Reduce Model Uncertainty Research Articles

Learning Robust LQ-Controllers Using Application Oriented Exploration

This letter concerns the problem of learning robust LQ-controllers, when the dynamics of the linear system are unknown. First, we propose a robust control synthesis method to minimize the worst-case LQ cost, with probability 1 - δ, given empirical observations of the system. Next, we propose an approximate dual controller that simultaneously regulates the system and reduces model uncertainty. The objective of the dual controller is to minimize the worst-case cost attained by a new robust controller, synthesized with the reduced model uncertainty. The dual controller is subject to an exploration budget in the sense that it has constraints on its worst-case cost with respect to the current model uncertainty. In our numerical experiments, we observe better performance of the proposed robust LQ regulator over the existing methods. Moreover, the dual control strategy gives promising results in comparison with the common greedy random exploration strategies.

Bayesian joint inversion of controlled source electromagnetic and magnetotelluric data to image freshwater aquifer offshore New Jersey

SUMMARY Joint inversion of multiple electromagnetic data sets, such as controlled source electromagnetic and magnetotelluric data, has the potential to significantly reduce uncertainty in the inverted electrical resistivity when the two data sets contain complementary information about the subsurface. However, evaluating quantitatively the model uncertainty reduction is made difficult by the fact that conventional inversion methods—using gradients and model regularization—typically produce just one model, with no associated estimate of model parameter uncertainty. Bayesian inverse methods can provide quantitative estimates of inverted model parameter uncertainty by generating an ensemble of models, sampled proportional to data fit. The resulting posterior distribution represents a combination of a priori assumptions about the model parameters and information contained in field data. Bayesian inversion is therefore able to quantify the impact of jointly inverting multiple data sets by using the statistical information contained in the posterior distribution. We illustrate, for synthetic data generated from a simple 1-D model, the shape of parameter space compatible with controlled source electromagnetic and magnetotelluric data, separately and jointly. We also demonstrate that when data sets contain complementary information about the model, the region of parameter space compatible with the joint data set is less than or equal to the intersection of the regions compatible with the individual data sets. We adapt a trans-dimensional Markov chain Monte Carlo algorithm for jointly inverting multiple electromagnetic data sets for 1-D earth models and apply it to surface-towed controlled source electromagnetic and magnetotelluric data collected offshore New Jersey, USA, to evaluate the extent of a low salinity aquifer within the continental shelf. Our inversion results identify a region of high resistivity of varying depth and thickness in the upper 500 m of the continental shelf, corroborating results from a previous study that used regularized, gradient-based inversion methods. We evaluate the joint model parameter uncertainty in comparison to the uncertainty obtained from the individual data sets and demonstrate quantitatively that joint inversion offers reduced uncertainty. In addition, we show how the Bayesian model ensemble can subsequently be used to derive uncertainty estimates of pore water salinity within the low salinity aquifer.

Effect of temperature variations on the travel time of infiltrating water in the Amsterdam Water Supply Dunes (the Netherlands)

Travel time is one of the important criteria in the design of managed aquifer recharge systems for securing good drinking water quality. Traditionally, groundwater travel time has been modelled without considering the effect of temperature. In this study, a cross-sectional heat transport model was constructed for the Amsterdam dune filtration system (in the Netherlands) to analyse the effect of temperature on groundwater travel times. A groundwater flow model, a chloride transport model, and a heat transport model were iteratively calibrated with measured groundwater levels, chloride concentrations, and temperature series in order to improve model calibration and reduce model uncertainty. The coupled flow and heat transport model with temperature-dependent density and viscosity provided more accurate estimation of travel times. The results show that seasonal temperature fluctuations in the source water in the infiltration pond cause temperature variations in the shallow groundwater. Viscosity is more sensitive to temperature changes and has a larger effect on groundwater travel times. Groundwater travel time in the shallow sand aquifer increases from 60 days when computed with the traditional groundwater flow model to 73 days in the winter season and 95 days in the summer season when computed with the coupled model. Longer travel time is beneficial for water quality improvement. Thus, it is important to consider the effect of temperature variations on groundwater travel times for the design and operation of managed aquifer recharge systems.

