Model-data Fusion Research Articles

An Algorithm for Strapdown Airborne Gravity Disturbance Vector Measurement Based on High-Precision Navigation and EGM2008.

Attitude errors, accelerometer bias, the gravity disturbance vector, and their coupling are the primary factors obstructing strapdown airborne vector gravimetry. This paper takes the geocentric inertial frame as a reference and solves the kinematic equations of its motion and its errors of the body frame and local geographic frame in the Lie group, respectively; the attitude accuracy is improved through a high-precision navigation algorithm. The constant accelerometer bias is estimated through Kalman filtering and is deducted from the accelerometer output to eliminate its influence. Based on the EGM2008 model, the low-frequency components of the gravity disturbance vector are corrected. The gravity disturbance vectors after model data fusion were low-pass filtered to obtain the ultimate results. This method was applied to flight experimental data in the South China Sea, and a gravity anomaly accuracy of better than 0.5 mGal, a northward gravity disturbance accuracy of 0.85 mGal, and an eastward gravity disturbance accuracy of 4.0 mGal were obtained, with a spatial resolution of approximately 4.8 km.

Developing a new active canopy sensor- and machine learning-based in-season rice nitrogen status diagnosis and recommendation strategy

ContextTraditional critical nitrogen (N) dilution curve (CNDC) construction for N nutrition index (NNI) determination has limitations for in-season crop N diagnosis and recommendation under diverse on-farm conditions. ObjectivesThis study was conducted to (i) develop a new rice (Oryza sativa L.) critical N concentration (Nc) determination approach using vegetation index-based CNDCs; and (ii) develop an N recommendation strategy with this new Nc determination approach and evaluate its reliability and practicality. MethodsFive years of plot and on-farm experiments involving three japonica rice varieties were conducted at fourteen sites in Qixing Farm, Northeast China. Two machine learning (ML) methods, random forest (RF) and extended gradient boosting (XGBoost) regression, were used to fuse multi-source data including genotype, environment, management, growth stage, normalized difference vegetation index (NDVI) and normalized difference red edge (NDRE) from portable active canopy sensor RapidSCAN. The CNDC was established using NDVI and NDRE instead of aboveground biomass (AGB) measured by destructive sampling. A new in-season N diagnosis and recommendation strategy was further developed using direct and indirect NNI prediction using multi-source data fusion and ML models. ResultsThe new CNDC based on NDVI or NDRE explained 94−96 % of Nc variability in the evaluation dataset when it was coupled with environmental and agronomic factors using ML models. The ML-based PNC and NNI prediction models explained 85 % and 21–36 % more variability over simple regression models using NDVI or NDRE in the evaluation dataset, respectively. The new in-season N diagnosis strategy using the NDVI and NDRE-based CNDCs and plant N concentration (PNC) predicted with RF model and multi-source data fusion performed slightly better than direct NNI prediction, explaining 7 % more of NNI variability and achieving 89 % of the areal agreement for N diagnosis across all evaluation experiments. Integrating this new N management strategy into the precision rice management system (as ML_PRM) increased yield, N use efficiency (NUE) and economic benefits over farmer’s practice (FP) by 7–15 %, 11–71 % and 4–16 % (161–596 $ ha−1), respectively, and increased NUE by 11–26 % and economic benefits by 8–97 $ ha−1 than regional optimum rice management (RORM) under rice N surplus status under on-farm conditions. ConclusionsIn-season rice N status diagnosis can be improved using NDVI- and NDRE-based CNDC and PNC predicted by ML modeling with multi-source data fusion. ImplicationsThe active canopy sensor- and ML-based in-season N diagnosis and management strategy is more practical for applications under diverse on-farm conditions and has the potential to improve rice yield and ecological and economic benefits.

Permafrost carbon cycle and its dynamics on the Tibetan Plateau.

Our knowledge on permafrost carbon (C) cycle is crucial for understanding its feedback to climate warming and developing nature-based solutions for mitigating climate change. To understand the characteristics of permafrost C cycle on the Tibetan Plateau, the largest alpine permafrost region around the world, we summarized recent advances including the stocks and fluxes of permafrost C and their responses to thawing, and depicted permafrost C dynamics within this century. We find that this alpine permafrost region stores approximately 14.1 Pg (1 Pg=1015 g) of soil organic C (SOC) in the top 3 m. Both substantial gaseous emissions and lateral C transport occur across this permafrost region. Moreover, the mobilization of frozen C is expedited by permafrost thaw, especially by the formation of thermokarst landscapes, which could release significant amounts of C into the atmosphere and surrounding water bodies. This alpine permafrost region nevertheless remains an important C sink, and its capacity to sequester C will continue to increase by 2100. For future perspectives, we would suggest developing long-term in situ observation networks of C stocks and fluxes with improved temporal and spatial coverage, and exploring the mechanisms underlying the response of ecosystem C cycle to permafrost thaw. In addition, it is essential to improve the projection of permafrost C dynamics through in-depth model-data fusion on the Tibetan Plateau.

