Browsing by Author "Pasari, Sumanta"

Now showing 1 - 20 of 67

Application of Empirical Orthogonal Function on Geodetic Time-Series Data
(IEEE, 2021) Pasari, Sumanta
Growing amount of geodetic data through sophisticated data collection techniques in recent times has led to substantial challenges in data analysis and interpretation. The empirical orthogonal function (EOF), commonly known as principal component analysis (PCA), is one of the dominant tools in analyzing coherent space-time dataset. The EOF method belongs to the family of factor analysis and has application in dimensionality reduction and pattern extraction, especially in the field of Geophysics, Atmospheric Sciences and Oceanography. This paper provides some basic formulation of the EOF technique with an emphasis on the step-by-step implementation to extract dominant modes from any time-series data. For instance, the EOF-based results show that the deformation pattern of the 2016, M w 7.8, Kaikoura earthquake of New Zealand is in the North-East direction. A few variations of the conventional EOF method are also discussed.
Application of Spiking Neural Networks in Renewable Energy Forecasting
(Springer, 2024-12) Pasari, Sumanta
Considering the high consumption rates of the non-renewable energy sources as well as their adverse climatic impacts, renewable energy has become a widespread topic of discussion. Among the available renewable resources, solar and wind are the highest contributors. However, the high influence of atmospheric parameters and higher cost involved in energy production prevent the widespread use of renewable energy among common public. The location identification for optimum energy production is also a crucial step for setting up future energy plants. In this regard, here we propose a novel strategy to compare prediction results in terms of loss made by traditional convolutional neural network (CNN) with that of spiking neural network (SNN). Though the SNNs are popularly used for vision related tasks, here we evaluate their efficacy in analyzing time series data of solar irradiance and wind speed. In summary, we provide a comprehensive discussion on SNN and their significance on energy forecasting.
Application of Window Sliding ARIMA in Wind Speed and Solar Irradiance Forecasting
(IEEE, 2022) Pasari, Sumanta
For better management and integration of renewable energy to the existing grid system, its accurate prediction is an inevitable requirement. For such a purpose, in literature, the statistical ARIMA model is often suggested to analyze wind speed and solar irradiance values. The present study explores window sliding ARIMA (WSARIMA) for energy prediction and reports its performance with respect to the conventional ARIMA method. The wind speed and solar irradiance data (2000–2014) from two test sites, namely Dhanora (Madhya Pradesh) and Nowlaipalle (Telangana) are used for the demonstration. It is observed that both datasets for both variables (wind speed and global horizontal irradiance, GHI) exhibit weak stationarity. The parameters for the ARIMA method are obtained through grid-search technique. Then, the proposed WSARIMA approach is applied to both datasets and results are noted. Based on the RMSE values, the WSARIMA method is found to be superior for both wind speed and GHI prediction. The involvement of sliding windows essentially incorporates seasonal fluctuations more productively in both data variables–wind speed and GHI. Therefore, the present study strongly recommends the WSARIMA model for energy prediction.
Cluster analysis of rainfall patterns in Mamminasata: Validation of climate hazards group infrared precipitation with station dataset using observational comparisons
(2024) Pasari, Sumanta
This study aims to apply the cluster method to objectively classify rainfall patterns in the Mamminasata monsoon region based on observational data and Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) data. Validation was carried out using dichotomous and numerical comparisons with in situ observational data to strengthen the validity of CHIRPS data. The results showed that observational and CHIRPS data showed three different rainfall patterns. The first pattern shows rainfall at the beginning or end of the year ranging from 80-110 mm, decreasing after the 9th day of March, then increasing again in May to mid-June. The second pattern has a slightly lower rainfall intensity than the first one. The third pattern shows rainfall that increases beyond 100 mm at the end of each year, decreasing after February and reaching its lowest point in August. Towards the end of the year, there is an increase exponentially. CHIRPS rainfall predictions tend to overestimate in the southwest coastal area of Mamminasata, while underestimations are observed in the northern urban areas and eastern mountains. The accuracy of satellite rainfall estimates varies significantly across the Mamminasata region. In general, the performance of CHIRPS rainfall estimates is better in lowland areas than in mountainous areas.
