Results for “bias in machine learning” — Page 2

NASA NTRS2021

A Chlorophyll-a Algorithm for Landsat-8 Based on Mixture Density Networks

Retrieval of aquatic biogeochemical variables, such as the near-surface concentration of chlorophyll-a (Chla) in inland and coastal waters via remote observations, has long been regarded as a challenging task. This manuscript applies Mixture Density Networks (MDN) that use the visible spectral bands available by the Operational Land Imager (OLI) aboard Landsat-8 to estimate Chla. We utilize a database of co-located in situ radiometric and Chla measurements (N = 4,354), referred to as Type A data, to train and test an MDN model (MDN(A)). This algorithm’s performance, having been proven for other satellite missions, is further evaluated against other widely used machine learning models (e.g., support vector machines), as well as other domain-specific solutions (OC3), and shown to offer significant advancements in the field. Our performance assessment using a held-out test data set suggests that a 49% (median) accuracy with near-zero bias can be achieved via the MDN(A) model, offering improvements of 20 to 100% in retrievals with respect to other models. The sensitivity of the MDN(A) model and benchmarking methods to uncertainties from atmospheric correction (AC) methods, is further quantified through a semi-global matchup dataset (N = 3,337), referred to as Type B data. To tackle the increased uncertainties, alternative MDN models (MDN(B)) are developed through various features of the Type B data (e.g., Rayleigh-corrected reflectance spectra ρ(s)). Using held-out data, along with spatial and temporal analyses, we demonstrate that these alternative models show promise in enhancing the retrieval accuracy adversely influenced by the AC process. Results lend support for the adoption of MDN(B) models for regional and potentially global processing of OLI imagery, until a more robust AC method is developed. Index Terms—Chlorophyll-a, coastal water, inland water, Landsat-8, machine learning, ocean color, aquatic remote sensing.

Brandon Smith↗

NASA NTRS2022

Local Scale (3-M) Soil Moisture Mapping Using SMAP and Planet Superdove

A capability for mapping meter-level resolution soil moisture with frequent temporal sampling over large regions is essential for quantifying local-scale environmental heterogeneity and eco-hydrologic behavior. However, available surface soil moisture (SSM) products generally involve much coarser grain sizes ranging from 30 m to several 10s of kilometers. Hence a new method is proposed to estimate 3-m resolution SSM using a combination of multi-sensor fusion, machine- learning (ML) and Cumulative Distribution Function (CDF) matching approaches. This method established favorable SSM correspondence between 3-m pixels and overlying 9-km grid cells from overlapping Planet SuperDove (PSD) observations and NASA Soil Moisture Active-Passive (SMAP) mission products. The resulting 3-m SSM predictions showed improved accuracy by reducing ab- solute bias and RMSE by ~0.01 cm3/cm3 over the original SMAP data in relation to in-situ soil moisture measurements for the Australian Yanco region, while preserving the high sampling frequency (1-3 day global revisit) and sensitivity to surface wetness (R 0.865) from SMAP. Heterogeneous soil moisture distributions varying with vegetation biomass gradients and irrigation regimes were generally captured within a selected study area. Further algorithm refinement and implementation for regional applications will allow for improvement in water resources management, precision agriculture, and disaster forecasts and responses.

soil moisture↗

NASA NTRS1991

Acquiring Software Design Schemas: A Machine Learning Perspective

In this paper, we describe an approach based on machine learning that acquires software design schemas from design cases of existing applications. An overview of the technique, design representation, and acquisition system are presented. the paper also addresses issues associated with generalizing common features such as biases. The generalization process is illustrated using an example.

Harandi, Mehdi T.↗

NASA NTRSDate not supplied

Validation of Machine Learning Algorithms for Hyperspectral Inversion of Common Water Quality Indicators

The upcoming transition to a diverse suite hyperspectral airborne and orbiting optical sensors will provide an unprecedented opportunity to measure inland water quality characteristics at a fidelity not previously achievable. This presentation will assess prototype deep learning models trained on synthetic hyperspectral data and validated with collocated in-situ measurements. Synthesized data is becoming increasingly popular for use in data-driven approaches to complex problems, and can compliment real data to increase performance on complex and unusual phenomenon, reduce or test bias, and experiment to demonstrate explainability. We will present insights from hyperspectral inversions of Chlorophyl-a, Phycocyanin, and concentration of non-algal particles using selected orbiting and airborne sensors over diverse, optically complex aquatic scenarios. We analyze how various optical water types affect fidelity of results and where improvements can be made as we prototype for globally operational water quality algorithms which can be leveraged by upcoming hyperspectral missions such as the Surface Biology and Geology (SBG) mission.

