Big Microstructure Datasets for Materials Informatics: Using Statistically Conditioned Generative Models to Curate Big Datasets
Explore the source record for details and available documents.
Robertson, Andreas Euan↗
SEARCH · Search NASA
Results for “Dataset”
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.
Designing Large Datasets: Data-Scarce and Stable Deep Generative Models for Turning Sparse Experiments into Big Datasets in Materials Science
Explore the source record for details and available documents.
Robertson, Andreas Euan↗
Big Microstructure Datasets for Materials Informatics: Using Statistically Conditioned Generative Models to Curate Big Datasets
Explore the source record for details and available documents.
Robertson, Andreas Euan↗
MassIVE MSV000084488 - Development of an updated genome-scale metabolic model of Clostridium thermocellum and its application for integration of multi-omics datasets: Development of an updated genome-scale metabolic model of Clostridium thermocellum and its application for integration of multi-omics datasets
Explore the source record for details and available documents.
Giannone, Richard J.↗
NIR Dataset and Models (Near-Infrared Spectroscopy Calibration Dataset for Butanediol Fermentation and Multivariate Calibration Models) [SWR-22-60]
2,3-butanediol (2,3-BDO) is an economically important platform chemical that can be used in a variety of chemical feedstocks, liquid fuels, and biosynthetic building blocks. While 2,3-BDO can be efficiently produced by fermentation, the fermentation requires continuous monitoring and control to maximize 2,3-BDO yields and minimize inhibitory coproducts. Because of the time required for sampling and at-line measurement of fermentation samples with high pressure liquid chromatography (HPLC), the ability for operators to perform real-time modification to fermentation conditions is limited. To overcome this challenge, researchers from the National Renewable Energy Laboratory (NREL) have developed a calibration model which can predict the concentration of several analytes in real-time using near-infrared (NIR) spectra of the filtered fermentation broth. While significantly reducing the need for off-line sampling, NREL expects this technology to play a critical role in maximizing 2,3-BDO production. 2,3-butanediol (2,3-BDO) is a useful chemical platform that can be used to create a variety of products. For instance, 2,3-BDO can be (1) dehydrated and converted into methyl ethyl ketone, a liquid fuel additive or (2) deoxydehydrated into 1,3-butadiene for synthetic rubber, which can also be oligomerized in high yields to gasoline, diesel, and jet fuel. In order to maximize 2,3-BDO production, frequent measurement of fermentation samples is needed, as small changes in oxygen concentration can drive the fermentation to undesired products. For example, oxygen-deficient conditions result in glycerol production, while excess oxygen concentrations result in acetoin production. This results in the need for measuring dissolved oxygen, glucose, and xylose concentrations in order to optimize the aeration rate of the fermentation. Traditional monitoring methods occurs off-line and can take up to 30 minutes per sample. With multiple fermenters and high-pressure liquid chromatography (HPLC) injectors, resulting in the need for multiple samples, the sampling process can take hours to complete. NREL’s calibration model can predict the glucose, xylose, 2,3-BDO, acetoin, and glycerol concentrations from NIR spectra of filtered fermentation liquor samples. Using a partial least-squares (PLS) calibration model, NREL’s model can monitor the concentration of these analytes during subsequent fermentations at bench- and pilot-scale, demonstrating the utility of NIR spectroscopy combined with chemometrics for real-time, at-line monitoring of 2,3-BDO fermentations.
