Single Model Type Research Articles

Aggravation of global maize yield loss risk under various hot and dry scenarios using multiple types of prediction approaches

AbstractHigh temperature and drought are widely known to cause a reduction of crop yield, but the simultaneously occurring risks in major producing countries and the associated uncertainty across various climate change scenarios remain unclear at the global scale. Here, we evaluate global maize yield loss risk (i.e., the probability of yield reduction by over 10% relative to historical trend yield during 1981–2010) across 30 hot and dry scenarios using regression, machine learning and process‐based models. Besides examining yield loss risk in a single country, we predict the potential risks simultaneously occurring in the top two and top ten producing countries. The three approaches agree on the aggravation of yield loss risk under dry and hot scenarios, but show large discrepancy in the magnitude and sensitivities. Specifically, 2°C warming alone could lead to a global yield loss risk of 73%, 100% and 62% based on regression, long‐short term memory (LSTM) and process‐based models, respectively, and warming‐induced risks can be further aggravated by droughts especially in process models. Global yield loss by over 10% would even become the new norm (i.e., yield loss probability is 100%) when temperature increases by over 2°C in some models. Importantly, the probabilities of yield loss simultaneously occurring in the top two countries (i.e., United States and China) and top ten countries are unexpectedly high and could even become 100% under extreme hot and dry scenarios. Our results highlight the large risks that future climate change may bring to multiple exporting and importing countries simultaneously, thus threating global food market and security. We also emphasize the important value of using different types of prediction approaches for yield projection under hot and dry scenarios, which enables more realistic estimation of uncertainty range than a single type of model.

Bayesian model averaging to improve the yield prediction in wheat breeding trials

Accurate pre-harvest prediction of wheat yield through secondary traits helps to facilitate plant breeding and reduce costs. Machine learning (ML) algorithms are increasingly applied to grain yield with remote sensing data. However, the performance of individual ML algorithms varies for different species in different environments due to different sources of uncertainty. This study proposed a novel wheat yield prediction framework based on canopy hyperspectral reflectance (350–2500 nm) and adopted the ensemble Bayesian model averaging (EBMA) method to improve model performance. To develop the yield prediction models, important bands extracted by the Boruta feature selection method were fed into four linear ML models and four nonlinear ML models. Meanwhile, Bayesian model averaging (BMA) weights obtained based on model cross-validation performance were used to combine the predictions of individual ML models. Compared to the best-performing individual model, the EBMA models obtained a weak accuracy improvement by integrating only the linear models or the nonlinear models. Additionally, the integration of two linear models and two non-linear models simultaneously was analyzed. Results indicate that most EBMA combinations of mixed linear and non-linear models achieved higher prediction accuracy than those integrating a single type of model and the best-performing individual model. The advantage of the EBMA method is that it produces a prediction distribution that reflects the uncertainty associated with deterministic predictions. With full consideration of the model diversity of ensemble members, the EBMA modeling framework provides an alternative method for predicting grain yield in plant breeding trials.

Disentangling the separate and confounding effects of temperature and precipitation on global maize yield using machine learning, statistical and process crop models

Temperature impacts on crop yield are known to be dependent on concurrent precipitation conditions and vice versa. To date, their confounding effects, as well as the associated uncertainties, are not well quantified at the global scale. Here, we disentangle the separate and confounding effects of temperature and precipitation on global maize yield under 25 climate scenarios. Instead of relying on a single type of crop model, as pursued in most previous impact assessments, we utilize machine learning, statistical and process-based crop models in a novel approach that allows for reasonable inter-method comparisons and uncertainty quantifications. Through controlling precipitation, an increase in warming of 1 °C could cause a global yield loss of 6.88%, 4.86% or 5.61% according to polynomial regression, long short-term memory (LSTM) and process-based crop models, respectively. With a 10% increase in precipitation, such negative temperature effects could be mitigated by 3.98%, 1.05% or 3.10%, respectively. When temperature is fixed at the baseline level, a 10% increase in precipitation alone could lead to a global yield growth of 0.23%, 1.43% or 3.09% according to polynomial regression, LSTM and process-based crop models, respectively. Further analysis demonstrates substantial uncertainties in impact assessment across crop models, which show a larger discrepancy in predicting temperature impacts than precipitation effects. Overall, global-scale assessment is more uncertain under drier conditions than under wet conditions, while a diverse uncertainty pattern is found for the top ten maize producing countries. This study highlights the important role of climate interactions in regulating yield response to changes in a specific climate factor and emphasizes the value of using both machine learning, statistical and process crop models in a consistent manner for a more realistic estimate of uncertainty than would be provided by a single type of model.