Data-driven multistratum designs with the generalized Bayesian [formula omitted]-[formula omitted] criterion for highly uncertain models

Multistratum designs have gained much attention recently. Most criteria, such as the D criterion, select multistratum designs based on a given model that is assumed to be true by the experimenters. However, when the true model is highly uncertain, the model used for selecting the optimal design can be seriously misspecified. If this is the case, then the selected multistratum design will be not efficient for fitting the true model. To deal with the problem of high uncertain models, we propose the generalized Bayesian D-D (GBDD) criterion, which selects multistratum designs based on the experimental data. Under the framework of multistratum structures, we develop theorems and formula that are used for conducting Bayesian analysis and extracting information about the true model from the data to reduce model uncertainty. The GBDD criterion is easy and flexible in use. We provide several examples to demonstrate how to construct the GBDD-optimal split-plot, strip-plot, and staggered-level designs. By comparing with the D-optimal designs and one-stage generalized Bayesian D-optimal designs, we show that the GBDD-optimal designs have higher efficiency on fitting the true models. The extensions of the GBDD criterion for more complicated cases, such as more than two stages of experiments and more than one class of potential terms, are also developed.

Multi‐objective calibration of RothC using measured carbon stocks and auxiliary data of a long‐term experiment in Switzerland

Interactions between model parameters and low spatiotemporal resolution of available data mean that conventional soil organic carbon (SOC) models are often affected by equifinality, with consequent uncertainty in SOC forecasts. Estimation of belowground C inputs is another major source of uncertainty in SOC modelling. Models are usually calibrated on SOC stocks and fluxes from long‐term experiments (LTEs), whereas other point data are not used for constraining the model parameters. We used data from an agricultural long‐term (>65 years) fertilization experiment to test a multi‐objective parameter estimation approach on the RothC model, combining SOC data from different fertilization treatments with microbial biomass, basal respiration and Zimmermann's fractions data. We also compared two methods to estimate the belowground C inputs: a conventional scaling of belowground biomass from crop harvest yield and an alternative approach based on constant belowground C for cereals measured experimentally in the field. The resulting posterior parameter distributions still suffered from some equifinality; the most stable C pool kinetic constants and composition of exogenous organic matter were the most sensitive parameters. The use of fixed belowground C inputs for cereals improved the model performance, reducing the importance of treatment‐specific parameters and processes. The introduction of microbial biomass and basal respiration data was effective for increasing determination of the calibration, but also suggested a change in the model structure: the microbial biomass pool, which is proportional to the C inputs in the traditional models, could be represented by different microbial physiology functions.Highlights Multi‐objective calibration with SOC and microbial or soil fractionation data can reduce model uncertainty. We compared different methods to estimate belowground C inputs on a long‐term trial in Switzerland. Fixed belowground C inputs measured in the field gave the best model performance. Microbial data can improve model calibration, but a change in model structure is suggested.

Effects of Human Activities on Hydrological Components in the Yiluo River Basin in Middle Yellow River

