Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “COMPARATIVE TESTING”

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 1,081 records · Page 60

The EGS Collab Project – Summaries of Experiments 2 and 3: Experiments at 1.25 km depth at the Sanford Underground Research Facility

The EGS Collab project performed well-monitored rock stimulation and flow tests at the 10-m scale in an underground research laboratory to inform challenges in implementing enhanced geothermal system (EGS) technology. This project, supported by the US Department of Energy, gathered data and observations from the field tests and compared these to simulation results to understand processes and to build confidence in numerical modeling of the processes.

15 GEOTHERMAL ENERGY↗

Sister Rod Destructive Examinations (FY23) Appendix F: Cyclic Integrated Reversible-Bending Fatigue Tests

Spent nuclear fuel (SNF) assemblies must be shipped to other sites for processing and disposal. During shipment, the fuel is typically oriented horizontally, and the fuel rods are subject to periodic alternating loads related to the movement of the vehicle that results in alternating bending of the SNF fuel rods. The number of bending cycles is related to the length of the shipping route, with longer routes producing more cycles. Because it is well-known that cyclic loads can produce failures, even when the stress and strain imposed are below the material’s yield point, investigation of the SNF’s fatigue behavior is prudent. This report discusses the results of fatigue testing conducted at Oak Ridge National Laboratory (ORNL) using the Cyclic Integrated Reversible-Bending Fatigue Tester (CIRFT) created by the US Department of Energy Office of Nuclear Energy for the High Burnup Spent Fuel Data Project and its sister rods. The results of the tests are compared with results obtained over the last decade using the same CIRFT for the US Nuclear Regulatory Commission (NRC).

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Biofouling Analysis for Wave Energy Piston Design: Post-Access Test Report

This report summarizes the purpose, procedures, results, and conclusions of prolonged seawater tank testing of a prototype wave energy converter apparatus. Test objectives included determination of the effectiveness of a biofouling mitigation element in one test assembly compared to a control, and system force over many cycles of actuation over months.

16 TIDAL AND WAVE POWER↗

Prioritizing Off-Gas Metrics: A Guide for Comparable Off-Gas Capture Testing

The Material Recovery and Waste Form Development (MRWFD) off-gas team had a workshop, hosted by Idaho National Laboratory (INL), to align goals and expectations for off-gas research. The workshop included team members from four national laboratories. The workshop focused on defining distinct R&D phases with specific metrics, outlining standard test and measurement protocols for Iodine and Krypton/Xenon sorbents, brainstorming approaches to future disruptive technologies, and recognizing parameters with more inherent risk, requiring more rigorous evaluation. This report will serve as a guide for future off-gas work. Its purpose is to foster efficient collaboration across diverse research facilities and invite direct comparison of materials and results.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

4.5m GTEM RS105 Test Capability Commissioning

This work details the reconfiguration of the 4.5 m Gigahertz Transverse Electromagnetic test facility at Sandia National Laboratories to operate in accordance with the RS105 (radiated susceptibility) test from MIL-STD-461 representing a high-altitude electromagnetic pulse. This reconfiguration involved removal of the existing continuous wave source and connecting both a high voltage feed and a coaxial feed housing the Marx bank pulser. Marx control settings were calibrated for several voltage levels across two pulsers, and position-dependent measurements of the peak electric field were taken throughout the test volume for each pulser. The results showed field uniformity and purity across the test volume comparable to continuous wave operations, and field peaks were measured from 1.63 kV/m to 54.8 kV/m, with maximum capabilities expected to exceed 100 kV/m. Some challenges in consistent pulser operations at lower Marx bank voltages and high frequency reflections in the system were identified for future capability improvements.