Contemporary Earthquake Hazards in the West‐Northwest Himalaya: A Statistical Perspective through Natural Times
(Geoscience, 2020-08) Pasari, Sumanta
Himalayan earthquakes have deep societal and economic impact. In this article, we implement a surrogate method of nowcasting (Rundle et al., 2016) to determine the current state of seismic hazard from large earthquakes in a dozen populous cities from India and Pakistan that belong to the west‐northwest part of Himalayan orogeny. For this, we (1) perform statistical inference of natural times, intersperse counts of small‐magnitude events between pairs of succeeding large events, based on a set of eight probability distributions; (2) compute earthquake potential score (EPS) of 14 cities from the best‐fit cumulative distribution of natural times; and (3) carry out a sensitivity testing of parameters—threshold magnitude and area of city region. Formulation of natural time (Varostos et al., 2005) based on frequency–magnitude power‐law statistics essentially avoids the daunting need of seismicity declustering in hazard estimation. A retrospective analysis of natural time counts corresponding to M≥6 events for the Indian cities provides an EPS (%) as New Delhi (56), Chandigarh (86), Dehradun (83), Jammu (99), Ludhiana (89), Moradabad (84), and Shimla (87), whereas the cities in Pakistan observe an EPS (%) as Islamabad (99), Faisalabad (88), Gujranwala (99), Lahore (89), Multan (98), Peshawar (38), and Rawalpindi (99). The estimated nowcast values that range from 38% to as high as 99% lead to a rapid yet useful ranking of cities in terms of their present progression to the regional earthquake cycle of magnitude ≥6.0 events. The analysis inevitably encourages scientists and engineers from governments and industry to join hands for better policymaking toward land‐use planning, insurance, and disaster preparation in the west‐northwest part of active Himalayan belt.
Contemporary seismic moment budget along the Nepal Himalaya derived from high-resolution InSAR and GPS velocity field
(Springer Nature, 2024-07) Pasari, Sumanta
Throughout history, several large-magnitude earthquakes have caused damage to the Himalayan region and humanity. To understand the present-day strain rate distribution and associated seismic moment budget, a high-resolution velocity field is an essential component. The present study estimates the contemporary seismic moment budget along three spatial sections over the Nepal Himalaya using the state-of-the-art high-resolution velocity field. For this, (1) we integrate 5 years of InSAR data with 77 available GPS observations over the Nepal Himalaya; (2) we then calculate strain rate distribution (dilatational and maximum shear strain rates) from this integrated velocity field, and (3) at last, we compare the geodetic moment accumulation rate estimated from strain rate tensors with the seismic moment release rate based on an earthquake database of 500 years. The results reveal that: (1) the geodetic strain rate is not homogeneous over the Nepal Himalaya, rather along the main central thrust, a relatively higher strain rate is observed; (2) the geodetic moment rate from west to east across three sections ranges from to Nm/yr, with the minimum of Nm/yr in central Nepal, whereas the seismic moment rate varies between and Nm/yr, with the minimum of Nm/yr in central Nepal; (3) the difference between geodetic and seismic moment rates from west to east provides a moment deficit rate of to Nm/yr, with the minimum of Nm/yr in central Nepal, and more importantly, (4) the inferred moment deficit rate suggests that the western and eastern Nepal have an earthquake potential of magnitude 8.5 and 8.1, respectively, whereas the central Nepal has energy budget equivalent to an 7.9 event. In summary, the present study provides spatial distribution of earthquake potential in Nepal Himalaya using the most updated high-resolution InSAR and GPS velocity field, and the findings inevitably contribute to the time-dependent earthquake hazard analysis of the study region.
