Model Ensemble Approach Research Articles

Tropospheric Response to Gulf Stream Intrinsic Variability: A Model Ensemble Approach

AbstractThe Gulf Stream's (GS) impact on the marine boundary layer (MBL) is well established, yet the mechanisms and timescales through which it affects the upper‐troposphere and contributes to precipitation are debatable. Using a high‐resolution regional atmospheric model, we shed light on the impact of ocean intrinsic variability (OIV) along GS on midlatitude‐atmosphere. Taking advantage of a 24‐member ensemble of ocean model integrations, we devised a novel experimental setup where the same weather system feels different realizations of GS sea surface temperature (SST). We introduce the “Eddy Recharge‐Frontal Lift” (ERFL) mechanism, highlighting the joint importance of synoptic variability and boundary layer processes. ERFL mechanism proposes that OIV recharges/discharges MBL with moisture and heat, while convergence associated with passing atmospheric‐fronts uplifts these MBL‐trapped anomalies to upper‐troposphere and imprints on precipitation in surprisingly short periods (a month). The impact of OIV on precipitation depends on the background mean SST.

Generalizable disease detection using model ensemble on chest X-ray images

In the realm of healthcare, the demand for swift and precise diagnostic tools has been steadily increasing. This study delves into a comprehensive performance analysis of three pre-trained convolutional neural network (CNN) architectures: ResNet50, DenseNet121, and Inception-ResNet-v2. To ensure the broad applicability of our approach, we curated a large-scale dataset comprising a diverse collection of chest X-ray images, that included both positive and negative cases of COVID-19. The models’ performance was evaluated using separate datasets for internal validation (from the same source as the training images) and external validation (from different sources). Our examination uncovered a significant drop in network efficacy, registering a 10.66% reduction for ResNet50, a 36.33% decline for DenseNet121, and a 19.55% decrease for Inception-ResNet-v2 in terms of accuracy. Best results were obtained with DenseNet121 achieving the highest accuracy at 96.71% in internal validation and Inception-ResNet-v2 attaining 76.70% accuracy in external validation. Furthermore, we introduced a model ensemble approach aimed at improving network performance when making inferences on images from diverse sources beyond their training data. The proposed method uses uncertainty-based weighting by calculating the entropy in order to assign appropriate weights to the outputs of each network. Our results showcase the effectiveness of the ensemble method in enhancing accuracy up to 97.38% for internal validation and 81.18% for external validation, while maintaining a balanced ability to detect both positive and negative cases.

Assessing the accuracy of OpenET satellite-based evapotranspiration data to support water resource and land management applications

Remotely sensed evapotranspiration (ET) data offer strong potential to support data-driven approaches for sustainable water management. However, practitioners require robust and rigorous accuracy assessments of such data. The OpenET system, which includes an ensemble of six remote sensing models, was developed to increase access to field-scale (30 m) ET data for the contiguous United States. Here we compare OpenET outputs against data from 152 in situ stations, primarily eddy covariance flux towers, deployed across the contiguous United States. Mean absolute error at cropland sites for the OpenET ensemble value is 15.8 mm per month (17% of mean observed ET), mean bias error is −5.3 mm per month (6%) and r2 is 0.9. Results for shrublands and forested sites show higher inter-model variability and lower accuracy relative to croplands. High accuracy and multi-model convergence across croplands demonstrate the utility of a model ensemble approach, and enhance confidence among ET data practitioners, including the agricultural water resource management community.

PANACEA: a neural model ensemble for cyber-threat detection

Ensemble learning is a strategy commonly used to fuse different base models by creating a model ensemble that is expected more accurate on unseen data than the base models. This study describes a new cyber-threat detection method, called PANACEA, that uses ensemble learning coupled with adversarial training in deep learning, in order to gain accuracy with neural models trained in cybersecurity problems. The selection of the base models is one of the main challenges to handle, in order to train accurate ensembles. This study describes a model ensemble pruning approach based on eXplainable AI (XAI) to increase the ensemble diversity and gain accuracy in ensemble classification. We base on the idea that being able to identify base models that give relevance to different input feature sub-spaces may help in improving the accuracy of an ensemble trained to recognise different signatures of different cyber-attack patterns. To this purpose, we use a global XAI technique to measure the ensemble model diversity with respect to the effect of the input features on the accuracy of the base neural models combined in the ensemble. Experiments carried out on four benchmark cybersecurity datasets (three network intrusion detection datasets and one malware detection dataset) show the beneficial effects of the proposed combination of adversarial training, ensemble learning and XAI on the accuracy of multi-class classifications of cyber-data achieved by the neural model ensemble.