Wolfrum, Edward↗
Application of ML/AI for Identifying Earth Science Datasets in Research Publications
NASA Data Active Archive Centers, or DAACs, ingest, store and distribute data acquired from satellites, ground systems as well as modelling data. These data are organized by the datasets, each presenting collection of files usually associated with the certain mission, instrument, processing level, parameter(s), algorithm and/or model. The number of datasets offered by a single DAAC to the public varies. GES DISC, for example, currently offers for public use approximately ~1,300 datasets. While each publicly offered dataset comes with supporting documentation, it is challenging for novice and even experienced scientists to navigate among the datasets that offer similar parameters to find the datasets for their particular research application. Supplying dataset documentation with the scientific paper citations that refer to that dataset provides means for the dataset users to educate themselves with the application research that dataset is being used in. Collecting citations of the papers that use the datasets for their research yield valuable insights into application areas of those datasets, information about usage of the dataset groups for specific applications and those application topics. It also gives insights into the “deep metrics” of the dataset usage, as opposed to the common metrics of the dataset usage such as number of users who downloaded the dataset files and volumes of downloaded data. Association of a certain scientific paper with the dataset(s) presents a challenge because most of the paper authors do not properly cite the datasets, datasets usually have cryptic names and Digital Object Identifiers (DOIs) that are used for dataset identification were assigned to the datasets only few years ago. Simple Google or online library search do not provide even meaningful fraction of the results when performed by the dataset name or DOI, however they provide too many results when the search is done by more broader terms such as mission and instrument names. Attempts to create an AI system capable to identify dataset in the scientific papers have already been made using neural networks classifiers on the basis of the dataset mission, instrument and variable name. This method was applied to NASA SEDAC, which has 41 datasets in total. In GES DISC there can be as many as ~100 datasets per mission/instrument with some of the datasets consisting of multiple variables so there is a need for more differentiating parameters for dataset identification in the paper. The approach we are currently investigating is creating AI classifiers that are based on multiple dataset features, or keywords, extracted from the NASA Earthdata Common Dataset Repository (CMR). The features are weighted based on how precisely they can identify a dataset. The classifier uses preprocessed paper text as input and searches for the CMR datasets whose feature sets are the closest to the feature sets contained in the paper. The challenges of dataset identification include variety of ways the paper authors describe the datasets in their papers and incomplete tagging of the CMR dataset description (DIFs).
Irina Gerasimov↗
Discovering Research Areas in Dataset Applications Through Knowledge Graphs and Large Language Models
Scientific datasets are increasingly cited in peer-reviewed journal publications, facilitating easy access to research utilizing those datasets. Datasets undergo a life cycle where older versions of datasets are replaced by newer versions often due to improvements in data resolution, algorithms, and other factors. Unlike peer reviewed documents registered with a single Digital Unique Identifier (DOI), datasets can be updated over time and the newer version of the datasets are registered with a new DOI which is not necessarily linked to the previous version of the dataset. It is challenging when publications citing a dataset need to be traced over the entire life cycle of that dataset. We provide an innovative approach to link the dataset versions and publications using a knowledge graph (KG). KG can help to trace the dataset cited in publications over the entire dataset life cycle and shed light into dataset usage in various applied research areas. We fine-tuned the pretrained NASA IMPACTINDUS Large Language Model (LLM) on a set of labeled publications abstracts. Our results showed that 87% of the publications were classified into one of twenty applied research areas, while the remaining 13% were classified into non-applied research areas. By linking datasets to applied research areas through the KG and employing Global Change Master Directory(GCMD), a well-established controlled vocabulary of scientific keywords describing Earth science datasets, we contribute to a transparent and advanced search and discovery mechanism for datasets across the Earth data ecosystem. The integrated KG and LLM approach is now incorporated and operational in dataset publication management at one of NASA’s Earth science data archival centers.