The Current State of Earthquake Potential on Java Island, Indonesia
(Springer, 2021-07) Pasari, Sumanta
Between 2006 and 2020, earthquakes and other geohazards on volcano-dotted Java Island have caused about 7000 deaths, and another 1.8 million people were injured, displaced, or left homeless. In this study, we quantify the current state of earthquake hazard for 29 cities of Java, using seismicity statistics of a cumulative number of small events (natural times) between pairs of large earthquakes. This approach, known as earthquake nowcasting (Rundle et al., 2016), rests on the key concepts of elastic rebound and ergodic dynamics in earthquake fault networks. Our analysis of statistical inference shows that the estimated earthquake potential score (EPS) as on February 18, 2021 corresponding to M ≥ 6.5 events in a 300 km circular area ranges from 43 to 94%, with the scores of Jakarta (43), Surabaya (89), Bandung (69), Semarang (48), Serang (47), and Yogyakarta (59). This means, for example, that Surabaya has progressed significantly in the regional cycle of large earthquakes, whereas Yogyakarta is about midway in its seismic cycle. We observe that a change in magnitude threshold or geographic area has a consistent impact on the nowcast scores. These findings not only enable a rapid yet meaningful way to rank several cities based on their current exposure to earthquake hazards, but also empower earthquake scientists and policymakers towards better policymaking, land-use planning, earthquake insurance, disaster risk mitigation, and social awareness with respect to the seismically active island of Java.
Distribution of Earthquake Interevent Times in Northeast India and Adjoining Regions
(Springer, 2014-02) Pasari, Sumanta
This study analyzes earthquake interoccurrence times of northeast India and its vicinity from eleven probability distributions, namely exponential, Frechet, gamma, generalized exponential, inverse Gaussian, Levy, lognormal, Maxwell, Pareto, Rayleigh, and Weibull distributions. Parameters of these distributions are estimated from the method of maximum likelihood estimation, and their respective asymptotic variances as well as confidence bounds are calculated using Fisher information matrices. Three model selection criteria namely the Chi-square criterion, the maximum likelihood criterion, and the Kolmogorov–Smirnov minimum distance criterion are used to compare model suitability for the present earthquake catalog (Yadav et al. in Pure Appl Geophys 167:1331–1342, 2010). It is observed that gamma, generalized exponential, and Weibull distributions provide the best fitting, while exponential, Frechet, inverse Gaussian, and lognormal distributions provide intermediate fitting, and the rest, namely Levy, Maxwell Pareto, and Rayleigh distributions fit poorly to the present data. The conditional probabilities for a future earthquake and related conditional probability curves are presented towards the end of this article.
Earthquake cycle progression in major city regions of Taiwan through nowcasting technique
(Springer, 2025-05) Pasari, Sumanta
The complex tectonic framework of Taiwan makes it susceptible to devastating earthquakes that originate on both mapped faults, and at times, on unmapped faults. The unmapped faults especially highlight the limitation of conventional fault–based hazard assessment methods, emphasizing the need for alternative approaches. In this context, we implement a surrogate area–based earthquake nowcasting technique to assess the seismic cycle progression in 10 densely populated cities across Taiwan. We utilize the notion of natural times, the inter–event counts of small earthquakes between successive large events, to calculate the Earthquake Potential Score (EPS) for each city region. To derive natural time statistics, we analyze eight reference probability models, including exponential distribution and its variants, exponentiated group of distributions, and heavy–tailed distributions. Statistical inference of 114 observed natural times shows that the exponentiated exponential distribution provides the best fit. As of April 24, 2025, the EPS values (%) for M 6.0 earthquakes in the 10 cities range from 53% to 69%, with the following values: Taipei (69%), Hsinchu (68%), Keelung (67%), Hualien (59%), Nantou (58%), Taitung (57%), Chiayi (56%), Pingtung (55%), Tainan (54%), and Kaohsiung (53%). These EPS values indicate the progression in current earthquake cycle toward a M 6.0 earthquake in the corresponding city region. Moreover, there is a consistency in the nowcast scores despite some variations in threshold magnitudes and city regions. The studied approach and results therein offer valuable insights to decision makers to enhance earthquake preparedness and risk management across Taiwan.