Surface Roughness Prediction in Hard Turning (Finishing) of 16MnCr5 Using a Model Ensemble Approach

This paper investigates the predictive quality in finishing turning processes of 16MnCr5, specifically focusing on the use of model ensemble methods to predict surface roughness. Surface roughness plays a critical role in determining the quality and functionality of machined components. The objective of this study is to develop an accurate and robust predictive model for surface roughness after finishing turning operations. The proposed approach employs a model ensemble method, which combines multiple predictive models to enhance the accuracy and reliability of the predictions. The experimental setup includes collecting data using various types of in-situ sensors along with corresponding surface roughness measurements. The collected dataset is used to train and test the ensemble model, which integrates different machine learning algorithms and statistical techniques. The results demonstrate that the model ensemble method yields superior predictive performance compared to individual models, effectively capturing the complex relationships between process signals and surface roughness.

Research on application of model ensemble in sports image classification based on environmental information

In recent years, a deep convolutional neural network has been widely used in the field of image classification. However, training a satisfactory network is very arduous, not only to tune the hyper-parameters in the network but also to avoid the overfitting problem caused by a deep neural network. Another point is that it is difficult for a neural network to learn subtle details without human annotation. Therefore this paper proposed a preferable classification algorithm for sports classification tasks that combines the deep neural network with the object detection algorithm to obtain the prediction result. In this paper, the author compared the differences between classification directly using neural networks and a modified model ensemble approach to classification from a holistic perspective, as well as elaborating the advantages and disadvantages of the two approaches. The conclusion shows that the use of the improved model ensemble classification algorithm performs better than the direct use of neural networks and also achieves a high degree of accuracy in the test set.

Intergenerational actuarial fairness when longevity increases: Amending the retirement age

Continuous longevity improvements and population ageing have led countries to modify national public pension schemes by increasing standard and early retirement ages in a discretionary, scheduled, or automatic way, and making it harder for people to retire prematurely. To this end, countries have adopted alternative retirement age strategies, but our analyses show that the measures taken are often poorly designed and consequently misaligned with the pension scheme's ultimate goals. This paper discusses how to implement automatic indexation of the retirement age to life expectancy developments while respecting the principles of intergenerational actuarial fairness and neutrality among generations of the respective policy scheme design. With stable demographic conditions, we show in policy designs in which extended working lives translate into additional pension entitlements, the pension age must be automatically updated to keep the period in retirement constant. Alternatively, policy designs that pursue a fixed replacement rate are consistent with retirement age policies targeting a constant balance between active years in the workforce and years in retirement. Under conditions of population ageing, the statutory pension age will have to increase at a faster rate to meet the intergenerational equity criteria. The empirical strategy employed a Bayesian Model Ensemble approach to stochastic mortality modelling to address model risk and generate forecasts of intergenerationally and actuarially fair pension ages for 23 countries from 2000 to 2050. The findings show that the pension age increases needed to accommodate the effect of longevity developments on pay-as-you-go equilibrium and to reinstate equity between generations are sizeable and well beyond those employed and/or legislated in most countries. A new wave of pension reforms may be at the doorsteps.

Quantifying biological carbon pump pathways with a data-constrained mechanistic model ensemble approach

