Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “explainability”

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 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

Hysteretic temperature sensitivity of wetland CH4 fluxes explained by substrate availability and microbial activity: Model Archive

This Modeling Archive is in support of an NGEE Arctic publication "Hysteretic temperature sensitivity of wetland CH4 fluxes explained by substrate availability and microbial activity" in the Journal Biogeosciences (https://doi.org/10.5194/bg-17-5849-2020), which includes the model data used in the publication. CH4 emissions from terrestrial systems are posited to increase, which can offset mitigation efforts and accelerate climate change. Yet, the accuracy of modeled CH4 emissions is sensitive to the prescribed CH4 production (or emission) temperature dependencies that are currently uncertain. Here, we use a comprehensive biogeochemistry model (ecosys) to investigate factors modulating CH4 production and emission rates across a permafrost thaw gradient encompassing a partly thawed bog and a fully thawed fen. We find that seasonally varying substrate availability drives lower and higher modeled methanogen biomass and activity, and thereby CH4 production, during the earlier and later periods of the thawed season, respectively. Package follows the Model data archiving guidelines with data in a *.zip file with three subfolders containing *.csv files and raw model output files; a data dictionary table (data_dictionary.csv) and two model output description tables (ecosys_plantspecies_output_notes.csv and ecosys_soil_ouput_notes.csv) to explain the format and meaning of individual output variables; and a user guide as a *.pdf. A detailed model description can be found in the supplement of (Grant, 2013). The ecosys source code is available at https://doi:10.5281/zenodo.3906642. The Next-Generation Ecosystem Experiments: Arctic (NGEE Arctic), was a research effort to reduce uncertainty in Earth System Models by developing a predictive understanding of carbon-rich Arctic ecosystems and feedbacks to climate. NGEE Arctic was supported by the Department of Energy's Office of Biological and Environmental Research. The NGEE Arctic project had two field research sites: 1) located within the Arctic polygonal tundra coastal region on the Barrow Environmental Observatory (BEO) and the North Slope near Utqiagvik (Barrow), Alaska and 2) multiple areas on the discontinuous permafrost region of the Seward Peninsula north of Nome, Alaska. Through observations, experiments, and synthesis with existing datasets, NGEE Arctic provided an enhanced knowledge base for multi-scale modeling and contributed to improved process representation at global pan-Arctic scales within the Department of Energy's Earth system Model (the Energy Exascale Earth System Model, or E3SM), and specifically within the E3SM Land Model component (ELM).

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS1991

Explaining patterns in the ratification of global environmental treaties

A study was made of the ratification behavior of 160 countries with respect to 38 global environmental treaties. The study identifies and explains patterns in the ratification of treaties, providing two means of assessing the likelihood that any given country will support global environmental treaties. National ratification totals reveal a pattern of high ratification by countries in Western Europe, North America, Japan, Australia, and New Zealand. A country's standing within the range of high to low ratification rates can be explained by the statistical model developed in the study. This research allows one to identify countries likely to support global environmental treaties.

Cook, David W.↗
NASA NTRS2005

Laboratory Synthesized Calcium Oxide and Calcium Hydroxide Grains: A Candidate to Explain the 6.8 Micron Band

We will demonstrate that CaO and Ca(OH)2 are excellent candidates to explain the 6.8 microns feature, which is one of the most obscure features in young stellar objects. We discuss the condensation of CaO grains and the potential formation of a Ca(OH)2 surface layer. The infrared spectra of these grains are compared with the spectra of fifteen young stellar objects. We note that CaO-rich grains are seen in all meteoritic CAIs (calcium-aluminum-rich inclusions) and the 6.8 micron feature has only been observed in young stellar objects. Therefore, we consider CaO grains to be a plausible candidate to explain the 6.8 microns feature and hypothesize that they are produced in the hot interiors of young stellar environments.