Inferred drought-induced plant allocation shifts and their impact on drought legacy at a tropical forest site.

While droughts predominantly induce immediate reductions in plant carbon uptake, they can also exert long-lasting effects on carbon fluxes through associated changes in leaf area, soil carbon, etc. Among other mechanisms, shifts in carbon allocation due to water stress can contribute to the legacy effects of drought on carbon fluxes. However, the magnitude and impact of these allocation shifts on carbon fluxes and pools remain poorly understood. Using data from a wet tropical flux tower site in French Guiana, we demonstrate that drought-induced carbon allocation shifts can be reliably inferred by assimilating Net Biosphere Exchange (NBE) and other observations within the CARbon DAta MOdel fraMework. This model-data fusion system allows inference of optimized carbon and water cycle parameters and states from multiple observational data streams. We then examined how these inferred shifts affected the duration and magnitude of drought's impact on NBE during and after the extreme event. Compared to a static allocation scheme analogous to those typically implemented in land surface models, dynamic allocation reduced average carbon uptake during drought recovery by a factor of 2.8. Additionally, the dynamic model extended the average recovery time by 5 months. The inferred allocation shifts influenced the post-drought period by altering foliage and fine root pools, which in turn modulated gross primary productivity and heterotrophic respiration for up to a decade. These changes can create a bust-boom cycle where carbon uptake is enhanced some years after a drought, compared to what would have occurred under drought-free conditions. Overall, allocation shifts accounted for 65% [45%-75%] of drought legacy effects in modeled NBE. In summary, drought-induced carbon allocation shifts can play a substantial role in the enduring influence of drought on cumulative land-atmosphere CO2 exchanges and should be accounted for in ecosystem models.

A hybrid model-data method for seismic response reconstruction of instrumented buildings

This study presents a two-step hybrid (model-data fusion) method for reconstructing the seismic response of instrumented buildings at their non-instrumented floors. Over the past couple of decades, seismic data recorded within instrumented buildings have yielded invaluable insights into the behavior of civil structures, which were arguably impossible to obtain through numerical simulations, laboratory-scale experiments, or even in-situ testing. Recently, advances in sensing technology have opened new pathways for structural health monitoring (SHM) and rapid post-earthquake assessment. However, data-driven techniques tend to lack accuracy when structures have sparse instrumentation. In addition, creating detailed numerical models for the monitored structures is labor-intensive and time-consuming, often unsuitable for rapid post-event assessments. The common approach to address these challenges has been to use simple interpolation techniques over the sparse measurements. However, uncertainties associated with such estimates are usually overlooked, and these methods have certain physical limitations. In this study, we propose a two-step approach for reconstructing seismic responses. In the initial step, a coupled shear–flexural beam model is calibrated using data collected from instrumented floors. Next, the residual, representing the difference between measurements and the beam model’s predictions, is used to train a Gaussian process regression model. The combination of these two models provides both the mean and variance of the response at the non-instrumented floors. This new approach is verified by using simulated acceleration responses of a tall building. Validation is attained by using real seismic data recorded in two tall buildings and comparing the method’s predictions with actual measurements on floors not used for training. Finally, data recorded in a 52-story building during multiple earthquakes are used for demonstrating the practical application of the proposed approach in real-world scenarios.

Multi-type and fine-grained urban green space function mapping based on BERT model and multi-source data fusion

ABSTRACT Urban green space (UGS) is important to the urban ecological environment. It has physical characteristics and social function characteristics and plays an important role in urban climate change, sustainable development goals (SDG) and residents’ health. However, existing researches mostly focus on the extraction of UGS physical features, neglecting the importance of UGS social functions, resulting in the unresolved problem of multi-type and fine-grained functional mapping of UGS. Therefore, based on natural language processing (NLP) and multi-source data fusion, this paper proposes a multi-type and fine-grained UGS function mapping method. First, the social functional standards of UGS have been re-established, with a total of 19 categories. Second, the semantic information in the POI data name text is extracted using the deep learning model, and the reclassification of the UGS type of POI data is realized. Then, combined with multi-source data, 18 types of UGS are extracted. Finally, combining multi-source data to extract urban road green spaces (GS), a fine-grained UGS functional map of Shanghai is created. The results show that the overall accuracy rate of the method is 93.6%, and the Kappa coefficient is 0.93, which proves that the method has good performance in large-scale spatial UGS classification.

