Bayesian Prediction Model Research Articles

Multistate Markov modelling of long-term outcomes in immunotherapy for metastatic renal cell carcinoma (mRCC) using machine-learned dynamic Bayesian networks.

e14049 Background: Patient-level heterogeneity in response to treatment remains a major challenge in cancer immunotherapy. Long-term individual-level modelling of survival, tumour response and safety outcomes jointly can help improve efficacy and durable clinical benefits. Machine-learned Bayesian networks provide a solution to the “black box” problem of other machine-learning approaches. Objectives: To develop a dynamic Bayesian network model for multivariate risk prediction, survival modelling and long-term simulations of immunotherapy patients using a simulated trial dataset. Methods: A simulated randomized clinical trial dataset for second line immunotherapy of patients with renal cell carcinoma was used for analysis. We machine-learned a dynamic Bayesian network from censored data with incorporation of prior clinical knowledge. Classification performance, probability calibration, goodness-of-fit metrics and prognostic variables were calculated following TRIPOD guidelines. Results: The machine-learned graphical model encoded expected relationships between variables with minimal prior information. Visual and numerical goodness-of-fit checks for survival extrapolations showed that the model fitted data well and was appropriate. Probability calibration was < 10% from ideal. Classification performance for overall survival was high ( c-statistic ~0.85) soon after treatment initiation and gradually plateaued over time. Prognostic variables were calculated by treatment arm for overall survival and severe adverse events. Conclusions: Dynamic Bayesian network model was useful for transparently representing joint relationships between all variables in the trial dataset, for multivariate risk prediction and long-term extrapolations that can be used in economic evaluations for health technology assessments in oncology.

Feature Uncertainty Predicts Behavioral and Neural Responses to Combined Concepts.

The cognitive and neural structure of conceptual knowledge affects how concepts combine in language and thought. Examining the principles by which individual concepts (e.g., diamond, baseball) combine into more complex phrases (e.g., "baseball diamond") can illuminate not only how the brain combines concepts but also the key ingredients of conceptual structure. Here we specifically tested the role of feature uncertainty in the modulation of conceptual brightness evoked by adjective-noun combinations (e.g., "dark diamond") in male and female human subjects. We collected explicit ratings of conceptual brightness for 45 noun concepts and their "dark" and "light" combinations, resulting in a measure reflecting the degree of conceptual brightness modulation in each noun concept. Feature uncertainty was captured in an entropy measure, as well as in a predictive Bayesian model of feature modulation. We found that feature uncertainty (i.e., entropy) and the Bayesian model were both strong predictors of these behavioral effects. Using fMRI, we observed the neural responses evoked by the concepts and combinations in a priori ROIs. Feature uncertainty predicted univariate responses in left inferior frontal gyrus, and multivariate responses in left anterior temporal lobe were predicted by degree of conceptual brightness modulation. These findings suggest that feature uncertainty is a key ingredient of conceptual structure, and inform cognitive neuroscience theories of conceptual combination by highlighting the role of left inferior frontal gyrus and left anterior temporal lobe in the process of flexible feature modulation during comprehension of complex language.SIGNIFICANCE STATEMENT The meaning of a word depends on the words surrounding it. The challenge of understanding how flexible meaning emerges in language can be simplified by studying adjective-noun phrases. We tested whether the uncertainty of a feature (i.e., brightness) in a given noun concept (e.g., diamond) influences how the adjective and noun concepts combine. We analyzed feature uncertainty using two probabilistic measures, and found that feature uncertainty predicted people's explicit interpretations of adjective-noun phrases (e.g., "dark diamond"). Using fMRI, we found that combined concepts evoked responses in left inferior frontal gyrus and left anterior temporal lobe that related to our measures of feature modulation and uncertainty. These findings reveal the cognitive and neural processes supporting conceptual combination and complex language use.

Risk stratification in pulmonary arterial hypertension using Bayesian analysis.

