Search NASA⌕ Search

SEARCH · Search NASA

Results for “machine learning, Random Forest”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 145 records · Page 8

Machine learning strategies for systems with invariance properties

Here, in many scientific fields, empirical models are employed to facilitate computational simulations of engineering systems. For example, in fluid mechanics, empirical Reynolds stress closures enable computationally-efficient Reynolds-Averaged Navier-Stokes simulations. Likewise, in solid mechanics, constitutive relations between the stress and strain in a material are required in deformation analysis. Traditional methods for developing and tuning empirical models usually combine physical intuition with simple regression techniques on limited data sets. The rise of high-performance computing has led to a growing availability of high-fidelity simulation data, which open up the possibility of using machine learning algorithms, such as random forests or neural networks, to develop more accurate and general empirical models. A key question when using data-driven algorithms to develop these models is how domain knowledge should be incorporated into the machine learning process. This paper will specifically address physical systems that possess symmetry or invariance properties. Two different methods for teaching a machine learning model an invariance property are compared. In the first , a basis of invariant inputs is constructed, and the machine learning model is trained upon this basis, thereby embedding the invariance into the model. In the second method, the algorithm is trained on multiple transformations of the raw input data until the model learns invariance to that transformation. Results are discussed for two case studies: one in turbulence modeling and one in crystal elasticity. It is shown that in both cases embedding the invariance property into the input features yields higher performance with significantly reduced computational training costs.

97 MATHEMATICS AND COMPUTING↗

Modeling Spatial Distribution of Snow Water Equivalent by Combining Meteorological and Satellite Data with Lidar Maps

Abstract An accurate characterization of the water content of snowpack, or snow water equivalent (SWE), is necessary to quantify water availability and constrain hydrologic and land surface models. Recently, airborne observations (e.g., lidar) have emerged as a promising method to accurately quantify SWE at high resolutions (scales of ∼100 m and finer). However, the frequency of these observations is very low, typically once or twice per season in the Rocky Mountains of Colorado. Here, we present a machine learning framework that is based on random forests to model temporally sparse lidar-derived SWE, enabling estimation of SWE at unmapped time points. We approximated the physical processes governing snow accumulation and melt as well as snow characteristics by obtaining 15 different variables from gridded estimates of precipitation, temperature, surface reflectance, elevation, and canopy. Results showed that, in the Rocky Mountains of Colorado, our framework is capable of modeling SWE with a higher accuracy when compared with estimates generated by the Snow Data Assimilation System (SNODAS). The mean value of the coefficient of determination R 2 using our approach was 0.57, and the root-mean-square error (RMSE) was 13 cm, which was a significant improvement over SNODAS (mean R 2 = 0.13; RMSE = 20 cm). We explored the relative importance of the input variables and observed that, at the spatial resolution of 800 m, meteorological variables are more important drivers of predictive accuracy than surface variables that characterize the properties of snow on the ground. This research provides a framework to expand the applicability of lidar-derived SWE to unmapped time points. Significance Statement Snowpack is the main source of freshwater for close to 2 billion people globally and needs to be estimated accurately. Mountainous snowpack is highly variable and is challenging to quantify. Recently, lidar technology has been employed to observe snow in great detail, but it is costly and can only be used sparingly. To counter that, we use machine learning to estimate snowpack when lidar data are not available. We approximate the processes that govern snowpack by incorporating meteorological and satellite data. We found that variables associated with precipitation and temperature have more predictive power than variables that characterize snowpack properties. Our work helps to improve snowpack estimation, which is critical for sustainable management of water resources.

54 ENVIRONMENTAL SCIENCES↗

Machine learning models for rat multigeneration reproductive toxicity prediction

Reproductive toxicity is one of the prominent endpoints in the risk assessment of environmental and industrial chemicals. Due to the complexity of the reproductive system, traditional reproductive toxicity testing in animals, especially guideline multigeneration reproductive toxicity studies, take a long time and are expensive. Therefore, machine learning, as a promising alternative approach, should be considered when evaluating the reproductive toxicity of chemicals. We curated rat multigeneration reproductive toxicity testing data of 275 chemicals from ToxRefDB (Toxicity Reference Database) and developed predictive models using seven machine learning algorithms (decision tree, decision forest, random forest, k-nearest neighbors, support vector machine, linear discriminant analysis, and logistic regression). A consensus model was built based on the seven individual models. An external validation set was curated from the COSMOS database and the literature. The performances of individual and consensus models were evaluated using 500 iterations of 5-fold cross-validations and the external validation data set. The balanced accuracy of the models ranged from 58% to 65% in the 5-fold cross-validations and 45%–61% in the external validations. Prediction confidence analysis was conducted to provide additional information for more appropriate applications of the developed models. The impact of our findings is in increasing confidence in machine learning models. We demonstrate the importance of using consensus models for harnessing the benefits of multiple machine learning models (i.e., using redundant systems to check validity of outcomes). While we continue to build upon the models to better characterize weak toxicants, there is current utility in saving resources by being able to screen out strong reproductive toxicants before investing in vivo testing. The modeling approach (machine learning models) is offered for assessing the rat multigeneration reproductive toxicity of chemicals. Our results suggest that machine learning may be a promising alternative approach to evaluate the potential reproductive toxicity of chemicals.