Abstract. The ability to constrain the mechanisms that transport organic carbon into the deep ocean is complicated by the multiple physical, chemical, and ecological processes that intersect to create, transform, and transport particles in the ocean. In this paper we develop and parameterize a data-assimilative model of the multiple pathways of the biological carbon pump (NEMUROBCP). The mechanistic model is designed to represent sinking particle flux, active transport by vertically migrating zooplankton, and passive transport by subduction and vertical mixing, while also explicitly representing multiple biological and chemical properties measured directly in the field (including nutrients, phytoplankton and zooplankton taxa, carbon dioxide and oxygen, nitrogen isotopes, and 234Thorium). Using 30 different data types (including standing stock and rate measurements related to nutrients, phytoplankton, zooplankton, and non-living organic matter) from Lagrangian experiments conducted on 11 cruises from four ocean regions, we conduct an objective statistical parameterization of the model and generate 1 million different potential parameter sets that are used for ensemble model simulations. The model simulates in situ parameters that were assimilated (net primary production and gravitational particle flux) and parameters that were withheld (234Thorium and nitrogen isotopes) with reasonable accuracy. Model results show that gravitational flux of sinking particles and vertical mixing of organic matter from the euphotic zone are more important biological pump pathways than active transport by vertically migrating zooplankton. However, these processes are regionally variable, with sinking particles most important in oligotrophic areas of the Gulf of Mexico and California Current, sinking particles and vertical mixing roughly equivalent in productive coastal upwelling regions and the subtropical front in the Southern Ocean, and active transport an important contributor in the eastern tropical Pacific. We further find that mortality at depth is an important component of active transport when mesozooplankton biomass is high, but it is negligible in regions with low mesozooplankton biomass. Our results also highlight the high degree of uncertainty, particularly amongst mesozooplankton functional groups, that is derived from uncertainty in model parameters. Indeed, variability in BCP pathways between simulations for a specific location using different parameter sets (all with approximately equal misfit relative to observations) is comparable to variability in BCP pathways between regions. We discuss the implications of these results for other data-assimilation approaches and for studies that rely on non-ensemble model outputs.

A trait-based model ensemble approach to design rice plant types for future climate.

Crop models are powerful tools to support breeding because of their capability to explore genotype×environment×management interactions that can help design promising plant types under climate change. However, relationships between plant traits and model parameters are often model specific and not necessarily direct, depending on how models formulate plant morphological and physiological features. This hinders model application in plant breeding. We developed a novel trait-based multi-model ensemble approach to improve the design of rice plant types for future climate projections. We conducted multi-model simulations targeting enhanced productivity, and aggregated results into model-ensemble sets of phenotypic traits as defined by breeders rather than by model parameters. This allowed to overcome the limitations due to ambiguities in trait-parameter mapping from single modelling approaches. Breeders' knowledge and perspective were integrated to provide clear mapping from designed plant types to breeding traits. Nine crop models from the AgMIP-Rice Project and sensitivity analysis techniques were used to explore trait responses under different climate and management scenarios at four sites. The method demonstrated the potential of yield improvement that ranged from 15.8% to 41.5% compared to the current cultivars under mid-century climate projections. These results highlight the primary role of phenological traits to improve crop adaptation to climate change, as well as traits involved with canopy development and structure. The variability of plant types derived with different models supported model ensembles to handle related uncertainty. Nevertheless, the models agreed in capturing the effect of the heterogeneity in climate conditions across sites on key traits, highlighting the need for context-specific breeding programmes to improve crop adaptation to climate change. Although further improvement is needed for crop models to fully support breeding programmes, a trait-based ensemble approach represents a major step towards the integration of crop modelling and breeding to address climate change challenges and develop adaptation options.

Pricing participating longevity-linked life annuities: a Bayesian Model Ensemble approach

Participating longevity-linked life annuities (PLLA) in which benefits are updated periodically based on the observed survival experience of a given underlying population and the performance of the investment portfolio are an alternative insurance product offering consumers individual longevity risk protection and the chance to profit from the upside potential of financial market developments. This paper builds on previous research on the design and pricing of PLLAs by considering a Bayesian Model Ensemble of single population generalised age-period-cohort stochastic mortality models in which individual forecasts are weighted by their posterior model probabilities. For the valuation, we adopt a longevity option decomposition approach with risk-neutral simulation and investigate the sensitivity of results to changes in the asset allocation by considering a more aggressive lifecycle strategy. We calibrate models using Taiwanese (mortality, yield curve and stock market) data from 1980 to 2019. The empirical results provide significant valuation and policy insights for the provision of a cost effective and efficient risk pooling mechanism that addresses the individual uncertainty of death, while providing appropriate retirement income and longevity protection.