Current risk stratification tools in pulmonary arterial hypertension (PAH) are limited in their discriminatory abilities, partly due to the assumption that prognostic clinical variables have an independent and linear relationship to clinical outcomes. We sought to demonstrate the utility of Bayesian network-based machine learning in enhancing the predictive ability of an existing state-of-the-art risk stratification tool, REVEAL 2.0. We derived a tree-augmented naïve Bayes model (titled PHORA) to predict 1-year survival in PAH patients included in the REVEAL registry, using the same variables and cut-points found in REVEAL 2.0. PHORA models were validated internally (within the REVEAL registry) and externally (in the COMPERA and PHSANZ registries). Patients were classified as low-, intermediate- and high-risk (<5%, 5-20% and >10% 12-month mortality, respectively) based on the 2015 European Society of Cardiology/European Respiratory Society guidelines. PHORA had an area under the curve (AUC) of 0.80 for predicting 1-year survival, which was an improvement over REVEAL 2.0 (AUC 0.76). When validated in the COMPERA and PHSANZ registries, PHORA demonstrated an AUC of 0.74 and 0.80, respectively. 1-year survival rates predicted by PHORA were greater for patients with lower risk scores and poorer for those with higher risk scores (p<0.001), with excellent separation between low-, intermediate- and high-risk groups in all three registries. Our Bayesian network-derived risk prediction model, PHORA, demonstrated an improvement in discrimination over existing models. This is reflective of the ability of Bayesian network-based models to account for the interrelationships between clinical variables on outcome, and tolerance to missing data elements when calculating predictions.

A Bayesian In-Play Prediction Model for Association Football Outcomes

Point process models have made a significant contribution to the prediction of football association outcomes. It is conventionally the case that defence and attack capabilities have been assumed to be constant during a match and estimated against the average performance of all other teams in history. Drawing upon a Bayesian method, this paper proposes a dynamic strength model which relaxes assumption of the constant teams’ strengths and permits applying in-match performance information to calibrate them. An empirical study demonstrates that although the Bayesian model fails to achieve improvement in goal difference prediction, it registers clear achievements with regard to the prediction of the total number of goals and Win/Draw/Loss outcome prediction. When the Bayesian model bets against the SBOBet bookmaker, one of the most popular gaming companies among Asian handicaps fans, whose odds data were obtained from both the Win/Draw/Loss market and over–under market, it may obtain positive returns; this clearly contrasts with the process model with constant strengths, which fails to win money from the bookmaker.

SMOTE-Boost-based sparse Bayesian model for flood prediction

With a significant development of big data analysis and cloud-fog-edge computing, human-centered computing (HCC) has been a hot research topic worldwide. Essentially, HCC is a cross-disciplinary research domain, in which the core idea is to build an efficient interaction among persons, cyber space, and real world. Inspired by the improvement of HCC on big data analysis, we intend to involve related core and technologies to help solve one of the most important issues in the real world, i.e., flood prediction. To minimize the negative impacts brought by floods, researchers pay special attention to improve the accuracy of flood forecasting with quantity of technologies including HCC. However, historical flood data is essentially imbalanced. Imbalanced data causes machine learning classifiers to be more biased towards patterns with majority samples, resulting in poor classification of pattern with minority samples. In this paper, we propose a novel Synthetic Minority Over-sampling Technique (SMOTE)-Boost-based sparse Bayesian model to perform flood prediction with both high accuracy and robustness. The proposed model consists of three modules, namely, SMOTE-based data enhancement, AdaBoost training strategy, and sparse Bayes model construction. In SMOTE-based data enhancement, we adopt a SMOTE algorithm to effectively cover diverse data modes and generate more samples for prediction pattern with minority samples, which greatly alleviates the problem of imbalanced data by involving experts’ analysis and users’ intentions. During AdaBoost training strategy, we propose a specifically designed AdaBoost training strategy for sparse Bayesian model, which not only adaptively and inclemently increases prediction ability of Bayesian model, but also prevents its over-fitting performance. Essentially, the design of AdaBoost strategy helps keep balance between prediction ability and model complexity, which offers different but effective models over diverse rivers and users. Finally, we construct a sparse Bayesian model based on AdaBoost training strategy, which could offer flood prediction results with high rationality and robustness. We demonstrate the accuracy and effectiveness of the proposed model for flood prediction by conducting experiments on a collected dataset with several comparative methods.