consensus model↗

Probabilistic Forecasting of Generators Startups and Shutdowns in the MISO System Based on Random Forest

Solving security constrained unit commitment (SCUC) problems to plan an economical generation schedule for day-head electricity market has been an important research topic in recent years. Mixed integer programming method (MIP), the-state-of-art approach for solving SCUC problem, is known computationally hard when the number of binary status variables is large. In this paper, a machine learning-based algorithm - random forest (RF), was applied to forecast the startups (SU) and shutdowns (SD) hours of generators, based on historical hourly system condition observations in the Midcontinent Independent System Operator (MISO) system. The main purpose is to reduce the number of binary status variables, by fixing the SU/SD hours to a narrow range of high confidence. This would significantly reduce the size of the decision space, and therefore speed up SCUC solutions with reduced uncertainty.

Lin, Xinming↗

On the estimation of boundary layer heights: a machine learning approach

Abstract. The planetary boundary layer height (zi) is a key parameter used in atmospheric models for estimating the exchange of heat, momentum, and moisture between the surface and the free troposphere. Near-surface atmospheric and subsurface properties (such as soil temperature, relative humidity, etc.) are known to have an impact on zi. Nevertheless, precise relationships between these surface properties and zi are less well known and not easily discernible from the multi-year dataset. Machine learning approaches, such as random forest (RF), which use a multi-regression framework, help to decipher some of the physical processes linking surface-based characteristics to zi. In this study, a 4-year dataset from 2016 to 2019 at the Southern Great Plains site is used to develop and test a machine learning framework for estimating zi. Parameters derived from Doppler lidars are used in combination with over 20 different surface meteorological measurements as inputs to a RF model. The model is trained using radiosonde-derived zi values spanning the period from 2016 through 2018 and then evaluated using data from 2019. Results from 2019 showed significantly better agreement with the radiosonde compared to estimates derived from a thresholding technique using Doppler lidars only. Noteworthy improvements in daytime zi estimates were observed using the RF model, with a 50 % improvement in mean absolute error and an R2 of greater than 85 % compared to the Tucker method zi. We also explore the effect of zi uncertainty on convective velocity scaling and present preliminary comparisons between the RF model and zi estimates derived from atmospheric models.

54 ENVIRONMENTAL SCIENCES↗

Towards fast and accurate predictions of radio frequency power deposition and current profile via data-driven modelling: applications to lower hybrid current drive

Three machine learning techniques (multilayer perceptron, random forest and Gaussian process) provide fast surrogate models for lower hybrid current drive (LHCD) simulations. A single GENRAY/CQL3D simulation without radial diffusion of fast electrons requires several minutes of wall-clock time to complete, which is acceptable for many purposes, but too slow for integrated modelling and real-time control applications. The machine learning models use a database of more than 16 000 GENRAY/CQL3D simulations for training, validation and testing. Latin hypercube sampling methods ensure that the database covers the range of nine input parameters ( $n_{e0}$ , $T_{e0}$ , $I_p$ , $B_t$ , $R_0$ , $n_{\|}$ , $Z_{{\rm eff}}$ , $V_{{\rm loop}}$ and $P_{{\rm LHCD}}$ ) with sufficient density in all regions of parameter space. The surrogate models reduce the inference time from minutes to $\sim$ ms with high accuracy across the input parameter space.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Towards Fast and Accurate Predictions of Radio Frequency Power Deposition and Current Profile via Data-driven Modeling