Earthquake Forecasting in the Himalayas Artificial Neural Networks
(Springer, 2022) Pasari, Sumanta
Earthquake is a natural phenomenon that causes huge loss in both life and property. Improvement of seismic hazard assessment requires integrated techniques such as geodetic, stochastic, and machine learning models. Forecasting of the time of the event, magnitude, and location of the epicenter of future events has been the major focus of several efforts in recent years. Many methods have been proposed to forecast the occurrence of earthquakes like statistical methods and other modeling approaches. Such methods are based on either the study of electric or magnetic signals or microseismicity patterns in which changes are experienced due to an upcoming event. In this study, our aim is to forecast earthquakes using neural networks, based on some seismicity indicators which capture the intrinsic information of the earthquake events. For this, an effective neural network architecture is created with different deep learning optimization algorithms and the results showed that the eight seismicity indicators have essentially captured most of the information of earthquake events. It is observed that neural networks are an effective tool for forecasting earthquakes as the neural networks well capture the nonlinearity and heterogeneity of inherent mechanisms with appropriate weights. The proposed network provides 90% accuracy and an F1-score of 0.89. It is hoped that this study shall provide useful information to the industry, academia, and government agencies to develop new standards of monitoring and mitigation measures of earthquake hazard.
Earthquake forecasting using artificial neural networks
(ISPRS, 2018) Pasari, Sumanta
Earthquake is one of the most devastating natural calamities that takes thousands of lives and leaves millions more homeless and deprives them of the basic necessities. Earthquake forecasting can minimize the death count and economic loss encountered by the affected region to a great extent. This study presents an earthquake forecasting system by using Artificial Neural Networks (ANN). Two different techniques are used with the first focusing on the accuracy evaluation of multilayer perceptron using different inputs and different set of hyper-parameters. The limitation of earthquake data in the first experiment led us to explore another technique, known as nowcasting of earthquakes. The nowcasting technique determines the current progression of earthquake cycle of higher magnitude earthquakes by taking into account the number of smaller earthquake events in the same region. To implement the nowcasting method, a Long Short Term Memory (LSTM) neural network architecture is considered because such networks are one of the most recent and promising developments in the time-series analysis. Results of different experiments are discussed along with their consequences.
Earthquake interevent time distribution in Kachchh, Northwestern India
(Springer, 2015-08) Pasari, Sumanta
Statistical properties of earthquake interevent times have long been the topic of interest to seismologists and earthquake professionals, mainly for hazard-related concerns. In this paper, we present a comprehensive study on the temporal statistics of earthquake interoccurrence times of the seismically active Kachchh peninsula (western India) from thirteen probability distributions. Those distributions are exponential, gamma, lognormal, Weibull, Levy, Maxwell, Pareto, Rayleigh, inverse Gaussian (Brownian passage time), inverse Weibull (Frechet), exponentiated exponential, exponentiated Rayleigh (Burr type X), and exponentiated Weibull distributions. Statistical inferences of the scale and shape parameters of these distributions are discussed from the maximum likelihood estimations and the Fisher information matrices. The latter are used as a surrogate tool to appraise the parametric uncertainty in the estimation process. The results were found on the basis of two goodness-of-fit tests: the maximum likelihood criterion with its modification to Akaike information criterion (AIC) and the Kolmogorov-Smirnov (K-S) minimum distance criterion. These results reveal that (i) the exponential model provides the best fit, (ii) the gamma, lognormal, Weibull, inverse Gaussian, exponentiated exponential, exponentiated Rayleigh, and exponentiated Weibull models provide an intermediate fit, and (iii) the rest, namely Levy, Maxwell, Pareto, Rayleigh, and inverse Weibull, fit poorly to the earthquake catalog of Kachchh and its adjacent regions. This study also analyzes the present-day seismicity in terms of the estimated recurrence interval and conditional probability curves (hazard curves). The estimated cumulative probability and the conditional probability of a magnitude 5.0 or higher event reach 0.8–0.9 by 2027–2036 and 2034–2043, respectively. These values have significant implications in a variety of practical applications including earthquake insurance, seismic zonation, location identification of lifeline structures, and revision of building codes.