Landslide Susceptibility Prediction Based on the Information Value-Logistic Regression Model and Geographic Information System

The landslide susceptibility prediction (LSP) is generally implemented using certain types of single models; however, certain drawbacks exist in the single models; e.g., it is difficult to clearly reflect the weights of landslide-related environmental factors if only the information value (IV) model is adopted. To overcome these limitations, this study proposes an IV-logistic regression (IV-LR) model for LSP. The landslides that occurred in the southern part of Chongyi County, China, are used as study cases. Nine environmental factors—elevation, slope, plane curvature, profile curvature, relief amplitude, distance to river, lithology, normalized difference vegetable index, and normalized difference built-up index—are adopted based on remote sensing and geographic information system. Certain landslide grid units and the same number of non-landslide grid units are used as the output variables of these models. The IV, LR, and IV-LR models are used to implement the LSP in the southern part of Chongyi County. The predicted landslides susceptibility in Chongyi County mostly occurred in areas with low elevations, close distance to rivers, carbonate lithology, low vegetation coverage rate, and densely populated areas. The results show that the prediction rate of the IV-LR model (80.4%) is higher than that of the LR model (76.8%), followed by the IV model (72.8%); they further demonstrate that the IV-LR model has its unique superiority and rationality compared with the IV and LR models.

Fast likelihood calculation for multivariate Gaussian phylogenetic models with shifts

Phylogenetic comparative methods (PCMs) have been used to study the evolution of quantitative traits in various groups of organisms, ranging from micro-organisms to animal and plant species. A common approach has been to assume a Gaussian phylogenetic model for the trait evolution along the tree, such as a branching Brownian motion (BM) or an Ornstein–Uhlenbeck (OU) process. Then, the parameters of the process have been inferred based on a given tree and trait data for the sampled species. At the heart of this inference lie multiple calculations of the model likelihood, that is, the probability density of the observed trait data, conditional on the model parameters and the tree. With the increasing availability of big phylogenetic trees, spanning hundreds to several thousand sampled species, this approach is facing a two-fold challenge. First, the assumption of a single Gaussian process governing the entire tree is not adequate in the presence of heterogeneous evolutionary forces acting in different parts of the tree. Second, big trees present a computational challenge, due to the time and memory complexity of the model likelihood calculation.Here, we explore a sub-family, denoted GLInv, of the Gaussian phylogenetic models, with the transition density exhibiting the properties that the expectation depends Linearly on the ancestral trait value and the variance is Invariant with respect to the ancestral value. We show that GLInv contains the vast majority of Gaussian models currently used in PCMs, while supporting an efficient (linear in the number of nodes) algorithm for the likelihood calculation. The algorithm supports scenarios with missing data, as well as different types of trees, including trees with polytomies and non-ultrametric trees. To account for the heterogeneity in the evolutionary forces, the algorithm supports models with “shifts” occurring at specific points in the tree. Such shifts can include changes in some or all parameters, as well as the type of the model, provided that the model remains within the GLInv family. This contrasts with most of the current implementations where, due to slow likelihood calculation, the shifts are restricted to specific parameters in a single type of model, such as the long-term selection optima of an OU process, assuming that all of its other parameters, such as evolutionary rate and selection strength, are global for the entire tree.We provide an implementation of this likelihood calculation algorithm in an accompanying R-package called PCMBase. The package has been designed as a generic library that can be integrated with existing or novel maximum likelihood or Bayesian inference tools.

Economic Modelling of Chronic Kidney Disease: A Systematic Literature Review to Inform Conceptual Model Design.