Farm Exit among Smallhoder Farmers of Nepal: A Bayesian Logistic Regression Models Approach.

To leave or not to leave farming? This is a dilemma facing a large number of farm households in a rural agrarian setting of Nepal where nearly two-thirds of the population is small holder farmers. Using the uniquely detailed retrospective panel data collected in 2015 from farming households, we examine the influence of the access to cultivated land holding and land tenure on subsequent farm exit. We address the statistical modeling issue of complete separation by developing a robust Bayesian predictive model to predict the probability of farm exit. We use the Bayesian framework with weakly informative prior to carry out the logistic regression model and compare it with other available binary response models. Our results show that the size of cultivated land has a strong negative influence on farm exit, net of all other controls. Moreover, farmers who rented farmland from others or who rented farmland to others were significantly more likely to exit farming. We estimate that a farm household required at least 5 Katha of land (one sixth of a hectare) per year to stay in farming.

Artificial Intelligence-Enabled Predictive Insights for Ameliorating Global Malnutrition: A Human-Centric AI-Thinking Approach

According to the World Health Organization (WHO) and the World Bank, malnutrition is one of the most serious but least-addressed development challenges in the world. Malnutrition refers to the malfunction or imbalance of nutrition, which could be influenced not only by under-nourishment, but also by over-nourishment. The significance of this paper is that it shows how artificial intelligence (AI) can be democratized to enable analysts who are not trained in computer science to also use human-centric explainable-AI to simulate the possible dynamics between malnutrition, health and population indicators in a dataset collected from 180 countries by the World Bank. This AI-based human-centric probabilistic reasoning approach can also be used as a cognitive scaffold to educe (draw out) AI-Thinking in analysts to ask further questions and gain deeper insights. In this study, a rudimentary beginner-friendly AI-based Bayesian predictive modeling approach was used to demonstrate how human-centric probabilistic reasoning could be utilized to analyze the dynamics of global malnutrition and optimize conditions for achieving the best-case scenario. Conditions of the worst-case “Black Swan” scenario were also simulated, and they could be used to inform stakeholders to prevent them from happening. Thus, the nutritional and health status of vulnerable populations could be ameliorated.

A Bayesian learning model for design-phase service mashup popularity prediction

Using web services as building blocks to develop software applications, i.e., service mashups, not only reuses software development efforts to minimize development cost, but also leverages user groups and marketing efforts of those services to attract users and improve profits. This has significantly encouraged the development of a large number of service mashups in various domains. However, using existing services, even popular ones, does not guarantee the success of a mashup. In fact, a large portion of existing mashups fail to attract a good number of users, making the mashup development effort less effective. Design-phase popularity prediction can help avoid unpromising mashup developments by providing early-on insight into the potential popularity of a mashup. In this paper, we investigate the factors that can affect the popularity of a mashup through a comprehensive analysis on one of the largest mashup repository (i.e., ProgrammableWeb). We further propose a novel Bayesian approach that offers early-on insight to developers into the potential popularity of a mashup using design-phase features only. Besides identifying those relevant features, the Bayesian learning model can provide a confidence level for each prediction. This provides useful guidance to developers for successful mashup development. Experimental results demonstrate that the proposed approach achieves high prediction accuracy and outperforms competitive models.

