Object-Based Image Analysis for Land Cover Mapping in an Urbanized Watershed

Abstract

Land cover maps are one of the thematic information sought among the remote sensing community and beyond for various applications. We intend to generate the land cover map of a watershed that encompasses Addis Ababa city to evaluate its contribution for flood hazard occurrence and quantification of flood risks. The study used Object Based Image Analysis (OBIA) to derive five land cover classes using Landsat 8 satellite images acquired in 25 March 2016. The satellite image was undergone to pre-processing, processing and post-processing steps involving radiometric and atmospheric corrections, fuzzy logic classification, and accuracy assessment. The accuracy assessment was made by comparing the reference data collected from Google earth and the classified image using various accuracy assessment measurements. The land cover map was generated with an overall accuracy of 94%, and with 92% Kappa Index of agreement (KIA). Land cover class specific accuracy measurements such as producers, users, Helden, short and KIA per class were also generated with values ranging between 79%-100%. The higher level of accuracy obtained is attributed to the image discontinuity and steeper gradients of the dominantly urbanized landscape which makes them easier to extract, and the robustness of the OBIA.