Surface Biology and Geology (SBG)↗

NASA NTRSDate not supplied

Using Machine Learning to Infer Material Properties of Debris Fragments from X-ray Images in the DebriSat Project

The DebriSat project is a collaboration effort with the NASA Orbital Debris Program Office, the U.S. Space Force Space Systems Command Center, The Aerospace Corporation, and the University of Florida. To date, over 200,000 fragments from this ground-based, hypervelocity impact experiment have been collected, and processing is underway to determine their physical characteristics, such as material, shape, color, characteristic length, and average cross-sectional area. The x-ray process is primarily used to identify the location of the fragments and estimated size for extraction, so that these physical characteristics can be assessed. This paper proposes a machine learning-based approach to characterize materials from x-ray images of debris fragments embedded in soft-catch foam used in the DebriSat project. The novel methodology discussed in this paper will highlight the use of x-ray imagery data to characterize these fragments without extraction or a human-in-the-loop. Both supervised and unsupervised machine learning techniques are utilized with this approach to infer the physical parameters of the fragments embedded in the soft-catch foam panels used in the impact experiment based on x-ray images of the foam panels. Additionally, 3D reconstructions of the extracted fragments are created with images taken from two different angles using the structure from motion (SfM) method. The characteristic lengths and shape from the 3D reconstruction, alongside the physical characteristics of the debris, are used in the inference of the material type. To develop and test the approach, a dataset of x-ray images of debris fragments of varying sizes and materials is collected. Supervised learning methods such as convolutional neural networks (CNNs), support vector machines (SVM), decision trees, and random forest classifiers are used due to the high-dimensional feature spaces of the debris and nonlinear decision boundaries for material categorization. Given the limited pre-labeled data of embedded debris materials smaller than 10 mm, unsupervised machine learning techniques such as clustering algorithms and autoencoders are used, in addition to supervised learning methods. The clustering algorithms group similar fragments together based on their physical properties, and autoencoders reduce the dimensionality of the x ray images and extract relevant features. The performance of the proposed approach's is analyzed using a range of statistical methods, including confusion matrices, receiver operating characteristic curves, and precision-recall curves. The results are compared with those obtained using a baseline approach that relies on manual identification and classification of debris fragments. To evaluate the effectiveness of different machine learning methods, statistical tests such as t-tests, ANOVA, and cross-validation are performed, comparing the performance of CNNs, SVMs, clustering algorithms, and autoencoders. Additional analysis needs to be conducted to identify any sources of bias or variability that may affect the results, such as variations in imaging conditions or fragmentation patterns. Other topics explored are limitations, refinements, and the potential use of semi-supervised learning techniques, such as self-training to label unlabeled datasets and co-training using x-ray images taken from two different angles as two different models.

Saik Anam Siam↗

NASA NTRS2023

Using Machine Learning to Infer Material Properties of Debris Fragments from X-ray Images in the DebriSat Project

The DebriSat project is a collaboration effort with the NASA Orbital Debris Program Office, the U.S. Space Force Space Systems Command Center, The Aerospace Corporation, and the University of Florida. To date, over 200,000 fragments from this ground-based, hypervelocity impact experiment have been collected, and processing is underway to determine their physical characteristics, such as material, shape, color, characteristic length, and average cross-sectional area. The x-ray process is primarily used to identify the location of the fragments and estimated size for extraction, so that these physical characteristics can be assessed. This paper proposes a machine learning-based approach to characterize materials from x-ray images of debris fragments embedded in soft-catch foam used in the DebriSat project. The novel methodology discussed in this paper will highlight the use of x-ray imagery data to characterize these fragments without extraction or a human-in-the-loop. Both supervised and unsupervised machine learning techniques are utilized with this approach to infer the physical parameters of the fragments embedded in the soft-catch foam panels used in the impact experiment based on x-ray images of the foam panels. Additionally, 3D reconstructions of the extracted fragments are created with images taken from two different angles using the structure from motion (SfM) method. The characteristic lengths and shape from the 3D reconstruction, alongside the physical characteristics of the debris, are used in the inference of the material type. To develop and test the approach, a dataset of x-ray images of debris fragments of varying sizes and materials is collected. Supervised learning methods such as convolutional neural networks (CNNs), support vector machines (SVM), decision trees, and random forest classifiers are used due to the high-dimensional feature spaces of the debris and nonlinear decision boundaries for material categorization. Given the limited pre-labeled data of embedded debris materials smaller than 10 mm, unsupervised machine learning techniques such as clustering algorithms and autoencoders are used, in addition to supervised learning methods. The clustering algorithms group similar fragments together based on their physical properties, and autoencoders reduce the dimensionality of the x ray images and extract relevant features. The performance of the proposed approach's is analyzed using a range of statistical methods, including confusion matrices, receiver operating characteristic curves, and precision-recall curves. The results are compared with those obtained using a baseline approach that relies on manual identification and classification of debris fragments. To evaluate the effectiveness of different machine learning methods, statistical tests such as t-tests, ANOVA, and cross-validation are performed, comparing the performance of CNNs, SVMs, clustering algorithms, and autoencoders. Additional analysis needs to be conducted to identify any sources of bias or variability that may affect the results, such as variations in imaging conditions or fragmentation patterns. Other topics explored are limitations, refinements, and the potential use of semi-supervised learning techniques, such as self-training to label unlabeled datasets and co-training using x-ray images taken from two different angles as two different models.