BackgroundChronic kidney disease (CKD) is a progressive condition that leads to irreversible damage to the kidneys and is associated with an increased incidence of cardiovascular events and mortality. As novel interventions become available, estimates of economic and clinical outcomes are needed to guide payer reimbursement decisions.ObjectiveThe aim of the present study was to systematically review published economic models that simulated long-term outcomes of kidney disease to inform cost-effectiveness evaluations of CKD treatments.MethodsThe review was conducted across four databases (MEDLINE, Embase, the Cochrane library and EconLit) and health technology assessment agency websites. Relevant information on each model was extracted. Transition and mortality rates were also extracted to assess the choice of model parameterisation on disease progression by simulating patient’s time with end-stage renal disease (ESRD) and time to ESRD/death. The incorporation of cardiovascular disease in a population with CKD was qualitatively assessed across identified models.ResultsThe search identified 101 models that met the criteria for inclusion. Models were classified into CKD models (n = 13), diabetes models with nephropathy (n = 48), ESRD-only models (n = 33) and cardiovascular models with CKD components (n = 7). Typically, published models utilised frameworks based on either (estimated or measured) glomerular filtration rate (GFR) or albuminuria, in line with clinical guideline recommendations for the diagnosis and monitoring of CKD. Generally, two core structures were identified, either a microsimulation model involving albuminuria or a Markov model utilising CKD stages and a linear GFR decline (although further variations on these model structures were also identified). Analysis of parameter variability in CKD disease progression suggested that mean time to ESRD/death was relatively consistent across model types (CKD models 28.2 years; diabetes models with nephropathy 24.6 years). When evaluating time with ESRD, CKD models predicted extended ESRD survival over diabetes models with nephropathy (mean time with ESRD 8.0 vs. 3.8 years).DiscussionThis review provides an overview of how CKD is typically modelled. While common frameworks were identified, model structure varied, and no single model type was used for the modelling of patients with CKD. In addition, many of the current methods did not explicitly consider patient heterogeneity or underlying disease aetiology, except for diabetes. However, the variability of individual patients’ GFR and albuminuria trajectories perhaps provides rationale for a model structure designed around the prediction of individual patients’ GFR trajectories. Frameworks of future CKD models should be informed and justified based on clinical rationale and availability of data to ensure validity of model results. In addition, further clinical and observational research is warranted to provide a better understanding of prognostic factors and data sources to improve economic modelling accuracy in CKD.Electronic supplementary materialThe online version of this article (10.1007/s40273-019-00835-z) contains supplementary material, which is available to authorized users.

Automatic generation of evolutionary hypotheses using mixed Gaussian phylogenetic models

Phylogenetic comparative methods are widely used to understand and quantify the evolution of phenotypic traits, based on phylogenetic trees and trait measurements of extant species. Such analyses depend crucially on the underlying model. Gaussian phylogenetic models like Brownian motion and Ornstein-Uhlenbeck processes are the workhorses of modeling continuous-trait evolution. However, these models fit poorly to big trees, because they neglect the heterogeneity of the evolutionary process in different lineages of the tree. Previous works have addressed this issue by introducing shifts in the evolutionary model occurring at inferred points in the tree. However, for computational reasons, in all current implementations, these shifts are "intramodel," meaning that they allow jumps in 1 or 2 model parameters, keeping all other parameters "global" for the entire tree. There is no biological reason to restrict a shift to a single model parameter or, even, to a single type of model. Mixed Gaussian phylogenetic models (MGPMs) incorporate the idea of jointly inferring different types of Gaussian models associated with different parts of the tree. Here, we propose an approximate maximum-likelihood method for fitting MGPMs to comparative data comprising possibly incomplete measurements for several traits from extant and extinct phylogenetically linked species. We applied the method to the largest published tree of mammal species with body- and brain-mass measurements, showing strong statistical support for an MGPM with 12 distinct evolutionary regimes. Based on this result, we state a hypothesis for the evolution of the brain-body-mass allometry over the past 160 million y.

LAmbDA: label ambiguous domain adaptation dataset integration reduces batch effects and improves subtype detection.