data provenance↗
DEEPEN Global Standardized Categorical Exploration Datasets for Magmatic Plays
DEEPEN stands for DE-risking Exploration of geothermal Plays in magmatic ENvironments. As part of the development of the DEEPEN 3D play fairway analysis (PFA) methodology for magmatic plays (conventional hydrothermal, superhot EGS, and supercritical), weights needed to be developed for use in the weighted sum of the different favorability index models produced from geoscientific exploration datasets. This was done using two different approaches: one based on expert opinions, and one based on statistical learning. This GDR submission includes the datasets used to produce the statistical learning-based weights. While expert opinions allow us to include more nuanced information in the weights, expert opinions are subject to human bias. Data-centric or statistical approaches help to overcome these potential human biases by focusing on and drawing conclusions from the data alone. The drawback is that, to apply these types of approaches, a dataset is needed. Therefore, we attempted to build comprehensive standardized datasets mapping anomalies in each exploration dataset to each component of each play. This data was gathered through a literature review focused on magmatic hydrothermal plays along with well-characterized areas where superhot or supercritical conditions are thought to exist. Datasets were assembled for all three play types, but the hydrothermal dataset is the least complete due to its relatively low priority. For each known or assumed resource, the dataset states what anomaly in each exploration dataset is associated with each component of the system. The data is only a semi-quantitative, where values are either high, medium, or low, relative to background levels. In addition, the dataset has significant gaps, as not every possible exploration dataset has been collected and analyzed at every known or suspected geothermal resource area, in the context of all possible play types. The following training sites were used to assemble this dataset: - Conventional magmatic hydrothermal: Akutan (from AK PFA), Oregon Cascades PFA, Glass Buttes OR, Mauna Kea (from HI PFA), Lanai (from HI PFA), Mt St Helens Shear Zone (from WA PFA), Wind River Valley (From WA PFA), Mount Baker (from WA PFA). - Superhot EGS: Newberry (EGS demonstration project), Coso (EGS demonstration project), Geysers (EGS demonstration project), Eastern Snake River Plain (EGS demonstration project), Utah FORGE, Larderello, Kakkonda, Taupo Volcanic Zone, Acoculco, Krafla. - Supercritical: Coso, Geysers, Salton Sea, Larderello, Los Humeros, Taupo Volcanic Zone, Krafla, Reyjanes, Hengill. **Disclaimer: Treat the supercritical fluid anomalies with skepticism. They are based on assumptions due to the general lack of confirmed supercritical fluid encounters and samples at the sites included in this dataset, at the time of assembling the dataset. The main assumption was that the supercritical fluid in a given geothermal system has shared properties with the hydrothermal fluid, which may not be the case in reality. Once the datasets were assembled, principal component analysis (PCA) was applied to each. PCA is an unsupervised statistical learning technique, meaning that labels are not required on the data, that summarized the directions of variance in the data. This approach was chosen because our labels are not certain, i.e., we do not know with 100% confidence that superhot resources exist at all the assumed positive areas. We also do not have data for any known non-geothermal areas, meaning that it would be challenging to apply a supervised learning technique. In order to generate weights from the PCA, an analysis of the PCA loading values was conducted. PCA loading values represent how much a feature is contributing to each principal component, and therefore the overall variance in the data.
15 GEOTHERMAL ENERGY↗
An Assessment of High-Resolution Gridded Temperature Datasets over California
Abstract High-resolution gridded datasets are in high demand because they are spatially complete and include important finescale details. Previous assessments have been limited to two to three gridded datasets or analyzed the datasets only at the station locations. Here, eight high-resolution gridded temperature datasets are assessed two ways: at the stations, by comparing with Global Historical Climatology Network–Daily data; and away from the stations, using physical principles. This assessment includes six station-based datasets, one interpolated reanalysis, and one dynamically downscaled reanalysis. California is used as a test domain because of its complex terrain and coastlines, features known to differentiate gridded datasets. As expected, climatologies of station-based datasets agree closely with station data. However, away from stations, spread in climatologies can exceed 6°C. Some station-based datasets are very likely biased near the coast and in complex terrain, due to inaccurate lapse rates. Many station-based datasets have large unphysical trends (>1°C decade−1) due to unhomogenized or missing station data—an issue that has been fixed in some datasets by using homogenization algorithms. Meanwhile, reanalysis-based gridded datasets have systematic biases relative to station data. Dynamically downscaled reanalysis has smaller biases than interpolated reanalysis, and has more realistic variability and trends. Dynamical downscaling also captures snow–albedo feedback, which station-based datasets miss. Overall, these results indicate that 1) gridded dataset choice can be a substantial source of uncertainty, and 2) some datasets are better suited for certain applications.
