Sources Of Model Error Research Articles

An evaluation of common stock assessment diagnostic tools for choosing among state-space models with multiple random effects processes

State-space models have received increasing attention in fisheries stock assessments given their flexibility to incorporate multiple sources of process errors. Identifying which process errors to include is important because incorrectly including some process errors can induce bias in management quantities. Existing model selection tools commonly applied in traditional statistical catch-at-age models may not perform as well for state-space models. We evaluated the efficacy of common diagnostic tools for correctly identifying the presence, absence, and magnitude of three process errors (survival, selectivity, and natural mortality) in a simulation–estimation experiment. No model diagnostic tools could consistently identify the correct process error structure in all situations. Incorrectly attributing the process error from natural mortality to other processes, or vice versa, led to relatively large bias in management quantities. Furthermore, incorrectly including an additional source of process error in the assessment models exhibited similar performance to the correct model and generally showed unbiased estimates of management quantities; incorrectly excluding a source of process error, however, generated large biases. Thus, despite not having generally reliable model diagnostic tools for state-space assessments, practitioners should err on the side of using overly complex models, except for natural mortality unless there is external corroborating evidence of changing natural mortality.

PLAXIS 3D numerical analysis of complex geotechnical problems of colossal built heritage

The paper aims to contribute to the preservation of high valuable historic masonry structures and historic urban landscapes through the combination of geotechnical, structural engineering. The main objective of the study is to conduct finite element analysis (FEA) of bearing saturated soft clay soil problems and induced structural failure mechanisms. This analysis is based on experimental and numerical studies using coupled PLAXIS 3D FE models. The paper presents a geotechnical analytical model for the measurement of stresses, deformations, and differential settlement of saturated clay soils under colossal stone/brick masonry structures. The study also discusses the behavior of soft clay soils under Qasr Yashbak through numerical analysis, which helps in understanding the studied behavior and the loss of soil-bearing capacity due to moisture content or ground water table (G.W.T) changes. The paper presents valuable insights into the behavior of soft clay soils under colossal stone/ brick masonry structures. The present study summarized specific details about the limitations and potential sources of error in Finite Element Modeling (FEM). Further field research and experimental analysis may be required to address these limitations and enhance the understanding of the studied soft clay soil behavior. The geotechnical problems in historic monuments and structures such as differential settlement are indeed important issues for their conservation since it may induce serious damages. It deserves more in-depth researches.

Impacts of Nudged Sea Surface Temperature on Tropical Precipitation, Moisture, and Vertical Velocity in an Earth System Model

Abstract Understanding the effect of the nudged sea surface temperature (SST) on precipitation simulation is important in Earth system models. In this study, we analyzed the role of nudged SST and air–sea interaction in precipitation simulation by using the Chinese Academy of Sciences Earth System Model version 2 (CAS-ESM2). CAS-ESM2 fully coupled runs with and without nudged tropical SST to observations, and the atmosphere-only runs (AMIP) are comprehensively evaluated against observations and reanalysis over the tropical ocean from 1982 to 2015. The precipitation is affected by the nudged SST and air–sea interaction by means of changes in both thermodynamic and dynamic processes. Simulation with the nudged SST dramatically improves the climatological precipitation fields over the tropics and four regional areas. As the simulated SSTs are nudged toward the observations, CAS-ESM2 well reproduces the mean annual values in column water vapor (CWV), vertical structure of specific humidity, and large-scale circulation. The nudged SST run also captures a wide range of the complex interactions between SST, precipitation, moisture, and large-scale circulations, as measured by the monthly relationships and bivariate probability distribution function. The biases in heavy precipitation associated with the CWV induced under moist conditions are decreased by applying the nudging technique. In particular, the nudging run performs better in reproducing the SST–precipitation covariability than the control and AMIP runs. However, CAS-ESM2 with nudged tropical SST simulates the degradation of strong ascending motions, which is similar to the AMIP runs and reflects the deviation of the atmospheric model. Our results confirm that the fully coupled run using the nudged SST is a method for improving the precipitation simulation across the tropical oceans. Significance Statement Understanding the sources of precipitation errors in Earth system models is challenging because of the complex error compensation and air–sea interaction. When the simulated SST is nudged toward observations, few studies have focused on the changes in both thermodynamic and dynamic processes via atmospheric large-scale circulation, moisture, and changes in the SST–precipitation relationship at the same time. Here, we carry out two CMIP6 historical experiments (a fully coupled control run and an SST-nudged run) and the atmosphere-only runs by using CAS-ESM2 from 1982 to 2015. Then, we validate and analyze the climatology of SST, precipitation, moisture, large-scale circulation, and their complex interactions over the tropical oceans and different regional areas. The fully coupled nudged-SST run outperforms the fully free coupled run in reproducing the climatology and complex interactions between SST, precipitation, moisture, and large-scale circulations across the tropics. Moreover, the biases in the SST–precipitation covariability are significantly alleviated in the nudging experiments. We confirmed that the SST nudging method is a potential way to improve precipitation simulation in CAS-ESM2.