Saik Anam Siam↗

NASA NTRSDate not supplied

Dense Feature Tracking of Atmospheric Winds with Deep Optical Flow

Atmospheric winds are a key physical phenomenon impacting natural hazards, energy transport, ocean currents, large-scale circulation, and ecosystem fluxes. Observing winds is a complex process and presents a large gap in NASA’s Earth Observation System. Atmospheric motion vectors (AMVs) aim to fill this gap by making numerical estimates of cloud movement between sequences of multi-spectral satellite images, tracking clouds and water vapor. Recent imaging hardware and software advancements have enabled the use of numerical optical flow techniques to produce accurate and dense vector fields outperforming traditional methods. This work presents WindFlow as the first machine learning based system for feature tracking atmospheric motion using optical flow. Due to the lack of large-scale satellite-based observations, we leverage high-resolution numerical simulations from NASA's GEOS-5 Nature Run to perform supervised learning and transfer to satellite images. We demonstrate that our approach using deep learning based optical flow scales to ultra-high-resolution images of size 2881x5760 with less than 1 m/s bias and 2.5 m/s average error. Four network and learning architectures are compared and it is found that recurrent all-pairs field transforms (RAFT) produces the lowest errors on all metrics for wind speed and direction. Results on held out numerical outputs shows RAFT's good performance in each of the spatial, temporal, and physical dimensions. A comparison between WindFlow and an operational AMV product against rawinsonde observations show that RAFT transfers across simulations and thermal infrared satellite observations. This work shows that machine learning based optical flow is an efficient approach to generating robust feature tracking for AMVs consistently over large regions.

Atmospheric winds↗

NASA NTRS1993

Representation in incremental learning

Work focused on two areas in machine learning: representation for inductive learning and how to apply concept learning techniques to learning state preferences, which can represent search control knowledge for problem solving. Specifically, in the first area the issues of the effect of representation on learning, on how learning formalisms are biased, and how concept learning can benefit from the use of a hybrid formalism are addressed. In the second area, the issues of developing an agent to learn search control knowledge from the relative values of states, of the source of that qualitative information, and of the ability to use both quantitative and qualitative information in order to develop an effective problem-solving policy are examined.

Source record↗

NASA NTRS2020

Building a landslide hazard indicator with machine learning and land surface models

The U.S. Pacific Northwest has a history of frequent and occasionally deadly landslides caused by various factors. Using a multivariate, machine-learning approach, we combined a Pacific Northwest Landslide Inventory with a 36-year gridded hydrologic dataset from the National Climate Assessment – Land Data Assimilation System to produce a landslide hazard indicator (LHI) on a daily 0.125-degree grid. The LHI identified where and when landslides were most probable over the years 1979–2016, addressing issues of bias and completeness that muddy the analysis of multi-decadal landslide inventories. The seasonal cycle was strong along the west coast, with a peak in the winter, but weaker east of the Cascade Range. This lagging indicator can fill gaps in the observational record to identify the seasonality of landslides over a large spatiotemporal domain and show how landslide hazard has responded to a changing climate.