42 ENGINEERING↗

The Circular Home: Development and Demonstration of a Net Negative Carbon, Reusable Residence

This project started the development of an innovative modular building system intended for residential construction. The project was centered around single-family homes that were carbon-negative cradle-to-grave over a 100 year time frame, which is approximately double the current standard for operational life. The project sought to accomplish this objective by designing the modular home in a manner that ensures circularity, where the main house components can be used over several consecutive 50-year lifespans. To accomplish these objectives, this project utilized integrated design with the inclusion of life-cycle assessment to design the single-family house for architectural, structural, energy, mechanical, thermal, and moisture demands, while ensuring carbon negativity and annual net-zero energy use. The core technology of this project was the use of cross-laminated timber and biogenic materials, such as wood-fiber insulation, in the construction of the modular building units. The robustness and factory manufacturing ability of cross-laminated timber allow for factory construction of most of the home, which minimizes on-site time, saving money and reducing construction waste. During this project, initial milestones were met that delivered the architectural plans for the circular home and an initial structural testing matrix. Compared to current code-built homes, which average 13 kg CO2eq. / ft2 and are demolished at their end of life, the circular home has an estimated -30 kg CO2eq. / ft2 of embodied carbon emissions during its first build iteration. It is estimated that approximately 60%-70% of the total building mass could be reused and/or recycled during subsequent rebuilds. This project was concluded at approximately the 1/3 point and a separate project was established to conclude the remaining milestones. This project promises to benefit the public by delivering another option for single-family, and eventually multi-family, housing using a novel building construction system. The system of reusable modular construction facilitates not only lower emissions during the first building iteration, but also lower emissions during subsequent iterations that drastically reduce waste and help society meet its climate goals. Many other industries, such as clothing and technology sectors, are starting to focus on circularity and this project adds the residential building construction industry to that list.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Hydraulic Testing of the Lost Lake Aquifer Near Recovery Well RWM001R

An aquifer pumping test was conducted on the Lost Lake Aquifer Zone (LLAZ) at the recently installed recovery well RWM001R in accordance with the approved test plan (Dixon, 2024). The objective of the testing was to determine baseline well performance parameters and aquifer hydraulic conductivity. This testing consisted of two step-drawdown tests to determine well performance properties and two constant pumping rate aquifer tests with recovery periods to determine aquifer hydraulic properties. Well performance parameters determined included specific capacity, well efficiency, and head loss coefficients. The results of the step-drawdown testing are presented in Tables ES1 and ES2 whereas the results of the constant pumping rate testing are presented in Table ES3. The average specific capacity of RWM001R operating at pump capacity (~54 gpm) was 2.2 gpm/ft using data from the final step of both tests. The average well efficiency determined from both step-drawdown tests was 56% using data from the final step of both tests. For both step-drawdown tests, well efficiency was inversely related to the pumping rate with the average efficiency decreasing from approximately 79% to 56%. Aquifer head loss coefficients from both tests were comparable (Table ES2).

54 ENVIRONMENTAL SCIENCES↗

Experimental Investigation of Plasma Processing for PIP-II SSR2 Cavities

Field emission and multipacting are critical factors that limit the achievable operational gradient of superconducting radio frequency (SRF) cavities. Plasma processing (PP) has been demonstrated as an effective in-situ technique for mitigating hydrocarbon-induced field emission and multipacting across a range of cavity geometries. In this work, we present the initial development and subsequent application of plasma processing for PIP-II type II spoke resonators (SSR2). The process was carried out at room temperature on an SSR2 cavity equipped with the CM-style high-power coupler and installed in the cryostat. Cavity performance was evaluated through comparative cold tests conducted before and after plasma processing.

Giaccone, Bianca [Fermilab] (ORCID:000000027275846↗

Environmentally assisted cracking in light water reactors.

This report summarizes work performed by Argonne National Laboratory on fatigue and environmentally assisted cracking (EAC) in light water reactors (LWRs) from January to December 2002. Topics that have been investigated include: (a) environmental effects on fatigue crack initiation in carbon and low-alloy steels and austenitic stainless steels (SSs), (b) irradiation-assisted stress corrosion cracking (IASCC) of austenitic SSs in BWRs, (c) evaluation of causes and mechanisms of irradiation-assisted cracking of austenitic SS in PWRs, and (d) cracking in Ni-alloys and welds. A critical review of the ASME Code fatigue design margins and an assessment of the conservation in the current choice of design margins are presented. The existing fatigue {var_epsilon}-N data have been evaluated to define the effects of key material, loading, and environmental parameters on the fatigue lives of carbon and low-alloy steels and austenitic SSs. Experimental data are presented on the effects of surface roughness on fatigue crack initiation in these materials in air and LWR environments. Crack growth tests were performed in BWR environments on SSs irradiated to 0.9 and 2.0 x 10{sup 21} n x cm{sup -2}. The crack growth rates (CGRs) of the irradiated steels are a factor of {approx}5 higher than the disposition curve proposed in NUREG-0313 for thermally sensitized materials. The CGRs decreased by an order of magnitude in low-dissolved oxygen (DO) environments. Slow-strain-rate tensile (SSRT) tests were conducted in high-purity 289 C water on steels irradiated to {approx}3 dpa. The bulk S content correlated well with the susceptibility to intergranular SCC in 289 C water. The IASCC susceptibility of SSs that contain >0.003 wt. % S increased drastically. bend tests in inert environments at 23 C were conducted on broken pieces of SSRT specimens and on unirradiated specimens of the same materials after hydrogen charging. The results of the tests and a review of other data in the literature indicate that IASCC in 289 C water is dominated by a crack-tip grain-boundary process that involves S. An initial IASCC model has been proposed. A crack growth test was completed on mill annealed Alloy 600 in high-purity water at 289 C and 320 C under various environmental and loading conditions. The results from this test are compared with data obtained earlier on several other heats of Alloy 600.

