Effectiveness of DOS (Dark-Object Subtraction) method and water index techniques to map wetlands in a rapidly urbanising megacity with Landsat 8 data

The objectives of this work were to examine the applicability of the Dark-Object Subtraction (DOS) atmospheric correction method and water-based index techniques to map wetlands in Dhaka megacity using Landsat 8 data. With the use of both raw data and DOScorrected imagery, the analysis revealed that DOScorrected images performed better in discriminating wetland areas. Furthermore, the Modified Normalised Water Index (MNDWI) was the most superior technique whilst the Normalised Difference Water Index (NDWI) was the least suitable in identifying the spatial locations of wetlands in a rapidly urbanising environment such as Dhaka. Copyright © by the paper’s authors. Copying permitted only for private and academic purposes. In: B. Veenendaal and A. Kealy (Eds.): Research@Locate'15, Brisbane, Australia, 10-12 March 2015, published at http://ceur-ws.org Introduction Wetlands comprise roughly 6–9 percent of the Earth’s surface (Zedler and Kercher, 2005). The role of wetlands in maintaining environmental quality is well recognised (Ozesmi and Bauer, 2002), and includes the storage of global terrestrial carbon (Mitsch and Gosselink, 2007). In addition, their influence on many aspects of ecology, economy and human welfare has been well documented (Klemas, 2011; Ma et al. 2007). Furthermore, wetlands act as an oasis in an urban area which is important in the reduction of surrounding surface air temperature (Sun and Chen, 2012). Changes in the distribution of wetlands either by natural factors or anthropogenic activities could significantly affect the ecosystem services (Barducci et al. 2009) mentioned above. Although they are an important environmental resource, they are heavily abused due to a lack of understanding (Smardon, 2009), particularly in developing countries. Accurate mapping and precise area statistics are therefore of paramount importance in the prevention and management of wetlands and related ecosystem services (Klemas, 2011). Satellite remote sensing data have extensively been used to delineate wetlands across the world with a wide range of techniques, including a per-pixel classifier (e.g. supervised classification), semi-automated (e.g. image segmentation) method and spectral water index (e.g. normalised difference water index) (Mwita et al. 2013; Sun et al. 2012; Song et al. 2012; Jiang et al. 2012; Lu et al. 2011; Zhou et al. 2010; Islam et al. 2008; Shanmugam et al. 2006; Lira, 2006; Ouma and Tateishi, 2006). Among these techniques, water-based indices including single-band density slicing (Knight et al. 2009; Frazier and Page, 2000) techniques and band ratios comprising of two reflective bands, are found to perform better in discriminating water features such as wetlands from non-water features (Sun et al. 2012; Xu, 2006). However, deciding the optimal threshold value in isolating wetlands from the surrounding urban and land features remains an inordinate challenge (Zhang et al. 2009). In addition to single-band and band ratio techniques, new automated water-based indices such as the Automated Water Extraction Index (AWEI) has been developed and tested with several sensors in different areas however its applicability to distinguishing wetland areas within a rapidly urbanizing environments has only undergone minor testing. Research@Locate '15 100 Various methods have been developed to correct atmospheric influence on remote sensing data. Whilst absolute atmospheric correction methods require in-situ information, image base techniques known as relative scattering correction, on the other hand, are handy and relatively easier to subdue scattering problem in an image. A study by Song et al. (2000) suggests that atmospheric correction of remote sensing data is always not necessary and depends on the nature of the work. However, some researchers strongly favour reducing the effects of atmospheric scattering caused by light scattering (Weng, 2012), particularly in the visible region of the electromagnetic spectrum for all studies involving remotely sensed imagery. The Dark Object Subtraction (DOS) method is an image-based technique to cancel out the haze component caused by additive scattering from remote sensing data (Chavez Jr, 1988). This method is found to be data dependent and well accepted by the geospatial community to correct light scattering in remote sensing data (Song et al. 2000). However, the DOS method has been developed for early generation Landsat sensors (e.g. TM) and may not work effectively for the new generation data such as Landsat 8 which started delivering data from early 2013. It may be noted that band composition according to electromagnetic energy of Landsat 8 differs from its predecessor Landsat sensors (e.g. TM/ETM+), hence little is known about the effectiveness of DOS method with respect to the scattering correction of the new Landsat 8 sensor. Since this new generation Landsat is expected to deliver data consistently over the next several years, research on the application of the DOS method in analysing Landsat 8 data deserves further examination. Dhaka megacity, the capital of the people’s republic of Bangladesh, has evolved into a rapidly urbanizing mega city as it attempts to accommodate large numbers of people migrating from rural areas since the independence of the country in 1971 (Dewan and Corner, 2014). With a total population of more than 14 million people according to 2011 population and housing census, the city is facing severe environmental degradation, including the rapid decline in natural wetlands due to unplanned urban expansion and related socioeconomic development (Dewan and Yamaguchi, 2009). Studies for example, show that the rapid conversion of wetlands to urban areas aggravated flooding during the monsoon seanson(Dewan et al. 2012), thus increasing vulnerability of urban dwellers to severe floods. Although a number of studies on the mapping of wetlands in Dhaka have been conducted (Islam, 2009; Sultana et al. 2009), they all are based on a smaller study area. Moreover, none of the studies considered advanced techniques to accurately map wetlands in the megacity. Hence, this paper is expected to contribute significantly to the existing knowledge-base on the spatial locations of wetlands in the Dhaka Metropolitan Development Plan (DMDP) area which is a recently developed planning unit by the policy makers, enforcing local organisations to preserve remaining wetland ecosystems. Considering the above facts, the objectives of the work are: (i) to understand the effect of DOS correction technique on Landsat 8 in estimating wetlands; and (ii) to analyse the suitability of water-based index in assessing the spatial locations of wetlands in a rapidly urbanising megacity. Data and Image Pre-Processing The imagery was sourced from the USGS Earth Explorer web service, with two images having a path-row of 137-43 and 137-44 required as the study area is split between them. The two images were mosaicked with the resultant image clipped to the study area. As this study examines the impact of the DOS algorithm (Chavez Jr., 1988) on Landsat 8 imagery, a copy of the raw imagery was made and underwent DOS correction. Song et al (2001) describes how the DOS algorithm assumes the existence of ‘dark objects’, which are pixels having zero to very small reflectance numbers, within a Landsat scene. Therefore the minimum DN (digital number) value in the Figure 1 Location map of the DMDP area (source: Google Earth) Research@Locate '15 101 histogram is considered to be the effect of atmospheric scattering and is subtracted from all pixels within the scene, thus creating ‘dark objects’ with a DN value of zero. Elexclis ENVI software was used for performing the DOS correction as it is an automated process which produces a corrected multispectral image. The two corrected images were then mosaicked and co-registered to the raw image was returning a RMSE of < 0.5 pixels which ensured that pixels from both images were positioned almost perfectly on top of each other.