Three machine learning techniques (multilayer perceptron, random forest, and Gaussian process) provide fast surrogate models for lower hybrid current drive (LHCD) simulations. A single GENRAY/CQL3D simulation without radial diffusion of fast electrons requires several minutes of wall-clock time to complete, which is acceptable for many purposes, but too slow for integrated modeling and real-time control applications. The machine learning models use a database of 16,000+ GENRAY/CQL3D simulations for training, validation, and testing. Latin hypercube sampling methods ensure that the database covers the range of 9 input parameters ($n_{e0}$, $T_{e0}$, $I_p$, $B_t$, $R_0$, $n_{||}$, $Z_{eff}$, $V_{loop}$, $P_{LHCD}$) with sufficient density in all regions of parameter space. The surrogate models reduce the inference time from minutes to ~ms with high accuracy across the input parameter space.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

A Machine Learning Approach to Quantitative Analysis of Enamel Microstructure from Scanning Electron Microscopy Images

Dental enamel, the outermost tissue of mammalian teeth, must withstand a lifetime of wear and cyclic contact. To meet this demand, enamel possesses a combination of high hardness and resistance to fracture, properties that are typically mutually exclusive. The impressive damage tolerance has been attributed largely to decussation of the enamel rods, the principal unit of its microstructure. As such, enamel is inspiring the design of next‐generation structural materials. However, quantitative descriptions of the decussated enamel rod microstructure remain limited due to challenges encountered in applying computed tomography and in acquiring quality images appropriate for traditional digital processing methods. Here, a machine learning segmentation method is applied to images of the enamel obtained using scanning electron microscopy to support quantitative analysis of the microstructure. A pretrained convolutional neural network is used to expand the input training image dataset to allow the training of a random forest classifier, which ultimately segments the image with a very small training set ( n = 3 images). A validation of this segmentation method is presented, in addition to its application to calculate relevant microstructural parameters for images of tooth enamel from selected mammalian species. The methodology applied here is equally applicable to other hard tissues.

36 MATERIALS SCIENCE↗

A comparative study of machine learning models for predicting the state of reactive mixing

Mixing phenomena are important mechanisms controlling flow, species transport, and reaction processes in fluids and porous media. Accurate predictions of reactive mixing are critical for many Earth and environmental science problems such as contaminant fate and remediation, macroalgae growth, and plankton biomass evolution. Here, to investigate the evolution of mixing dynamics under different scenarios (e.g., anisotropy, fluctuating velocity fields), a finite-element-based numerical model was built to solve the fast, irreversible bimolecular reaction-diffusion equations to simulate a range of reactive-mixing scenarios. A total of 2,315 simulations were performed using different sets of model input parameters comprising various spatial scales of vortex structures in the velocity field, time-scales associated with velocity oscillations, the perturbation parameter for the vortex-based velocity, anisotropic dispersion contrast (i.e., ratio of longitudinal-to-transverse dispersion), and molecular diffusion. The outputs comprised concentration profiles of reactants and products. The inputs to and outputs from these simulations were concatenated into feature and label matrices, respectively, to train 20 different machine learning (ML) models intended to emulate system behavior. These 20 ML emulators, based on linear methods, Bayesian methods, ensemble learning methods, and multilayer perceptrons (MLPs), were trained to classify the state of mixing and predict three quantities of interest (QoIs) characterizing species production, decay (i.e., average concentration, square of average concentration), and degree of mixing (i.e., variances of species concentration). Unsurprisingly, linear classifiers and regressors failed to reproduce the QoIs; however, ensemble methods (classifiers and regressors) and the MLP model accurately classified the state of reactive mixing and the QoIs. Among ensemble methods, random forest and decision-tree-based AdaBoost faithfully predicted the QoIs. At run time, trained ML emulators produced results times faster than the finite-element simulations. Due to their low computational expense and high accuracy, ensemble and MLP models are excellent emulators for these numerical simulations and great utilities in uncertainty quantification exercises, which can require 1,000s of forward model runs.

97 MATHEMATICS AND COMPUTING↗

Towards fast, accurate predictions of RF simulations via data-driven modeling: Forward and lateral models