08 HYDROGEN↗

Modeling and Simulation of Fluid Mixing Laser Experiments and Supernova

The three year plan for this project is to develop novel theories and advanced simulation methods leading to a systematic understanding of turbulent mixing. A primary focus is the comparison of simulation models (both Direct Numerical Simulation and subgrid averaged models) to experiments. The comprehension and reduction of experimental and simulation data are central goals of this proposal. We will model 2D and 3D perturbations of planar interfaces. We will compare these tests with models derived from averaged equations (our own and those of others). As a second focus, we will develop physics based subgrid simulation models of diffusion across an interface, with physical but no numerical mass diffusion. We will conduct analytic studies of mix, in support of these objectives. Advanced issues, including multiple layers and reshock, will be considered.

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEF↗

Modeling and Simulation of Fluid Mixing Laser Experiments and Supernova

The three year plan for this project was to develop novel theories and advanced simulation methods leading to a systematic understanding of turbulent mixing. A primary focus is the comparison of simulation models (Direct Numerical Simulation (DNS), Large Eddy Simulations (LES), full two fluid simulations and subgrid averaged models) to experiments. The comprehension and reduction of experimental and simulation data are central goals of this proposal. We model 2D and 3D perturbations of planar or circular interfaces. We compare these tests with models derived from averaged equations (our own and those of others). As a second focus, we develop physics based subgrid simulation models of diffusion across an interface, with physical but no numerical mass diffusion. Multiple layers and reshock are considered here.

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEF↗

EFRT M-12 Issue Resolution: Comparison of Filter Performance at PEP and CUF Scale

Pacific Northwest National Laboratory (PNNL) has been tasked by Bechtel National Inc. (BNI) on the River Protection Project-Hanford Tank Waste Treatment and Immobilization Plant (RPP-WTP) project to perform research and development activities to resolve technical issues identified for the Pretreatment Facility (PTF). The Pretreatment Engineering Platform (PEP) was designed and constructed and is to be operated as part of a plan to respond to issue M12, Undemonstrated Leaching Processes. The PEP is a 1/4.5-scale test platform designed to simulate the WTP pretreatment caustic leaching, oxidative leaching, ultrafiltration solids concentration, and slurry washing processes. The PEP replicates the WTP leaching processes using prototypic equipment and control strategies. The PEP also includes non-prototypic ancillary equipment to support the core processing. Two operating scenarios are currently being evaluated for the ultrafiltration process (UFP) and leaching operations. The first scenario has caustic leaching performed in the UFP-2 ultrafiltration feed vessels (i.e., vessel UFP-VSL-T02A in the PEP and vessels UFP-VSL-00002A and B in the WTP PTF). The second scenario has caustic leaching conducted in the UFP-1 ultrafiltration feed-preparation vessels (i.e., vessels UFP-VSL-T01A and B in the PEP; vessels UFP-VSL-00001A and B in the WTP PTF). In both scenarios, 19-M sodium hydroxide solution (NaOH, caustic) is added to the waste slurry in the vessels to leach solid aluminum compounds (e.g., gibbsite, boehmite). Caustic addition is followed by a heating step that uses direct injection of steam to accelerate the leach process. Following the caustic leach, the vessel contents are cooled using vessel cooling jackets and/or external heat exchangers. The main difference between the two scenarios is that for leaching in UFP1, the 19-M NaOH is added to un-concentrated waste slurry (3 to 8 wt% solids), while for leaching in UFP2, the slurry is concentrated to nominally 20 wt% solids using cross-flow ultrafiltration before the addition of caustic. The work described in this report presents filter flux results obtained at two different scales based on tests performed with a Hanford tank waste simulant. The tests were made at the lab-bench scale on a cold (i.e., designated for non-radioactive simulant test materials) Cells Unit Filter [CUF]) and in the Pretreatment Engineering Platform (PEP), which is a 1/4.5-scale mock-up of key PTF process equipment. One set of tests was conducted with the simulant feed (low solids), and one test was conducted at a relatively high solids concentration. The results of these tests are compared to support a scale factor for use in the WTP.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Contam airflow models of three large buildings: Model descriptions and validation