A Sequential Bayesian Updated Wiener Process Model for Remaining Useful Life Prediction

Wiener processes have been extensively used to model the degradation processes exhibiting a linear trend for predicting the remaining useful life (RUL) of degrading components. To incorporate the real-time degradation monitoring information into degradation modeling, the Bayesian method has been frequently utilized to update the model parameter, particularly for the drift parameter in Wiener process. However, due to the inherent independent increment and Markov properties of Wiener process, the Bayesian updated drift parameter only utilizes the current degradation measurement and cannot incorporate the whole degradation measurements up to now. As such, once the updated degradation model in this way is used to predict the RUL, the obtained result may be dominated by partial degradation observations or lower the prognosis accuracy. In this paper, we propose a sequential Bayesian updated Wiener process model for RUL prediction. First, a Wiener process model with random drift efficient is used to model the degradation process with the linear trend. To estimate the model parameters, the historical degradation measurements are used to determine the initial model parameters based on the maximum likelihood estimation (MLE) method. Then, for the degrading component in service, a sequential Bayesian method is proposed to update the random drift parameter in Wiener process model. Differing from existing studies using the Bayesian method, the proposed sequential method uses the Bayesian estimate for random drift parameter in the last time as the prior of the next time. As such, the Bayesian estimate for random drift parameter in the current time is dependent on the whole degradation measurements up to current time, and thus the problem of depending only on the current degradation measurement is solved. Finally, we derive the analytical expressions of the RUL distribution based on the concept of the first passage time (FPT). Two case studies associated with the gyroscope drift data and lithium-ion battery data are provided to show the effectiveness and superiority of the proposed method. The results indicate that the proposed method can improve the RUL prediction accuracy.

Dynamic Bayesian network state prediction based on variable relationship

In order to improve the accuracy of the state prediction model, a dynamic Bayesian network state prediction model based on the relationship of prediction variables is designed. The prediction model of dynamic Bayesian network structure learning algorithm was improved, integrated into the Gibbs sampling algorithm model prediction, joined the predicted relationship between different factors affecting the node, is given based on the variable relationship between the dynamic Bayesian network structure design, using a moment on the different nodes and state influence factors to predict the probability distribution of the moment state nodes. The experimental results show that the model is simple in structure, more accurate than the traditional learning method of Bayesian network structure, and more practical.

Understanding the vulnerability of beef producers in Australia to an FMD outbreak using a Bayesian Network predictive model

Effective and adaptable biosecurity and surveillance systems are crucial for maintaining and increasing Australia’s competitive advantages in international markets, and for the production of high quality, safe animal products. These systems are continuously strengthened by ongoing government and industry investment. However, a better understanding of evolving disease risks and the country’s capacity to respond to these risks is needed. This study developed a vulnerability framework based on characteristics and behaviours of livestock producers that impact exposure and response capacity to an emergency animal disease (EAD) outbreak among beef producers in Australia, with a focus on foot and mouth disease (FMD). This framework articulated producer vulnerability typologies to better inform surveillance resource allocation and future research direction. A cross-sectional study of beef producers in Australia was conducted to gather information on producers’ demographics, husbandry characteristics, biosecurity and animal health management practices and beliefs, including those specific to FMD risk and response capacity. A Bayesian Network (BN) model was developed from the vulnerability framework, to investigate the complex interrelationships between variables and identify producer typologies. A total of 375 usable responses were obtained from the cross-sectional study. Regarding EAD exposure, producers implemented appropriate biosecurity practices for incoming stock, such as isolation (72.0 %), inspection for disease (88.7 %) and the use of vendor declarations (78.5 %); however, other biosecurity practices were limited, such as restriction of visitor access, visitor biosecurity requirements or feral animal control. In relation to response capacity, a moderate uptake of practices was observed. Whilst daily or weekly visual inspection of animals was reported by most producers (90.1 %), physical inspection was less frequent. Most producers would call a private veterinarian in response to unusual signs of disease in their cattle; however, over 40 % of producers did not cite calling a government veterinarian as a priority action. Most producers believe an FMD outbreak would have extremely serious consequences; however, their level of concern was moderate and their confidence in identifying FMD symptoms was low. The BN analysis identified six vulnerability typologies, with three levels of exposure (high, moderate, low) and two levels of response capacity (high, low), as described by producer demographics and practices. The model identified property size, number of cattle and exposure variables as the most influential to the overall producer vulnerability. Results from this study can inform how to best use current biosecurity and surveillance resources and identify where opportunities exist for improving Australia’s preparedness for future EAD incursions.

