In recent decades, different remote sensing methods such as aerial photography and satellite imaging have become widely available as a source of data for mapping and monitoring land use and land cover. Due to rapid urbanization, land cover around many of the world’s urban areas changes more quickly than ever before. The data obtained from remote sensing platforms provide up-to-date information and a general view of landscape characteristics and changes in urban areas [1
Bare land, whether seasonal or permanently bare, plays an important role in many ecosystems, for instance, as a source of Aeolian deposits during sandstorms. However, the differentiation of base land from urban areas in arid and semi-arid environments has often proven difficult. There are different methods and techniques that have been used to identify urban areas from other land cover types, for example, classification into urban, rural and other classes. Night-light satellite data can also be used to distinguish between urban and non-urban areas [2
]. The Normalized Difference Built-up Index (NDBI) was proposed in 2003 as a new method which automatically maps built-up areas. This method was first applied in practice to extract data for a built-up area of Nanjing city in eastern China [1
]. Although the NDBI index was able to distinguish between built-up and vegetated or green and wet surroundings for the city of Nanjing, it was not successful in distinguishing between built-up and other land cover such as bare and dry soil that surround the city, due to the overlapping spectral reflectance for these land cover types [3
]. Although several techniques to observe land use and land cover have been developed, an evaluation of the performance and accuracy of these techniques in dry climate regions is rarely available. We have conducted several tests using NDBI and NDBaI derived from Landsat TM, ETM+ and OLI for LULC of Erbil. The NDBI results produced a high value for bare soil; in contrast, NDBaI produced a high value for built-up. Generally, NDBI is more efficient in places where the NDVI value is greater than 0 [1
]. Therefore, this method is not suitable for cities located in dry climates (Table 1
Spectral indices are a good approach to distinguish land cover types [4
]. Chen [5
] proposed a bare soil index (BI) primarily for bare land extraction from satellite data. The normalized difference soil index (NDSI) has been used to detect signature changes in un-mixing coastal marsh from satellite images [6
]. A normalized difference bareness index (NDBaI) was proposed by Chen et al. [4
] to distinguish bare land from other land use classes using Landsat data. As the output of this method showed a higher NDBaI value of built-up areas than bare land, this index appears not to be appropriate for use in cities in semi-arid environments. As-syakur et al. [7
] proposed an enhanced built-up and bareness index (EBBI) and provided a case study of Denpasar city in Bali, Indonesia, with 90.5% overall accuracy of bare land detection [8
]. A normalized difference bare land index (NBLI) and unsupervised classification was used by Li et al. [8
] to automatically map bare land from Landsat images. Bare soil indices, in particular most commonly NDBaI, are frequently used in humid regions with high accuracy while bare soil in dry areas requires improved indices.
Impervious surface materials and bare soil are both parts of the land surface [9
]. Sub-pixel and layered classification was proposed by Ji and Jensen [10
] for coastal and urban environments. This method was later modified by Flanagan and Civco [11
] by using artificial neutral networks to retrieve the impervious surface fraction. Lu and Weng [12
] developed a new method for urban land use classification based on the combined use of impervious surface and population density. The Urban Index (UI) [13
] performs better in the identification of urban areas but is unable to distinguish bare land and built-up areas accurately [14
]. Wu and Murry [15
] employed a linear mixture pixel analysis (LMPA) to extract impervious distribution. Similarly, Xu [16
] proposed a normalized impervious surface index (NDISI) for the estimation of impervious surfaces. It was demonstrated that NDISI can extract impervious surfaces efficiently. Based on NDBI, a continuous built-up index (BUc
) was developed by He et al. [3
]. Stathakis et al. [17
] suggested a vegetation index built-up index (VIBI) to segment urban areas efficiently. Deng and Wu [18
] proposed a biophysical composition index (BCI). Land in arid and semi-arid regions has different spectral characteristics from other climatic regions. Therefore, appropriate remotely sensed indicators of land use and land cover types need to be defined for arid and semi-arid lands, as indices developed for other climatic regions may not give plausible results in those regions.
In 2013, Landsat 8 was launched with two sensors, the Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS). To monitor long-term land cover and land use change, Landsat data are crucial. To minimize atmospheric absorption features, the width of several OLI bands in Landsat 8 was enhanced [19
]. In 2014, Bhatti and Tripathi proposed a built-up area extraction method (BAEM), and similarly, Zhou et al. used a built-up and bare-land index (BBI) to extract built-up and bare soil from Landsat 8 [20
]. Piyoosh and Ghosh [22
] developed a normalized ratio urban index (NRUI) and modified normalized difference soil index (MNDSI) to distinguish between urban areas and soil from Landsat 8.