XGBoost↗

NASA NTRSDate not supplied

Global SO 2 Data Record from OMPS Instruments on the JPSS Constellation

NASA’s Earth Observing System (EOS) SO 2 climate data record (CDR) started in 2004, with the launch of the Aura/Ozone Monitoring Instrument (OMI) and is now being continued with the SNPP/Ozone Mapping and Profiler Suite (OMPS) launched in 2011. Both OMI and SNPP/OMPS SO 2 CDRs are produced with the Goddard principal component analysis (PCA) spectral fitting algorithm. An advantage of the data-driven PCA retrieval technique is that it enables highly consistent retrievals from different instruments, by inherently accounting for various instrumental factors. To further extend the EOS SO 2 CDR, we are implementing the PCA SO 2 retrieval algorithm with the L1B measurements from OMPS instruments flying on the Joint Polar Satellite System (JPSS) constellation. In this presentation, we will provide an update on our progress in NOAA-20 (launched in 2017) and NOAA-21 (launched in 2022) PCA SO2 retrievals. We will focus on our new NOAA-20/OMPS PCA SO 2 EOS continuity product, to be publicly released in fall of 2023. We will present statistical analyses on the quality of NOAA-20 PCA SO 2 product, including retrieval noise, biases over background areas, and long-term stability. We will compare our PCA SO 2 retrievals from NOAA-20 with those from OMI, SNPP/OMPS, and S5P/TROPOMI (TROPOspheric Monitoring Instrument) for anthropogenic sources as well as large volcanic plumes. We will also discuss the application of a new machine learning technique that helps to further reduce the noise of NOAA-20 SO 2 retrievals. In addition, we will present preliminary PCA SO 2 retrievals from NOAA-21/OMPS, including those from direct readout implementation for aviation disaster avoidance. Finally, we will share some first results applying the PCA algorithm to NASA’s geostationary TEMPO (Tropospheric Emissions: Monitoring of Pollution) instrument to obtain hourly, high resolution SO 2 data over North America.

SO2↗

NASA NTRSDate not supplied

Assessing Long-Term Health (LTH) Outcomes in Astronauts

NASA has linked potential Long-Term Health (LTH) risks of astronauts to their spaceflight experience, including the effects of space radiation, microgravity, and stressors such as isolation, sleep deprivation, and stress. Therefore, assessing the LTH outcomes for astronauts exposed to these hazards becomes a critical investigation in ensuring any deleterious effects are minimal. Prior research conducted into LTH outcomes of astronauts has not revealed increases in non-accidental mortality rates when compared to the U.S. general population, analog healthy cohorts, and professional athletes. One study did report an increased incidence of non-fatal cardiovascular disease events (Hazard Ratio=2.41, 95% Confidence Interval 1.26 to 4.63). Additional evidence indicates that astronauts may have increases in incidences of melanoma, prostate, and hematologic cancers, though the former is consistent with rates observed among aircraft pilots and the latter two are likely caused by detection bias due to increased screening on former astronauts.

long term health↗

NASA NTRSDate not supplied

Predicting Team Functioning in Long Term Space Missions Using Acoustic and Linguistic Measures

Maintaining optimal team functioning is critical for long-duration space exploration missions, yet traditional monitoring methods, such as self-reports and wearable sensors, often impose operational burdens or suffer from bias. This paper investigates a non-intrusive speech-based artificial intelligence (AI) framework to predict degradations in team functioning using data from the Human Exploration Research Analog (HERA) of the U.S. National Aeronautics and Space Administration (NASA). Using acoustic features, linguistic descriptors, and semantic embeddings, we evaluate static non-linear and temporal machine learning models to predict both objective (task accuracy) and subjective (self-reported efficacy and cohesion) team functioning outcomes. Results indicate that temporal models outperform static approaches, with prediction of objective task accuracy in Team Interaction Battery (TIB) improving from near chance to 71%. Self-reported outcomes, including team efficacy and cohesion, are predicted more reliably than task performance, achieving balanced accuracies of up to 85.56% and 78.12%, respectively, and are found to be most strongly associated with acoustic features. In a second interdependent task, the MMSEV–EVA, accuracies of up to 78% are achieved using temporal models with acoustic features. Furthermore, incorporating just 1–2 days of team-specific historical data systematically improved performance, and acoustic markers from informal pre-task interactions provided modest predictive gains. Finally, while automated preprocessing yielded viable accuracy, humancorrected data provided moderate performance gains, though transcription error rates did not significantly correlate with model performance. These findings highlight the potential of speech as a passive, high-fidelity monitoring tool for autonomous habitats.