Bayesian predictive analysis for Weibull-Pareto composite model with an application to insurance data

Aminzadeh and Deng, respectively provide Bayesian predictive models for Exponential-Pareto and Inverse Gamma-Pareto composite distributions which are one-parameter models. The purpose of this article is to develop an alternative Bayesian predictive model (two-parameter) which can be used to compute important risk measures that are not defined via the above predictive models. Bayesian predictive density for the Weibull-Pareto composite distribution is developed and is used to compute risk measures such as Value at Risk (VaR), Conditional Tail Expectation (CTE), Predictive Expectation (PE), Limited Predictive Expected value (LPE), Limited Predictive Variance (LPV), and Limited Predictive Tail-VaR (LPCTE). Accuracy of parameter estimates as well as the risk measures are assessed via simulation studies. It is shown that the informative Bayes estimates are consistently more accurate than ML and the non-informative Bayes estimates. Backtesting for the risk measures is performed and goodness-of-fit of Weibull-Pareto among other composite models to the Danish fire data is assessed.

A Bayesian network model for prediction and analysis of possible forest fire causes

Possible causes of a forest fire ignition could be human-caused (arson, smoking, hunting, picnic fire, shepherd fire, stubble burning) or natural-caused (lightning strikes, power lines). Temperature, relative humidity, tree species, distance from road, wind speed, distance from agricultural land, amount of burnt area, month and distance from settlement are the risk factors that may affect the occurrence of forest fires. This study introduces the use of Bayesian network model to predict the possible forest fire causes, as well as to perform an analysis of the multilateral interactive relations among them. The study was conducted in Mugla Regional Directorate of Forestry area located in the southwest of Turkey. The fire data, which were recorded between 2008 and 2018 in the area, were provided by General Directorate of Forestry. In this study, after applying some different structural learning algorithms, a Bayesian network, which is built on the nodes relative humidity, temperature, wind speed, month, distance from settlement, amount of burnt area, distance from agricultural land, distance from road and tree species, was estimated. The model showed that month is the first and temperature is the second most effective factor on the forest fire ignitions. The Bayesian network model approach adopted in this study could also be used with data obtained from different areas having different sizes.

A Bayesian Spatial Categorical Model for Prediction to Overlapping Geographical Areas in Sample Surveys

Summary Motivated by the Australian National University poll, we consider a situation where survey data have been collected from respondents for several categorical variables and a primary geographic classification, e.g. postcode. Here, a common and important problem is to obtain estimates for a second target geography that overlaps with the primary geography but has not been collected from the respondents. We examine this problem when areal level census information is available for both geographic classifications. Such a situation is challenging from a small area estimation perspective for several reasons: there is a misalignment between the census and survey information as well as the geographical classifications; the geographic areas are potentially small and so prediction can be difficult because of the sparse or spatially missing data issue; and there is the possibility of non-stationary spatial dependence. To address these problems we develop a Bayesian model using latent processes, underpinned by a non-stationary spatial basis that combines Moran's I and multiresolution basis functions with a small but representative set of knots. The study results based on simulated data demonstrate that the model can be highly effective and gives more accurate estimates for areas defined by the target geography than several existing models. The model also performs well for the Australian National University poll data to predict on a second geographic classification: statistical area level 2.

Accrual Prediction Program: A web-based clinical trials tool for monitoring and predicting accrual for early-phase cancer studies.

Monitoring subject recruitment is key to the success of a clinical trial. Accordingly, researchers have developed accrual-monitoring tools to support the design and conduct of trials. At an institutional level, delays in identifying studies with high risk of accrual failure can lead to inefficient and costly trials with little chances of meeting study objectives. Comprehensive accrual monitoring is necessary to the success of the research enterprise. This article describes the design and implementation of the University of Kansas Cancer Center Accrual Prediction Program, a web-based platform was developed to support comprehensive accrual monitoring and prediction for all active clinical trials. The Accrual Prediction Program provides information on accrual, including the predicted completion date, predicted number of accrued subjects during the pre-specified accrual period, and the probability of achieving accrual targets. It relies on a Bayesian accrual prediction model to combine protocol information with real-time trial enrollment data and disseminates results via web application. First released in 2016, the Accrual Prediction Program summarizes enrollment information for active studies categorized by various trial attributes. The web application supports real-time evidence-based decision making for strategic resource allocation and study management of over 120 ongoing clinical trials at the University of Kansas Cancer Center. The Accrual Prediction Program makes accessing comprehensive accrual information manageable at an institutional level. Cancer centers or even entire institutions can reproduce the Accrual Prediction Program to achieve real-time comprehensive monitoring and prediction of subject accrual to aid investigators and administrators in the design, conduct, and management of clinical trials.