Research on Underwater Wireless Optical Communication Channel Model and Its Application

Underwater wireless optical communication (UWOC) is an emerging technology with wide-ranging applications in marine exploration, offshore industries, environmental monitoring, and underwater robotics. In order to investigate the application of UWOC in environments of different water quality, this study establishes a model of an optical communication channel and analyzes the impact of water quality on communication range. Our experimental design is employed to validate the effectiveness of the model and analyze the sources of model errors. Furthermore, this research introduces the concept of an “effective communication space” for underwater optical wireless communication and constructs an experimental platform to test the effective communication space under various water quality conditions. In addition, the application methods and workflow of wireless optical communication on underwater mobile platforms are discussed, and field tests are conducted in a practical lake environment to verify the application value of the effective communication space. This research offers valuable guidance for advancing the study and engineering applications of UWOC technology.

Phenological Mapping of Invasive Insects: Decision Support for Surveillance and Management.

Readily accessible and easily understood forecasts of the phenology of invasive insects have the potential to support and improve strategic and tactical decisions for insect surveillance and management. However, most phenological modeling tools developed to date are site-based, meaning that they use data from a weather station to produce forecasts for that single site. Spatial forecasts of phenology, or phenological maps, are more useful for decision-making at area-wide scales, such as counties, states, or entire nations. In this review, we provide a brief history on the development of phenological mapping technologies with a focus on degree-day models and their use as decision support tools for invasive insect species. We compare three different types of phenological maps and provide examples using outputs of web-based platforms that are presently available for real-time mapping of invasive insects for the contiguous United States. Next, we summarize sources of climate data available for real-time mapping, applications of phenological maps, strategies for balancing model complexity and simplicity, data sources and methods for validating spatial phenology models, and potential sources of model error and uncertainty. Lastly, we make suggestions for future research that may improve the quality and utility of phenological maps for invasive insects.

An along-track Biogeochemical Argo modelling framework: a case study of model improvements for the Nordic seas

Abstract. We present a framework that links in situ observations from the Biogeochemical Argo (BGC-Argo) array to biogeochemical models. The framework minimizes the technical effort required to construct a Lagrangian-type 1D modelling experiment along BGC-Argo tracks. We utilize the Argo data in two ways: (1) to drive the model physics and (2) to evaluate the model biogeochemistry. BGC-Argo physics data are used to nudge the model physics closer to observations to reduce the errors in the biogeochemistry stemming from physics errors. This allows us to target the model biogeochemistry and, by using the Argo biogeochemical dataset, we identify potential sources of model errors, introduce changes to the model formulation, and validate model configurations. We present experiments for the Nordic seas and showcase how we identify potential BGC-Argo buoys to model, prepare forcing, design experiments, and approach model improvement and validation. We use the ECOSMO II(CHL) model as the biogeochemical component and focus on chlorophyll a. The experiments reveal that ECOSMO II(CHL) requires improvements during low-light conditions, as the comparison to BGC-Argo reveals that ECOSMO II(CHL) simulates a late spring bloom and does not represent the deep chlorophyll maximum layer formation in summer periods. We modified the productivity and chlorophyll a relationship and statistically documented decreased bias and error in the revised model when using BGC-Argo data. Our results reveal that nudging the model temperature and salinity closer to BGC-Argo data reduces errors in biogeochemistry, and we suggest a relaxation time period of 1–10 d. The BGC-Argo data coverage is ever-growing and the framework is a valuable asset, as it improves biogeochemical models by performing efficient 1D model configurations and evaluation and then transferring the configurations to a 3D model with a wide range of use cases at the operational, regional/global and climate scales.