Land use and land cover change (LUCC) and water resource utilization behavior and policy (WRUBAP) affect the hydrological cycle in different ways. Their effects on streamflow and hydrological balance components were analyzed in the Yiluo River Basin using the delta method and the Soil and Water Assessment Tool (SWAT). The multivariable (runoff and actual evapotranspiration) calibration and validation method was used to reduce model uncertainty. LUCC impact on hydrological balance components (1976–2015) was evaluated through comparison of simulated paired land use scenarios. WRUBAP impact on runoff was assessed by comparing natural (simulated) and observed runoff. It showed that urban area reduction led to decreased groundwater, but increased surface runoff and increased water area led to increased evaporation. LUCC impact on annual runoff was found limited; for instance, the difference under the paired scenarios was <1 mm. Observed runoff was 34.7–144.1% greater than natural runoff during November–June because of WRUBAP. The effect of WRUBAP on wet season runoff regulation was limited before the completion of the Guxian Reservoir, whereas WRUBAP caused a reduction in natural runoff of 21.6–35.0% during the wet season (July–October) after its completion. The results suggest that WRUBAP has greater influence than LUCC on runoff in the Yiluo River Basin. Based on existing drought mitigation measures, interbasin water transfer measures and deep groundwater exploitation could reduce the potential for drought attributable to predicted future climate extremes. In addition to reservoir regulation, conversion of farmland to forestry in the upstream watershed could also reduce flood risk.

Experiments in unmanned aerial vehicle/unmanned ground vehicle radiation search

AbstractThis paper discusses the results of a field experiment conducted at Savannah River National Laboratory to test the performance of several algorithms for the localization of radioactive materials. In this multirobot system, both an unmanned aerial vehicle, a custom hexacopter, and an unmanned ground vehicle (UGV), the ClearPath Jackal, equipped with γ‐ray spectrometers, were used to collect data from two radioactive source configurations. Both the Fourier scattering transform and the Laplacian eigenmap algorithms for source detection were tested on the collected data sets. These algorithms transform raw spectral measurements into alternate spaces to allow clustering to detect trends within the data which indicate the presence of radioactive sources. This study also presents a point source model and accompanying information‐theoretic active exploration algorithm. Field testing validated the ability of this model to fuse aerial and ground collected radiation measurements, and the exploration algorithm’s ability to select informative actions to reduce model uncertainty, allowing the UGV to locate radioactive material online.

Assimilating INSAT‐3D Thermal Infrared Window Imager Observation With the Particle Filter: A Case Study for Vardah Cyclone

AbstractA big challenge in the satellite data assimilation is the effective use of InfraRed (IR) window channel radiances in the high‐resolution weather model. A hybrid data‐assimilation method is used in this study for very severe cyclonic storm “Vardah,” in which three‐dimensional variational method is used to assimilate control observations, and particle filter method is used to assimilate Indian geostationary satellite INSAT‐3D data. In the context of imperfect weather model, various particles (or ensembles) are prepared with different combinations of model physics. To implement particle filter, INSAT‐3D thermal IR window channel 1 (TIR‐1; center wavelength 11 μm) measured brightness temperature (BT) and cloud mask product are used to assign appropriate weights to different particles to reduce model uncertainties. This step is followed by resampling step in which new particles are generated from high weight particles using stochastic kinetic‐energy backscatter scheme method and dynamical variables are perturbed into the model physics. Results suggest that simulated TIR‐1 BT analysis and forecasts from WPF (with INSAT‐3D data using particle filter experiments) are closer to INSAT‐3D measured TIR‐1 BT in comparison to WCNT (without INSAT‐3D data using particle filter) experiments. Furthermore, approximately 10% to 50% improvements are found in mean track error forecasts in WPF experiments. An improvement of ~10% is noticed in cyclone center position in analysis. Prediction of storm intensity is also improved after assimilation. Results also suggest that vertical structure of WPF simulated humidity, temperature, wind speed, and surface pressure is improved over WCNT runs.

Turbulence Modeling in the Age of Data

Data from experiments and direct simulations of turbulence have historically been used to calibrate simple engineering models such as those based on the Reynolds-averaged Navier–Stokes (RANS) equations. In the past few years, with the availability of large and diverse data sets, researchers have begun to explore methods to systematically inform turbulence models with data, with the goal of quantifying and reducing model uncertainties. This review surveys recent developments in bounding uncertainties in RANS models via physical constraints, in adopting statistical inference to characterize model coefficients and estimate discrepancy, and in using machine learning to improve turbulence models. Key principles, achievements, and challenges are discussed. A central perspective advocated in this review is that by exploiting foundational knowledge in turbulence modeling and physical constraints, researchers can use data-driven approaches to yield useful predictive models.

