Urbanization and its Impact on Hydrologic and Environmental Parameters of Urban Indian Catchments

Abstract

The principal factor responsible for the hydrologic and climatic alterations in a tropical country like India is associated with every increasing proportionate area under impervious surface. Determination of imperviousness, which is defined by total impervious area (TIA) and effective impervious area (EIA), is mandatory for hydrological modelling of water quantity and quality in urban areas. In this study, a multilayer deep learning model Convolutional Neural it automatically detects the spatial features from image input and is gaining attention for their capability of achieving better classification accuracy. With estimation of TIA, its impact on different environmental parameters (Land surface temperature, Pollution (air/water/solid waste) parameters and Diurnal temperature range (DTR)) and hydrological parameters (groundwater recharge, rainfall extreme indices and surface runoff) were also studied. A more realistic automated method is suggested in this study to determine EIA by integrating the remote sensing data, the digital format of the drainage network, and a digital elevation model (DEM) of the study areas. A graphical user interface (GUI) called EIA estimator is developed for automatic creation of EIA maps. An effort is made to derive a relationship between TIA and EIA and it is generalized based on some common parameters like TIA growth, drainage density, stream order, flow length and topographic wetness index (TWI) of study areas. Based on these different categories several power and linear relationships were obtained for easily measurable TIA and hydraulically relevant EIA in urban catchments of India. There were no such relationships available in the literature for an urban Indian catchment and it would aid planners and decision makers with quick initial estimate for surface water quantity and quality problems. This research concludes with implementation of different deep learning models (RNN, LSTM and GRU) for time series forecasting of environmental and hydrological datasets univariate time series prediction and from the error analysis it was observed that LSTM performed very well than other two model algorithms. On the other hand, a single model could not produce the most accurate results and might be vulnerable to mistakes like bias and volatility. To cut down on these inaccuracies and enhance predictions, several models were merged into a single one. This technique of training different machine learning models and integrating their outputs is known as ensemble learning. In this study, the simplest ensemble technique i.e., averaging is implemented. The final prediction is made using the average of prediction outputs from Vanilla LSTM, Stacked LSTM and Bidirectional LSTM were averaged parameters for the year 2025. These deep learning models were first implemented with DTR Network (CNN) is implemented for estimating TIA of urban Indian cities during 1995-2021 as and denoted as Ensembled LSTM (ELSTM). ELSTM was applied to other hydrological and growing urbanization as later is unescapable.