Using probabilistic machine learning to better model temporal patterns in parameterizations: a case study with the Lorenz 96 model

Abstract. The modelling of small-scale processes is a major source of error in weather and climate models, hindering the accuracy of low-cost models which must approximate such processes through parameterization. Red noise is essential to many operational parameterization schemes, helping model temporal correlations. We show how to build on the successes of red noise by combining the known benefits of stochasticity with machine learning. This is done using a recurrent neural network within a probabilistic framework (L96-RNN). Our model is competitive and often superior to both a bespoke baseline and an existing probabilistic machine learning approach (GAN, generative adversarial network) when applied to the Lorenz 96 atmospheric simulation. This is due to its superior ability to model temporal patterns compared to standard first-order autoregressive schemes. It also generalizes to unseen scenarios. We evaluate it across a number of metrics from the literature and also discuss the benefits of using the probabilistic metric of hold-out likelihood.

Virtual sensing of subsoil strain response in monopile-based offshore wind turbines via Gaussian process latent force models

Virtual sensing techniques have gained traction in applications to the structural health monitoring of monopile-based offshore wind turbines, as the strain response below the mudline, which is a primary indicator of fatigue damage accumulation, is impractical to measure directly with physical instrumentation. The Gaussian process latent force model (GPLFM) is a generalized Bayesian virtual sensing technique which combines a physics-driven model of the structure with a data-driven model of latent variables of the system to extrapolate unmeasured strain states. In the GPLFM, unknown sources of excitation are modeled as a Gaussian process (GP) and endowed with a structured covariance relationship with response states, using properties of the GP covariance kernel as well as correlation information supplied by the mechanical model. It is shown that posterior inference of the latent inputs and states is performed by Gaussian process regression of measured accelerations, computed efficiently using Kalman filtering and Rauch–Tung–Striebel smoothing in an augmented state-space model. While the GPLFM has been previously demonstrated in numerical studies to improve upon other virtual sensing techniques in terms of accuracy, robustness, and numerical stability, this work provides one of the first cases of in-situ validation of the GPLFM. The predicted strain response by the GPLFM is compared to subsoil strain data collected from an operating offshore wind turbine in the Westermeerwind Park in the Netherlands. A number of test cases are conducted, where the performance of the GPLFM is evaluated for its sensitivity to varying operational and environmental conditions, to the instrumentation scheme of the turbine, and to the fidelity of the mechanical model. In particular, this paper discusses the capacity of the GPLFM to achieve relatively robust strain predictions under high model uncertainty in the soil-foundation system of the offshore wind turbine by attributing sources of model error to the estimated stochastic input.

Modelling Green Volume Using Sentinel-1, -2, PALSAR-2 Satellite Data and Machine Learning for Urban and Semi-Urban Areas in Germany.

Urban Green Infrastructure (UGI) provides ecosystem services such as cooling of temperatures and is majorly important for climate change adaptation. Green Volume (GV) describes the 3-D space occupied by vegetation and is highly useful for the assessment of UGI. This research uses Sentinel-2 (S-2) optical data, vegetation indices (VIs), Sentinel-1 (S-1) and PALSAR-2 (P-2) radar data to build machine learning models for yearly GV estimation on large scales. Our study compares random and stratified sampling of reference data, assesses the performance of different machine learning algorithms and tests model transferability by independent validation. The results indicate that stratified sampling of training data leads to improved accuracies when compared to random sampling. While the Gradient Tree Boost (GTB) and Random Forest (RF) algorithms show generally similar performance, Support Vector Machine (SVM) exhibits considerably greater model error. The results suggest RF to be the most robust classifier overall, achieving highest accuracies for independent and inter-annual validation. Furthermore, modelling GV based on S-2 features considerably outperforms using only S-1 or P-2 based features. Moreover, the study finds that underestimation of large GV magnitudes in urban forests constitutes the biggest source of model error. Overall, modelled GV explains around 79% of the variability in reference GV at 10 m resolution and over 90% when aggregated to 100 m resolution. The research shows that accurately modelling GV is possible using openly available satellite data. Resulting GV predictions can be useful for environmental management by providing valuable information for climate change adaptation, environmental monitoring and change detection.

Assessment of Engineering Turbulence Models in Buoyant Diabatic Turbulent Flow

Turbulent heat transfer in buoyancy-dominated flows is a challenging problem for computational fluid dynamics (CFD). Many authors attribute model error in these conditions to the Reynolds analogy. We leverage a brand-new direct numerical simulation database to evaluate the performance of several popular turbulence models in buoyant diabatic channel flow. We find that heat transfer results are relatively accurate, with a Nusselt number error less than 20%. However, the turbulent flow solution is very inaccurate, with wall shear overpredicted by up to 100%. This indicates significant turbulence model error in such flows. We determined that the dominant sources of model error are missing physics in the algebraic Reynolds stress framework and the simple buoyancy production term used in industrial CFD. We suggest that future modeling efforts focus on these two sources of model error. We demonstrate that the Reynolds analogy is not the dominant source of model error.