Kimura, Yuki↗
NASA NTRS2012

Explaining Polarization Reversals in STEREO Wave Data

Recently Breneman et al. reported observations of large amplitude lightning and transmitter whistler mode waves from two STEREO passes through the inner radiation belt (L<2). Hodograms of the electric field in the plane transverse to the magnetic field showed that the transmitter waves underwent periodic polarization reversals. Specifically, their polarization would cycle through a pattern of right-hand to linear to left-hand polarization at a rate of roughly 200 Hz. The lightning whistlers were observed to be left-hand polarized at frequencies greater than the lower hybrid frequency and less than the transmitter frequency (21.4 kHz) and right-hand polarized otherwise. Only righthand polarized waves in the inner radiation belt should exist in the frequency range of the whistler mode and these reversals were not explained in the previous paper. We show, with a combination of observations and simulated wave superposition, that these polarization reversals are due to the beating of an incident electromagnetic whistler mode wave at 21.4 kHz and linearly polarized, symmetric lower hybrid sidebands Doppler-shifted from the incident wave by +/-200 Hz. The existence of the lower hybrid waves is consistent with the parametric decay mechanism of Lee and Kuo whereby an incident whistler mode wave decays into symmetric, short wavelength lower hybrid waves and a purely growing (zero-frequency) mode. Like the lower hybrid waves, the purely growing mode is Doppler-shifted by 200 Hz as observed on STEREO. This decay mechanism in the upper ionosphere has been previously reported at equatorial latitudes and is thought to have a direct connection with explosive spread F enhancements. As such it may represent another dissipation mechanism of VLF wave energy in the ionosphere and may help to explain a deficit of observed lightning and transmitter energy in the inner radiation belts as reported by Starks et al.

Breneman, A.↗
NASA NTRS2011

Can a Double Component Outflow Explain the X-Ray and Optical Lightcurves of Swift Gamma-Ray Bursts?

An increasing sample of Gamma-Ray Bursts (GRBs) observed by Swift show evidence of 'chromatic breaks', i.e. breaks that are present in the X-ray but not in the optical. We find that in a significant fraction of these GRB afterglows the X-ray and the optical emission cannot be produced by the same component. We propose that these afterglow lightcurves are the result of a two-component jet, in which both components undergo energy injection for the whole observation and the X-ray break is due to a jet break in the narrow outflow. Bursts with chromatic breaks also explain another surprising finding, the paucity of late achromatic breaks. We propose a model that may explain the behaviour of GRB emission in both X-ray and optical bands. This model can be a radical and noteworthy alternative to the current interpretation for the 'canonical' XRT and UVOT lightcurves, and it bears fundamental implications for GRB physics.

De Pasquale, Massimiliano↗
DOE OSTI2023

Explainable AI (XAI)-driven vibration sensing scheme for surface quality monitoring in a smart surface grinding process

Local Interpretable and Model-agnostic Explanation (LIME), an explainable artificial intelligence (XAI) approach is adapted to identify the globally important time-frequency bands for predicting average surface roughness (Ra) in a smart grinding process. The smart grinding setup consisted of a Supertech CNC precision surface grinding machine, instrumented with a Dytran piezoelectric accelerometer attached to the tailstock quill along the tangential direction (Y-axis). For every grinding pass, vibration signatures were captured, and the ground truth surface roughness values were recorded using a Mahr Marsurf M300C portable surface roughness profilometer. The roughness values ranged from 0.06 to 0.14 microns over the complete set of experiments. Time-frequency domain spectrogram frames were extracted for each of the vibration signals collected during the grinding process. Convolutional Neural Networks (CNNs) were modeled to predict the surface roughness based on these spectrogram frames and their image augmentations. The best CNN model was able to predict the roughness values with an overall R2-score of 0.95, training R2-score of 0.99, and testing R2-score of 0.81 with only 80 sets of vibration signals corresponding to 4 experiments with 20 trials each. Although the data size is not large enough to guarantee such performance metrics in real-world scenarios, one can extract statistically consistent explanations underlying the relationships these complex deep learning models capture. Further, the LIME methodology was implemented on the developed surface roughness CNN model to identify the important time-frequency bands (i.e., the superpixels of a spectrogram) influencing the predictions. Based on the identified important regions on the spectrogram frames, the corresponding frequency characteristics were determined that influence the surface roughness predictions. The important frequency range based on LIME results was approximately 11.7 to 19.1 kHz. The power of XAI was demonstrated by cutting down the sampling rate from 160 kHz to 30, 20, 10, and 5 kHz based on the important frequency range and considering Nyquist criteria. Separate CNN models were developed for these ranges by only extracting time-frequency contents below their corresponding Nyquist cut-offs. A proper data acquisition strategy is proposed by comparing the model performances to argue the selection of a sufficient sampling rate to capture the grinding process successfully and robustly.