Airflow and pollutant transport models are useful for several reasons, including protection from or response to biological terrorism. In recent years they have been used for deciding how many biological agent samplers are needed in a given building to detect the release of an agent; to figure out where those samplers should be located; to predict the number of people at risk in the event of a release of a given size and location; to devise response strategies in the event of a release; to determine optimal trade-offs between sampler characteristics (such as detection limit and response time); and so on. For some of these purposes it is necessary to model a specific building of interest: if you are trying to determine optimal sampling locations, you must have a model of your building and not some different building. But for many purposes generic or 'prototypical' building models would suffice. For example, for determining trade-offs between sampler characteristics, results from one building will carry over other, similar buildings. Prototypical building models are also useful for comparing or testing different algorithms or computational pproaches: different researchers can use the same models, thus allowing direct comparison of results in a way that is not otherwise possible. This document discusses prototypical building models developed by the Airflow and Pollutant Transport Group at Lawrence Berkeley National Laboratory. The models are implemented in the Contam v2.4c modeling program, available from the National Institutes for Standards and Technology. We present Contam airflow models of three virtual buildings: a convention center, an airport terminal, and a multi-story office building. All of the models are based to some extent on specific real buildings. Our goal is to produce models that are realistic, in terms of approximate magnitudes, directions, and speeds of airflow and pollutant transport. The three models vary substantially in detail. The airport model is the simplest; the onvention center model is more detailed; and the large office building model is quite complicated. We give several simplified floor plans in this document, to explain basic features of the buildings. The actual models are somewhat more complicated; for instance, spaces that are represented as rectangles in this document sometimes have more complicated shapes in the models. (However, note that the shape of a zone is irrelevant in Contam). Consult the Contam models themselves for detailed floor plans. Each building model is provided with three ventilation conditions, representing mechanical systems in which 20%, 50%, or 80% of the building air is recirculated and the rest is provided from outdoors. Please see the section on 'Use of the models' for important information about issues to consider if you wish to modify the models to provide no mechanical ventilation or eliminate provision of outdoor air.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Investigations into the origins of polyatomic ions in inductively coupled plasma-mass spectrometry