54 ENVIRONMENTAL SCIENCES↗
A Hybrid Approach to Labeling Datasets in Earth Science Publications
NASA Data Centers provide the public with thousands of datasets that result in published papers, reports, and conference proceedings. Collecting accurate metrics on usage of these datasets is key to connecting different areas of knowledge and evaluating the datasets’ impact. While most of the datasets have Digital Object Identifiers (DOIs) assigned, most publications do not cite them hampering the automated search of these publications. Instead, articles mention attributes like organization, instrument, mission, variable, or a publication describing the dataset. Often only domain experts can deduce the dataset that was used in the publication text. The lack of a citation slows the spread of information and reduces the research’s impact. With thousands of papers produced each year, an automated means of labeling datasets is critical. This paper explores a hybrid approach of heuristics and a Natural Language Processing (NLP) Named Entity Recognition (NER) model to find and label the datasets used within Earth Science papers. Heuristics are used to produce the labelled sentences and any potential dataset candidates that can be derived from a sentence. The heuristic labels the sentences with the names of mission, instrument, re-analysis models, and science keywords taken from the Global Change Master Directory (GCMD) ontology. Additionally, it uses those labels to generate the dataset citation candidates. If the mission, instrument, and variable are sufficient to create the citation for the dataset the citation and the label the domain expert reviews the output without going through the NLP model. If the extracted label is not sufficient to label the dataset on its own, the sentence and its associated dataset labels will be inputted into the NER model. The model outputs the labeled sentence and the potential dataset candidates with their associated probabilities. The domain expert then reviews the NER model’s output and the correct labels are determined. The newly labelled papers can then be used as additional training data. This creates an iterative process for the approach to continuously improve. Because all the possible mentions are gathered by the model, the domain expert can quickly and easily label the papers resulting in large time savings.
Jacob Atkins↗
Improved Estimates of Pentad Precipitation through the Merging of Independent Precipitation Datasets
Three independent, quasi-global, gridded datasets of precipitation (a rain gauge-based dataset, the satellite-only component of the NASA Integrated Multi-satellitE Retrievals for Global Precipitation Measurement mission [IMERG] Final Run precipitation product, and precipitation estimates derived from NASA Soil Moisture Active Passive [SMAP] soil moisture retrievals), are objectively combined into a single pentad precipitation dataset at 36-km resolution using a unique approach based on extended triple collocation. The quality of each of the four datasets is then evaluated against independent observations. When a global land surface model at 36-km resolution is integrated four times, once utilizing the merged precipitation forcing and once with each of the three contributing datasets, the near-surface soil moisture variations produced with the merged forcing validate best against independent satellite-based soil moisture fields. In addition, the merged dataset is found to be more consistent, relative to each contributor, with estimates of air temperature variations across the globe. The merged dataset thus appears to draw successfully on the complementary strengths of each contributor: the particularly high quality of the rain gauge-based dataset in areas of high gauge density, the more uniform accuracy across the globe of the IMERG data, and the moderate accuracy, particularly in semi-arid regions, of the soil moisture retrieval-based data. Plain Language Summary Obtaining measurements of precipitation across the globe can be challenging. Rain gauges in some ways provide the most accurate measurements, but gauges are absent in many parts of the world, and even where they exist, they only measure precipitation at the gauge itself and therefore may not provide an accurate large-scale average. Satellite-based estimates of precipitation largely overcome these problems, but such data have their own issues, notably a “snapshot” (rather than a time-average) character of the measurements and difficulty associated with interpreting the measured radiances in the presence of complex land surfaces. In the present paper, we use a novel approach to generate a “merged” dataset, one that optimally combines the gauge precipitation information and the satellite-based precipitation information with a third set of estimates derived from soil moisture retrievals. The merged precipitation dataset and each of the three contributors (aggregated here to 5-day averages at a spatial resolution of about 36-km) are then evaluated for consistency with independent geophysical fields. The merged dataset is found to perform best, a clear indication that it takes proper advantage of the complementary strengths of each contributor and, accordingly, that the presented approach for merging the different contributors is indeed viable.
Precipitation↗
Overcoming small minirhizotron datasets using transfer learning
Minirhizotron technology is widely used to study root growth and development. Yet, standard approaches for tracing roots in minirhiztron imagery is extremely tedious and time consuming. Machine learning approaches can help to automate this task. However, lack of enough annotated training data is a major limitation for the application of machine learning methods. Transfer learning is a useful technique to help with training when available datasets are limited. In this paper, we investigated the effect of pre-trained features from the massives-cale, irrelevant ImageNet dataset and a relatively moderate-scale, but relevant peanut root dataset on switchgrass root imagery segmentation applications. We compiled two minirhizotron image datasets to accomplish this study: one with 17,550 peanut root images and another with 28 switchgrass root images. Both datasets were paired with manually labeled ground truth masks. Deep neural networks based on the U-net architecture were used with different pre-trained features as initialization for automated, precise pixel-wise root segmentation in minirhizotron imagery. We observed that features pre-trained on a closely related but relatively moderate size dataset like our peanut dataset were more effective than features pre-trained on the large but unrelated ImageNet dataset. Here, we achieved high quality segmentation on peanut root dataset with 99.04% accuracy at the pixel-level and overcame errors in human-labeled ground truth masks. By applying transfer learning technique on limited switchgrass dataset with features pre-trained on peanut dataset, we obtained 99% segmentation accuracy in switchgrass imagery using only 21 images for training (fine tuning). Furthermore, the peanut pre-trained features can help the model converge faster and have much more stable performance.