Active Learning Through Multi-Standard Optimization

Active learning selects the most critical instances and obtains their labels through interaction with an oracle. Selecting either informative or representative unlabeled instances may result in sampling bias or cluster dependency. In this paper, we propose a multi-standard optimization active learning (MSAL) algorithm that considers the informativeness, representativeness, and diversity of instances. Informativeness is measured by the soft-max predicted entropy, whereas representativeness is measured by the probability density function obtained by a non-parametric estimation. The multiplex of the two is used as an optimization objective to reduce model uncertainty and explore the distribution of unlabeled data. Diversity is measured by the difference between the selected critical instances. This is used as a constraint to prevent the selection of instances that are too similar. Learning experiments were performed with 12 datasets from various domains. The results of significance tests verify the effectiveness of MSAL and its superiority over state-of-the-art active learning algorithms.

Using plant, microbe, and soil fauna traits to improve the predictive power of biogeochemical models

AbstractProcess‐based models describing biogeochemical cycling are crucial tools to understanding long‐term nutrient dynamics, especially in the context of perturbations, such as climate and land‐use change. Such models must effectively synthesize ecological processes and properties. For example, in terrestrial ecosystems, plants are the primary source of bioavailable carbon, but turnover rates of essential nutrients are contingent on interactions between plants and soil biota. Yet, biogeochemical models have traditionally considered plant and soil communities in broad terms. The next generation of models must consider how shifts in their diversity and composition affect ecosystem processes.One promising approach to synthesize plant and soil biodiversity and their interactions into models is to consider their diversity from a functional trait perspective. Plant traits, which include heritable chemical, physical, morphological and phenological characteristics, are increasingly being used to predict ecosystem processes at a range of scales, and to interpret biodiversity–ecosystem functional relationships. There is also emerging evidence that the traits of soil microbial and faunal communities can be correlated with ecosystem functions such as decomposition, nutrient cycling, and greenhouse gas production.Here, we draw on recent advances in measuring and using traits of different biota to predict ecosystem processes, and provide a new perspective as to how biotic traits can be integrated into biogeochemical models. We first describe an explicit trait‐based model framework that operates at small scales and uses direct measurements of ecosystem properties; second, an integrated approach that operates at medium scales and includes interactions between biogeochemical cycling and soil food webs; and third, an implicit trait‐based model framework that associates soil microbial and faunal functional groups with plant functional groups, and operates at the Earth‐system level. In each of these models, we identify opportunities for inclusion of traits from all three groups to reduce model uncertainty and improve understanding of biogeochemical cycles.These model frameworks will generate improved predictive capacity of how changes in biodiversity regulate biogeochemical cycles in terrestrial ecosystems. Further, they will assist in developing a new generation of process‐based models that include plant, microbial, and faunal traits and facilitate dialogue between empirical researchers and modellers.

Linking big models to big data: efficient ecosystem model calibration through Bayesian model emulation