42 ENGINEERING↗
DOE OSTI2023

Reprint of: Explainable AI (XAI)-driven vibration sensing scheme for surface quality monitoring in a smart surface grinding process

Local Interpretable and Model-agnostic Explanation (LIME), an explainable artificial intelligence (XAI) approach is adapted to identify the globally important time-frequency bands for predicting average surface roughness (Ra) in a smart grinding process. The smart grinding setup consisted of a Supertech CNC precision surface grinding machine, instrumented with a Dytran piezoelectric accelerometer attached to the tailstock quill along the tangential direction (Y-axis). For every grinding pass, vibration signatures were captured, and the ground truth surface roughness values were recorded using a Mahr Marsurf M300C portable surface roughness profilometer. The roughness values ranged from 0.06 to 0.14 microns over the complete set of experiments. Time-frequency domain spectrogram frames were extracted for each of the vibration signals collected during the grinding process. Convolutional Neural Networks (CNNs) were modeled to predict the surface roughness based on these spectrogram frames and their image augmentations. The best CNN model was able to predict the roughness values with an overall R2-score of 0.95, training R2-score of 0.99, and testing R2-score of 0.81 with only 80 sets of vibration signals corresponding to 4 experiments with 20 trials each. Although the data size is not large enough to guarantee such performance metrics in real-world scenarios, one can extract statistically consistent explanations underlying the relationships these complex deep learning models capture. Further, the LIME methodology was implemented on the developed surface roughness CNN model to identify the important time-frequency bands (i.e., the superpixels of a spectrogram) influencing the predictions. Based on the identified important regions on the spectrogram frames, the corresponding frequency characteristics were determined that influence the surface roughness predictions. The important frequency range based on LIME results was approximately 11.7 to 19.1 kHz. The power of XAI was demonstrated by cutting down the sampling rate from 160 kHz to 30, 20, 10, and 5 kHz based on the important frequency range and considering Nyquist criteria. Separate CNN models were developed for these ranges by only extracting time-frequency contents below their corresponding Nyquist cut-offs. A proper data acquisition strategy is proposed by comparing the model performances to argue the selection of a sufficient sampling rate to capture the grinding process successfully and robustly.

47 OTHER INSTRUMENTATION↗
DOE OSTI2023

Discovery of multi-functional polyimides through high-throughput screening using explainable machine learning

Polyimides have been widely used in modern industries because of their excellent mechanical and thermal properties, e.g., high-temperature fuel cells, displays, and aerospace composites. However, it usually takes decades of experimental efforts to develop a successful product. Aiming to expedite the discovery of high-performance polyimides, we utilize computational methods of machine learning (ML) and molecular dynamics (MD) simulations. Our study provides compelling evidence for the effectiveness of a data-driven approach in discovering novel polyimides. We first build a comprehensive library of more than 8 million hypothetical polyimides based on the polycondensation of existing dianhydride and diamine/diisocyanate molecules. Then we establish multiple ML models for the thermal and mechanical properties of polyimides based on their experimentally reported values, including glass transition temperature, Young’s modulus, and tensile yield strength. The obtained ML models demonstrate excellent predictive performance in identifying the key chemical substructures influencing the thermal and mechanical properties of polyimides. The use of explainable machine learning describes the effect of chemical substructures on individual properties, from which human experts can understand the cause of the ML model decision. Applying the well-trained ML models, we obtain property predictions of the 8 million hypothetical polyimides. Then, we screen the whole hypothetical dataset and identify three (3) best-performing novel polyimides that have better-combined properties than existing ones through Pareto frontier analysis. For an easy query of the discovered high-performing polyimides, we also create an online platform https://polyimide-explorer.herokuapp.com/ that embeds the developed ML model with interactive visualization. Furthermore, we validate the ML predictions through all-atom MD simulations and examine their synthesizability. The MD simulations are in good agreement with the ML predictions and the three novel polyimides are predicted to be easy to synthesize via Schuffenhauer’s synthetic accessibility score. Following the proposed ML guidance, we successfully synthesized a novel polyimide and the experimentally obtained high glass transition/thermal decomposition temperature demonstrated its excellent thermal stability. Here our study demonstrates an efficient way to expedite the discovery of novel polymers using ML prediction and MD validation. The high-throughput screening of a large computational dataset can serve as a general approach for new material discovery in other polymeric material exploration problems, such as organic photovoltaics, polymer membranes, and dielectrics.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