Rapid advances in single cell RNA sequencing (scRNA-seq) have produced higher-resolution cellular subtypes in multiple tissues and species. Methods are increasingly needed across datasets and species to (i) remove systematic biases, (ii) model multiple datasets with ambiguous labels and (iii) classify cells and map cell type labels. However, most methods only address one of these problems on broad cell types or simulated data using a single model type. It is also important to address higher-resolution cellular subtypes, subtype labels from multiple datasets, models trained on multiple datasets simultaneously and generalizability beyond a single model type. We developed a species- and dataset-independent transfer learning framework (LAmbDA) to train models on multiple datasets (even from different species) and applied our framework on simulated, pancreas and brain scRNA-seq experiments. These models mapped corresponding cell types between datasets with inconsistent cell subtype labels while simultaneously reducing batch effects. We achieved high accuracy in labeling cellular subtypes (weighted accuracy simulated 1 datasets: 90%; simulated 2 datasets: 94%; pancreas datasets: 88% and brain datasets: 66%) using LAmbDA Feedforward 1 Layer Neural Network with bagging. This method achieved higher weighted accuracy in labeling cellular subtypes than two other state-of-the-art methods, scmap and CaSTLe in brain (66% versus 60% and 32%). Furthermore, it achieved better performance in correctly predicting ambiguous cellular subtype labels across datasets in 88% of test cases compared with CaSTLe (63%), scmap (50%) and MetaNeighbor (50%). LAmbDA is model- and dataset-independent and generalizable to diverse data types representing an advance in biocomputing. github.com/tsteelejohnson91/LAmbDA. Supplementary data are available at Bioinformatics online.

Reconstructing forest-cover change in the Swiss Alps between 1880 and 2010 using ensemble modelling

Understanding drivers of forest-cover change is essential for a broad range of ecosystem properties. In this work, we assessed changes in forest cover using physical, climatic and socio-economic drivers, including forest neighbourhood effects representing spatial interactions within and between two time periods (1880–1940; 1940–2010), for a mountainous study area located in eastern Switzerland. The robust assessment relied on an ensemble modelling approach that combined projections of six statistical models (GAM, CART, ANN, RF, GBM, GLM). A generic neighbourhood variable (distance to forest edge) explained most of the forest-cover change (variable importance of >90%) for both forest gain and loss, whereas socio-economic, physical and climatic variables were of least importance. The performance of models of forest loss was consistently higher and less varied (TSS of model ensembles 0.65–0.82) compared to models of forest gain (TSS of model ensembles 0.2–0.62) independent of the time period considered. We concluded that (a) the relative importance of drivers for the simulated processes is independent of the time period considered; (b) ensemble modelling proves a powerful tool to assess projection robustness by considering a suite of models rather than a single model type and (c) the inclusion of generic neighbourhood variables such as distance to forest edge improves model performance significantly.

Assessing coastal species distribution models through the integration of terrestrial, oceanic and atmospheric data

Aim When considering multiple species in distribution models, environmental variables should be selected that describe the group of species' environmental requirements. This can, however, be challenging in locations such as coastal areas, where different species may respond to terrestrial, oceanic and/or atmospheric conditions. Here, we evaluate the use of remotely sensed (RS) terrestrial, oceanic and interpolated climate variables, as well as a more detailed shore-zone data set, as a means of modelling the distributions of coastal bird species with diverse habitat requirements. Location Coastal British Columbia, Canada. Methods Boosted regression trees were used to model the distributions of 60 species of coastal birds using each environmental variable group, run individually and in combination. Models were assessed for their predictive ability and model fit, as well as for model overfitting. Results Models that incorporated terrestrial data were found to produce the highest model fit and predictive ability of the broad-scale environmental groups. Incorporating the fine-scale shore-zone data offered little improvement, as did selecting the best model in terms of predictive ability by species. Model fit and predictive ability were also found to vary by functional feeding group, with insectivores and benthivores being the best-modelled. Main conclusions Having access to both broad-scale RS environmental data and more detailed coastal shore-zone data produced modest improvements over employing RS environmental data alone. Models using only the terrestrial data set, however, performed similarly to the best single model type (terrestrial + shore-zone), indicating that broad-scale environmental data also offer an effective means of estimating coastal bird distributions. Testing multiple environmental variable groups in different combinations and selecting the best model allowed models to be optimized by species; conversely, the results of a ‘one model fits all’ approach were comparable to those of the best models, indicating that a single-model approach is also valid.