Abstract. Data-model integration plays a critical role in assessing and improving our capacity to predict ecosystem dynamics. Similarly, the ability to attach quantitative statements of uncertainty around model forecasts is crucial for model assessment and interpretation and for setting field research priorities. Bayesian methods provide a rigorous data assimilation framework for these applications, especially for problems with multiple data constraints. However, the Markov chain Monte Carlo (MCMC) techniques underlying most Bayesian calibration can be prohibitive for computationally demanding models and large datasets. We employ an alternative method, Bayesian model emulation of sufficient statistics, that can approximate the full joint posterior density, is more amenable to parallelization, and provides an estimate of parameter sensitivity. Analysis involved informative priors constructed from a meta-analysis of the primary literature and specification of both model and data uncertainties, and it introduced novel approaches to autocorrelation corrections on multiple data streams and emulating the sufficient statistics surface. We report the integration of this method within an ecological workflow management software, Predictive Ecosystem Analyzer (PEcAn), and its application and validation with two process-based terrestrial ecosystem models: SIPNET and ED2. In a test against a synthetic dataset, the emulator was able to retrieve the true parameter values. A comparison of the emulator approach to standard brute-force MCMC involving multiple data constraints showed that the emulator method was able to constrain the faster and simpler SIPNET model's parameters with comparable performance to the brute-force approach but reduced computation time by more than 2 orders of magnitude. The emulator was then applied to calibration of the ED2 model, whose complexity precludes standard (brute-force) Bayesian data assimilation techniques. Both models are constrained after assimilation of the observational data with the emulator method, reducing the uncertainty around their predictions. Performance metrics showed increased agreement between model predictions and data. Our study furthers efforts toward reducing model uncertainties, showing that the emulator method makes it possible to efficiently calibrate complex models.

Identifying the Sensitivity of Ensemble Streamflow Prediction by Artificial Intelligence

Sustainable water resources management is facing a rigorous challenge due to global climate change. Nowadays, improving streamflow predictions based on uneven precipitation is an important task. The main purpose of this study is to integrate the ensemble technique concept into artificial neural networks for reducing model uncertainty in hourly streamflow predictions. The ensemble streamflow predictions are built following two steps: (1) Generating the ensemble members through disturbance of initial weights, data resampling, and alteration of model structure; (2) consolidating the model outputs through the arithmetic average, stacking, and Bayesian model average. This study investigates various ensemble strategies on two study sites, where the watershed size and hydrological conditions are different. The results help to realize whether the ensemble methods are sensitive to hydrological or physiographical conditions. Additionally, the applicability and availability of the ensemble strategies can be easily evaluated in this study. Among various ensemble strategies, the best ESP is produced by the combination of boosting (data resampling) and Bayesian model average. The results demonstrate that the ensemble neural networks greatly improved the accuracy of streamflow predictions as compared to a single neural network, and the improvement made by the ensemble neural network is about 19–37% and 20–30% in Longquan Creek and Jinhua River watersheds, respectively, for 1–3 h ahead streamflow prediction. Moreover, the results obtained from different ensemble strategies are quite consistent in both watersheds, indicating that the ensemble strategies are insensitive to hydrological and physiographical factors. Finally, the output intervals of ensemble streamflow prediction may also reflect the possible peak flow, which is valuable information for flood prevention.

Integration of satellite remote sensing data in ecosystem modelling at local scales: Practices and trends

Abstract Spatiotemporal ecological modelling of terrestrial ecosystems relies on climatological and biophysical Earth observations. Due to their increasing availability, global coverage, frequent acquisition and high spatial resolution, satellite remote sensing (SRS) products are frequently integrated to in situ data in the development of ecosystem models (EMs) quantifying the interaction among the vegetation component and the hydrological, energy and nutrient cycles. This review highlights the main advances achieved in the last decade in combining SRS data with EMs, with particular attention to the challenges modellers face for applications at local scales (e.g. small watersheds). We critically review the literature on progress made towards integration of SRS data into terrestrial EMs: (1) as input to define model drivers; (2) as reference to validate model results; and (3) as a tool to sequentially update the state variables, and to quantify and reduce model uncertainty. The number of applications provided in the literature shows that EMs may profit greatly from the inclusion of spatial parameters and forcings provided by vegetation and climatic‐related SRS products. Limiting factors for the application of such models to local scales are: (1) mismatch between the resolution of SRS products and model grid; (2) unavailability of specific products in free and public online repositories; (3) temporal gaps in SRS data; and (4) quantification of model and measurement uncertainties. This review provides examples of possible solutions adopted in recent literature, with particular reference to the spatiotemporal scales of analysis and data accuracy. We propose that analysis methods such as stochastic downscaling techniques and multi‐sensor/multi‐platform fusion approaches are necessary to improve the quality of SRS data for local applications. Moreover, we suggest coupling models with data assimilation techniques to improve their forecast abilities. This review encourages the use of SRS data in EMs for local applications, and underlines the necessity for a closer collaboration among EM developers and remote sensing scientists. With more upcoming satellite missions, especially the Sentinel platforms, concerted efforts to further integrate SRS into modelling are in great demand and these types of applications will certainly proliferate.