An Approach for Fast-Charging Lithium-Ion Batteries State of Health Prediction Based on Model-Data Fusion

Abstract Fast charging has become the norm for various electronic products. The research on the state of health prediction of fast-charging lithium-ion batteries deserves more attention. In this paper, a model-data fusion state of health prediction method which can reflect the degradation mechanism of fast-charging battery is proposed. First, based on the Arrhenius model, the log-power function (LP) model and log-linear (LL) model related to the fast-charging rate are established. Second, combined with Gaussian process regression prediction, a particle filter is used to update the parameters of models in real-time. Compared with the single Gaussian process regression, the average root-mean-square error of LP and LL is reduced by 71.56% and 69.11%, respectively. Finally, the sensitivity and superiority of the two models are analyzed by using Sobol method, Akaike and Bayesian information criterion. The results show that the two models are more suitable for fast-charging lithium batteries than the traditional Arrhenius model, and LP model is better than LL model.

Center of Gravity Position Estimation of Counterbalanced Forklift Truck Based on Multi Model Data Fusion

Center of Gravity Position Estimation of Counterbalanced Forklift Truck Based on Multi Model Data Fusion

Inversion and forward estimation with process-based models: An investigation into cost functions, uncertainty-based weights and model-data fusion

Process-based models for inversion and forward estimation have supported our understanding of earth, environment, and hazards processes. These methods are often applied on spatial data without accounting for their spatial nature and uncertainty. Using an example of reconstructing past volcanic eruption characteristics and associated tephra fallout from different sets of field observation, we demonstrate the importance of making the best use of data-related uncertainty and spatial information in inversion and forward estimation. We present strategies for: (1) the selection of appropriate cost functions accounting for their behaviour and implied distribution of residuals, (2) the treatment of differential uncertainty when combining multiple data, and (3) the leveraging of both model and data when estimating the spatial distribution of output. Results show that a data-informed choice of cost function and accounting for uncertainty and spatial characteristics of data leads to consistent improvements in model predictive performance for both inversion and forward models.

Improved quantification of cover crop biomass and ecosystem services through remote sensing-based model–data fusion

Cover crops have long been seen as an effective management practice to increase soil organic carbon (SOC) and reduce nitrogen (N) leaching. However, there are large uncertainties in quantifying these ecosystem services using either observation (e.g. field measurement, remote sensing data) or process-based modeling. In this study, we developed and implemented a model–data fusion (MDF) framework to improve the quantification of cover crop benefits in SOC accrual and N retention in central Illinois by integrating process-based modeling and remotely-sensed observations. Specifically, we first constrained and validated the process-based agroecosystem model, ecosys, using observations of cover crop aboveground biomass derived from satellite-based spectral signals, which is highly consistent with field measurements. Then, we compared the simulated cover crop benefits in SOC accrual and N leaching reduction with and without the constraints of remotely-sensed cover crop aboveground biomass. When benchmarked with remote sensing-based observations, the constrained simulations all show significant improvements in quantifying cover crop aboveground biomass C compared with the unconstrained ones, with R 2 increasing from 0.60 to 0.87, and root mean square error (RMSE) and absolute bias decreasing by 64% and 97%, respectively. On all study sites, the constrained simulations of aboveground biomass C and N at termination are 29% and 35% lower than the unconstrained ones on average. Correspondingly, the averages of simulated SOC accrual and N retention net benefits are 31% and 23% lower than the unconstrained simulations, respectively. Our results show that the MDF framework with remotely-sensed biomass constraints effectively reduced the uncertainties in cover crop biomass simulations, which further constrained the quantification of cover crop-induced ecosystem services in increasing SOC and reducing N leaching.