59 BASIC BIOLOGICAL SCIENCES↗
Novel insights enabled by combining mouse muscle datasets from the Rodent Research-1 mission
Biological space experiments are often expensive and difficult to conduct. As such, it is critical to maximize the value of the data that is collected during these experiments. One way to do this is to combine multiple–previously separate–datasets. This can increase the number of replicates for the conditions of interest (and hence statistical power), allow new multi-factor questions to be asked, and potentially highlight new patterns that otherwise would not have been identified from single-dataset studies. However, the process of combining datasets introduces noise due to inherent technical variations between experiments. To better understand the insights that can be gained from multi-dataset analyses and the problems that may arise from joining multiple datasets, several mouse muscle RNA-Seq datasets from the Rodent Research-1 mission were first selected. Then, using the R package DESeq2, principal component analysis (PCA) plots and differentially expressed gene (DEG) lists between ground and flight muscle samples were generated for individual datasets and for different pairwise combinations of datasets. Several new DEGs were identified in the combined datasets, and patterns in the PCA plots were affected depending on which datasets were joined. Understanding the results of this work will be critical for future studies that seek to perform multi-dataset analyses.
spaceflight↗
Daily evaluation of 26 precipitation datasets using Stage-IV gauge-radar data for the CONUS
New precipitation (P) datasets are released regularly, following innovations in weather forecasting models, satellite retrieval methods, and multi-source merging techniques. Using the conterminous US as a case study, we evaluated the performance of 26 gridded (sub-)daily P datasets to obtain insight into the merit of these innovations. The evaluation was performed at a daily timescale for the period 2008–2017 using the Kling–Gupta efficiency (KGE), a performance metric combining correlation, bias, and variability. As a reference, we used the high-resolution (4 km) Stage-IV gauge-radar P dataset. Among the three KGE components, the P datasets performed worst overall in terms of correlation (related to event identification). In terms of improving KGE scores for these datasets, improved P totals (affecting the bias score) and improved distribution of P intensity (affecting the variability score) are of secondary importance. Among the 11 gauge-corrected P datasets, the best overall performance was obtained by MSWEP V2.2, underscoring the importance of applying daily gauge corrections and accounting for gauge reporting times. Several uncorrected P datasets outperformed gauge-corrected ones. Among the 15 uncorrected P datasets, the best performance was obtained by the ERA5-HRES fourth-generation reanalysis, reflecting the significant advances in earth system modeling during the last decade. The (re)analyses generally performed better in winter than in summer, while the opposite was the case for the satellite-based datasets. IMERGHH V05 performed substantially better than TMPA-3B42RT V7, attributable to the many improvements implemented in the IMERG satellite P retrieval algorithm. IMERGHH V05 outperformed ERA5-HRES in regions dominated by convective storms, while the opposite was observed in regions of complex terrain. The ERA5-EDA ensemble average exhibited higher correlations than the ERA5-HRES deterministic run, highlighting the value of ensemble modeling. The WRF regional convection-permitting climate model showed considerably more accurate P totals over the mountainous west and performed best among the uncorrected datasets in terms of variability, suggesting there is merit in using high-resolution models to obtain climatological P statistics. Our findings provide some guidance to choose the most suitable P dataset for a particular application.