Model uncertainty reduction for real-time flood control by means of a flexible data assimilation approach and reduced conceptual models

Recently, a combination of model predictive control and a reduced genetic algorithm (RGA-MPC) has shown to be an efficient control technique for real-time flood control, making use of fast conceptual river models. This technique was so far only tested under ideal circumstances of perfect model predictions. Prediction errors originating from hydrodynamic model mismatches, however, result in a deterioration of the real-time control performance. Therefore, this paper presents two extensions of the RGA-MPC technique. First, a new type of conceptual model is introduced to further increase the computational efficiency. This reduced conceptual model is specially tailored for real-time flood control applications by eliminating all unnecessary intermediate calculations to obtain the flood control objectives and by introducing a new transport element by means of flow matrices. Furthermore, the RGA-MPC technique is extended with a flexible data assimilation approach that analyzes the past observed errors and applies an appropriate error prediction scheme. The proposed approach largely compensates for the loss in control performance due to the hydrodynamic model uncertainty.

Partitioning global land evapotranspiration using CMIP5 models constrained by observations

The ratio of plant transpiration to total terrestrial evapotranspiration (T/ET) captures the role of vegetation in surface–atmosphere interactions. However, its magnitude remains highly uncertain at the global scale. Here we apply an emergent constraint approach that integrates CMIP5 Earth system models (ESMs) with 33 field T/ET measurements to re-estimate the global T/ET value. Our observational constraint strongly increases the original ESM estimates (0.41 ± 0.11) and greatly alleviates intermodel discrepancy, which leads to a new global T/ET estimate of 0.62 ± 0.06. For all the ESMs, the leaf area index is identified as the primary driver of spatial variations of T/ET, but to correct its bias generates a larger T/ET underestimation than the original ESM output. We present evidence that the ESM underestimation of T/ET is, instead, attributable to inaccurate representation of canopy light use, interception loss and root water uptake processes in the ESMs. These processes should be prioritized to reduce model uncertainties in the global hydrological cycle.

Model predictive control with active learning under model uncertainty: Why, when, and how

Optimal control relies on a model, which is generally uncertain because of incomplete knowledge of the system and changes in the dynamics over time. Probing the system under closed‐loop control can reduce the model uncertainty through generating input‐output data that is more informative than the data generated from normal operation. This paper addresses the problem of model predictive control (MPC) with active learning, with a particular focus on how incorporating probing in the control action can reduce model uncertainty. We discuss some of the central theoretical questions in this problem, and demonstrate the potential of active learning for maintaining MPC performance in the presence of uncertainty in model parameters and structure. Simulation results show that active learning is particularly beneficial when a system undergoes abrupt changes (such as the sudden occurrence of a fault) that can compromise operational safety, reliability, and profitability. © 2018 American Institute of Chemical Engineers AIChE J, 64: 3071–3081, 2018

Effects of desiccation stress on adult female longevity in Aedes aegypti and Ae. albopictus (Diptera: Culicidae): results of a systematic review and pooled survival analysis