An inductively coupled plasma-mass spectrometer (ICP-MS) is an elemental analytical instrument capable of determining nearly all elements in the periodic table at limits of detection in the parts per quadrillion and with a linear analytical range over 8-10 orders of magnitude. Three concentric quartz tubes make up the plasma torch. Argon gas is spiraled through the outer tube and generates the plasma powered by a looped load coil operating at 27.1 or 40.6 MHz. The argon flow of the middle channel is used to keep the plasma above the innermost tube through which solid or aqueous sample is carried in a third argon stream. A sample is progressively desolvated, atomized and ionized. The torch is operated at atmospheric pressure. To reach the reduced pressures of mass spectrometers, ions are extracted through a series of two, approximately one millimeter wide, circular apertures set in water cooled metal cones. The space between the cones is evacuated to approximately one torr. The space behind the second cone is pumped down to, or near to, the pressure needed for the mass spectrometer (MS). The first cone, called the sampler, is placed directly in the plasma plume and its position is adjusted to the point where atomic ions are most abundant. The hot plasma gas expands through the sampler orifice and in this expansion is placed the second cone, called the skimmer. After the skimmer traditional MS designs are employed, i.e. quadrupoles, magnetic sectors, time-of-flight. ICP-MS is the leading trace element analysis technique. One of its weaknesses are polyatomic ions. This dissertation has added to the fundamental understanding of some of these polyatomic ions, their origins and behavior. Although mainly continuing the work of others, certain novel approaches have been introduced here. Chapter 2 includes the first reported efforts to include high temperature corrections to the partition functions of the polyatomic ions in ICP-MS. This and other objections to preceeding papers in this area were addressed. Errors in the measured T gas values were found for given errors in the experimental and spectroscopic values. The ionization energy of the neutral polyatomic ion was included in calculations to prove the validity of ignoring more complicated equilibria. Work was begun on the question of agreement between kinetics of the plasma and interface and the increase and depletion seen in certain polyatomic ions. This dissertation was also the first to report day to day ranges for T gas values and to use a statistical test to compare different operating conditions. This will help guide comparisons of previous and future work. Chapter 4 was the first attempt to include the excited electronic state 2 in the partition function of ArO + as well as the first to address the different dissociation products of the ground and first electronic levels of ArO + . Chapter 5 reports an interesting source of memory in ICP-MS that could affect mathematical corrections for polyatomic ions. For future work on these topics I suggest the following experiments and investigations. Clearly not an extensive list, they are instead the first topics curiosity brings to mind. (1) Measurement of T gas values when using the flow injection technique of Appendix B. It was believed that there was a fundamental difference in the plasma when the auto-sampler was used versus a continuous injection. Is this reflected in T gas values? (2) The work of Chapter 3 can be expanded and supplemented with more trials, new cone materials (i.e. copper, stainless steel) and more cone geometries. Some of this equipment is already present in the laboratory, others could be purchased or made. (3) T gas values from Chapter 3 could be correlated with instrument pressures during the experiment. Pressures after the skimmer cone were recorded for many days but have yet to be collated with the measured T gas values. (4) The work in Chapter 5 could be expanded to include more metals. Does the curious correlation between measured T gas and element boiling point persist? (5) Investigate non-linear correlations to T gas values of the MO + memory in Chapter 5. Temperatures along the skimmer walls are not a linear gradient. Ring deposits have been observed on the cone and photographs of the interface show light intensities shaping a sort of tailing peak along the outside skimmer wall. Is there a physical property of the metals or metal oxides that would give this peak with the T gas values? (6) Chemical state speciation of the metal deposits on the skimmers of Chapter 5. There may be a more logical correlation between Tgas and a physical property of the deposit-ing chemical if all the metals do not deposit in the same form. (7) A collaboration with our computational collegues would be most welcome. Newer calculations for ArO + and RuO + would be very helpful.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Improving Residential Building Simulations Through Large-Scale Empirical Validation

Residential building energy simulations are increasingly used for energy-efficient building design, codes and standards analysis, home certifications and ratings, utility programs, and technology assessments. Various software tools exist to perform residential building simulations, and these tools often use different models, inputs, and assumptions. This leads to inconsistencies that can undermine confidence in the predicted results. Validation of these tools can increase confidence by ensuring their accuracy and consistency. One way to validate simulation tools is through empirical testing, which compares predicted energy usage to measured utility billing data. This paper describes the process of data collection, data standardization, and empirical validation, and illustrates its use with our residential EnergyPlus (R)-based software. The data and process can be extended to other simulation tools and contribute to improving residential building simulations more broadly.

empirical validation↗

Enhancing segmentation fairness through curriculum learning and progressive loss: a centralized and federated perspective on radiograph analysis