Earthquake Magnitude Prediction in Chile Using Neural Network
(IEEE, 2022) Pasari, Sumanta
In this study, we implement an earthquake magnitude prediction model using a neural network for a test region in Chile. For this, the epicenter of earthquake is located on a mesh with dimensions of 1°×1°. We adopt a zonation scheme originally proposed by Reyes and Cardenas [1]. The scheme uses increments in b−value and other input parameters to incorporate G-R linear relation and Bath’s law. The model enables the prediction of the maximum magnitude for a given cell within the next five days. Common seismological parameters are used for the performance evaluation of the model. Results show satisfactory performance of the proposed model in comparison to other existing models.
Earthquake Prediction Using Deep Neural Networks
(IEEE, 2022) Pasari, Sumanta
Reliable prediction of earthquakes has numerous societal and engineering benefits. In recent years, the exponentially rising volume of seismic data has led to the development of several automatic earthquake detection algorithms through machine learning approaches. In this study, we propose a fully functional and efficient earthquake detector cum forecaster based on deep neural networks of long-short-term memory (LSTM) units. The model captures inherent temporal characteristics of earthquake data. For illustration, we consider an earthquake catalog from the Himalaya and its neighboring regions. The proposed LSTM model shows satisfactory performance for small to medium-sized earthquakes. We also implement a baseline artificial neural network (ANN) model to perform a suitable comparison. It is observed that both ANN and LSTM models fail to produce desired result for large events.
Earthquake prediction using long short term memory on spatio-temporally segmented data
(IEEE, 2023) Pasari, Sumanta
This paper describes a machine learning model for predicting earthquakes on the basis of past earthquake data. In particular, this study uses the Long Short Term Memory (LSTM) model, a neural network model designed to operate on time-series data with long-term dependencies. Here, the instrumental earthquake data is considered from three selected locations in Indonesia. First, the dataset is pre-processed by segmenting it into time intervals and space grids. The multi-dimensional time-series data is then fed into the network to output the probability of an earthquake in the next interval. This method was originally introduced by Wang et al. [1] and achieved an accuracy close to 85% on a dataset from Mainland China (1966–2016). To the best of our knowledge, no subsequent works have attempted to reproduce their results on different datasets, or introduce enhancements. This research work has implemented the same model on three different datasets. Further, the softmax activation function is replaced with the sigmoid activation function. This ensures that the probability values of earthquakes occurring in the segmented grids are independent of each other and are not rendered mutually exhaustive or exclusive events. Finally, a failure mode of this model is mentioned by showing that it performs poorly to predict large earthquakes.
Earthquake Prediction using Long Short Term Memory on Spatio-Temporally Segmented Data
(IEEE, 2023) Pasari, Sumanta
This paper describes a machine learning model for predicting earthquakes on the basis of past earthquake data. In particular, this study uses the Long Short Term Memory (LSTM) model, a neural network model designed to operate on time-series data with long-term dependencies. Here, the instrumental earthquake data is considered from three selected locations in Indonesia. First, the dataset is pre-processed by segmenting it into time intervals and space grids. The multi-dimensional time-series data is then fed into the network to output the probability of an earthquake in the next interval. This method was originally introduced by Wang et al. [1] and achieved an accuracy close to 85% on a dataset from Mainland China (1966–2016). To the best of our knowledge, no subsequent works have attempted to reproduce their results on different datasets, or introduce enhancements. This research work has implemented the same model on three different datasets. Further, the softmax activation function is replaced with the sigmoid activation function. This ensures that the probability values of earthquakes occurring in the segmented grids are independent of each other and are not rendered mutually exhaustive or exclusive events. Finally, a failure mode of this model is mentioned by showing that it performs poorly to predict large earthquakes.
