Comparative evaluation of potential of optical and SAR data for the detection of human settlements using digital classification

Abstract

Delection of human settlements on satelUte imagery is an important activity as it is the first step in population estimation and analysis of settlement patterns using satellite data. Radar data is particularly suited for this purpose as settlements are generally well accentuated due to the presence and effects of horizontal and vertical structures (dihedral and trihedral reflectors). Optical and IR sensors data is also useful for this purpose since settlemenls tone, shape and assosciation/connectivity with rail/road network allows a settlement to be clearly seen on the false color composite(FCC). However, in case of optical and IR data, the detection is mostly based upon the visual interpretation due to the mixing signatures of settlements with fallow land and few wasteland categories including gulled/ ravenous land. Hence use of digital classification of optical and IR data for the detection of human settlements leads to relatively poor classification accuracies in terms of omission and commission errors. In contrast to the dependence of optical sensors on spectral signatures, backscattered energy sensed by SAR is strongly dependent on the physical properties of an object like roughness, orientation and its dielectric properties. Hence, SAR can play an important role in the detection of human settlements. However, so far use of SAR data in detection and detailed urban mapping has received less attention as compared to the use of SAR data in agricultural, forestry, geology, geomorphology and hydrological studies. This paper attempts to demonstrate the potential of higher incidence angle SAR data in detecting human settlements using digital classification technique. Study indicated that human settlement could be digitally classified with classification accuracy as high as 98.23% using SAR data alone. However due to the mixing of few categories of wasteland with fallow land, overall classification accuracy as low as 75.59% is achieved when SAR data alone is used. It is observed that synergic use of SAR and optical data acquired in the month of February proved to be the best showing highest overall classification accuracy of 89.33% with classification accuracy of human settlement to be 93.10%.