Three machine learning techniques (multilayer perceptron, random forest, and Gaussian process) provide fast surrogate models for lower hybrid current drive (LHCD) simulations. A single GENRAY/CQL3D simulation without radial diffusion of fast electrons requires several minutes of wall-clock time to complete, which is acceptable for many purposes, but too slow for integrated modeling and real-time control applications. More accurate simulations with fast electron diffusion are even slower, requiring multiple hours of run time with parallel processing. The machine learning models use a database of 16,000+ GEN-RAY/CQL3D simulations for training, validation, and testing. Latin hypercube sampling methods implemented in πScope ensure that the database covers the range of 9 input parameters (n e0 , T e0 , I p , B t , R 0 , n ∥︀ , Z e f f , V loop , P LHCD ) with sufficient density in all regions of parameter space. The surrogate models reduce the computation time from minutes-hours to ms with high accuracy across the input parameter space. Data-driven surrogate models also allow for solving inverse and “lateral” problems. A surrogate model for the inverse problem maps from a desired current drive or power deposition profile to a set of input parameters that would result in such a profile, while a surrogate model for the lateral problem maps from a measured experimental quantity such as hard x-ray emission to a current drive or power deposition profile. In conclusion, the πScope database creation workflow is flexible and applicable to other RF simulation codes such as TORIC.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Random Forests as a Viable Method to Select and Discover High-redshift Quasars

We present a method of selecting quasars up to redshift ≈6 with random forests, a supervised machine-learning method, applied to Pan-STARRS1 and WISE data. We find that, thanks to the increasing set of known quasars, we can assemble a training set that enables supervised machine-learning algorithms to become a competitive alternative to other methods up to this redshift. We present a candidate set for the redshift range 4.8–6.3, which includes the region around z = 5.5 where selecting quasars is difficult due to their photometric similarity to red and brown dwarfs. We demonstrate that, under our survey restrictions, we can reach a high completeness (66% ± 7% below redshift 5.6/83{sub -9}{sup +6}% above redshift 5.6) while maintaining a high selection efficiency (78{sub -8}{sup +10}%/94{sub -8}{sup +5}%). Our selection efficiency is estimated via a novel method based on the different distributions of quasars and contaminants on the sky. The final catalog of 515 candidates includes 225 known quasars. We predict the candidate catalog to contain additional 148{sub -33}{sup +41} new quasars below redshift 5.6 and 45{sub -8}{sup +5} above, and we make the catalog publicly available. Spectroscopic follow-up observations of 37 candidates led us to discover 20 new high redshift quasars (18 at 4.6 ≤ z ≤ 5.5, 2 z ~ 5.7). These observations are consistent with our predictions on efficiency. We argue that random forests can lead to higher completeness because our candidate set contains a number of objects that would be rejected by common color cuts, including one of the newly discovered redshift 5.7 quasars.

79 ASTRONOMY AND ASTROPHYSICS↗

IRNet: A General Purpose Deep Risidual Regression Framework for Materials Discovery

Materials discovery is crucial for making scientific advances in many domains. Collections of data from experiments and first-principle computations have spurred interest in applying machine learning methods to create predictive models capable of mapping from composition and crystal structures to materials properties. Generally, these are regression problems with the input being a 1D vector composed of numerical attributes representing the material composition and/or crystal structure. While neural networks consisting of fully connected layers have been applied to such problems, their performance often suffers from the vanishing gradient problem when network depth is increased. Hence, predictive modeling for such tasks has been mainly limited to traditional machine learning techniques such as Random Forest. In this paper, we study and propose design principles for building deep regression networks composed of fully connected layers with numerical vectors as input. We introduce a novel deep regression network with individual residual learning, IRNet, that places shortcut connections after each layer so that each layer learns the residual mapping between its output and input. We use the problem of learning properties of inorganic materials from numerical attributes derived from material composition and/or crystal structure to compare IRNet’s performance against that of other machine learning techniques. Using multiple datasets from the Open Quantum Materials Database (OQMD) and Materials Project for training and evaluation, we show that IRNet provides significantly better prediction performance than the state-of-the-art machine learning approaches currently used by domain scientists. We also show that IRNet’s use of individual residual learning leads to better convergence during the training phase than when shortcut connections are between multi-layer stacks while maintaining the same number of parameters.

deep learning↗

A general spatial-temporal framework for short-term building temperature forecasting at arbitrary locations with crowdsourcing weather data

