Mississippi. The objective of the project is to understand the sources of model uncertainty in GCS, and if possible, to quantify these uncertainties through comparison of the different conceptual models and also through comparison with observed data from the S-3 site. In this paper, we compare six different conceptual models of the S-3 site, and present a preliminary uncertainty analysis of these six models using a generalized linear model approach. We show that differences in model conceptualization and interpretation of site characterization data cause a significant range in predictions.

Seismic Risk Analysis of Concrete Gravity Dams under Near-Fault Ground Motions

There are a large number of concrete dams worldwide. Some of the dams are in areas prone to seismicity and were built many years ago with minimal consideration to seismic loads. Dam safety during and after an earthquake, is the aim of the present study. The failure of a dam during an earthquake will be catastrophic in terms of human life and financial losses. In the present work, an analytical fragility analysis was performed in order to characterize the seismic vulnerability of concrete gravity dams by using a probabilistic method to model sources of uncertainty that could impact dam performance. The assessment of the seismic vulnerability of concrete gravity dams under near-fault ground motions was performed to assess their performance against seismic hazards. A case study was considered, it is about the dam of Oued el Fodda on the Oued Chelif River, West Algeria. This dam was designed in the early 1930s.

A study of enhancive parameter correction with coupled data assimilation for climate estimation and prediction using a simple coupled model

ABSTRACTUncertainties in physical parameters of coupled models are an important source of model bias and adversely impact initialisation for climate prediction. Data assimilation using error covariances derived from model dynamics to extract observational information provides a promising approach to optimise parameter values so as to reduce such bias. However, effective parameter estimation in a coupled model is usually difficult because the error covariance between a parameter and the model state tends to be noisy due to multiple sources of model uncertainties. Using a simple coupled model consisting of the 3-variable Lorenz model and a slowly varying slab ‘ocean’, this study first investigated how to enhance the signal-to-noise ratio in covariances between model states and parameters, and then designed a data assimilation scheme for enhancive parameter correction (DAEPC). In DAEPC, parameter estimation is facilitated after state estimation reaches a ‘quasi-equilibrium’ where the uncertainty of coupled model states is sufficiently constrained by observations so that the covariance between a parameter and the model state is signal dominant. The observation-updated parameters are applied to improving the next cycle of state estimation and the refined covariance of parameter and model state further improves parameter correction. Performing dynamically adaptive state and parameter estimations with speedy convergence, DAEPC provides a systematic way to estimate the whole array of coupled model parameters using observations, and produces more accurate state estimates. Forecast experiments show that the DAEPC initialisation with observation-estimated parameters greatly improves the model predictability – while valid ‘atmospheric’ forecasts are extended two times longer, the ‘oceanic’ predictability is almost tripled. The simple model results here provide some insights for improving climate estimation and prediction with a coupled general circulation model.

Impact of dumped sediment structures on hydrological modelling in the artificial Chicken Creek catchment, Germany

Revealing the hydrological impact of sediment structures promises a better understanding of the influence of the spatial variability of sediment properties on the hydrological patterns and processes at the catchment scale. To improve the knowledge of structure-process interactions in initial ecosystems, the 6-ha artificial Chicken Creek Catchment in Germany was investigated by the Transregional Collaborative Research Centre 38 (SFB/TRR 38). Sediment structures called pour-ribs, which are dumped by stackers during the construction process, lead to differently compacted sediment zones, which increase the spatial variability of sediments’ hydraulic properties. Although levelled afterwards by bulldozers, the majority of these structures remain in the subsurface. To analyse the effects of pour-ribs on the hydrological catchment’s behaviour, the process-based spatially distributed Water balance Simulation Model (WaSiM-ETH) was applied. The results show that the consideration of pour-ribs improves the runoff simulation and significantly affects the simulated soil moisture patterns and, thereby, the initial stage of the ecosystem development. Compacted zones act as hydraulic barriers and inhibit subsurface lateral water flow, whereas non-compacted zones constitute areas with increased water storage capacity. Both effects cause reduced catchment runoff. Moreover, disregarding of the pour-ribs was identified as a source of model uncertainty in previous studies. A further outcome of this study is the importance of a global sensitivity analysis as a tool for model improvement. Finally, the results stress the importance of considering the variability of sediment properties for hydrological modelling at the catchment scale.

