Extraction of impervious features from spectral indices using artificial neural network

Abstract

An urban area comprises a complex mix of diverse land cover types and materials. Urban ecology and environment is significantly influenced by the proportion of impervious cover that is increasing considerably with time due to the continuous influx of people into urban areas. Therefore, it is of vital importance to determine the spatiotemporal pattern and magnitude of urbanization. In the present study, we have employed a supervised backpropagation neural network in order to extract the impervious features using five spectral indices, such as one vegetation index—Soil-Adjusted Vegetation Index (SAVI), one water index—Modified Normalized Water Index (MNDWI), and three urban indices—Normalized Difference Built-up Index (NDBI), Built-up Index (BUI), and Index-Based Built-up Index (IBI). The study has been performed using Landsat Thematic Mapper data of November, 2011, of the rapidly urbanizing city of Ranchi, capital of Jharkhand state, India. Using different combinations of these spectral indices while keeping SAVI and MNDWI constant, seven composite images were built, and from each of these composites, impervious features were classified and its accuracy assessed with reference to high-resolution images provided by Microsoft Bing Imagery and adequate ground truthing. It was observed that along with SAVI and MNDWI, whenever IBI was used in any combination, it decreased the classification efficiency. On the other hand, NDBI and BUI, individually or when used together, discriminated the impervious features from the others with high accuracy with the combination of SAVI, MNDWI, and BUI achieving the highest accuracy of 90.14 %.