Explainable machine learning reveals that local structural motifs encode the thermodynamic state across the CuZr metallic glass-forming range

Metallic glasses derive their properties from the statistics of local atomic motifs rather than from long-range order, yet a quantitative, chemistry-specific link between motif populations and the underlying glassy state has remained elusive. In this work we combine large-scale molecular dynamics, Voronoi tessellation, deep neural networks, and SHapley Additive exPlanations (SHAP) to identify which local structural motifs define the glassy state of Cu—Zr metallic glasses. A dataset of 17,180 atomistic configurations spanning ten compositions (Cu 20 Zr 80 –Cu 80 Zr 20 ) and four quench rates (10 9 –10 12 K/s) is used to train a feed-forward neural network that regresses temperature across the 50–2000 K liquid–supercooled–glass range, achieving a mean absolute error of 19.89 K and R 2 = 0.9974, confirming that the local structural state is faithfully encoded in motif-level structure. SHAP analysis then reveals that a tightly coupled near-icosahedral family of motifs (coordination numbers (CN) 11–13, including the full icosahedron 001200 and its single-atom-perturbation sibling 10930) collectively encodes the thermodynamic state of the system across the full glass-forming range. The CN = 11–13 ordered members carry negative SHAP values at high populations, tracking the most deeply-quenched configurations, while 10930 shows the reversed signature consistent with its role as a soft-spot host whose population shrinks as the icosahedral network deepens. The analysis demonstrates that explainable machine learning can isolate the minimal motif vocabulary defining the glassy state and recovers the near-icosahedral building blocks previously identified by data-driven analyses of Cu—Zr. The approach provides a general, chemistry-specific route for characterizing the structural state of disordered materials.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Explainable artificial intelligence relates perovskite luminescence images to current-voltage metrics

As the demand for low-cost, high-efficiency solar energy technologies grows, metal halide perovskite (MHP) solar cells have emerged as a promising candidate for next-generation photovoltaics due to their high power conversion efficiencies. However, their poor durability and issues with manufacturing consistency remain significant barriers to commercialization. In this work, we develop deep learning models to support materials characterization and provide insight into features and processes influencing performance. The models are trained using transfer learning of a pretrained model to predict relevant current-voltage (IV) metrics based on different combinations of input electroluminescence (EL) and photoluminescence (PL) images of MHP devices. We examine which image types are most informative in accurately predicting different IV metrics. Additionally, we use explainable artificial intelligence (XAI) techniques to provide insights into specific spatial features in the devices that drive differences in performance. We find that stabilized luminescence images (e.g. those collected after biasing the devices for at least 1 min) are better for predicting metrics of open-circuit voltage (by PL) and short-circuit current (by PL with EL), but that predicting fill factor and overall power output may use the time-evolution of EL images. Based on attribution masks generated by integrated gradients for each device performance metric, we further suggest different loss mechanisms associated with categories of large and small spatial defects. Overall, this case study highlights the potential applicability of XAI methodology for streamlining MHP device analysis and accelerating detailed understanding of the relationships between spatial defects and impacts on performance.

14 SOLAR ENERGY↗
DOE OSTI2025

Applications of explainable artificial intelligence in renewable energy research

Researchers in renewable energy are applying deep learning (DL) to a variety of problems from diverse renewable energy domains, such as biofuels, wind, solar, power systems, buildings, vehicles, and transportation systems. Improvements in accuracy may be demonstrated using DL in laboratory settings. However, the lack of interpretability of DL models poses a practical limitation to their utility in advancing scientific knowledge and in the deployment of DL models in safety-critical energy systems. In this article, we discuss explainable artificial intelligence (XAI) as one pathway toward more interpretable DL models. We explore a brief timeline of U.S. national laboratory interest in XAI, an overview and taxonomy of methods in the field of XAI, and a selection of applications across renewable energy research domains. We conclude by highlighting pivotal areas where XAI can accelerate innovation in artificial intelligence for renewable energy research and other essential future directions.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Stream Temperature Prediction in a Shifting Environment: Explaining the Influence of Deep Learning Architecture