Improving statistical projections of ocean dynamic sea-level change using pattern recognition techniques

Abstract. Regional emulation tools based on statistical relationships, such as pattern scaling, provide a computationally inexpensive way of projecting ocean dynamic sea-level change for a broad range of climate change scenarios. Such approaches usually require a careful selection of one or more predictor variables of climate change so that the statistical model is properly optimized. Even when appropriate predictors have been selected, spatiotemporal oscillations driven by internal climate variability can be a large source of statistical model error. Using pattern recognition techniques that exploit spatial covariance information can effectively reduce internal variability in simulations of ocean dynamic sea level, significantly reducing random errors in regional emulation tools. Here, we test two pattern recognition methods based on empirical orthogonal functions (EOFs), namely signal-to-noise maximizing EOF pattern filtering and low-frequency component analysis, for their ability to reduce errors in pattern scaling of ocean dynamic sea-level change. We use the Max Planck Institute Grand Ensemble (MPI-GE) as a test bed for both methods, as it is a type of initial-condition large ensemble designed for an optimal characterization of the externally forced response. We show that the two methods tested here more efficiently reduce errors than conventional approaches such as a simple ensemble average. For instance, filtering only two realizations by characterizing their common response to external forcing reduces the random error by almost 60 %, a reduction that is only achieved by averaging at least 12 realizations. We further investigate the applicability of both methods to single-realization modeling experiments, including four CMIP5 simulations for comparison with previous regional emulation analyses. Pattern filtering leads to a varying degree of error reduction depending on the model and scenario, ranging from more than 20 % to about 70 % reduction in global-mean root mean squared error compared with unfiltered simulations. Our results highlight the relevance of pattern recognition methods as a tool to reduce errors in regional emulation tools of ocean dynamic sea-level change, especially when one or only a few realizations are available. Removing internal variability prior to tuning regional emulation tools can optimize the performance of the statistical model, leading to substantial differences in emulated dynamic sea level compared to unfiltered simulations.

Uncertainty-Based Calibration Method for Environmental Sensors—Application to Chlorine and pH Monitoring With Carbon Nanotube Sensor Array

Societal demands in the field of air and water pollution monitoring require the ability to simultaneously detect a wide variety of chemical elements at very low concentrations in complex environments using compact and low-cost sensor devices. Although nanomaterial-based sensors have long been proposed as a solution to these exacting requirements, their detection accuracy is generally degraded in real-world conditions compared to laboratory conditions due to the effects of various interferents. To manage the related uncertainties and to associate confidence to the estimations, it seems natural to formulate the calibration-estimation problem in a probabilistic framework. This probabilistic formulation and its successful application to the monitoring of pH and active chlorine in drinking water are the main contributions of this work. While other tested calibration methods only allowed the monitoring of active chlorine, our solution enables the monitoring of both active chlorine and pH with quantified and reasonable uncertainties. Its success relies mainly on two adaptations of standard calibration methods: the consideration of the sensor inputs measurement errors (and not only the errors associated with the sensor outputs as it is generally done), and the introduction of two sources of model error (ME), one accounting for the approximate character of the calibration model, the other for unmeasured—and possibly unknown—interferents in the calibration environment.

Photovoltaic power electricity generation nowcasting combining sky camera images and learning supervised algorithms in the Southern Spain

The alternation between cloudy and clear skies alters the photovoltaic production. This makes it necessary to anticipate these disturbances hours in advance for the correct operation of the electricity distribution plants and networks. In this paper, two short-term forecasting models (3 h) are developed to forecast the photovoltaic production in an integrated plant in the CIESOL building of the University of Almería. The methodology used is based on sky camera images and Artificial Intelligence techniques. Two models have been developed and compared applying artificial neural network (ANN) and support vector machine (SVM) techniques. The global irradiance predicted using sky camera images is used as an input variable in both models. In addition, the operational status of the plants has been included as an input parameter through the performance ratio. The results have shown that the errors made by ANN and SVM are very similar. For all sky conditions, the uncertainty of the production forecast differs by less than 2% from the uncertainty of the solar resource, which is the main source of error in the production models developed.