Uncertainty in computations of the spread of warm water in a river – lessons from Environmental Impact Assessment case study

Abstract. The present study aims at the evaluation of sources of uncertainty in modelling of heat transport in a river caused by the discharge coming from a cooling system of a designed gas-stem power plant. This study was a part of an Environmental Impact Assessment and was based on two-dimensional modelling of temperature distribution in an actual river. The problems with the proper description of the computational domain, velocity field and hydraulic characteristics were considered in the work. An in-depth discussion on the methods of evaluation of the dispersion coefficients in the model comprising of all four components of the dispersion tensor was carried out. It was shown that in natural rivers all components of a dispersion tensor should be taken into account to qualitatively reflect the proper shape of temperature distributions. The results considerably depend on the 2-D velocity field as well as hydraulic and morphometric characteristics of the flow. Numerical methods and their influence on the final results of computations were also discussed. All computations were based upon a real case study performed in Vistula River in Poland.

Improving Identifiability in Model Calibration Using Multiple Responses

In physics-based engineering modeling, the two primary sources of model uncertainty, which account for the differences between computer models and physical experiments, are parameter uncertainty and model discrepancy. Distinguishing the effects of the two sources of uncertainty can be challenging. For situations in which identifiability cannot be achieved using only a single response, we propose to improve identifiability by using multiple responses that share a mutual dependence on a common set of calibration parameters. To that end, we extend the single response modular Bayesian approach for calculating posterior distributions of the calibration parameters and the discrepancy function to multiple responses. Using an engineering example, we demonstrate that including multiple responses can improve identifiability (as measured by posterior standard deviations) by an amount that ranges from minimal to substantial, depending on the characteristics of the specific responses that are combined.

Model Uncertainty and Robustness Tests: Towards a New Logic of Statistical Inference

We augment the increasingly common practice of typically ad hoc robustness tests into a research methodology that allows reliable inferences when researchers do not know the true data-generating process. We identify three principal sources of model uncertainty. First, theories simplify and aim at isolating causal mechanisms under ceteris paribus conditions. Theories thus cannot sufficiently specify the data-generating process. Second, empirical researchers are uncertain about what constitutes the population – the set of cases for which a theory claims validity – from which to draw a representative sample as required for econometric estimation. Finally, model uncertainty also results from the fact that many concepts used in social science theories cannot be observed and therefore can only be measured by proxy variables, which deviate in unknown, but potentially systematic ways from the theoretical concepts. These three causes of model uncertainty result in at least 16 specific aspects of uncertainty about model specification that can be subjected to various robustness tests. Defining robustness as stability of causal inferences, we argue against the misplaced focus on single point estimates from any specific test. Instead, we suggest scholars employ model averaging techniques to assess robustness in terms of the entire set of empirical evidence from the baseline and all robustness test models. This method avoids the type-II error bias of current practice and encourages researchers to undertake a larger set of robustness tests instead of carefully hand-selecting only those tests that do not deviate from their baseline model results. We illustrate our proposal using an example of an existing study by the authors on the political determinants of famine mortality, which contained only two robustness tests and thus begs the question whether the causal inferences are fragile. Subjecting our baseline model to a battery of additional tests, we conclude that our causal inferences are in fact robust: democracies respond more elastically than autocracies to the simultaneous presence of international food aid and a large share of the population being affected, which results in lower famine mortality.

Clouds continue to plague latest generation climate models

In anticipation of the 2013 publication of the Intergovernmental Panel on Climate Change's fifth assessment report, many in the climate modeling community came together in 2008 under the banner of the Fifth Coupled Model Intercomparison Project (CMIP5). An international effort, this project sought to identify crucial research directions and laid down standardized parameters under which atmosphere‐ocean coupled general circulation models (AOGCMs) should be tested so that they can be compared on equal footing. Andrews et al. assessed 15 models participating in the CMIP5 to find the most up‐to‐date measure of climate sensitivity and the remaining sources of model uncertainty. To test the AOGCMs, the authors looked at how the modeled top‐of‐the‐atmosphere (TOA) radiative emissions from the Earth respond to an imposed change in atmospheric carbon dioxide concentration. Because TOA emissions depend on the global mean atmospheric and surface temperatures, this technique provides an effective way to investigate the feedback mechanisms built into each of the models. The authors assessed model runs for pre‐industrial atmospheric carbon dioxide levels as well as those that simulated a sudden quadrupling in the concentration of atmospheric carbon dioxide—the amount that could be reached given a “do nothing” approach to mitigating carbon emissions.