Stream temperature is a fundamental control on ecosystem health. Recent efforts incorporating process guidance into deep learning models for predicting stream temperature have been shown to outperform existing statistical and physical models. This performance is in part because deep learning architectures can actively learn spatiotemporal relationships that govern how water and energy propagate through a river network. However, exploration of how spatiotemporal awareness and process guidance influence a model's generalizability under shifting environmental conditions such as climate change is limited. Here, we use Explainable Artificial Intelligence (XAI) to interrogate how differing deep learning architectures affect a model's learned spatial and temporal dependencies, and how those learned dependencies affect a model's ability to maintain high accuracy when applied to unseen environmental conditions. Using the Delaware River Basin in the northeastern United States as a test case, we compare two spatiotemporally aware process–guided deep learning models for predicting stream temperature (a recurrent graph convolution network—RGCN, and a temporal convolution graph model—Graph WaveNet). Both models achieve equally high predictive performance when testing data are well represented in the training data (test root mean squared errors of 1.64°C and 1.65°C); however, Graph WaveNet significantly outperforms RGCN in 4 out of 5 experiments where test partitions represent different types of unseen environmental conditions. XAI results show that the architecture of Graph WaveNet leads to learned spatial relationships with greater fidelity to physical processes, and that this fidelity improves the generalizability of the model when applied to shifting and/or unseen environmental conditions.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Explainable physics-based constraints on reinforcement learning for accelerator optimization

We present a reinforcement learning (RL) framework for optimizing particle accelerator experiments that builds explainable physics-based constraints on agent behavior. The goal is to increase transparency and trust by letting users verify that the agent’s decision-making process incorporates suitable physics. Our algorithm uses a learnable surrogate function for physical observables, such as energy, and uses them to fine-tune how actions are chosen. This surrogate can be represented by a neural network or by an interpretable sparse dictionary model. We test our algorithm on a range of particle accelerator optimization environments designed to emulate the Continuous Electron Beam Accelerator Facility at Jefferson Lab. By examining the mathematical form of the learned constraint function, we are able to confirm the agent has learned to use the established physics of each environment. In addition, we find that the introduction of a physics-based surrogate enables our RL algorithms to reliably converge for difficult high-dimensional accelerator optimization environments.

explainability↗
DOE OSTI2023

Assessing decadal variability of subseasonal forecasts of opportunity using explainable AI

Abstract Identifying predictable states of the climate system allows for enhanced prediction skill on the generally low-skill subseasonal timescale via forecasts with higher confidence and accuracy, known as forecasts of opportunity. This study takes a neural network approach to explore decadal variability of subseasonal predictability, particularly during forecasts of opportunity. Specifically, this work quantifies subseasonal prediction skill provided by the tropics within the Community Earth System Model Version 2 (CESM2) Large Ensemble and assesses how this skill evolves on decadal timescales. Utilizing the networks’ confidence and explainable artificial intelligence, physically meaningful sources of predictability associated with periods of enhanced skill are identified. Using these networks, we find that tropically-driven subseasonal predictability varies on decadal timescales during forecasts of opportunity. Further, we investigate the drivers of the low frequency modulation of the tropical-extratropical teleconnection and discuss the implications. Analysis is extended to ECMWF Reanalysis v5 data, revealing that the relationships learned within the CESM2-Large Ensemble holds in modern reanalysis data. These results indicate that the neural networks are capable of identifying predictable decadal states of the climate system within CESM2 that are useful for making confident, accurate subseasonal precipitation predictions in the real world.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Explainable AI for Multivariate Time Series Pattern Exploration: Latent Space Visual Analytics With Temporal Fusion Transformer and Variational Autoencoders in Power Grid Event Diagnosis

