Abstract Natural calamities like drought cause misery to human lives as well as environment in a variety of ways. The huge adverse consequences and globally predicted climate change accentuate the importance of an effective drought assessment and management system. Lack of universality of available drought indices emphasize the need of an automated system that works globally. Internet of Things (IoT) is highly appropriated for ubiquitous monitoring, acquisition and evaluation of causing parameters for reliable prediction of drought situations. The framework includes dimensionality reduction algorithm at Fog layer through which only variance rich data passes. This data is evaluated at the Cloud layer to determine the drought severity level employing Artificial Neural Network (ANN), ANN optimized with Genetic Algorithm (ANN-GA), DNN (Deep Neural Network) and their performance is compared. Support Vector Regression (SVR) method predicts the drought conditions for three different climate blocks and three different time frames. Experimentation reveals that DNN outperformed with high accuracy, sensitivity, specificity, precision, f-measure with values 95.361%, 91.584%, 96.834%, 91.857%, 91.72% respectively and with effective execution time.
Paper
Full text
Deep learning based drought assessment and prediction framework
Semantic Scholar · Environmental Science · 2020
Abstract
Abstract Natural calamities like drought cause misery to human lives as well as environment in a variety of ways. The huge adverse consequences and globally predicted climate change accentuate the importance of an effective drought assessment and management system. Lack of universality of available drought indices emphasize the need of an automated system that works globally. Internet of Things (IoT) is highly appropriated for ubiquitous monitoring, acquisition and evaluation of causing parameters for reliable prediction of drought situations. The framework includes dimensionality reduction algorithm at Fog layer through which only variance rich data passes. This data is evaluated at the Cloud layer to determine the drought severity level employing Artificial Neural Network (ANN), ANN optimized with Genetic Algorithm (ANN-GA), DNN (Deep Neural Network) and their performance is compared. Support Vector Regression (SVR) method predicts the drought conditions for three different climate blocks and three different time frames. Experimentation reveals that DNN outperformed with high accuracy, sensitivity, specificity, precision, f-measure with values 95.361%, 91.584%, 96.834%, 91.857%, 91.72% respectively and with effective execution time.