• Sunita Srivastava

      Articles written in Journal of Earth System Science

    • A Hidden Markov Model for avalanche forecasting on Chowkibal–Tangdhar road axis in Indian Himalayas

      Jagdish Chandra Joshi Sunita Srivastava

      More Details Abstract Fulltext PDF

      A numerical avalanche prediction scheme using Hidden Markov Model (HMM) has been developed for Chowkibal–Tangdhar road axis in J&K, India. The model forecast is in the form of different levels of avalanche danger (no, low, medium, and high) with a lead time of two days. Snow and meteorological data (maximum temperature, minimum temperature, fresh snow, fresh snow duration, standing snow) of past 12 winters (1992–2008) have been used to derive the model input variables (average temperature, fresh snow in 24 hrs, snow fall intensity, standing snow, Snow Temperature Index (STI) of the top layer, and STI of buried layer). As in HMMs, there are two sequences: a state sequence and a state dependent observation sequence; in the present model, different levels of avalanche danger are considered as different states of the model and Avalanche Activity Index (AAI) of a day, derived from the model input variables, as an observation. Validation of the model with independent data of two winters (2008–2009, 2009–2010) gives 80% accuracy for both day-1 and day-2. Comparison of various forecasting quality measures and Heidke Skill Score of the HMM and the NN model indicate better forecasting skill of the HMM.

    • Optimisation of Hidden Markov Model using Baum–Welch algorithm for prediction of maximum and minimum temperature over Indian Himalaya

      J C Joshi Tankeshwar Kumar Sunita Srivastava Divya Sachdeva

      More Details Abstract Fulltext PDF

      Maximum and minimum temperatures are used in avalanche forecasting models for snow avalanche hazard mitigation over Himalaya. The present work is a part of development of Hidden Markov Model (HMM) based avalanche forecasting system for Pir-Panjal and Great Himalayan mountain ranges of the Himalaya. In this work, HMMs have been developed for forecasting of maximum and minimum temperatures for Kanzalwan in Pir-Panjal range and Drass in Great Himalayan range with a lead time of two days. The HMMs have been developed using meteorological variables collected from these stations during the past 20 winters from 1992 to 2012. The meteorological variables have been used to define observations and states of the models and to compute model parameters (initial state, state transition and observation probabilities). The model parameters have been used in the Forward and the Viterbi algorithms to generate temperature forecasts. To improve the model forecasts, the model parameters have been optimised using Baum–Welch algorithm. The models have been compared with persistence forecast by root mean square errors (RMSE) analysis using independent data of two winters (2012–13, 2013–14). The HMM for maximum temperature has shown a 4–12% and 17–19% improvement in the forecast over persistence forecast, for day-1 and day-2, respectively. For minimum temperature, it has shown 6–38% and 5–12% improvement for day-1 and day-2, respectively.

    • Hidden Markov Model for quantitative prediction of snowfall and analysis of hazardous snowfall events over Indian Himalaya

      J C Joshi K Tankeshwar Sunita Srivastava

      More Details Abstract Fulltext PDF

      A Hidden Markov Model (HMM) has been developed for prediction of quantitative snowfall in Pir-Panjal and Great Himalayan mountain ranges of Indian Himalaya. The model predicts snowfall for two days in advance using daily recorded nine meteorological variables of past 20 winters from 1992–2012. There are six observations and six states of the model. The most probable observation and state sequence has been computed using Forward and Viterbi algorithms, respectively. Baum–Welch algorithm has been used for optimizing the model parameters. The model has been validated for two winters (2012–2013 and 2013–2014) by computing root mean square error (RMSE), accuracy measures such as percent correct (PC), critical success index (CSI) and Heidke skill score (HSS). The RMSE of the model has also been calculated using leave-one-out cross-validation method. Snowfall predicted by the model during hazardous snowfall events in different parts of the Himalaya matches well with the observed one. The HSS of the model for all the stations implies that the optimized model has better forecasting skill than random forecast for both the days. The RMSE of the optimized model has also been found smaller than the persistence forecast and standard deviation for both the days.

  • Journal of Earth System Science | News

© 2017-2019 Indian Academy of Sciences, Bengaluru.