Uncertainty in East Antarctic Firn Thickness Constrained Using a Model Ensemble Approach

AbstractMass balance assessments of the East Antarctic ice sheet (EAIS) are highly sensitive to changes in firn thickness, causing substantial disagreement in estimates of its contribution to sea‐level. To better constrain the uncertainty in recent firn thickness changes, we develop an ensemble of 54 model scenarios of firn evolution between 1992 and 2017. Using statistical emulation of firn‐densification models, we quantify the impact of firn compaction formulation, differing climatic forcing, and surface snow density on firn thickness evolution. At basin scales, the ensemble uncertainty in firn thickness change ranges between 0.2 and 1.0 cm yr−1 (15%–300% relative uncertainty), with the choice of climate forcing having the largest influence on the spread. Our results show the regions of the ice sheet where unexplained discrepancies exist between observed elevation changes and an extensive set of modeled firn thickness changes estimates, marking an important step toward more accurately constraining ice sheet mass balance.

A Model Ensemble Approach for Few-Shot Learning Using Aggregated Classifiers

A Model Ensemble Approach for Few-Shot Learning Using Aggregated Classifiers

A model ensemble approach to determine the humus building efficiency of organic amendments in incubation experiments

AbstractOrganic amendments are important to sustain soil organic matter (SOM) and soil functions in agricultural soils. Information about the contribution of organic amendments to SOM can be derived from incubation experiments. In this study, data from 72 incubated organic amendments including plant residues, digestates and manure were analysed. The incubation data was compiled from three experimental setups with varying incubation times, soils and incubation temperatures, in which CO2 release was measured continuously. The analysis of the incubation data was performed with an approach relying on conceptual parts of C‐TOOL, CCB, Century, ICBM, RothC and Yasso which are all well‐approved first‐order carbon models that differ in structure and abstraction level. All models are an approximation of reality, whereby each model differs in understanding of the processes involved in soil carbon dynamics. To accumulate the advantages from each model a model ensemble was performed for each substrate. With the ability of each carbon model to compute the distribution of carbon into specific SOM pools a new approach for evaluating organic amendments in terms of humus building efficiency is presented that, depends on the weighted model fit of each ensemble member. Depending on the organic substrate added to the soil, the time course of CO2 release in the incubation studies was predicted with different accuracy by the individual model concepts. Averaging the output of the individual models leads to more robust prediction of SOM dynamics. The EHUM value is easy to interpret and the results are in accordance with the literature.

Conceptual uncertainties in groundwater and porewater fluxes estimated by radon and radium mass balances

AbstractRadium isotopes and radon are routinely used as tracers to quantify groundwater and porewater fluxes into coastal and freshwater systems. However, uncertainties associated with the determination of the tracer flux are often poorly addressed and often neglect all the potential errors associated with the conceptualization of the system (i.e., conceptual uncertainties). In this study, we assess the magnitude of some of the key uncertainties related to the determination of the radium and radon inputs supplied by groundwater and porewater fluxes into a waterbody (La Palme Lagoon, France). This uncertainty assessment is addressed through a single model ensemble approach, where a tracer mass balance is run multiple times with variable sets of assumptions and approaches for the key parameters determined through a sensitivity test. In particular, conceptual uncertainties linked to tracer concentration, diffusive fluxes, radon evasion to the atmosphere, and change of tracer inventory over time were considered. The magnitude of porewater fluxes is further constrained using a comparison of independent methods: (1) 224Ra and (2) 222Rn mass balances in overlying waters, (3) a model of 222Rn deficit in sediments, and (4) a fluid‐salt numerical transport model. We demonstrate that conceptual uncertainties are commonly a major source of uncertainty on the estimation of groundwater or porewater fluxes and they need to be taken into account when using tracer mass balances. In the absence of a general framework for assessing these uncertainties, this study provides a practical approach to evaluate key uncertainties associated to radon and radium mass balances.