Temporal modeling↗

NASA NTRS2024

Low-Cost Sensor Performance Intercomparison, Correction Factor Development, and 2+ Years of Ambient PM2.5 Monitoring in Accra, Ghana

Particulate matter air pollution is a leading cause of global mortality, particularly in Asia and Africa. Addressing the high and wide-ranging air pollution levels requires ambient monitoring, but many low- and middle-income countries (LMICs) remain scarcely monitored. To address these data gaps, recent studies have utilized low-cost sensors. These sensors have varied performance, and little literature exists about sensor intercomparison in Africa. By colocating 2 QuantAQ Modulair-PM, 2 PurpleAir PA-II SD, and 16 Clarity Node-S Generation II monitors with a reference-grade Teledyne monitor in Accra, Ghana, we present the first intercomparisons of different brands of low-cost sensors in Africa, demonstrating that each type of low-cost sensor PM2.5 is strongly correlated with reference PM2.5, but biased high for ambient mixture of sources found in Accra. When compared to a reference monitor, the QuantAQ Modulair-PM has the lowest mean absolute error at 3.04 μg/m3, followed by PurpleAir PA-II (4.54 μg/m3) and Clarity Node-S (13.68 μg/m3). We also compare the usage of 4 statistical or machine learning models (Multiple Linear Regression, Random Forest, Gaussian Mixture Regression, and XGBoost) to correct low-cost sensors data, and find that XGBoost performs the best in testing (R2: 0.97, 0.94, 0.96; mean absolute error: 0.56, 0.80, and 0.68 μg/m3 for PurpleAir PA-II, Clarity Node-S, and Modulair-PM, respectively), but tree-based models do not perform well when correcting data outside the range of the colocation training. Therefore, we used Gaussian Mixture Regression to correct data from the network of 17 Clarity Node-S monitors deployed around Accra, Ghana, from 2018 to 2021. We find that the network daily average PM2.5 concentration in Accra is 23.4 μg/m3, which is 1.6 times the World Health Organization Daily PM2.5 guideline of 15 μg/m3. While this level is lower than those seen in some larger African cities (such as Kinshasa, Democratic Republic of the Congo), mitigation strategies should be developed soon to prevent further impairment to air quality as Accra, and Ghana as a whole, rapidly grow.

Humidity↗

NASA NTRSDate not supplied

Creating Benchmark Data for Artificial Intelligence and Machine Learning Space Biology Research

To identify an appropriate AI/ML approach for a specific problem, the best practice is to measure algorithm performance through the benchmarking process. A scientific benchmark consists of an AI-ready dataset and a reference implementation on a specific scientific question. The NASA Science Mission Directorate (SMD) has started the “Benchmark Initiative for AI/ML to create scientific benchmark datasets in three applications: 1) scientific benchmarking, which finds the best algorithm for a specific problem; 2) application benchmarking, which measures algorithm performance against a set of parameters; and 3) system benchmarking, which evaluates performance of hardware and software architecture. Currently, there are no standardized datasets available to benchmark AI/ML algorithms in the domain of space biology. In this work, we constructed two AI/ML-ready biological datasets from experiments in space-flown mice: cellular imaging and RNA-seq. First, radiation-exposed immune cells harbor DNA damage foci that can be fluorescently marked to visualize the amount of damage following exposure to ionizing radiation. However, such large datasets are difficult to analyze visually, due to imaging inconsistencies and human bias, and classical image processing approaches can fail on imaging artifacts. AI/ML are therefore exciting alternative, providing the speed of machines and the accuracy of humans. We have made this dataset available at https://registry.opendata.aws/bps_microscopy/. Second, high-throughput nucleic acid sequencing (DNA-seq, RNA-seq) has become widespread in biomedical research due to the growing availability and affordability of these assays. However, most sequencing datasets suffer from high dimensionality and low sample count. In this work, we used a generative adversarial network to synthesize a standardized, AI-ready, publicly available benchmark dataset for space biology RNA-seq data with sufficient space-flown and ground control mouse liver samples from NASA GeneLab. This dataset is available at https://registry.opendata.aws/bps_rnaseq/. These datasets are now fully open the Space Biology community to test their favorite AI/ML approaches.