Effectiveness of PNL technique in disaster damage assessment: evidence from selective case studies
(IOP, 2022-06) Pasari, Sumanta
Natural disasters often cause large scale infrastructural damage and disruption of services. In such cases, a rapid damage assessment technique might prove beneficial to assist the disaster management efforts. Percent of normal light (PNL) can be one such technique. In this study, we test the effectiveness of PNL technique in rapid damage assessment from four selected case studies, namely the 2014 cyclone Hudhud, 2015 Gorkha (Nepal) earthquake, 2016 Central Italy earthquake, and the 2018 flood in Kerala. The dataset used has been taken from the Visible Infrared Imaging Radiometer Suite–Day/Night Band (VIIRS-DNB) scan data from the Joint Polar Satellite System (JPSS-1). The change in radiance values from the VIIRS-DNB dataset enables PNL computation to map the disaster affected regions. The results depict that PNL can be a viable alternative for rapid damage assessment.
Efficacy and application of the window-sliding ARIMA for daily and weekly wind speed forecasting
(AIP, 2022-10) Pasari, Sumanta
Accurate forecasting of renewable energy resources has a deep societal and environmental impact. In this work, we investigate the efficacy and applicability of the Window-Sliding ARIMA (WS-ARIMA) method for daily and weekly forecasting of wind speed. The WS-ARIMA technique with a fixed or variable window length belongs to the class of adaptive models. Particularly, the sliding windows of fixed length are popular in the areas of finance, energy, and traffic management, where the dataset of necessity exhibits a seasonal pattern. To carry out the proposed analysis, the following processes were done: (1) we first perform a stationarity test on the wind speed data and observe weak stationarity; (2) we then apply a grid search method to obtain the optimal parameters of the ARIMA model; (3) we implement the WS-ARIMA method for both daily and weekly wind speed data and compare the results with the conventional ARIMA model, and (4) finally, we perform a residual analysis as a post processing step to examine any systematic bias in the implemented models. The experimental results based on 15 years (2000–2014) of daily and weekly wind speed data collected at four different locations in India reveal that the WS-ARIMA method consistently outperforms the conventional ARIMA method. The inclusion of window sliding in ARIMA has resulted in the overall RMSE reduction up to 75% in daily wind speed data and 50% in the weekly data. Therefore, we recommend the WS-ARIMA model as one of the potential techniques in wind speed forecasting at daily and weekly time horizons.
Estimation of Current Earthquake Hazard Through Nowcasting Method
(Springer, 2022) Pasari, Sumanta
In several tectonically active regions of the world, large magnitude earthquakes on fault systems are observed to occur in near-repetitive cycles as a consequence of stress accumulation and moment release. Since absolute measurements of stress–strain is unavailable through direct observations at all regions of interest, the area-based nowcasting method based on earthquake data is a potential alternative to estimate the uncertain current state of earthquake hazard in a defined region. Using the concept of natural-time counts, the nowcasting result comprises time-dependent earthquake potential score—a numerical quantification of earthquake-cycle progression since the last major event in the region. The nowcast score may be linked to the instantaneous risk of large events. This paper summarizes some basic formulation and key concepts of earthquake nowcasting with a demonstration of its applicability in disaster preparation and risk estimation. A case study from Java, Indonesia, is considered for illustration.
Exploration of Solar Irradiance in Thar Desert Using Time Series Model
(Springer, 2023) Pasari, Sumanta
The present study concentrates on the exploration of solar irradiance in the Thar desert at eight selected locations, including Bhadla and Dhirubhai Ambani solar parks. For this, we first perform daily, weekly, and monthly solar irradiance prediction using five time-series models, namely AR, MA, ARMA, ARIMA, and seasonal ARIMA (SARIMA). The dataset of necessity includes hourly solar irradiance values of 21 yr (2001–2021) from NASA’s power project. As these time series models turn out to be inadequate to capture seasonality across temporal resolution, we additionally implement the window sliding ARIMA (WSARIMA) and LSTM to incorporate possible nonlinearity and seasonality in the dataset. Based on the three standard indicators, namely RMSE, MAPE, and MAE, we observe that LSTM outperforms other models at daily and weekly time resolution, whereas ARMA turns out to be the best on monthly dataset. The emanated results suggest that all locations reveal a high potential for harnessing solar power. The present analysis, therefore, enables solar irradiance exploration in the Thar desert through different time series models.