Bias in medical image segmentation can lead to unequal performance across demographic subgroups, raising concerns about fairness and reliability in clinical AI systems. While deep learning models have achieved high segmentation accuracy, ensuring equitable performance across race and gender remains a significant challenge, particularly in privacy-sensitive healthcare environments. This study investigates fairness-aware medical image segmentation for hip and knee radiographs using deep learning models evaluated in both centralized and Federated Learning (FL) settings. We introduce Curriculum Learning (CL) strategies and Progressive Loss (PL) functions to regulate sample difficulty during training. In addition, we propose two novel fairness-oriented federated learning algorithms, Federated Intersection over Union (FedIoU) and Federated Intersection over Union with Outlier Analysis (FedIoUoutlier). Experiments are conducted using multiple segmentation backbones and simulated multi-site data partitions derived from the Osteoarthritis Initiative dataset. Model performance is evaluated using Intersection over Union (IoU), IoU standard deviation, Skewed Error Ratio (SER), and Min-Max Disparity across race and gender subgroups. Statistical significance was verified using paired t-tests to compare per-sample IoU performance against baseline configurations. Across both hip and knee segmentation tasks, curriculum learning and progressive loss strategies consistently improved segmentation accuracy and reduced demographic performance disparities in centralized training. In federated settings, fairness-aware aggregation further enhanced performance. Notably, FedIoUoutlier combined with balanced curriculum learning and tiered progressive loss achieved the highest mean IoU while yielding the lowest SER and Min-Max Disparity, indicating improved fairness without sacrificing accuracy. In several configurations, federated models matched or exceeded the performance of optimized centralized models, with statistically significant improvements in per-sample IoU over baseline configurations. The results demonstrate that structured training strategies and fairness-aware federated aggregation can jointly improve accuracy, stability, and demographic fairness in medical image segmentation. By integrating curriculum learning, progressive loss, and novel FL algorithms, this work provides a practical pathway toward equitable and privacy-preserving AI systems for medical imaging.

97 MATHEMATICS AND COMPUTING↗

A Two-Step PCR Protocol Enabling Flexible Primer Choice and High Sequencing Yield for Illumina MiSeq Meta-Barcoding

High-throughput amplicon sequencing that primarily targets the 16S ribosomal DNA (rDNA) (for bacteria and archaea) and the Internal Transcribed Spacer rDNA (for fungi) have facilitated microbial community discovery across diverse environments. A three-step PCR that utilizes flexible primer choices to construct the library for Illumina amplicon sequencing has been applied to several studies in forest and agricultural systems. The three-step PCR protocol, while producing high-quality reads, often yields a large number (up to 46%) of reads that are unable to be assigned to a specific sample according to its barcode. Here, we improve this technique through an optimized two-step PCR protocol. We tested and compared the improved two-step PCR meta-barcoding protocol against the three-step PCR protocol using four different primer pairs (fungal ITS: ITS1F-ITS2 and ITS1F-ITS4, and bacterial 16S: 515F-806R and 341F-806R). We demonstrate that the sequence quantity and recovery rate were significantly improved with the two-step PCR approach (fourfold more read counts per sample; determined reads ≈90% per run) while retaining high read quality (Q30 > 80%). Given that synthetic barcodes are incorporated independently from any specific primers, this two-step PCR protocol can be broadly adapted to different genomic regions and organisms of scientific interest.

16S rDNA↗

A Comparison of Machine Learning Methods to Forecast Tropospheric Ozone Levels in Delhi

Ground-level ozone is a pollutant that is harmful to urban populations, particularly in developing countries where it is present in significant quantities. It greatly increases the risk of heart and lung diseases and harms agricultural crops. This study hypothesized that, as a secondary pollutant, ground-level ozone is amenable to 24 h forecasting based on measurements of weather conditions and primary pollutants such as nitrogen oxides and volatile organic compounds. We developed software to analyze hourly records of 12 air pollutants and 5 weather variables over the course of one year in Delhi, India. To determine the best predictive model, eight machine learning algorithms were tuned, trained, tested, and compared using cross-validation with hourly data for a full year. The algorithms, ranked by R2 values, were XGBoost (0.61), Random Forest (0.61), K-Nearest Neighbor Regression (0.55), Support Vector Regression (0.48), Decision Trees (0.43), AdaBoost (0.39), and linear regression (0.39). When trained by separate seasons across five years, the predictive capabilities of all models increased, with a maximum R 2 of 0.75 during winter. Bidirectional Long Short-Term Memory was the least accurate model for annual training, but had some of the best predictions for seasonal training. Out of five air quality index categories, the XGBoost model was able to predict the correct category 24 h in advance 90% of the time when trained with full-year data. Separated by season, winter is considerably more predictable (97.3%), followed by post-monsoon (92.8%), monsoon (90.3%), and summer (88.9%). These results show the importance of training machine learning methods with season-specific data sets and comparing a large number of methods for specific applications.

54 ENVIRONMENTAL SCIENCES↗