Intergenerational Actuarial Fairness When Longevity Increases: Amending the Retirement Age

Continuous longevity improvements and population ageing have led countries to modify national public pension schemes by increasing the standard and early retirement ages in a discretionary, scheduled, or automatic way, and by making it harder for people to retire prematurely. To this end, countries have adopted alternative retirement age strategies, but our analyses show that the measures taken are often poorly designed and consequently misaligned with the pension scheme’s ultimate goals. In addition, our analyses demonstrate that countries risk falling short of their goals given their use of projection methods that underestimate life expectancy. This paper discusses how to implement automatic indexation of the retirement age to life expectancy developments while respecting the principles of intergenerational actuarial fairness and neutrality among generations. We show that in policy designs in which extended working lives translate into additional pension entitlements, the pension age must be automatically updated to keep the period in retirement constant. Alternatively, policy designs that pursue a fixed replacement rate are consistent with retirement age policies targeting a constant balance between active years in the workforce and years in retirement. The empirical strategy employed to project the relevant cohort life expectancy uses a Bayesian Model Ensemble approach to stochastic mortality modelling to generate forecasts of intergenerationally and actuarially fair pension ages for 23 countries and regions from 2000 to 2050. The empirical results show that the pension age increases needed to accommodate the effect of longevity developments on pay-as-you-go equilibrium and to reinstate equity between generations are sizeable and well beyond those employed and/or legislated in most countries. A new wave of pension reforms may be at the doorsteps.

Elimination of Multicollinearity in Continuous Non-Invasive Blood Pressure Approximation by Information Criterion

Continuous and unobtrusive method of measuring blood pressure has been gaining more attention in the healthcare community. With the application of data analysis techniques in biosignals like Electrocardiogram (ECG) and Photoplethysmogram (PPG), several predictors are obtained that correlates well with the blood pressure. But the BP approximation regression models formed using these predictors suffers from multicollinearity (higher correlation between predictors). The article proposes the use of information criterion-based model ensemble approach to reduce the effect of multicollinearity in the continuous BP estimation. The study focuses on forming pool of candidate models from feature subsets. The best performing models are selected based on information criterion and combined to form the ensemble model. Experiments with performed with MIMIC-II dataset that consists of 104 records with simultaneously recorded PPG and arterial BP. The results suggest that the technique achieves Mean Absolute Error (MAE) of 5.81 mm Hg and 3.35 mm Hg for systolic and diastolic BP and Root Mean Square Error (RMSE) of 6.08 mm Hg and 4.12 mm Hg for systolic and diastolic BP respectively. The error measures conform to the standards set by American Association of Medical Instrumentation (AAMI). The method reveals that the ensemble model based on information criterion outperforms well compared to the usage of single model.

Assessing the sensitivity of three Alaska marine food webs to perturbations: an example of Ecosim simulations using Rpath

Ecosystem modelling is a useful tool for exploring the potential outcomes of policy options and conducting experiments that would otherwise be impractical in the real world. However, ecosystem models have been limited in their ability to engage in the management of living marine resources due in part to high levels of uncertainty in model parameters and model outputs. Additionally, for multispecies or food web models, there is uncertainty about the predator-prey functional response, which can have implications for population dynamics. In this study, we evaluate the sensitivity of large marine food webs in Alaska to parameter uncertainty, including parameters that govern the predator-prey functional response. We use Rpath, an R implementation of the food web modeling program Ecopath with Ecosim (EwE), to conduct a series of mortality-based perturbations to examine the sensitivity and recovery time of higher trophic level groups in the eastern Chukchi Sea, eastern Bering Sea, and Gulf of Alaska. We use a Monte Carlo approach to generate thousands of plausible ecosystems by drawing parameter sets from the range of uncertainty around the base model parameters. We subjected the ecosystem ensembles to a series of mortality-based perturbations to identify which functional groups the higher trophic level groups are most sensitive to when their mortality was increased, whether the food webs returned to their unperturbed configurations following a perturbation, and how long it took to return to that state. In all three ecosystems, we found that the number of disrupted ensemble food webs was positively related to the biomass and the number of trophic links of the perturbed functional group, and negatively related to trophic level. The eastern Chukchi Sea was most sensitive to perturbations to benthic invertebrate groups, the eastern Bering Sea was most sensitive to shrimp and walleye pollock, and the Gulf of Alaska was most sensitive to shrimps, pelagic forage fish, and zooplankton. Recovery time to perturbations were generally less than 5 years in all three ecosystems. The recovery times when fish groups were perturbed were generally longer than when benthic invertebrates were perturbed, and recovery times were shortest when it was pelagic invertebrates. The single model ensemble approach produced simulation results that described a range of possible outcomes to the prescribed perturbations and provided a sense for how robust the results are to parameter uncertainty.