Population- versus Cohort-Based Modelling Approaches

While no single type of model can provide adequate answers under all circumstances, any modelling endeavour should incorporate three fundamental considerations in any decision-making question: the target population, the disease and the intervention characteristics. A target population is likely to be characterized by various types of heterogeneity and a dynamic evolution over time. It is therefore important to adequately capture these population effects on the results of a model. There are essentially two different approaches in modelling a population over time: a cohort-based approach and a population-based approach. In a cohort-based model, a closed group of individuals who have at least one specific characteristic or experience in common over a defined period of time is run through a state transition process. The cohort is generally composed of a hypothetical number of representative or 'average' individuals (i.e. the target population is considered to be a homogeneous group). The population-based approach projects the evolution of the estimated prevalent target population and intends to reflect as much as possible the demographic, epidemiological and clinical characteristics of the prevalent target population relevant for the decision problem. A cohort-based approach is generally used in most published healthcare decision models. However, this choice is rarely discussed by modellers. In this article, we challenge this assumption. To address the underlying decision problem, we affirm it is crucial that modellers consider the characteristics of the target population. Then, they could opt for using the most appropriate approach. Decision makers should also understand the impact on the results of both types of models in order to make informed healthcare decisions.

A Combined Forecast Method Integrating Contextual Knowledge

In the framework of TEI@I methodology, this paper proposes a combined forecast method integrating contextual knowledge (CFMIK). With the help of contextual knowledge, this method considers the effects of those factors that cannot be explicitly included in the forecast model, and thus it can efficiently decrease the forecast error resulted from the irregular events. Through a container throughput forecast case, this paper compares the performance of CFMIK, AFTER (a combined forecast method) and 3 types of single models (ARIMA, BP-ANN, exponential smoothing). The results show that the performance of CFMIK is better than that of the others.

TWO METHODS FOR TREE VOLUME ESTIMATION OF Acacia melanoxylon IN PORTUGAL

AbstractTwo models were developed to evaluate total volume under and over-bark of Blackwood (Acacia melanoxylon R. Br.) stand trees. By variance analysis regarding volume with and without bark it was confirmed that trees age is responsible for 48 to 50% of its volumetric behaviour. Considered the averages of those variables and regarding the three homogeneous groups of the studied sites, a single model type was used (v=β1dβ2hβ3) for estimating volumes. The critical errors are between 72 and 195 dm3 under-bark, and 82 and 212dm3 over-bark. The model that uses sites as dummy variables to the estimation of stand tree volumes is the same, and this method performs maximum expected errors with values of 155 dm3 (under-bark) and 169dm3 (over-bark), to mean maximum volumes of 1362 and 1459 dm3 respectively.The use of dummy variables is a more expedite and rapid modelling method, and its maximum expected errors perform lower values.Keywords: Blackwood; tree volume; modelling; ANOVA; dummy variables. ResumoDois métodos para estimar o volume em pé de árvores de Acacia melanoxylon em Portugal. Foram construídos dois modelos para estimar o volume sobre e sob casca de árvores em pé da acácia austrália (Acacia melanoxylon R. Br.). Por análise de variância do volume com e sem casca, foi confirmado que a idade das árvores é responsável por 48 a 50% do seu comportamento volumétrico. Consideradas as médias dessas variáveis e tendo em conta os três grupos homogêneos das “estações” estudadas, foi usado um único modelo tipo (v = β1dβ2hβ3) para estimar volumes. Os erros críticos situam-se entre 72 e 195 dm3 sobre casca, e 82 e 212 dm3 sob casca. O modelo que usa as “estações” como variáveis dummy para estimar volumes de árvores em pé é o mesmo e gera erros máximos esperados com valores de 155 dm3 (sob casca) e de 169 dm3 (sobre casca), para volumes máximos médios de 1362 e 1459 dm3, respectivamente. A utilização de variáveis dummy é um método de modelação rápido e mais expedito, e os seus valores de erros máximos esperados são inferiores.Palavras-chave: Austrália; volume de árvore; modelação; ANOVA; variáveis dummy.