Hylke E. Beck↗
BuildingsBench: A Large-Scale Dataset of 900K Buildings and Benchmark for Short-Term Load Forecasting
The BuildingsBench datasets consist of: - Buildings-900K: A large-scale dataset of 900K buildings for pretraining models on the task of short-term load forecasting (STLF). Buildings-900K is statistically representative of the entire U.S. building stock. - 7 real residential and commercial building datasets for benchmarking two downstream tasks evaluating generalization: zero-shot STLF and transfer learning for STLF. Buildings-900K can be used for pretraining models on day-ahead STLF for residential and commercial buildings. The specific gap it fills is the lack of large-scale and diverse time series datasets of sufficient size for studying pretraining and finetuning with scalable machine learning models. Buildings-900K consists of synthetically generated energy consumption time series. It is derived from the NREL End-Use Load Profiles (EULP) dataset (see link to this database in the links further below). However, the EULP was not originally developed for the purpose of STLF. Rather, it was developed to "...help electric utilities, grid operators, manufacturers, government entities, and research organizations make critical decisions about prioritizing research and development, utility resource and distribution system planning, and state and local energy planning and regulation." Similar to the EULP, Buildings-900K is a collection of Parquet files and it follows nearly the same Parquet dataset organization as the EULP. As it only contains a single energy consumption time series per building, it is much smaller (~110 GB). BuildingsBench also provides an evaluation benchmark that is a collection of various open source residential and commercial real building energy consumption datasets. The evaluation datasets, which are provided alongside Buildings-900K below, are collections of CSV files which contain annual energy consumption. The size of the evaluation datasets altogether is less than 1GB, and they are listed out below: 1. ElectricityLoadDiagrams20112014 2. Building Data Genome Project-2 3. Individual household electric power consumption (Sceaux) 4. Borealis 5. SMART 6. IDEAL 7. Low Carbon London A README file providing details about how the data is stored and describing the organization of the datasets can be found within each data lake version under BuildingsBench.
Array↗
Maximizing efficiency of dataset compression for machine learning potentials with information theory
Machine learning interatomic potentials (MLIPs) balance high accuracy and lower costs compared to density functional theory calculations, but their performance often depends on the size and diversity of training datasets. Large datasets improve model accuracy and generalization but are computationally expensive to produce and train on, while smaller datasets risk discarding rare but important atomic environments and compromising MLIP accuracy/reliability. Here, we develop an information-theoretical framework to quantify the efficiency of dataset compression methods and propose an algorithm that maximizes this efficiency. By framing atomistic dataset compression as an instance of the minimum set cover (MSC) problem over atom-centered environments, our method identifies the smallest subset of structures that contains as much information as possible from the original dataset while pruning redundant information. The approach is extensively demonstrated on the GAP-20 and TM23 datasets and validated on 64 varied datasets from the ColabFit repository. Across all cases, MSC consistently retains outliers, preserves dataset diversity, and reproduces the long-tail distributions of forces even at high compression rates, outperforming other subsampling methods. Furthermore, MLIPs trained on MSC-compressed datasets exhibit reduced error for out-of-distribution data even in low-data regimes. We explain these results using an outlier analysis and show that such quantitative conclusions could not be achieved with conventional dimensionality reduction methods. The algorithm is implemented in the open-source QUESTS package and can be used for several tasks in atomistic modeling, from data subsampling, outlier detection, and training improved MLIPs at a lower cost.
36 MATERIALS SCIENCE↗
Downscaled hyper-resolution (400 m) gridded datasets of daily precipitation and temperature (2008–2019) for the East–Taylor subbasin (western United States)
Abstract. High-resolution gridded datasets of meteorological variables are needed in order to resolve fine-scale hydrological gradients in complex mountainous terrain. Across the United States, the highest available spatial resolution of gridded datasets of daily meteorological records is approximately 800 m. This work presents gridded datasets of daily precipitation and mean temperature for the East–Taylor subbasin (in the western United States) covering a 12-year period (2008–2019) at a high spatial resolution (400 m). The datasets are generated using a downscaling framework that uses data-driven models to learn relationships between climate variables and topography. We observe that downscaled datasets of precipitation and mean temperature exhibit smoother spatial gradients (while preserving the spatial variability) when compared to their coarser counterparts. Additionally, we also observe that when downscaled datasets are upscaled to the original resolution (800 m), the mean residual error is almost zero, ensuring no bias when compared with the original data. Furthermore, the downscaled datasets are observed to be linearly related to elevation, which is consistent with the methodology underlying the original 800 m product. Finally, we validate the spatial patterns exhibited by downscaled datasets via an example use case that models lidar-derived estimates of snowpack. The presented dataset constitutes a valuable resource to resolve fine-scale hydrological gradients in the mountainous terrain of the East–Taylor subbasin, which is an important study area in the context of water security for the southwestern United States and Mexico. The dataset is publicly available at https://doi.org/10.15485/1822259 (Mital et al., 2021).
