2. Thesis and Dissertations

Permanent URI for this communityhttps://idr.nitk.ac.in/handle/1/10

Browse

Filters

20242

Settings

Has files: YesSubject: search.filters.subject.Drought

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Assessment of Meteorological and Hydrological Droughts Using Stationary and Non-Stationary Indices for two Contrasting Climate Regions in India
    (National Institute of Technology Karnataka, Surathkal, 2024) SAJEEV, ARYA; KUNDAPURA, SUBRAHMANYA
    Only a few researchers have incorporated climate change in drought indices calculations. This research attempts to build non-stationary indices for assessing meteorological drought in two different climate zones of India: the arid Saurashtra and Kutch and humid-tropical Coastal Karnataka. Time and climate indices are considered as covariates to develop non-stationary models using the Generalized Additive Model in Location, Scale, and Shape (GAMLSS) for the period, 1951-2004. A comparative study has been conducted to assess the statistical performance of stationary and non-stationary models on various time scales (3-,6-,12- and 24- months). The best model is selected to conduct copula-based bivariate drought analysis. For this purpose, drought properties such as drought severity, duration, and peak are calculated. The annual and seasonal rainfall departures are also analysed, and more rainfall-deficient years are detected in Saurashtra and Kutch regions than in Coastal Karnataka. The non-stationary index performed better in capturing drought properties in statistical analysis over both the study areas at all time scales. The non-stationary drought index shows better consistency with historical drought and flood events than the stationary index. The impact of rainfall and drought on the yield of major crops in study areas is also analysed. The yield loss rate of bajra significantly correlates with Non-stationary Standardized Precipitation Index (NSPI) in Saurashtra and Kutch, whereas rice yield has no significant correlation with the index in Coastal Karnataka. Co-occurrence and joint return periods are calculated and compared with univariate return periods. A significant difference is observed between bivariate and univariate return periods, and more risk is detected in Saurashtra and Kutch than in Coastal Karnataka. Drought forecasting is crucial in water resource management and agricultural planning, particularly in regions vulnerable to water scarcity. Hence, the efficacy of various time-series forecasting models, including Autoregressive Integrated Moving Average (ARIMA), Feed-forward Neural Network (FNN), Recurrent Neural Network (RNN), as well as hybrid combinations such as ARIMA-FNN, ARIMA-RNN, FNN-ARIMA, and RNN-ARIMA, for predicting drought indices at different time scales (3, 6, 12, and 24 months) is performed in Saurashtra and Kutch. The effectiveness of the models is evaluated through Correlation Coefficient (CC), R-squared (R2), Mean ii Square Error (MSE), Mean Absolute Error (MAE), and Relative Absolute Error (RAE). FNN exhibits superior performance as a standalone model across all time scales considered, and scale 24 was the best-performing time scale with a Correlation Coefficient of 0.874 and R2 of 0.911. However, further improvements in forecast accuracy are observed at all time scales when incorporating ARIMA as a post-processing step in the hybrid FNN-ARIMA model. Notably, FNN-ARIMA emerges as the top-performing model among all evaluated approaches, demonstrating its effectiveness in capturing the complex temporal dynamics of drought phenomena. This research emphasizes the significance of hybrid forecasting techniques, especially the combination of neural networks with traditional time-series models, in enhancing drought prediction accuracy. The findings contribute to the advancement of forecasting methodologies for better-informed decision-making in water resource management and agricultural sectors, thereby aiding in mitigating the impacts of drought events on vulnerable regions. Comparative analyses of meteorological and hydrological droughts using non-stationary indices have not been explored yet. The other objective of this research is to develop non-stationary indices for assessing meteorological and hydrological droughts in the Shetrunji River basin in Saurashtra region, India, from 1971 to 2015. The statistical performance of stationary and non-stationary models has been compared across various time scales (3-,6-,12- and 24- months), and the results indicate that non-stationary models more effectively capture meteorological and hydrological drought events than stationary models. The drought and flood events detected by non-stationary indices are compared with historical episodes to assess the robustness of the indices. The results are also compared with drought events obtained from rainfall and streamflow departures. The annual and seasonal departures in rainfall and streamflow show the highest deficiency of rainfall and streamflow in 1987. The probability of different drought classes is calculated, and a higher likelihood of severe to extreme dry conditions is observed compared to very wet and extreme wet conditions in the basin. Investigation has been conducted on the impact of meteorological drought on hydrological drought and a