James Casaletto↗

NASA NTRSDate not supplied

Development of Digital Twin Technologies for Climate Projections

Climate projections are increasingly needed for adaptation, climate resilience and related decision making. However, existing projections have systematic biases, are limited in scope, and are not readily available for most potential users. While the ideal of an observational data-driven ‘digital twin’ for climate is initially attractive, there is only a very limited set of climate data available with which to train such a tool. Nonetheless, we are confident that there is a role for ‘digital twin technologies’ in removing biases, increasing computational efficiency, expanding scenarios and data accessibility.

digital twins↗

NASA NTRSDate not supplied

Continuing Long-term Global SO 2 Data Record with JPSS OMPS Instruments

NASA’s long-term Earth Observing System (EOS) SO 2 climate data record (CDR) started with Aura/Ozone Monitoring Instrument (OMI, launched in 2004) and is now being continued with the SNPP/Ozone Mapping and Profiler Suite (OMPS, launched in 2011). Both OMI and SNPP/OMPS SO 2 CDRs are produced with the Goddard principal component analysis (PCA) spectral fitting algorithm. By inherently accounting for various instrumental factors, the PCA technique enables highly consistent retrievals between different instruments. In this presentation, we will provide an overview on our effort to further extend the EOS SO 2 CDR, by implementing the PCA SO 2 algorithm with multiple OMPS instruments flying on the Joint Polar Satellite System (JPSS) constellation, including NOAA-20 (launched in 2017) and NOAA-21 (launched in 2022). We will present results analyzing our new NOAA-20/OMPS PCA SO 2 EOS continuity product, to be publicly released in fall of 2023. We will show statistical analyses on the quality of NOAA-20 PCA SO 2 product, such as retrieval noise, biases over background areas, and long-term stability. We will employ a previously established top-down method to estimate SO2 emissions from selected large point sources, using NOAA-20 SO 2 retrievals and assimilated wind fields as input. The SO 2 emission estimates derived from NOAA-20 retrievals will be compared with those from OMI, SNPP/OMPS, and S5P/TROPOMI (TROPOspheric Monitoring Instrument). We will also demonstrate the application of a new machine learning technique that further reduces the noise of NOAA-20 SO 2 retrievals. Finally, we will present preliminary PCA SO 2 retrievals from recently launched satellite sensors, including NOAA-21/OMPS and NASA’s geostationary TEMPO (Tropospheric Emissions: Monitoring of Pollution) instrument.

SO2↗

NASA NTRS1989

Learning in stochastic neural networks for constraint satisfaction problems

Researchers describe a newly-developed artificial neural network algorithm for solving constraint satisfaction problems (CSPs) which includes a learning component that can significantly improve the performance of the network from run to run. The network, referred to as the Guarded Discrete Stochastic (GDS) network, is based on the discrete Hopfield network but differs from it primarily in that auxiliary networks (guards) are asymmetrically coupled to the main network to enforce certain types of constraints. Although the presence of asymmetric connections implies that the network may not converge, it was found that, for certain classes of problems, the network often quickly converges to find satisfactory solutions when they exist. The network can run efficiently on serial machines and can find solutions to very large problems (e.g., N-queens for N as large as 1024). One advantage of the network architecture is that network connection strengths need not be instantiated when the network is established: they are needed only when a participating neural element transitions from off to on. They have exploited this feature to devise a learning algorithm, based on consistency techniques for discrete CSPs, that updates the network biases and connection strengths and thus improves the network performance.

Johnston, Mark D.↗

NASA NTRSDate not supplied

Linking OH Variability to Observable Variables, Meteorology and Transport

The hydroxyl radical (OH) plays a vital role in tropospheric chemistry, as it provides the dominant sink for a multitude of pollutants and climate-relevant gases such as methane. Observational constraints on the global distribution and temporal variability of OH are limited, and models simulate a wide range of OH distributions. While OH itself has a short atmospheric lifetime, OH is photochemically coupled to longer-lived species that undergo atmospheric transport. Here, we investigate how much of the OH variability within and between models can be explained by differences in observable species to develop diagnostics for OH differences. We find that NO 2 and water vapor together explain much of the spatial and temporal variability in simulated OH, and we use satellite observations to identify biases in these variables. The OH response to ENSO also differs between models, and we investigate potential causes of these differences such as differences in convection or lightning NOx. We also explore the potential of idealized tracers to represent the OH distribution. Within a single model, meteorological variables such as humidity and idealized tracers of transport can explain a significant portion of the OH spatial variability. We use a Gradient Boosted Regression Trees, a type of machine learning, to account for non-linear relationships between OH and the input variables.

Meteorology↗