54 ENVIRONMENTAL SCIENCES↗
Xanthos-Lake Dataset
The Xanthos-Lake v1.0 dataset provides the input data, trained machine-learning models, and simulation outputs needed to characterize lake water balance, snow and ice conditions, and mixing-layer temperature within the Xanthos global hydrological modeling framework. The dataset supports lake representation across a wide range of lake sizes and hydroclimatic conditions by combining xLSIM, a basin-specific machine-learning emulator of lake snow, ice, ice-cover fraction, and mixing-layer temperature, with the Xanthos-Lake water-balance model. The archive contains NetCDF datasets used to train and evaluate xLSIM, trained model weights, processed meteorological and lake-property inputs, and basin- and lake-category-specific simulation outputs. These materials are organized into four primary data groups, described below. Snowice_model_inputs: Contains the NetCDF input data used to train xLSIM. The xLSIM machine-learning framework uses three lake-based datasets. The meteorological forcing dataset provides monthly relative humidity, specific humidity, surface wind speed, maximum and minimum air temperature, downward longwave and shortwave radiation, snowfall, surface air pressure, and total precipitation. Lake surface area is included as an additional static predictor. The target-state dataset provides lake ice thickness, snow depth, snow cover, and lake mixing-layer temperature, while a companion lake-surface dataset provides the lake ice-cover fraction. Before training, ice thickness and snow depth are converted from meters to centimeters, mixing-layer temperature is converted from kelvin to degrees Celsius and constrained to nonnegative values, and ice-cover fraction is converted from a fraction to a percentage. The predictor variables are normalized using statistics calculated across the selected lakes and time steps. Snowice_model_outputs: Contains the NetCDF outputs generated by xLSIM. For each basin, xLSIM produces a file containing observed and predicted lake-state variables for the training, validation, and testing periods. The modeled variables include lake ice thickness, snow depth, snow cover, mixing-layer temperature, and lake ice-cover fraction. For basins without a sufficiently persistent snow-and-ice signal, the emulator predicts only mixing-layer temperature. The outputs also include training and validation loss histories, the selected model configuration, identifiers of the lakes used in training, and SHAP-based feature-importance information at the global, lake, and seasonal-regime levels. The trained machine-learning model weights are provided separately within the dataset archive. Together, these files support model evaluation and subsequent coupling with the Xanthos-Lake water-balance framework. XanthosLAKES: Contains the NetCDF input data used by the Xanthos-Lake framework. Monthly meteorological inputs include relative and specific humidity, downward shortwave and longwave radiation, mean, maximum, and minimum air temperature, wind speed, precipitation, snowfall, and surface air pressure. Static lake-property datasets provide lake identifiers, geographic locations, surface area, volume, mean depth, elevation, drainage area, fetch, outlet-routing information, and associated Xanthos grid-cell attributes. Separate bathymetric datasets provide the coefficients of the area–depth and volume–depth relationships for each aggregated lake unit. GLEV-based records provide observed lake surface area and evaporation data used to initialize lake states, define reference conditions, and calibrate and evaluate the model. Xanthos-Lake Outputs: Contains the basin- and lake-category-specific NetCDF outputs generated by Xanthos-Lake. Monthly variables include lake surface area, storage volume, outlet discharge, evaporation rate, evaporation volume, lake–groundwater exchange, lake inflow, ice thickness, snow depth, snow-cover fraction, ice-cover fraction, and mixing-layer temperature. The files also contain lake-specific calibration and validation statistics, including normalized root-mean-square error, mean absolute error, Nash–Sutcliffe efficiency, Kling–Gupta efficiency, and percent bias. Stored calibrated and derived parameters include the weir discharge coefficient, fractional freeboard, groundwater exchange coefficient, reference water level, corresponding reference surface area and storage volume, weir-width adjustment factor, and the fraction of routed inflow entering the lake. Basin identifiers, lake category, simulation period, calibration and validation periods, and parameter-schema information are retained as NetCDF metadata.
Abeshu, Guta [Pacific Northwest National Laborator↗