Weather forecasting has been a critical component to predict and control building energy consumption for better building energy management. Without accessibility to other data sources, the onsite observed temperatures or the airport temperatures are used in forecast models. In this paper, we present a novel approach by utilizing the crowdsourcing weather data from neighboring personal weather stations (PWS) to improve the weather forecast accuracy around buildings using a general spatial-temporal modeling framework. The final forecast is based on the ensemble of local forecasts for the target location using neighboring PWSs. Our approach is distinguished from existing literature in various aspects. First, we leverage the crowdsourcing weather data from PWS in addition to public data sources. In this way, the data is at much finer time resolution (e.g., at 5-minute frequency) and spatial resolution (e.g., arbitrary location vs grid). Second, our proposed model incorporates spatial-temporal correlation information of weather variables between the target building and a set of neighboring PWSs so that underlying correlations can be effectively captured to improve forecasting performance. Here, we demonstrate the performance of the proposed framework by comparing to the benchmark models on temperature forecasting for a building located at an arbitrary location at San Antonio, Texas, USA. In general, the proposed model framework equipped with machine learning technique such as Random Forest can improve forecasting by 50% compares with persistent model and has 90% chance to outperform airport forecast in short-term forecasting. In a real-time setting, the proposed model framework can provide more accurate temperature forecasting results compared with using airport temperature forecast for most forecast horizon. Moreover, we analyze the sensitivity of model parameters to gain insights on how crowdsourcing data from the neighboring personal weather stations impacts forecasting performance. Finally, we implement our model in other cities such as Syracuse and Chicago to test the model's performance in different landforms and climate types.

54 ENVIRONMENTAL SCIENCES↗

Machine learning to predict final fire size at the time of ignition

Fires in boreal forests of Alaska are changing, threatening human health and ecosystems. Given expected increases in fire activity with climate warming, insight into the controls on fire size from the time of ignition is necessary. Such insight may be increasingly useful for fire management, especially in cases where many ignitions occur in a short time period. Here we investigated the controls and predictability of final fire size at the time of ignition. Using decision trees, we show that ignitions can be classified as leading to small, medium or large fires with 50.4 ± 5.2% accuracy. This was accomplished using two variables: vapour pressure deficit and the fraction of spruce cover near the ignition point. The model predicted that 40% of ignitions would lead to large fires, and those ultimately accounted for 75% of the total burned area. Other machine learning classification algorithms, including random forests and multi-layer perceptrons, were tested but did not outperform the simpler decision tree model. Applying the model to areas with intensive human management resulted in overprediction of large fires, as expected. This type of simple classification system could offer insight into optimal resource allocation, helping to maintain a historical fire regime and protect Alaskan ecosystems.

54 ENVIRONMENTAL SCIENCES↗

Comparison of four learning-based methods for predicting groundwater redox status

Knowing the location where groundwater denitrification occurs, or by proxy the groundwater redox status (oxic, mixed, and anoxic), is valuable information for assessing and managing potential agricultural land-use impacts on freshwater quality. We compare the efficacy of supervised (Linear Discriminant Analysis LDA; Boosted Regression Trees, BRT; and Random Forest, RF) and unsupervised (Modified Self-Organizing Map, MSOM) learning-based methods to predict groundwater redox status in the agriculturally dominated Tasman, Waikato, and Wellington regions of New Zealand. Thresholds applied to regional groundwater-quality samples provide redox status variables and learn heuristics constrained by these variables and applied to spatial factors (climate, elevation, geologic, hydrology soils, and well depth) identify optimal sets of regional predictor variables. A split-sample approach is used to train and test the learning methods ability to predict redox status using the optimal predictor variables. Overall, the supervised methods demonstrate a prediction bias toward oxic conditions and inability to perform statistically well when using independent regional data; for example, consider kappa statistics for BRT (Tasman: 0.42, Waikato: 0.38, Wellington: 0.17), RF (Tasman: 0.42, Waikato: 0.47, Wellington: 0.17 and LDA (Tasman: 0.46, Waikato: 0.32, Wellington: 0.17). By contrast, the unsupervised method performs statistically well when predicting oxic, mixed, and anoxic conditions and corresponding depths when using independent regional data; for example, consider MSOM kappa statistics for Tasman: 0.78, Waikato: 0.80, Wellington: 0.76. The unsupervised learning method provides the added benefits of being (1) able to combine predictions into 3D regional anoxic probability plots for interpreting the spatial influence of paleosols and groundwater flowpaths on redox status, and (2) readily extended to map 3D redox status across New Zealand and other countries despite data bias and sparsity.