Hydrogeochemical imprints and spatio-temporal health risk assessment of lead in drinking water sources of Abeokuta, south-western Nigeria

Water samples were obtained from 63 boreholes around Kuto, Abeokuta, Nigeria, and analysed for some physicochemical parameters and heavy metals. The mean concentrations of TDS, Alka, F−, Fe, Mn, Pb, Zn, Ca2+, K+, Mg2+, Na+, SO42−, SiO2, HCO3−, Cl−, and CO32− were 1136, 27.45, 1.232, 0.521, 0.035, 0.013, 0.286, 78.48, 5.796, 89.91, 193.5, 298.3, 3.671, 411.3, 181.3, and 15.81 mg/L, respectively. Furthermore, the hydrochemical signatures influencing the dissolution of ions in the water samples were obtained, and the empirical Bayesian Kriging prediction model was used to validate the spatial distribution of the HM contaminants within the study region. 42.9% of the groundwater samples has lead concentrations exceeding the WHO threshold for drinking water. The PC/FA produced principal factors, and the hydrochemical facies verified the ion exchange mechanism, linking water quality parameters to geogenic sources (weathering, dissolution, leaching), and anthropogenic emissions from agricultural activities. Health risk investigation revealed that the highest hazard quotient (HRQtot) values for the specified heavy metals were observed at SMP18, SMP5, SMP10, and SMP21, respectively. Carcinogenic risks for adults after the exposure were unacceptable (CRcum > 1E−6) for SMP4, SMP6, SMP7, SMP10, and SMP21. SMP21 presented the highest CRcum (2.62E−6 ± 8.71E−4), while the cumulative carcinogenic risk (CRcum) for children that exceeded the limit was as follows: SMP21 < SMP9 < SMP13 < SMP5 < SMP11 < SMP10 < SMP7 < SMP6 < SMP4 < SMP21. Thus, the scenario calls for adequate water treatment strategies in order to safeguard the health of residents.

Did the movie Finding Dory increase demand for blue tang fish?

Representations of wildlife in television and films have long been hypothesized to shape human-wildlife interactions. A recent example is Pixar’s film Finding Dory, which featured a blue tang fish (Paracanthurus hepatus) as the main character and was widely reported in the popular press to have increased the number of such fish in the pet trade. We use Bayesian posterior predictive counterfactual models to evaluate the movie’s effect on three metrics of societal behaviour. Although there was an increase in global online searches for the blue tang 2–3 weeks after the movie, we find no substantial evidence for an increase in imports of blue tang fish into the US, or in number of visitors to US aquaria compared to counterfactual expectations. It is vital that an evidence-based discourse is used when communicating potential impacts of popular culture on human-wildlife relationships to avoid loss of credibility and misdirection of conservation resources.

Autism and psychosis as diametrical disorders of embodiment.