The reviewed literature on spectral indices leads us to conclude that these indices have not entirely successfully addressed the confusion between impervious surface and bare soil [18
]. Hence, this study focusses on identifying appropriate remotely sensed indices for arid and semi-arid regions derived from Landsat 8 data. It addresses the difference between bare soil and built-up classes that are inherent in other indices for mapping land use land cover classes in cities in dry climates in anticipation that this will provide better information than that obtainable from previously published spectral indices. Erbil city in Iraq was selected for this research and a Landsat 8 image was obtained in the dry season for examining the effectiveness of the proposed techniques. The objective of the current study was to determine the accuracy of the newly proposed indices (dry built-up index (DBI) and dry bare-soil index (DBSI)) in mapping built-up and bare soil areas in dry climate from Landsat 8.
Erbil is located 412 m above sea level within 36°08′ N–36°14′ N and 43°57′ E–44°03′ E [24
]. The study area is located in the central part of northern Iraq serving as the capital of Kurdistan region (Figure 1
). The population of the Erbil Governorate was estimated to be around 1,500,000 inhabitants in 2015 [25
The city has a semi-arid continental climate, a rainy cool winter and dry and warm summer. The seasonal distribution of the precipitation in the study area varies and falls mostly as rain in winter and autumn [26
]. The average annual precipitation is 380.26 ± 108.88 mm. The annual air temperature is 22 °C, and July and August are the hottest months of the year [24
]. The dominant land use type in the city is residential land use and concrete blocks are the main material for buildings. In general, winter grains are the most common form of land cultivation in the area and depend on rainfall; therefore, in the summer, the majority of croplands are dry [27
To distinguish urban areas and bare land using spectral indices is commonly associated with low accuracy due to the high degree of homogeneity [7
]. Due to low moisture content in green areas surrounding cities, built-up and bare land indices that were developed for humid regions can often not sufficiently differentiate between built-up and bare land in drylands (Table 1
and Table 2
; Figure 5
and Figure 6
). In past studies, NDBI identified built-up areas accurately in cites in humid climates, for instance, in Colombo city in Sri Lanka [32
], Montreal in Canada [30
], São José dos Campos city in Brazil [23
], Beijing and Guangzhou city in China [33
]. In contrast, NDBI has performed poorly for mapping built-up areas in the semi-arid cities of Urumqi and Shihezi in western China [35
]. Zhou et al. [21
] also reported a low accuracy (57.4%) of NDBI for Zhengzhou city in China from Landsat 8 OLI data.
Using Blue with TIR1 and SWIR1 with Green bands, the proposed DBI and DBSI techniques can map built-up and bare land in dry climate zones at an accuracy of 93% and 92%, respectively. The results from DBI and DBSI are highly similar to land use classes from high-resolution satellite images. This result confirms that the proposed indices are able to detect built-up land from high DBI values and bare-soil land from high DBSI values. Applying the proposed indices in dry climate zones is shown to be more suitable. Thus, using the proposed indices for different backgrounds of urban areas may produce better and more accurate classification. This could be replicated in many fields of research, for example, in urban expansion and sustainability of cities.
Built-up areas have higher spectral values than bare land in the Blue band (Band 2), and in contrast, in the TIR1
band they have lower spectral values. The built-up index is accurate for distinguishing between land use/land cover classes using the TIR channels that display high emissivity in urban areas [7
]. The main disadvantage of use thermal band is it has lower spatial resolution; however, because in Erbil and similar semi-arid areas, built-up areas generally exhibit higher values in the Blue and lower in the TIR band, it improves the accuracy of classification. Diminishing NDVI in the DBI and DBSI indices significantly improved accuracy of built-up mapping and addresses the ambiguity in areas with high NDVI (i.e., green areas) that are often confused with built-up areas [3
]. Generally, high values of DBI and DBSI represent intense built-up and bare-soil, respectively, but the threshold to extract built-up areas and bare soil may vary between study sites.
Appropriate remotely sensed indicators of land use and land cover need to be defined for arid and semi-arid regions because these areas have different reflectance characters. As a consequence, indices developed for temperate regions may not give adequate results in dry climate areas. Since indices as NDBI and NDBaI are unable to distinguish between built-up areas and bare land that often surrounds cities in dry climates, this study proposes the application of two new spectral indices, DBI and DBSI derived from Landsat 8 data for mapping built-up and bare soil areas in dry climates.
The developed DBI and DBSI were applied to map urban areas and bare land in the city of Erbil, Iraq. The accuracy assessment shows an overall accuracy of 93% (κ = 0.86) and 92% (κ = 0.84) for DBI and DBSI, respectively. The results indicate that the proposed indices can be reliably used to differentiate built-up and bare land from other land use classes in arid and semi-arid climate.
The suggested DBI index can separate only general built-up areas; it cannot separate finer urban cover types such as commercial and residential areas in detail. These indices focus on urban areas in dry climate, thus they may not be suitable for cities in humid regions or those surrounded by green spaces (positive NDVI). In this paper, the indices have demonstrated a potential for using Landsat 8 images to delineate between built-up areas and bare land within cities in dry climates. However, this study may have some limitations, thus more investigation into the application of different remote sensing imageries using these techniques in other study sites is required.