machine learning algorithms, Random Forest, booste↗

Information Content of JWST NIRSpec Transmission Spectra of Warm Neptunes

Warm Neptunes offer a rich opportunity for understanding exo-atmospheric chemistry. With the upcoming James Webb Space Telescope (JWST), there is a need to elucidate the balance between investments in telescope time versus scientific yield. We use the supervised machine-learning method of the random forest to perform an information content (IC) analysis on a 11-parameter model of transmission spectra from the various NIRSpec modes. The three bluest medium-resolution NIRSpec modes (0.7–1.27 μm, 0.97–1.84 μm, 1.66–3.07 μm) are insensitive to the presence of CO. The reddest medium-resolution mode (2.87–5.10 μm) is sensitive to all of the molecules assumed in our model: CO, CO{sub 2}, CH{sub 4}, C{sub 2}H{sub 2}, H{sub 2}O, HCN, and NH{sub 3}. It competes effectively with the three bluest modes on the information encoded on cloud abundance and particle size. It is also competitive with the low-resolution prism mode (0.6–5.3 μm) on the inference of every parameter except for the temperature and ammonia abundance. We recommend astronomers to use the reddest medium-resolution NIRSpec mode for studying the atmospheric chemistry of 800–1200 K warm Neptunes; its corresponding high-resolution counterpart offers diminishing returns. We compare our findings to previous JWST IC analyses that favor the blue orders and suggest that the reliance on chemical equilibrium could lead to biased outcomes if this assumption does not apply. A simple, pressure-independent diagnostic for identifying chemical disequilibrium is proposed based on measuring the abundances of H{sub 2}O, CO, and CO{sub 2}.

79 ASTRONOMY AND ASTROPHYSICS↗

Classification Analysis of Southwest Pacific Tropical Cyclone Intensity Changes Prior to Landfall

This study evaluates the ability of a random forest classifier to identify tropical cyclone (TC) intensification or weakening prior to landfall over the western region of the Southwest Pacific Ocean (SWPO) basin. For both Australia mainland and SWPO island cases, when a TC first crosses land after spending ≥24 h over the ocean, the closest hour prior to the intersection is considered as the landfall hour. If the maximum wind speed (V max ) at the landfall hour increased or remained the same from the 24-h mark prior to landfall, the TC is labeled as intensifying and if the V max at the landfall hour decreases, the TC is labeled as weakening. Geophysical and aerosol variables closest to the 24 h before landfall hour were collected for each sample. The random forest model with leave-one-out cross validation and the random oversampling example technique was identified as the best-performing classifier for both mainland and island cases. The model identified longitude, initial intensity, and sea skin temperature as the most important variables for the mainland and island landfall classification decisions. Incorrectly classified cases from the test data were analyzed by sorting the cases by their initial intensity hour, landfall hour, monthly distribution, and 24-h intensity changes. TC intensity changes near land strongly impact coastal preparations such as wind damage and flood damage mitigations; hence, this study will contribute to improve identifying and prioritizing prediction of important variables contributing to TC intensity change before landfall.

54 ENVIRONMENTAL SCIENCES↗

Estimating soybean yields from high-temporal-resolution multi-source data using deep learning

Accurate and timely crop yield prediction is crucial for ensuring food security and maintaining stable agricultural markets. In recent years, there has been a surge in interest in leveraging high-temporal-resolution, multi-source data for effective crop growth monitoring and yield estimation. A notable challenge arises from the difficulty in capturing the intricate interactions between variables across different time steps within these high-temporal-resolution time series datasets. This complexity hinders the reliable extraction of yield information from voluminous and often noisy datasets, especially during periods of extreme weather events. Here, in this study, we propose an Attention and Graph Isomorphism Network-enhanced Bi-directional Long Short-Term Memory network (AGB-LSTM) for estimating county-level soybean yield in the United States. This model integrates a diverse set of remote sensing data, including Near-Infrared Reflectance of Vegetation (NIRv), Sun-Induced chlorophyll Fluorescence (SIF), and Gross Primary Productivity (GPP), along with environmental covariates. The AGB-LSTM effectively leverages information related to crop yield from high-temporal-resolution time series data (5-days), achieving an accuracy of R²= 0.67 and rRMSE = 14.46%. This approach significantly outperforms traditional machine learning methods such as Random Forest (RF) (R²= 0.52, rRMSE = 17.36%) and Bi-LSTM (R²= 0.58, rRMSE = 16.17%). Sensitivity experiments with different time steps and ranges demonstrated that our model could accurately and stably predict yields 1 to 2 months before harvest. Moreover, data with a finer temporal resolution consistently improved prediction performance, resulting in an approximately 20% increase in and an approximately 20% decrease in rRMSE compared to using monthly composites. We also evaluated the robustness of the model under extreme climate events and observed strong performance (R²= 0.50, rRMSE = 21.32%). Finally, yield mapping for major soybean-producing regions in North America in 2023 revealed spatial patterns that closely matched USDA yield reports. Our findings suggest that the AGB-LSTM model is a promising and effective method for estimating yield and has notable potential for global crop yield forecasting.

Deep learning↗