Detecting and analyzing complex patterns in multivariate time-series data is crucial for decision-making in urban and environmental system operations. However, challenges arise from the high dimensionality, intricate complexity, and interconnected nature of complex patterns, which hinder the understanding of their underlying physical processes. Existing AI methods often face limitations in interpretability, computational efficiency, and scalability, reducing their applicability in real-world scenarios. This paper proposes a novel visual analytics framework that integrates two generative AI models, Temporal Fusion Transformer (TFT) and Variational Autoencoders (VAEs), to reduce complex patterns into lower-dimensional latent spaces and visualize them in 2D using dimensionality reduction techniques such as PCA, t-SNE, and UMAP with DBSCAN. These visualizations, presented through coordinated and interactive views and tailored glyphs, enable intuitive exploration of complex multivariate temporal patterns, identifying patterns’ similarities and uncover their potential correlations for a better interpretability of the AI outputs. The framework is demonstrated through a case study on power grid signal data, where it identifies multi-label grid event signatures, including faults and anomalies with diverse root causes. Additionally, novel metrics and visualizations are introduced to validate the models and assess the performance, efficiency, and consistency of latent maps generated by VAE, which have been utilized in prior studies for latent space cartography and used as a benchmark in this study, and the emerging TFT architecture under various configurations. These analyses provide actionable insights for model parameter tuning and reliability improvements. Comparative results highlight that TFT achieves shorter run times and superior scalability to diverse time-series data shapes compared to VAE. This work advances fault diagnosis in multivariate time series, fostering explainable AI to support critical system operations.

Explainable AI↗
DOE OSTI2021

Leveraging Community and Author Context to Explain the Performance and Bias of Text-Based Deception Detection Models

Deceptive news posts shared in online communities can be detected with NLP models, and much recent research has focused on the development of such models. In this work, we use characteristics of online communities and authors --- the context of how and where content is posted --- to explain the performance of a neural network deception detection model and identify sub-populations who are disproportionately affected by model accuracy or failure. We examine who is posting the content, and where the content is posted to. We find that while author characteristics are better predictors of deceptive content than community characteristics, both characteristics are strongly correlated with model performance. Traditional performance metrics such as F1 score may fail to capture poor model performance on isolated sub-populations such as specific authors, and as such, more nuanced evaluation of deception detection models is critical.

machine learning (ML), machine learning explanatio↗
DOE OSTI2025

Explainable Machine Learning for Functional Data

Black-box machine learning models are recognized as useful tools for prediction applications, but the algorithmic complexity of some models causes interpretation challenges. Explainability methods have been proposed to provide insight into these models, but there is little research focused on supervised modeling with functional data inputs. We argue that, especially in applications of high consequence, it is important to explicitly model the functional dependence in a black-box analysis to not obscure or misrepresent patterns in explanations. As such, we propose the V ariable importance E xplainable E lastic S hape A nalysis (VEESA) pipeline for training supervised machine learning models with functional inputs. The pipeline is an analysis process that includes the data preprocessing, modeling, and post-hoc explanations. The preprocessing is done using elastic functional principal components analysis, which accounts for vertical and horizontal variability in functional data and, ultimately, allows for explanations in the original data space that identify the important functional variability without bias due to correlated variables. Here, we demonstrate the pipeline on two high-consequence applications: explosives classification for national security and inkjet printer identification in forensic science. The applications exhibit the VEESA pipeline’s ability to provide an understanding of the characteristics of the functional data useful for prediction. Code for implementing the pipeline is available in the veesa R package (and supplemental python code).

Elastic Shape Analysis↗
NASA NTRS2023

Explainable deep learning for insights in El Niño and river flows

The El Niño Southern Oscillation (ENSO) is a semi-periodic fluctuation in sea surface temperature (SST) over the tropical central and eastern Pacific Ocean that influences interannual variability in regional hydrology across the world through long-range dependence or teleconnections. Recent research has demonstrated the value of Deep Learning (DL) methods for improving ENSO prediction as well as Complex Networks (CN) for understanding teleconnections. However, gaps in predictive understanding of ENSO-driven river flows include the black box nature of DL, the use of simple ENSO indices to describe a complex phenomenon and translating DL-based ENSO predictions to river flow predictions. Here we show that eXplainable DL (XDL) methods, based on saliency maps, can extract interpretable predictive information contained in global SST and discover SST information regions and dependence structures relevant for river flows which, in tandem with climate network constructions, enable improved predictive understanding. Our results reveal additional information content in global SST beyond ENSO indices, develop understanding of how SSTs influence river flows, and generate improved river flow prediction, including uncertainty estimation. Observations, reanalysis data, and earth system model simulations are used to demonstrate the value of the XDL-CN based methods for future interannual and decadal scale climate projections.

SST↗