ARMAsel for Detection and Correction of Outliers in Univariate Stochastic Data

For stationary random data, an automatic estimation algorithm can now select a time series model with a spectral accuracy close to the Cramer-Rao lower bound. The parameters of that selected time series model accurately represent the spectral density and the autocovariance function of the data. That is all the possible information for Gaussian data, as well as the most important information for arbitrarily distributed data. A single model type and order is selected from many candidate time series models by looking for the smallest prediction error. The single selected model precisely includes only the statistically significant details that are present in the data. The residuals of the automatically selected time series model reveal the location of outliers or other irregularities that may not be visible in the measured signal. The program requires no user interaction and can be incorporated into automatic measurement instruments and protocols.

Combining Microsimulation with CGE and Macro Modelling for Distributional Analysis in Developing and Transition Countries

This paper overviews recent work that has attempted to bring together microsimulation, Computable General Equilibrium (CGE) and macro models to perform distributional analysis in developing and transition countries. Particular attention is paid to applications relating to aspects of economic growth and political economy. Applications in which macro, CGE and microsimulation models are either layered or integrated are considered. It is demonstrated that different combinations of such models, including those where only a single model-type is used, are appropriate for different problems. For short- run impact analysis, microsimulation on its own may be appropriate. For longer-run analyses, where interest is in the interrelationship between changes in disposable income, consumption and labour supply, these models need to be supplemented a combination of microsimulation on the one hand, and general equilibrium price changes or changes in macro variables on the other hand. In the case of national subregions, or countries embedded in free-trade areas, it is argued that microsimulation may adequately be combined with pure macro models. That is, CGE modelling may not be necessary. For distinct national economies, however, the first step beyond microsimulation should likely be integration with CGE modelling. Whilst much promising work has been undertaken on dynamic integrated CGE microsimulation work in developing countries, CGE work is most advanced for Less Developed Countries. At present several groups of development researchers are found to be putting these two approaches together, and in some cases are adding macroeconomic and financial modelling as well. In contrast, with a few conspicuous exceptions, little such work is being done for the transition economies.

FROM DATABASE TO OPERATION DESIGN USING A MULTIPLE MODEL APPROACH

We address the problem of estimating the input conditions (mixture compositions, operating temperature, pressure, etc.) that will result in manufacturing a product with the required properties using a data-based approach. A single-model type paradigm is shown to be unsuitable for all data sets or even for all output variables in a particular data set. A multiple-model type paradigm is therefore advocated. The input-output map is first obtained using one or more of the several modeling approaches. These models are then inverted to suggest operational strategies that will meet production goals. Several examples are presented to demonstrate the efficacy of the proposed methodology.

Downscaling heavy precipitation over the United Kingdom: a comparison of dynamical and statistical methods and their future scenarios

AbstractSix statistical and two dynamical downscaling models were compared with regard to their ability to downscale seven seasonal indices of heavy precipitation for two station networks in northwest and southeast England. The skill among the eight downscaling models was high for those indices and seasons that had greater spatial coherence. Generally, winter showed the highest downscaling skill and summer the lowest. The rainfall indices that were indicative of rainfall occurrence were better modelled than those indicative of intensity. Models based on non‐linear artificial neural networks were found to be the best at modelling the inter‐annual variability of the indices; however, their strong negative biases implied a tendency to underestimate extremes. A novel approach used in one of the neural network models to output the rainfall probability and the gamma distribution scale and shape parameters for each day meant that resampling methods could be used to circumvent the underestimation of extremes. Six of the models were applied to the Hadley Centre global circulation model HadAM3P forced by emissions according to two SRES scenarios. This revealed that the inter‐model differences between the future changes in the downscaled precipitation indices were at least as large as the differences between the emission scenarios for a single model. This implies caution when interpreting the output from a single model or a single type of model (e.g. regional climate models) and the advantage of including as many different types of downscaling models, global models and emission scenarios as possible when developing climate‐change projections at the local scale. Copyright © 2006 Royal Meteorological Society