Thumbnail Image

Theses

Browse

20191
Reset filters
Subject: Soil and Water Engineering × Author: Amit Kumar × End date: 2019 × Type: Thesis × Thesis Type: M.Tech. ×
Reset filters

Search Results

Now showing 1 - 1 of 1
  • Thumbnail Image
    ThesisItemOpen Access
    Comparison of MLP-ANN and W-ANN for SPI forecasting to assess meteorological drought
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-07) Amit Kumar; Singh, Pravin Vikram
    The accurate assessment of drought is an essential component for effective water resource management planning to mitigate adverse consequences of drought. The Standardized Precipitation Index (SPI) is a widely used index to characterize meteorological drought on a varying time scale. Information about Standardized Precipitation Index (SPI) at a place is vital for the assessment of drought. In this study, an approach to forecast Standardized Precipitation Index (SPI) has been attempted to assess meteorological drought in drought prone area of the country at different time scales. This approach involved application of Multi-Layer Perceptron Artificial Neural Networks (MLP-ANN) and Wavelet Artificial Neural Network (W-ANN) to generate Standardized Precipitation Index values for different scales and denoted as, SPI-1, SPI-3, SPI-6, SPI-9, SPI-12 and SPI-24. To generate SPI values using these two models, the data set of Prabhani district in the state of Maharashtra was considered. The total data set of calculated values of SPI during 1971 to 2014 at various time scales was divided into three sets; (i) a training set, consisting of first 36 years data from January, 1971 to December, 2006; and (ii) a testing set, consisting of 4 years data from January, 2007 to December, 2010; and (ⅲ) a validation set, consisting of remaining 4 years data from January, 2011 to December 2014 for both the approaches. The SPI values at previous six-month lag were used to forecast current month SPI values and gamma test was used to decide the best combination of inputs for SPI forecasting. Both MLP-ANN and W-ANN models trained with the Levenberg Marquardt (LM) back propagation algorithm were developed using single hidden layer. The Root Mean Square Error (RMSE), Correlation Coefficient (r) and Coefficient of Efficiency (CE) statistical indices were adopted to evaluate the performance of these models. The SPI values generated by using best developed MLP-ANN and W-ANN models were compared with calculated values of SPI. The forecasted results indicate that for SPI-1, the performance of both MLP-ANN and WANN models was not satisfactory, however, MLP-ANN based model performed better than W-ANN model. For SPI-3, 6 and 9, the performance of W-ANN model was found to be better than MLP-ANN based model. In case of SPI-12 hand SPI-24, both the models were found to be performing satisfactorily, however, WANN model has a little bit edge over MLP-ANN. Interestingly, it was observed that the performance of both these models was found to be improving with increasing SPI time scale.