सीएसआईआर-राष्ट्रीय डेटा विज्ञान एवं कृत्रिम बुद्धिमत्ता संस्थान

Modeling and forecasting atmospheric Carbon Dioxide concentrations at Bengaluru city in India

by Iranna Gogeri, K. C. Gouda & T. Sumathy

Atmospheric carbon dioxide (CO2) is considered as most significant greenhouse gas (GHG) in terms of its global warming potential and human-caused emissions. In Indian context, CSIR Fourth Paradigm Institute has established continuous monitoring of GHG stations at Hosakote nearer to Bengaluru, India. In the present study, seasonal autoregressive integrated moving average (SARIMA) model has been developed based on evaluation criteria, considering its statistical performance and ability to capture the trend and patterns in CO2 concentration at station scale which will be useful in regional studies. The present work addresses to find an optimized SARIMA configuration for the accurate prediction of the CO2 concentration and to provide a projection of the same in the near future i.e., till 2027. The proposed model is being trained with data from October 2016 to December 2020 and the performance is validated by comparing simulated results with the observed data for recent 2 years (2021 and 2022). The statistical measures for the model’s performance has been evaluated by computing the important error parameters and correlation analysis.The study demonstrated that the SARIMA model is very much suitable for forecasting the CO2 concentrations, as it provided accurate predictions by showing good agreement with the observed values.The predicted data align well with observations and capture the seasonal variations, suggesting that model developed in this study has captured the patterns and trends in the CO2 concentration data effectively and can be considered a reliable tool for predicting future CO2 concentrations. Also, the projection made till 2027 using the optimised model and it is predicted that there is slight increase in the CO2 concentrations over the Bengaluru region in the near future time i.e., 2023–2027. These inferences will contribute for understanding the CO2 concentration trends in at station scale and provide valuable insights for policymakers, researchers, and stakeholders involved in climate change mitigation and environmental planning.

Source: https://doi.org/10.1007/s00477-023-02629-4 

Implications of seismic and GNSS strain rates in Himachal, Kashmir and Ladakh

by T. S. Shrungeshwara, Bhavani Narukula, Sridevi Jade, Sapna Ghavri, Chiranjeevi G. Vivek & I. A. Parvez 

We report the present day GNSS velocities (2015–2021) and strain rates in Himachal, Kashmir and Ladakh Himalaya covering the rupture zones of the 2005 Muzaffarabad earthquake and the 1905 Kangra earthquake. Geodetic strain rates estimated from GPS velocities of about 58 sites spanning last two decades of measurements indicate a mean compression rate of − 32.5 ns/year (nanostrain/year) and dilatation of − 37.3 ns/year. Seismic strain rates are estimated using both the instrumental period (1964–2021) and historical earthquakes since 1500 AD in this region. Seismic strain rates during the instrumental period of the past 50 + years indicate a mean compression rate of − 28.1 ns/year and it slightly decreases to − 21.7 ns/year after including the historical earthquakes of the past 520 years. The Azimuth of the seismic strain tensor for the instrumental and historic periods and geodetic strain tensor is broadly consistent with orientation of major faulting in this region suggesting uniform compression over a long-time interval justifying combined analysis of the strain rate field to determine the seismic potential of the region. Composite analysis of geodetic and seismic strain rates and the associated moments estimate the accumulated strain budget of ~ 1E + 21 Nm in the past 520 years which has a potential of generating future earthquake of Mw > 8 in this segment of Northwest Himalaya.

Source: https://doi.org/10.1038/s41598-023-48997-3 

Projected changes in heat wave characteristics over India

by Neethu C & K V Ramesh

Heat waves (HW) are increasing in intensity, frequency, and duration. The projected changes in the characteristics of HW at the regional level are essential input to develop mitigation strategies to minimize social risks in densely populated regions. In this study, we examine the projected spatiotemporal changes in heat wave characteristics under different climate change scenarios using simulations of Coupled Model Intercomparison Project phase 6 (CMIP6) in seven temperature homogeneous zones of India, i.e., North West (NW), North Central (NC), West Coast (WC), East Coast (EC), Interior Peninsula (IP), Western Himalaya (WH) and North East (NE). The results show that the area of occurrence of a daily maximum temperature above 43∘C is projected to increase about 16-fold over WC, 10-fold over EC, and in other zones in the range of 1-3 fold. The warm days are projected to increase fivefold over WC and threefold over NW, EC, IP, and WH. In India, HW days are projected to increase by 7-8 days in the near future (2025-2050) and by 10-17 days in the far future (2076-2100), while under SSP585 over WH (24 days), NW (19 days), and other zones 12-15 days in the far future. EC and WC are plausible to be more vulnerable under SSP370 and SSP585, with an increase in HW intensity (>1.5∘C). The area of occurrence of long-lasting heat waves over WC is expected to have a drastic increase of more than 20-fold under all scenarios, while increasing 12-fold over IP and 8-fold over NC, EC, and WH under SSP585. The projected HW days will be more intense in the coastal zones and more frequent over WH and NW.

Improved learning by using a modified activation function of a Convolutional Neural Network in multi-spectral image classification

by Ramesh Kalidhasan Vasanthakumari, Rakesh Vasudevan Nair and Venkatesh Gowda Krishnappa

The Convolutional Neural Network (CNN) algorithm is used to classify multispectral images of labelled EuroSAT data from Sentinel-2 satellite. The main objective of this study to examine the role of newly proposed activation function, Modified Rectified Linear Unit (MReLU) in improving multi-spectral image classification accuracy. The experimental design deployed a 5-layer CNN with activation function Rectified Linear Unit (ReLU) and the MReLU having 27 learning rates ranged from 0.0001 to 0.01 with 10 % increment and 8 optimizers. ReLU has a diminishing gradient problem since it does not consider the significant information present in the negative values of the convolution layer for different spectral bands. In order to overcome this short coming, we have designed the MReLU which allows the absolute values of the small negative gradient and thereby introducing higher nonlinearity in the output. The experimental results show that the image classification accuracy of the CNN with ReLU and MReLU activation functions is definitely higher than the pre-trained ResNet-50 model which is a complex and computationally expensive architecture. The MReLU has advantage over ReLU in land use classification, particularly in the Industrial, Herbaceous Vegetation, and Permanent Crop classes where accuracies were typically low. In summary, this paper highlights the importance of deploying a suitable activation function and hyperparameter tuning in the CNN model for improving the accuracy of multispectral image classification.

Source: https://doi.org/10.1016/j.mlwa.2023.100502