Scale variance in the carbon dynamics of fragmented, mixed-use landscapes estimated using model–data fusion

Abstract. Many terrestrial landscapes are heterogeneous. Mixed land cover and land use generate a complex mosaic of fragmented ecosystems at fine spatial resolutions with contrasting ecosystem stocks, traits, and processes, each differently sensitive to environmental and human factors. Representing spatial complexity within terrestrial ecosystem models is a key challenge for understanding regional carbon dynamics, their sensitivity to environmental gradients, and their resilience in the face of climate change. Heterogeneity underpins this challenge due to the trade-off between the fidelity of ecosystem representation within modelling frameworks and the computational capacity required for fine-scale model calibration and simulation. We directly address this challenge by quantifying the sensitivity of simulated carbon fluxes in a mixed-use landscape in the UK to the spatial resolution of the model analysis. We test two different approaches for combining Earth observation (EO) data into the CARDAMOM model–data fusion (MDF) framework, assimilating time series of satellite-based EO-derived estimates of ecosystem leaf area and biomass stocks to constrain estimates of model parameters and their uncertainty for an intermediate complexity model of the terrestrial C cycle. In the first approach, ecosystems are calibrated and simulated at pixel level, representing a “community average” of the encompassed land cover and management. This represents our baseline approach. In the second, we stratify each pixel based on land cover (e.g. coniferous forest, arable/pasture) and calibrate the model independently using EO data specific to each stratum. We test the scale dependence of these approaches for grid resolutions spanning 1 to 0.05∘ over a mixed-land-use region of the UK. Our analyses indicate that spatial resolution matters for MDF. Under the community average baseline approach biological C fluxes (gross primary productivity, Reco) simulated by CARDAMOM are relatively insensitive to resolution. However, disturbance fluxes exhibit scale variance that increases with greater landscape fragmentation and for coarser model domains. In contrast, stratification of assimilated data based on fine-resolution land use distributions resolved the resolution dependence, leading to disturbance fluxes that were 40 %–100 % higher than the baseline experiments. The differences in the simulated disturbance fluxes result in estimates of the terrestrial carbon balance in the stratified experiment that suggest a weaker C sink compared to the baseline experiment. We also find that stratifying the model domain based on land use leads to differences in the retrieved parameters that reflect variations in ecosystem function between neighbouring areas of contrasting land use. The emergent differences in model parameters between land use strata give rise to divergent responses to future climate change. Accounting for fine-scale structure in heterogeneous landscapes (e.g. stratification) is therefore vital for ensuring the ecological fidelity of large-scale MDF frameworks. The need for stratification arises because land use places strong controls on the spatial distribution of carbon stocks and plant functional traits and on the ecological processes controlling the fluxes of C through landscapes, particularly those related to management and disturbance. Given the importance of disturbance to global terrestrial C fluxes, together with the widespread increase in fragmentation of forest landscapes, these results carry broader significance for the application of MDF frameworks to constrain the terrestrial C balance at regional and national scales.

Forest productivity recovery or collapse? Model-data integration insights on drought-induced tipping points.

More frequent and severe droughts are driving increased forest mortality around the globe. We urgently need to describe and predict how drought affects forest carbon cycling and identify thresholds of environmental stress that trigger ecosystem collapse. Quantifying the effects of drought at an ecosystem level is complex because dynamic climate-plant relationships can cause rapid and/or prolonged shifts in carbon balance. We employ the CARbon DAta MOdel fraMework (CARDAMOM) to investigate legacy effects of drought on forest carbon pools and fluxes. Our Bayesian model-data fusion approach uses tower observed meteorological forcing and carbon fluxes to determine the response and sensitivity of aboveground and belowground ecological processes associated with the 2012-2015 California drought. Our study area is a mid-montane mixed conifer forest in the Southern Sierras. CARDAMOM constrained with gross primary productivity (GPP) estimates covering 2011-2017 show a ~75% reduction in GPP, compared to negligible GPP change when constrained with 2011 only. Precipitation across 2012-2015 was 45% (474 mm) lower than the historical average and drove a cascading depletion in soil moisture and carbon pools (foliar, labile, roots, and litter). Adding 157 mm during an especially stressful year (2014, annual rainfall = 293 mm) led to a smaller depletion of water and carbon pools, steering the ecosystem away from a state of GPP tipping-point collapse to recovery. We present novel process-driven insights that demonstrate the sensitivity of GPP collapse to ecosystem foliar carbon and soil moisture states-showing that the full extent of GPP response takes several years to arise. Thus, long-term changes in soil moisture and carbon pools can provide a mechanistic link between drought and forest mortality. Our study provides an example for how key precipitation threshold ranges can influence forest productivity, making them useful for monitoring and predicting forest mortality events.

Mid-Level Data Fusion Combined with the Fingerprint Region for Classification DON Levels Defect of Fusarium Head Blight Wheat.