Humans have evolved an elaborate system of self-consciousness, self-identity, self-agency, and self-embodiment that is grounded in specific neurological structures including an expanded insula. Instantiation of the bodily self has been most-extensively studied via the ‘rubber hand illusion’, whereby parallel stimulation of a hidden true hand, and a viewed false hand, leads to the felt belief that the false hand is one’s own. Autism and schizophrenia have both long been regarded as conditions centrally involving altered development of the self, but they have yet to be compared directly with regard to the self and embodiment. Here, we synthesize the embodied cognition literature for these and related conditions, and describe evidence that these two sets of disorders exhibit opposite susceptibilities from typical individuals to the rubber hand illusion: reduced on the autism spectrum and increased in schizophrenia and other psychotic-affective conditions. Moreover, the opposite illusion effects are mediated by a consilient set of associated phenomena, including empathy, interoception, anorexia risk and phenotypes, and patterns of genetic correlation. Taken together, these findings: (i) support the diametric model of autism and psychotic-affective disorders, (ii) implicate the adaptive human system of self-embodiment, and its neural bases, in neurodevelopmental disorders, and suggest new therapies and (iii) experimentally ground Bayesian predictive coding models with regard to autism compared with psychosis.Lay summary: Humans have evolved a highly developed sense of self and perception of one’s own body. The ‘rubber hand illusion’ can be used to test individual variation in sense of self, relative to connection with others. We show that this illusion is reduced in autism spectrum disorders, and increased in psychotic and mood disorders. These findings have important implications for understanding and treatment of mental disorders.

Coupled flow accumulation and atmospheric blocking govern flood duration

We present a physically based Bayesian network model for inference and prediction of flood duration that allows for a deeper understanding of the nexus of antecedent flow regime, atmospheric blocking, and moisture transport/release mechanisms. Distinct scaling factors at the land surface and regional atmospheric levels are unraveled using this Bayesian network model. Land surface scaling explains the variability in flood duration as a function of cumulative exceedance index, a new measure that represents the evolution of the flood in the basin. Dynamic atmospheric scaling explains the cumulative exceedance index using the interaction between atmospheric blocking system and the synergistic model of wind divergence and atmospheric water vapor. Our findings underline that the synergy between a large persistent low-pressure blocking system and a higher rate of divergent wind often triggers a long-duration flood, even in the presence of moderate moisture supply in the atmosphere. This condition in turn causes an extremely long-duration flood if the basin-wide cumulative flow prior to the flood event was already high. Thus, this new land-atmospheric interaction framework integrates regional flood duration scaling and dynamic atmospheric scaling to enable the coupling of ‘horizontal’ (for example, streamflow accumulation inside the basin) and ‘vertical’ flow of information (for example, interrelated land and ocean-atmosphere interactions), providing an improved understanding of the critical forcing of regional hydroclimatic systems. This Bayesian model approach is applied to the Missouri River Basin, which has the largest system of reservoirs in the United States. Our predictive model can aid in decision support systems for the protection of national infrastructure against long-duration flood events.

Bayesian inference and predictive performance of soil respiration models in the presence of model discrepancy

Abstract. Bayesian inference of microbial soil respiration models is often based on the assumptions that the residuals are independent (i.e., no temporal or spatial correlation), identically distributed (i.e., Gaussian noise), and have constant variance (i.e., homoscedastic). In the presence of model discrepancy, as no model is perfect, this study shows that these assumptions are generally invalid in soil respiration modeling such that residuals have high temporal correlation, an increasing variance with increasing magnitude of CO2 efflux, and non-Gaussian distribution. Relaxing these three assumptions stepwise results in eight data models. Data models are the basis of formulating likelihood functions of Bayesian inference. This study presents a systematic and comprehensive investigation of the impacts of data model selection on Bayesian inference and predictive performance. We use three mechanistic soil respiration models with different levels of model fidelity (i.e., model discrepancy) with respect to the number of carbon pools and the explicit representations of soil moisture controls on carbon degradation; therefore, we have different levels of model complexity with respect to the number of model parameters. The study shows that data models have substantial impacts on Bayesian inference and predictive performance of the soil respiration models such that the following points are true: (i) the level of complexity of the best model is generally justified by the cross-validation results for different data models; (ii) not accounting for heteroscedasticity and autocorrelation might not necessarily result in biased parameter estimates or predictions, but will definitely underestimate uncertainty; (iii) using a non-Gaussian data model improves the parameter estimates and the predictive performance; and (iv) accounting for autocorrelation only or joint inversion of correlation and heteroscedasticity can be problematic and requires special treatment. Although the conclusions of this study are empirical, the analysis may provide insights for selecting appropriate data models for soil respiration modeling.