correlation analysis between both types of droughts. A significant correlation is observed between meteorological and hydrological drought at iii all analysed time scales. Meteorological drought impacts surface water resources with a one-month lag at all time scales, with the highest response rate obtained at 6-month scale (91.13%). The study also examines the impact of drought on yield loss in Kharif (Bajra) and Rabi (Wheat) crops. Bajra and wheat yield loss rates strongly correlate with non-stationary drought indices, with a more significant effect of drought on bajra yield than wheat during major drought events. The hydrological drought analysis in the humid Netravathi River basin is also conducted using stationary and non-stationary indices. This drought analysis provides feasible results in both arid and humid regions in a changing environment. This novel dimension of drought studies provides practical insights into semi-arid regions in a changing environment. The findings can be utilized by various sectors, including drought management, agricultural planners, and policymakers, to reduce crop loss due to drought.
  • Thumbnail Image
    Item
    Assessment of Drought Indices for the Meteorological Subdivisions of India Using Machine Learning Techniques
    (National Institute of Technology Karnataka, Surathkal, 2024) KIKON, AYILOBENI; DODAMANI, B M
    Drought is a highly damaging natural event having a significant impact on the environment, agriculture, economy and public health resulting in a cascade of vulnerabilities across several sectors. Drought occurs in all climatic zones mainly because of deficit in precipitation for a prolonged period. Every year significant areas and population around the globe are affected by drought which can last anywhere between weeks to years. Understanding the numerous climatic parameters affecting the variability of rainfall and outbreaks of drought is a major scientific challenge. Due to changes in the climate and activities by people, there is a need to understand the various catastrophe causing due to drought and adopt measures to overcome and prevent the drought consequences. Drought prediction emerges as one of the crucial tools that can provide helpful information and may be used to mitigate drought impacts. Meteorological drought is a type of drought that results from inadequate amounts of rainfall in any region. The study has been conducted in the Indian region consisting of thirty-four meteorological subdivisions. The study aims to analyse the rainfall and drought indices trend using the monthly precipitation data from 1958-2017. The Mann-Kendall test has been applied to determine the trends in rainfall and drought indices. The Effective Drought Index (EDI) and Standardized Precipitation Index with 9-month and 12-month timescale are the meteorological drought indices that are assessed using monthly rainfall data. These meteorological drought indices are predicted using machine learning algorithms such as the Genetic Algorithm-Adaptive Neural Fuzzy Inference System (GA-ANFIS), Particle Swarm Optimization-Adaptive Neural Fuzzy Inference System (PSO-ANFIS), and Generalized Regression Neural Network (GRNN), and the obtained results are compared. The Mann-Kendall test results showed a clear indication that rainfall has been consistently decreasing during the study period, leading to water shortages and dry conditions. Understanding both rainfall patterns and drought trends is therefore essential for efficient planning and control of the numerous impacts of drought. The ii machine learning algorithms employed in this work show they are capable of predicting meteorological drought indices under various climatic situations. Based on performance measures such as coefficient of determination (R2), Nash-Sutcliffe Efficiency (NSE) and Normalized root mean square error (NRMSE), comparative study of the models shows that hybrid machine learning models (GA-ANFIS and PSO-ANFIS) perform better than the non-hybrid model (GRNN). Notably, it has been observed that, as the timescale for the drought index increases, it shows a better performance with more accuracy of the performance metrics. Based on the study findings, it emphasizes in assessing the rainfall and drought trend could be beneficial in understanding the drought behaviour and identify drought prone locations and develop mitigation strategies to overcome the drought impacts. Overall, this study plays a significant role in understanding the rainfall pattern and its distribution for water management and planning for future water use. Adopting hybrid machine learning algorithms for predicting of meteorological drought indices may provide a better outcome for drought assessment. Also, assessing the historical droughts provides a better understanding and management of past drought occurrences. Future research attempts could be focused on improving drought vulnerability mapping by modelling and probabilistic climate data analysis. Additionally, understanding the dynamics of drought may also be improved by investigating at how drought occurrences begin and terminate. The exploration of alternative hybrid machine learning approaches and the incorporation of additional drought indices could contribute to more robust evaluations in assessing drought conditions.