Robust detection of exotic infectious diseases in animal herds: A comparative study of three decision methodologies under severe uncertainty

When animals are transported and pass through customs, some of them may have dangerous infectious diseases. Typically, due to the cost of testing, not all animals are tested: a reasonable selection must be made. How to test effectively whilst avoiding costly disease outbreaks? First, we extend a model proposed in the literature for the detection of invasive species to suit our purpose, and we discuss the main sources of model uncertainty, many of which are hard to quantify. Secondly, we explore and compare three decision methodologies on the problem at hand, namely, Bayesian statistics, info-gap theory and imprecise probability theory, all of which are designed to handle severe uncertainty. We show that, under rather general conditions, every info-gap solution is maximal with respect to a suitably chosen imprecise probability model, and that therefore, perhaps surprisingly, the set of maximal options can be inferred at least partly—and sometimes entirely—from an info-gap analysis.

Declining sea ice to lead to cloudier Arctic

Arctic sea ice has been declining over the past several decades as global climate has warmed. Moreover, sea ice has declined more quickly than many models predicted, indicating that climate models may not be correctly representing some processes controlling sea ice. One source of uncertainty in models is feedback from cloud cover. Sea ice can affect cloud cover, as melting sea ice and increased evaporation from the ocean surface can lead to increased cloud formation.

Ice in Clouds Experiment–Layer Clouds. Part II: Testing Characteristics of Heterogeneous Ice Formation in Lee Wave Clouds

Abstract Heterogeneous ice nucleation is a source of uncertainty in models that represent ice clouds. The primary goal of the Ice in Clouds Experiment–Layer Clouds (ICE-L) field campaign was to determine if a link can be demonstrated between ice concentrations and the physical and chemical characteristics of the ambient aerosol. This study combines a 1D kinematic framework with lee wave cloud observations to infer ice nuclei (IN) concentrations that were compared to IN observations from the same flights. About 30 cloud penetrations from six flights were modeled. The temperature range of the observations was −16° to −32°C. Of the three simplified ice nucleation representations tested (deposition, evaporation freezing, and condensation/immersion droplet freezing), condensation/immersion freezing reproduced the lee wave cloud observations best. IN concentrations derived from the modeling ranged from 0.1 to 13 L−1 compared to 0.4 to 6 L−1 from an IN counter. A better correlation was found between temperature and the ratio of IN concentration to the concentration of large aerosol (>500 nm) than between IN concentration and the large aerosol concentration or temperature alone.

Capital Punishment and Deterrence: Understanding Disparate Results

Investigate how different model assumptions have driven the conflicting findings in the literature on the deterrence effect of capital punishment. The deterrence effect of capital punishment is estimated across different models that reflect the following sources of model uncertainty: (1) the uncertainty about the probability model generating the aggregate murder rate equation, (2) the uncertainty about the determinants of an individual’s choice of committing a murder or not, (3) the uncertainty about state level heterogeneity, and (4) the uncertainty about the exchangeability between observations with zero murder case and those with positive murder cases. First, the estimated deterrence effects exhibit great dispersion across models. Second, a particular subset of models—linear models with constant coefficients—always predict a positive deterrence effect. All other models predict negative deterrence effects. Third, the magnitudes of the point estimates of deterrence effects differ mainly because of the choice of linear versus logistic specifications. The question about the deterrence effect of capital punishment cannot be answered independently from substantive assumptions on what determines individual behavior. The need for judgment cannot be escaped in empirical work.