Analysis of Pneumatic Artificial Muscles and the Inelastic Braid Assumption

Pneumatic artificial muscles (PAMs) are becoming an increasingly popular form of soft actuator due to their unique actuation characteristics. The creation of accurate PAM actuation models is important for their successful implementation. However, PAM studies often employ actuation models that use simplifying assumptions which make the models easier to formulate and use, but at the cost of reduced accuracy. One of the most commonly used assumptions, the inelastic braid assumption, suggests that the braid does not stretch, and therefore would not affect its geometry or actuation force. Although this assumption has often been cited as a likely source of model error, its use has persevered for decades due to researchers’ inability to directly measure the effects of braid elasticity. The recent development of a photogrammetric method to accurately measure PAM geometry now enables this analysis. This study seeks to assess the current default adoption of the inelastic braid assumption in PAM models by attempting to quantify the braid elasticity effects. This research finds that current models that use the inelastic braid assumption can underestimate PAM diameter by as much as 30%, and overestimate actuation force by as much as 70%. These results show that braid elasticity can have a substantial effect on the geometry and actuation force of PAMs, and demonstrates that the inelastic braid assumption may not be a suitable universal assumption for PAM modeling and analyses, especially when low-stiffness braid materials are used.

Error Propagation in Satellite Multi-Image Geometry

This paper describes an investigation of the source of geospatial error in digital surface models (DSMs) constructed from multiple satellite images. In this study the uncertainty in surface geometry is separated into two spatial components; global error that affects the absolute position of the surface, and local error that varies from surface point to surface point. The global error component is caused by inaccuracy in the satellite imaging process, mainly due to uncertainty in the satellite position and orientation (pose) during image collection. A key result of the investigation is a new algorithm for determining the absolute geoposition of the DSM that takes into account the pose covariance of each satellite during image collection. This covariance information is used to weigh the evidence from each image in the computation of the global position of the DSM. The use of covariance information significantly decreases the overall uncertainty in global position. The paper also describes an approach to the prediction of local error in the DSM surface. The observed variance in surface position within a single stereo surface reconstruction defines the local horizontal error. The variance in the fused set of elevations from multiple stereo pairs at a single DSM location defines the local vertical error. These accuracy predictions are compared to ground truth provided by LiDAR scans of the same geographic region of interest.

Physical modelling of reinforced concrete at a 1:40 scale using additively manufactured reinforcement cages

AbstractGlobal level assumptions of numerical models have received relatively less attention, but have been indicated to be a major source of error in numerical modeling of Reinforced Concrete (RC) structures. In parallel, it has been stated that a statistical approach involving many virgin specimens and ground motions is necessary for model validation. Such an approach would require very small‐scale testing. Then, the reinforcement fabrication becomes a major issue. This paper proposes using additive manufacturing to fabricate the reinforcement cage. It presents the results from cyclic tests on 1:40 RC cantilever members. The cages were manufactured using an SLM 3D printer able to print rebars with submillimeter diameters. Different longitudinal and transverse reinforcement configurations were tested. A numerical model using existing Opensees elements was built and its parameters were calibrated against material level small‐scale tests. It captured the cyclic response of the RC members with a reasonable accuracy. The cyclic behavior of the RC members resembles the behavior of full‐scale RC members indicating that such small‐scale specimens can be used for the statistical validation of the global level assumptions of numerical models.

Representing Microphysical Uncertainty in Convective‐Scale Data Assimilation Using Additive Noise

AbstractFor convective clouds and precipitation, model uncertainty in cloud microphysics is considered one of the most significant sources of model error. In this study, samples for model microphysical uncertainty are obtained by calculating the differences between simulations equipped with two‐ and one‐moment schemes during a one‐month training period. The samples are then added to convective‐scale ensemble data assimilation as additive noise and combined with large‐scale additive noise based on samples from climatological atmospheric background error covariance. Two experiments, including the combination and large‐scale error only, are conducted for a one‐week convective period. The results reveal that the simulation with a two‐moment scheme triggers more convection and has larger ice‐phase precipitation particles, which produce a stronger signal in the melting layer. During data assimilation cycling, although more water is introduced to the model, it is shown that the combination performs better for both background and analysis and significantly improves short‐term ensemble forecasts of radar reflectivity and hourly precipitation.