In this study, a method of mid-level data fusion with the fingerprint region was proposed, which was combined with the characteristic wavelengths that contain fingerprint information in NIR and FT-MIR spectra to detect the DON level in FHB wheat during wheat processing. NIR and FT-MIR raw spectroscopy data on normal wheat and FHB wheat were obtained in the experiment. MSC was used for pretreatment, and characteristic wavelengths were extracted by CARS, MGS and XLW. The variables that can effectively reflect fingerprint information were retained to build the mid-level data fusion matrix. LS-SVM and PLS-DA were applied to investigate the performance of the single spectroscopic model, mid-level data fusion model and mid-level data fusion with fingerprint information model, respectively. The experimental results show that mid-level data fusion with a fingerprint information strategy based on fused NIR and FT-MIR spectra represents an effective method for the classification of DON levels in FHB wheat samples.

Multiscale Representation of Radar Echo Data Retrieved through Deep Learning from Numerical Model Simulations and Satellite Images

Radar reflectivity data snapshot fine-grained atmospheric variations that cannot be represented well by numerical weather prediction models or satellites, which poses a limit for nowcasts based on model–data fusion techniques. Here, we reveal a multiscale representation (MSR) of the atmosphere by reconstructing the radar echoes from the Weather Research and Forecasting (WRF) model simulations and the Himawari-8 satellite products using U-Net deep networks. Our reconstructions generated the echoes well in terms of patterns, locations, and intensities with a root mean square error (RMSE) of 5.38 dBZ. We find stratified features in this MSR, with small-scale patterns such as echo intensities sensitive to the WRF-simulated dynamic and thermodynamic variables and with larger-scale information about shapes and locations mainly captured from satellite images. Such MSRs with physical interpretations may inspire innovative model–data fusion methods that could overcome the conventional limits of nowcasting.

Two-level eddy covariance measurements reduce bias in land-atmosphere exchange estimates over a heterogeneous boreal forest landscape

Estimates of land-atmosphere exchanges of carbon, energy, water vapor, and other greenhouse gases based on the eddy covariance (EC) technique rely on the fundamental assumption that the flux footprint area is homogeneous. We investigated the impact of source area heterogeneity on flux estimates in single-level EC measurements over a managed boreal forest landscape. For this purpose, we compared single-level measurements with those from a two-level approach consisting of concurrent EC measurements at 60 and 85 m above the ground. This two-level set-up provided a unique opportunity to obtain nearly congruent diel footprint areas by combining data from the higher and lower levels during day- and nighttime, respectively.We found that the variation in the averaged footprint area between day- and nighttime was reduced by up to 89% in the two-level approach compared to the single-level data at the higher level (85 m). Considering spring, summer, and fall months, the resulting relative potential bias in flux observations due to landscape heterogeneity was highest at short time steps (≤ daily) ranging between 35% and 325% for half-hourly data. During winter months, when stable atmospheric regimes prevailed during day and night, the footprints within the diel course nearly overlapped also at a given single level and hence no improvement of flux estimates was found. The absolute cumulated sums for the study period (excluding winter months) of gross primary production, ecosystem respiration, latent heat, and sensible heat flux were underestimated by about 28%, 52%, 5%, and 3%, respectively, whereas that of net ecosystem CO2 exchange was overestimated by about 109% in the single-level approach. Overall this study suggests that footprint heterogeneity may introduce considerable bias in single-level flux estimates — particularly at short time scales — with large implications for model-data fusion studies, site comparisons, and up- or downscaling of land-atmosphere exchange processes.

QOMS: A Comprehensive Observation Station for Climate Change Research on the Top of Earth

Abstract Mount Everest (Qomolangma), the highest mountain on Earth, is an unrivaled natural research platform for understanding multispheric interactions over heterogeneous landscapes. The land–atmosphere interactions in this iconic mountain region have paramount importance for weather and climate predictions at both regional and global scales; however, observing and modeling these interactions is inherently challenging due to the extreme environment. The scarcity of multiscale observations hinders progress in this field. Thus, establishing a comprehensive network to systematically observe the land–atmosphere interactions across multiscales in this unrivaled region, is the basis for gaining a better understanding of weather, climate, and climate change. As one of the 69 national observation and research stations in China, the Qomolangma Special Atmospheric Processes and Environmental Changes (QOMS) observation network of land–atmosphere interactions has been established over the northern slope of Mount Everest since 2005. This network consists of six sites with different underlying surfaces, which significantly improves the observational capabilities for the climate system. These observations have promoted the understanding of land–atmosphere interactions and their impacts on multiscale weather patterns, atmospheric circulations, and climate and have provided data support for informing and guiding model development and remote sensing monitoring. Facing an unprecedented opportunity with enormous development possibilities, we emphasize the considerable potential of these observations for understanding and predicting weather and climate in the Himalayas and beyond. Additionally, we expect to extend the future focus to model–data fusion and to societally relevant applications, such as natural disaster prevention and climate change mitigation and adaptation.

