A thirty-year (1st January 1995-31st December 2025) assessment of PM2.5 variability and meteorological controls over Bengaluru, India, was conducted using NASA MERRA-2 reanalysis data. Validation against CPCB observations, including a multi-station comparison across nine Continuous Ambient Air Quality Monitoring Stations (CAAQMS) during 2023-2024, demonstrated generally strong agreement between MERRA-2 and ground observations, supporting the suitability of the reanalysis product for long-term air-quality assessment. Mann-Kendall analysis revealed a highly significant increasing PM2.5 trend (Sen's slope = + 0.317 mu g m(-3) yr(-1); p < 0.0001), whereas key meteorological parameters exhibited no comparable long-term trends, indicating a dominant anthropogenic influence on the observed increase. Seasonally, PM2.5 concentrations peaked during the pre-monsoon season (22.1 mu g m(-3)) and reached minimum levels during the monsoon season (12.7 mu g m(-3)) owing to enhanced atmospheric ventilation and wet scavenging. Analysis of planetary boundary layer height (PBLH) revealed that the apparent positive PM2.5-PBLH relationship was seasonally confounded, with elevated PBLH ultimately promoting pollutant dilution above approximately 800 m. Extreme pollution episodes occurred predominantly during pre-monsoon and winter periods under low-moisture and weak-wind conditions. Random Forest analysis identified water vapour (27.1%) and wind speed (26.6%) as the dominant meteorological controls on PM2.5 variability, whereas PBLH contributed only 12.7%. Overall, the study provides one of the first integrated long-term assessments of PM2.5 variability over Bengaluru by combining satellite reanalysis, multi-station ground validation, trend analysis, nonlinear meteorological interpretation, and machine-learning attribution to distinguish anthropogenic forcing from meteorological variability.
The hybrid humidity generation model is a fundamental framework from which other models can be derived, such as the two-pressure, two-temperature, and two-flow humidity generator. With this type of generator, it is possible to generate airstreams of known relative humidity, dew point temperature and other quantities related to humidity. This model is derived from the mass conservation principle applied to a control volume and is regarded as a primary method; therefore, it is used as reference standard, in which uncertainty estimation is required. For its operation a dry air supply and the measurement of pressure, temperature, and mass flow rate throughout the main parts of the generator (saturator and test chamber) are required. Due to the presence of possible asymmetries, non-linearities or uncertainty sources from non-Gaussian distributions on the hybrid humidity generation model, it is necessary to carry out the uncertainty estimation using at least two different approaches. In this case we chose the approach given in the Guide to expression of Uncertainty of Measurement (GUM) and the Monte Carlo method (MCM). The latter is a stochastic method based on the propagation of probability distributions, which considers possible non-linearities in the hybrid humidity generation model and other conditions where the GUM does not work satisfactorily. Since MCM has a validity broader than GUM framework, it can be used for the validation of this approach. The obtained results show small differences between both methods, due to possible asymmetries in the model, from which it is not possible to validate the GUM approach in all cases.
Ultrasound is an important tool for many industrial applications, including in-line real-time process monitoring. However, the use of a metrological approach is not as widespread as it could be, for instance, for ultrasonic attenuation measurements. The main goal of this paper is to present a complete uncertainty budget for individual data points and the regression curve for the calibration of glycerol's ultrasonic attenuation as a function of frequency in the range of 1.0–5.0 MHz. Experimental data were acquired using a pulsed stepped-sine 20-cycle burst ultrasonic scheme. Attenuation was calculated using both theoretical and experimental approaches to validate the experimental setup and method. Experimental and theoretical attenuation variances are statistically homogeneous across all frequencies (Fisher test; p = 0.95), and attenuation pairs of results are statistically equivalent across all frequencies (t-Student test; p = 0.95). Theoretical and experimental attenuation coefficients were assessed using a regression curve of corrected attenuation results at 20 °C. For data from 1.0 to 5.0 MHz, the theoretical slope is 0.2847 dB cm−1 MHz−2, and the experimental slope is 0.2582 dB cm−1 MHz−2. Combining both slopes for a limited frequency range (1.0–3.5 MHz), the attenuation coefficient for pure glycerol at 20 °C is 0.249 dB⋅cm−1⋅MHz−2 with expanded uncertainty of 0.016 dB⋅cm−1⋅MHz−2 (p = 0.95). The presented method can be used in other chemometric approaches to assess, for instance, biofuel quality and purity.
Modal participation factors are fundamental to understanding modal dominance in vibrating structures, yet their estimation conventionally relies on contact sensors or laser-based instrumentation, limiting scalability and increasing experimental complexity. This study presents a vision-only framework for quantitative estimation of modal participation factors using a single smartphone camera, eliminating the need for physical sensors or specialised vibration measurement systems. The proposed method reconstructs full-field displacement responses from high-speed smartphone video through Harris corner detection–based feature tracking and performs modal decomposition by fitting analytical Euler–Bernoulli mode shape basis functions to the image-derived spatial displacement data via nonlinear least-squares optimisation. Unlike existing vision-based vibration studies that primarily report natural frequencies or qualitative mode shapes, this work enables time-resolved quantification of individual modal contributions under free and forced vibration conditions. Experiments are conducted on a cantilever beam subjected to free vibration and harmonic excitation near first and second resonance frequencies. Results show that during free vibration, the first bending mode dominates the response with a modal participation factor of approximately 0.80. Under forced excitation at 11 Hz, the response exhibits near single-mode behaviour, with the first mode contributing a modal participation factor exceeding 0.97. At higher excitation near 73 Hz, a clear transition from first-mode to second-mode dominance occurs, with the second bending mode becoming dominant with a modal participation factor of approximately 0.60, while higher-order modes remain secondary. Forced vibration tests performed over a duration of 10 s yielded a maximum displacement amplitude of 63.87 mm with a standard deviation of 2.34 mm, indicating stable excitation and predominantly linear system behaviour. The results demonstrate that modal participation factors—traditionally obtained using contact-based instrumentation, can be quantitatively extracted from video measurements alone, advancing vision-based vibration analysis from qualitative observation to practical modal assessment using readily available imaging devices.
The efficiency of GPS performance depends greatly on ionospheric interference, which is caused by solar terrestrial effects, resulting in delays and reduced accuracy. There are different influences such as variations in TEC, parameters of solar wind, and geomagnetic activities, which are responsible for GPS signal delay, and thus it is necessary to develop an accurate deep learning model for GPS error estimation. This paper suggests a new deep learning approach for forecasting GPS errors using multi-domain parameters of atmospheric and geomagnetic effects, such as TEC, solar wind speed, Kp index, IMF Bz, and RSSI, to make the input data representation complete. In contrast to previous studies that have employed only separate parameters or statistics-based models, our model uses multi-parameter fusion technology, which is more effective in predicting GPS errors due to complex nonlinear relationships between various environmental factors. Moreover, we employ the most effective deep sequential learning models like LSTM and bidirectional LSTM networks to detect complex temporal and long-term patterns of GPS signal delay evolution. The experimental results show that our model achieves the lowest RMSE and MAE with values 3.0925 and 2.5233 correspondingly, outperforming other LSTM and GRU-based models. The combination of real-time RSSI with space weather parameters improves the accuracy of prediction, which makes the system applicable in satellite-based navigation, such as intelligent transport systems, IoT-based systems, autonomous navigation, and real-time positioning.