Ensemble streamflow data assimilation using WRF-Hydro and DART: novel localization and inflation techniques applied to Hurricane Florence flooding

Abstract. Predicting major floods during extreme rainfall events remains an important challenge. Rapid changes in flows over short timescales, combined with multiple sources of model error, makes it difficult to accurately simulate intense floods. This study presents a general data assimilation framework that aims to improve flood predictions in channel routing models. Hurricane Florence, which caused catastrophic flooding and damages in the Carolinas in September 2018, is used as a case study. The National Water Model (NWM) configuration of the WRF-Hydro modeling framework is interfaced with the Data Assimilation Research Testbed (DART) to produce ensemble streamflow forecasts and analyses. Instantaneous streamflow observations from 107 United States Geological Survey (USGS) gauges are assimilated for a period of 1 month. The data assimilation (DA) system developed in this paper explores two novel contributions, namely (1) along-the-stream (ATS) covariance localization and (2) spatially and temporally varying adaptive covariance inflation. ATS localization aims to mitigate not only spurious correlations, due to limited ensemble size, but also physically incorrect correlations between unconnected and indirectly connected state variables in the river network. We demonstrate that ATS localization provides improved information propagation during the model update. Adaptive prior inflation is used to tackle errors in the prior, including large model biases which often occur in flooding situations. Analysis errors incurred during the update are addressed using posterior inflation. Results show that ATS localization is a crucial ingredient of our hydrologic DA system, providing at least 40 % more accurate (root mean square error) streamflow estimates than regular, Euclidean distance-based localization. An assessment of hydrographs indicates that adaptive inflation is extremely useful and perhaps indispensable for improving the forecast skill during flooding events with significant model errors. We argue that adaptive prior inflation is able to serve as a vigorous bias correction scheme which varies both spatially and temporally. Major improvements over the model's severely underestimated streamflow estimates are suggested along the Pee Dee River in South Carolina, and many other locations in the domain, where inflation is able to avoid filter divergence and, thereby, assimilate significantly more observations.

Assimilating atmospheric infrasound data to constrain atmospheric winds in a two‐dimensional grid

AbstractInfrasound waves travelling through atmospheric channels are affected by the conditions they encounter along their path. The shift in the back azimuth angle of a wavefront detected at the reception site depends on the cross‐winds it encountered. Estimating the original field from this integrated measurement is an (ill‐posed) inverse problem. By using a prior, this can be converted into a Bayesian estimation problem. In this work, we use the (ensemble) Kalman filter (EnKF) to tackle this problem. In particular, we provide an illustration of the setup and solution of the problem in a two‐dimensional grid, depending on both across‐track distance and height, which has not been done in previous works. We use a synthetic setup to discuss the details of the method. We show that one of the effects of along‐track averaging (done in previous studies to simplify the problem) is to overestimate the magnitude of the analysed values, and propose that this will be a source of model error. We also illustrate the process with real data corresponding to nine controlled ammunition explosions that took place in the summer of 2018. In these cases, the real infrasound waves we study seldom reach higher than km in height. However, the use of covariance‐based methods (e.g., the EnKF) allows for updates in higher regions where the wave did not travel and where traditional observations are sparse. In fact, the larger impacts from observations in these cases are in the region of 40–60 km, in agreement with previous works. This study contributes to paving the way towards the ultimate goal of assimilating information derived from infrasound waves into operational numerical weather forecasting. More studies in quality control of the observations and proper validation of the results are urgently needed.

Regularized machine learning on molecular graph model explains systematic error in DFT enthalpies

A major goal of materials research is the discovery of novel and efficient heterogeneous catalysts for various chemical processes. In such studies, the candidate catalyst material is modeled using tens to thousands of chemical species and elementary reactions. Density Functional Theory (DFT) is widely used to calculate the thermochemistry of these species which might be surface species or gas-phase molecules. The use of an approximate exchange correlation functional in the DFT framework introduces an important source of error in such models. This is especially true in the calculation of gas phase molecules whose thermochemistry is calculated using the same planewave basis set as the rest of the surface mechanism. Unfortunately, the nature and magnitude of these errors is unknown for most practical molecules. Here, we investigate the error in the enthalpy of formation for 1676 gaseous species using two different DFT levels of theory and the ‘ground truth values’ obtained from the NIST database. We featurize molecules using graph theory. We use a regularized algorithm to discover a sparse model of the error and identify important molecular fragments that drive this error. The model is robust to rigorous statistical tests and is used to correct DFT thermochemistry, achieving more than an order of magnitude improvement.