Biomes as evolutionary arenas: Convergence and conservatism in the trans‐continental succulent biome

AbstractAimHistorically, biomes have been defined based on their structurally and functionally similar vegetation, but there is debate about whether these similarities are superficial, and about how biomes are defined and mapped. We propose that combined assessment of evolutionary convergence of plant functional traits and phylogenetic biome conservatism provides a useful approach for characterizing biomes. We focus on the little‐known succulent biome, a trans‐continentally distributed assemblage of succulent‐rich, drought‐deciduous, fire‐free forest, thicket and scrub vegetation as a useful exemplar biome to gain insights into these questions.LocationGlobal lowland (sub)tropics.Time periodPresent.Major taxa studiedAngiosperms.MethodsWe use a model ensemble approach to model the distribution of 884 species of stem succulents, a plant functional group representing a striking example of evolutionary convergence. Using this model, phylogenies, and species occurrence data, we quantify phylogenetic succulent biome conservatism for 10 non‐succulent trans‐continental plant clades including prominent elements of the succulent biome, representing over 800 species.ResultsThe geographical and climatic distributions of stem succulents provide an objective and quantitative proxy for mapping the distribution of the succulent biome. High fractions of succulent biome occupancy across continents suggest all 10 non‐succulent study clades are phylogenetically conserved within the succulent biome.Main conclusionsThe trans‐continental succulent and savanna biomes both show evolutionary convergence in key biome‐related plant functional traits. However, in contrast to the savanna biome, which was apparently assembled via repeated local recruitment of lineages via biome shifts from adjacent biomes within continents, the succulent biome forms a coherent trans‐continental evolutionary arena for drought‐adapted tropical biome conserved lineages. Recognizing the important functional differences between the succulent‐rich, grass‐poor, fire‐free succulent biome and the grass‐dominated, succulent‐poor, fire‐prone savanna biome, and defining them as distinct seasonally dry tropical biomes, occupying essentially non‐overlapping distributions, provides critical insights into tropical biodiversity and the extent of biome stasis versus biome shifting.

The Use of Data Mining Methods for the Prediction of Dementia: Evidence From the English Longitudinal Study of Aging.

Dementia in older age is a major health concern with the increase in the aging population. Preventive measures to prevent or delay dementia symptoms are of utmost importance. In this study, a large and wide variety of factors from multiple domains were investigated using a large nationally representative sample of older people from the English Longitudinal Study of Ageing. Seven machine learning algorithms were implemented to build predictive models for performance comparison. A simple model ensemble approach was used to combine the prediction results of individual base models to further improve predictive power. A series of important factors in each domain area were identified. The findings from this study provide new evidence on factors that are associated with the dementia in later life. This information will help our understanding of potential risk factors for dementia and identify warning signs of the early stages of dementia. Longitudinal research is required to establish which factors may be causative and which factors may be a consequence of dementia.

Estimation of Gridded Population and GDP Scenarios with Spatially Explicit Statistical Downscaling

This study downscales the population and gross domestic product (GDP) scenarios given under Shared Socioeconomic Pathways (SSPs) into 0.5-degree grids. Our downscale approach has the following features. (i) It explicitly considers spatial and socioeconomic interactions among cities, (ii) it utilizes auxiliary variables, including road network and land cover, (iii) it endogenously estimates the influence from each factor by a model ensemble approach, and (iv) it allows us to control urban shrinkage/dispersion depending on SSPs. It is confirmed that our downscaling results are consistent with scenario assumptions (e.g., concentration in SSP1 and dispersion in SSP3). Besides, while existing grid-level scenarios tend to have overly-smoothed population distributions in nonurban areas, ours does not suffer from the problem, and captures the difference in urban and nonurban areas in a more reasonable manner. Our gridded dataset, including population counts and gross productivities by 0.5 degree grids by 10 years, are available from http://www.cger.nies.go.jp/gcp/population-and-gdp.html.