The Effect of Public Debt on Growth in Multiple Regimes

We employ a structural threshold regression methodology to investigate the heterogeneous effects of debt on growth using public debt as a threshold variable as well as several other plausible variables. Our methodology allows us to address three sources of model uncertainty that characterize cross-country growth data: parameter heterogeneity, theory uncertainty, and endogeneity. We find strong evidence for threshold effects based on democracy, which implies that higher public debt results in lower growth for countries in the Low-Democracy regime. Our results are consistent with the presence of parameter heterogeneity in the cross-country growth process due to fundamental determinants of economic growth proposed by the new growth theories.

Modelling gross primary production in the Heihe river basin and uncertainty analysis

Accurate and timely monitoring of gross primary production (GPP) at regional and global scales is necessary for understanding the terrestrial biosphere carbon balance. In this article, 8-day composite GPP is estimated using the region production efficiency model (REG-PEM) in the Heihe river basin from 2006 to 2008. The result indicates that GPP meets the seasonal cycle well and varies with different land covers. Analysis of uncertainty and sensitivity of the REG-PEM model are implemented by the Monte Carlo method. GPP is simulated with a test data set; the data set includes three groups, and each group has 8000 points. The three groups obey uniform distribution, normal distribution and beta distribution, respectively. Uncertainty is assessed by the quantification of mean values and standard deviation of outputs. The distribution of simulated GPP is slightly different from that of model-calculated GPP over the study sites. Among the input parameters, the land surface water index (LSWI) and air temperature (T) make less of a contribution to model output than photosynthetic active radiation (PAR) and the enhanced vegetation index (EVI). The main sources of model uncertainty are input data uncertainty and uncertainty from the specific model.

The bias on the Joule heating estimate: Small-scale variability versus resolved-scale model uncertainty and the correlation of electric field and conductance

[1] Estimating the Joule heating rate using models of electric field and conductance requires removal of various biases, which arise because the models have uncertainty. This problem has been referred to as electric field variability, although variability of the conductance is also at issue. Variability of the conductance can contribute to the bias on Joule heating if it is correlated with the variability of the squared electric field. The term “variability,” however, includes two distinct sources of model uncertainty: physical fluctuations on a small scale below the resolution of the model (small-scale variability) and uncertainty in the model's predictions even on the scale that it does resolve (resolved-scale model uncertainty). Previous work has tended to mix small-scale variability with resolved-scale model uncertainty in a way that may have led to confusion. In this work we separately analyze the small-scale variability and the resolved-scale model uncertainty using the Sondrestrom radar database, and we find that each gives a significant contribution to the bias on Joule heating. We also analyze the correlation of conductance and squared electric field, again in a way that distinguishes the effect of correlation over small scales from the effect of correlation of the resolved-scale model uncertainties. We explain that small-scale variability, being the result of actual physical fluctuations on small scales, is a more general quantity than resolved-scale model uncertainty; the latter is only defined with respect to a particular model.

Model Uncertainty in Finite-Element Analysis: Bayesian Finite Elements

In this paper, probabilistic models for structural analysis are put forward, with particular emphasis on model uncertainty. Context is provided by the finite-element method and the need for probabilistic prediction of structural performance in contemporary engineering. Sources of model uncertainty are identified and modeled. A Bayesian approach is suggested for the assessment of new model parameters within the element formulations. The expressions are formulated by means of numerical “sensors” that influence the model uncertainty, such as element distortion and degree of nonlinearity. An assessment procedure is proposed to identify the sensors that are most suitable to capture model uncertainty. This paper presents the general methodology and specific implementations for a general-purpose structural element. Two numerical examples are presented to demonstrate the methodology and its implications for probabilistic prediction of structural response.

Industrial Application of Robust Aeroelastic Analysis

EROELASTIC analysis and flight flutter testing (FFT) are central and important topics in aircraft airworthiness assessment. Confident results from numerical analysis are necessary in order to obtain an efficient and reliable flutter clearance process. There are, however, numerous sources of model uncertainties involvedinaeroelasticanalysis.Theaerodynamicsareoftenclaimed to be the most dominant source, but structural uncertainties in terms of variations in mass and stiffness properties may sometimes also be significant; see Pettit [1]. Considering FFT, this is an expensive activity that could be associated with high risk. Here as well, a systematic and reliable experimental procedure is required to obtain confident experimental data in a safe yet efficient manner. The number of gathered data points from fight testing is usually very limited. This fact often makes robust analysis methods based on statistics less usable in practice.