BackgroundTransmission dynamics of mosquito-borne viruses such as dengue, Zika and chikungunya are affected by the longevity of the adult female mosquito. Environmental conditions influence the survival of adult female Aedes mosquitoes, the primary vectors of these viruses. While the association of temperature with Aedes mortality has been relatively well-explored, the role of humidity is less established. The current study’s goals were to compile knowledge of the influence of humidity on adult survival in the important vector species Aedes aegypti and Ae. albopictus, and to quantify this relationship while accounting for the modifying effect of temperature.MethodsWe performed a systematic literature review to identify studies reporting experimental results informing the relationships among temperature, humidity and adult survival in Ae. aegypti and Ae. albopictus. Using a novel simulation approach to harmonize disparate survival data, we conducted pooled survival analyses via stratified and mixed effects Cox regression to estimate temperature-dependent associations between humidity and mortality risk for these species across a broad range of temperatures and vapor pressure deficits.ResultsAfter screening 1517 articles, 17 studies (one in semi-field and 16 in laboratory settings) met inclusion criteria and collectively reported results for 192 survival experiments. We review and synthesize relevant findings from these studies. Our stratified model estimated a strong temperature-dependent association of humidity with mortality in both species, though associations were not significant for Ae. albopictus in the mixed effects model. Lowest mortality risks were estimated around 27.5 °C and 21.5 °C for Ae. aegypti and Ae. albopictus, respectively, and mortality increased non-linearly with decreasing humidity. Aedes aegypti had a survival advantage relative to Ae. albopictus in the stratified model under most conditions, but species differences were not significant in the mixed effects model.ConclusionsHumidity is associated with mortality risk in adult female Ae. aegypti in controlled settings. Data are limited at low humidities, temperature extremes, and for Ae. albopictus, and further studies should be conducted to reduce model uncertainty in these contexts. Desiccation is likely an important factor in Aedes population dynamics and viral transmission in arid regions. Models of Aedes-borne virus transmission may be improved by more comprehensively representing humidity effects.

Combustion kinetic model development using surrogate model similarity method

An ideal combustion kinetic model needs to be validated by different experimental targets over a wide range of temperatures and pressures that represent operating conditions in real engines. However, conditions of laboratory experiments for model validation are often limited by the constraint of experimental techniques. In order to improve model predictions under certain conditions (for example, at a relatively higher pressure), it is often needed to use the experimental data obtained under other conditions. In this work, the surrogate model similarity (SMS) method is proposed to find the experimental conditions or targets for model optimisation under certain conditions where the experiments are hard to be conducted. The similarity coefficient is calculated by the cosine similarity between the characteristic coefficients (vectors) of the High Dimensional Model Representation (HDMR) models for different model predictions. A larger similarity coefficient represents a closer relationship between two model predictions. The experimental data with larger similarity coefficients could be more effective to model uncertainty reduction under the concerned conditions. To demonstrate this method, simulations were conducted for two selected combustion systems with hydrogen or methanol as the fuel. In addition to its strength in available experimental data selection for model optimization, this method can be used to screen out experimental targets with strong constraint effect beforehand, thus providing an effective way to maximise utilisation of experimental resources.

Using all data to improve seasonal sea surface temperature predictions: A combination‐based model forecast with unequal observation lengths

One way to reduce model uncertainty in climate predictions is to combine forecasts from several models. Recent multi‐model combination approaches combine model forecasts by pooling data for a time period, common across all the models, thus ignoring the additional data available or discarding altogether the models with the shorter time period. This results in the loss of some information which could otherwise be used while combining the models to possibly improve forecast skill. Our research explores this issue in the context of multi‐model sea surface temperature (SST) models predictions and proposes a novel concept that allows a framework for combining models with unequal time period. Here, the unequal time periods imply different range of start and end dates of available model forecasts. A qualitative standpoint of our multi‐model forecasting strategy is to reduce the uncertainty and improve the forecast skill. The utility of the approach is demonstrated by combining the global seasonal NDJ (November–January) SST predictions of two models and also as many as eight models, obtained using both equal and unequal time periods. The proposed approach shows improvement over 62–69% grid cells around the entire globe over the case when the common period of data length across the models is considered.