Synthesizing forecasts to inform decision‐making and advance ecological theory

Synthesizing forecasts to inform decision‐making and advance ecological theory

Insights into the tussock growth form with model-data fusion.

Some rhizomatous grass and sedge species form tussocks that impact ecosystem structure and function. Despite their importance, tussock development and size controls are poorly understood due to the decadal to centennial timescales over which tussocks form. We explored mechanisms regulating tussock development and size in a ubiquitous arctic tussock sedge (Eriophorum vaginatum) using field observations and a mass balance model coupled with a tiller population model. Model-data fusion was used to quantify parameter and prediction uncertainty, determine model sensitivity, and test hypotheses on the factors regulating tussock size. The model accurately captured the dynamics of tussock development, characteristics, and size observed in the field. Tussock growth approached maximal size within several decades, which was determined by feedbacks between the mass balance of tussock root necromass and density-dependent tillering. The model also predicted that maximal tussock size was primarily regulated by tiller root productivity and necromass bulk density and less so by tiller demography. These predictions were corroborated by field observations of tussock biomass and root characteristics. The study highlights the importance of belowground processes in regulating tussock development and size and enhances our understanding of the influence of tussocks on arctic ecosystem structure and function.

Soil Moisture Assimilation Improves Terrestrial Biosphere Model GPP Responses to Sub-Annual Drought at Continental Scale

Due to the substantial gross exchange fluxes with the atmosphere, the terrestrial carbon cycle plays a significant role in the global carbon budget. Drought commonly affects terrestrial carbon absorption negatively. Terrestrial biosphere models exhibit significant uncertainties in capturing the carbon flux response to drought, which have an impact on estimates of the global carbon budget. Through plant physiological processes, soil moisture tightly regulates the carbon cycle in the environment. Therefore, accurate observations of soil moisture may enhance the modeling of carbon fluxes in a model–data fusion framework. We employ the Carbon Cycle Data Assimilation System (CCDAS) to assimilate 36-year satellite-derived surface soil moisture observations in combination with flask samples of atmospheric CO2 concentrations. We find that, compared to the default model, the performance of optimized net ecosystem productivity (NEP) and gross primary productivity (GPP) has increased with the RMSEs reduced by 1.62 gC/m2/month and 10.84 gC/m2/month, which indicates the added value of the ESA-CCI soil moisture observations as a constraint on the terrestrial carbon cycle. Additionally, the combination of soil moisture and CO2 concentration in this study improves the representation of inter-annual variability of terrestrial carbon fluxes as well as the atmospheric CO2 growth rate. We thereby investigate the ability of the optimized GPP in responding to drought by comparing continentally aggregated GPP with the drought index. The assimilation of surface soil moisture has been shown to efficiently capture the influences of the sub-annual (≤9 months drought durations) and large-scale (e.g., regional to continental scales) droughts on GPP. This study highlights the significant potential of satellite soil moisture for constraining inter-annual models of the terrestrial biosphere’s carbon cycle and for illustrating how GPP responds to drought at a continental scale.

Model-data fusion domain adaptation for battery State of Health estimation with fewer data and simplified feature extractor

Accurate estimation of State of Health (SOH) with fewer data under different operation conditions is challenging for the lithium-ion battery. In this paper, a novel model-data fusion domain adaptation is initiated for SOH estimation. Distinguishing from existing model-based methods and data-driven methods, the proposed method not only can improve the accuracy of estimation with reliability and interpretability, but also can adapt to the different operation conditions with fewer data. Specifically, through introducing the battery models, more reliable and interpretable features are generated according to operation conditions and the layers of the feature extractor can be reduced. With the model-based features, the data-driven margin disparity discrepancy (MDD) based domain adaptation is introduced to transfer knowledge between the dataset of different operation conditions by reducing the probability distribution discrepancy. Therefore, the estimation performance with fewer training data in the target domain is improved with reliable and interpretable features, and with the transfer of knowledge learned from the source domain under various operation conditions. The effectiveness of the proposed method is verified by the NASA dataset and our experiment on the FST battery. The results demonstrate the estimation performance can be improved with fewer data and the reduction of layers in the feature extractor.