Big Scientific Data Project Zone II Versions EN1 Vol 4 (4) 2019
An impervious surface map of Hainan Island (2018)
: 2019 - 08 - 15
: 2019 - 12 - 27
: 2019 - 09 - 10
: 2019 - 12 - 31
283 5 0
Abstract & Keywords
Abstract: Along with rapid economic development and accelerated population growth, urban expansion has brought significant changes to the ecological environment of a city. As a major indicator of urban environment, impervious layers greatly impact ecological environment changes and urban development. By analyzing the spectral features of the 2018 remote sensing imagery, this study developed methods for extracting and mapping the impervious surface over Hainan Island. Using the Modified Normalized Difference Impervious Surface Index (M_NDISI) method and the Perpendicular Impervious Index (PII) method, we performed impervious layer extraction in Matlab environment to obtain the distribution of the impervious surfaces over Hainan Island. This dataset can be used to support the Island’s urban planning and development.
Keywords: impervious layer; impervious surface index; Hainan Island
Dataset Profile
TitleAn impervious surface map of Hainan Island (2018)
Data corresponding authorLi Qingwen (
Data authorLi Qingwen, Yan Dongmei, Ling Jianmei, Huang Qingqing
Time range2018Spatial resolution30 m
Geographical scopeHainan Island (19°20′N – 20°10′N, 108°21′E – 111°03′E)
Data format*.rarData volume767 KB
Data service system<>
Source of fundingKey Research and Development Plan of Hainan Province (Grant No. ZDYF2018001)
Dataset compositionThe dataset consists of one file only, namely, “MP_hainan.rar”, which records the distribution of Hainan Island’s impervious surfaces in 2018. It has a total data volume of 767 KB.
1.   Introduction
With its rapid economic development and population growth since the new millennium, China underwent an accelerated process of urbanization and urban expansion, driving the change of urban features such as land cover types, population distribution, building and road networks, etc. Among them is the impervious layer,1 a type of land cover surface preventing water from penetrating into the soil comprising concrete buildings, roads, parking lots, etc. It has been recognized as a major indicator of urban environment significantly impacting ecological environment changes and urban development.
Impervious layer is closely related to the ecological environment of a city.2 Impervious layer expansion, for example, would not only hinder the water cycle, leading to decreased water quality,3 but also prevent urban heat dissipation, causing heat island effect. Impervious layer distribution, on the other hand, reflects the planning and structure of a township, and its scale directly reflects a city’s development status. Studies of impervious layer distribution shed light on ecological environment changes and township distribution, which would support research on a city’s urban planning and design.
Compared with manual surveys, remote sensing technology provides a more efficient, accurate and informative means to study the impervious layers,4,5 which can be extracted through, most commonly, spectral hybrid analysis, index method, regression, object-oriented classification and decision tree-based method. Spectral hybrid analysis6 is to extract the impervious layers by establishing a linear hybrid model based on mixed pixels. Index method is to construct an impervious layer index based on its spectral responses. Regression method is to obtain the distribution of the impervious layers by establishing a regression equation. Wang Xiankai et al.,7 by using the Biophysical Composition Index (BCI) and Bare Soil Index (BI) methods, extracted the impervious surface over Sanya City of Hailan Island, which inspired our use of the index method here, though our study features a wider research scope and hopefully a more efficient method for impervious layer extraction.
Studies of the impervious layers are hitting a bottleneck. In low- and medium-resolution remote sensing images, as each pixel is a projection of composite surfaces, extraction has to be based on mixed pixels. Due to its diverse constituent materials, the impervious layers are known for their huge spectral disparities, which makes extraction work difficult. In high-resolution remote sensing images, the ground surface covered by the shadows of high buildings is unable to generate spectral information of itself. Considering Hainan Island’s vast geographical scope, high vegetation coverage and high cloud cover, this study used a combination of the Modified Normalized Difference Impervious Surface Index (M_NDISI) and Perpendicular Impervious Index (PII) method for extracting the impervious layers from the 2018 Landsat 8 remote sensing images of the Island.
2.   Data collection and processing
2.1   Geographic range
Hainan is the southernmost province of China, next to Guangdong Province in the north across Qiongzhou Strait, Beibu Gulf in the west, South China Sea and Taiwan Province in the east, and the Philippines, Brunei and Malaysia in the southeast and south. Under the jurisdiction of Hainan Province are Hainan Island, Xisha Islands, Nansha Islands, Zhongsha Islands, and the islands and reefs in their surrounding waters. Located between latitudes 18°10' N – 20°10' N and longitudes 108°37' E – 111°03' E, Hainan Province has a land area of 35,000 km2 and a sea area of about two million km2.
Hainan Island is China’s second largest island following Taiwan Island, with an area of 33,900 km2. Located in a typical tropical monsoon climatic zone, it has abundant rainfall, with an annual precipitation of 1,000 – 2,600 mm, and an average annual precipitation of 1,639 mm. The Island has a typical cascade structure and an overall dome mountain-shaped terrain that is high in the middle and low in the surroundings. Mt. Wuzhi in the central south is the highest peak of the Island.
2.2   Data preparation
This study aimed to map the impervious surface of Hainan Island by extracting the surface boundaries from existing databases. Landsat 8 satellite images was selected as the main data source for the following considerations. Launched by NASA on February 11, 2013, the Landsat 8 satellite was equipped with two sensors, namely, OLI (Operational Land Imager) and TIRS (Thermal Infrared Sensor). Landsat 8 band settings are shown in Table 1. The cirrus band provides good data support for cloud masking. The TIRS, on the other hand, provides more detailed thermal infrared band information which meets our band and resolution requirements.
Table 1   Landsat 8 band setting
SensorBandWavelength range (μm)Signal-to-noise ratioSpatial resolution (m)
2.3   Data processing
2.3.1   Data preprocessing
Considering possible deviations during the remote sensing image data collection and transmission (e.g., geometric distortion and radiation distortion), we performed geometrical correction, radiation calibration and atmospheric correction on the Landsat 8 data, to ensure the spectral information gives an accurate reflection of the ground features.
2.3.2   Masking
(1) Water body masking
Image masking is to control the scope of an image to be processed by blocking selected areas, terrains or objects on all or part of the image. As water bodies can interfere with the impervious layer extraction in the index method, water bodies have to be masked prior to image processing.
With a strong capacity to absorb sunlight, water bodies generally have a lower reflectance at the wavelength of visible light. The longer the wavelength, the smaller the reflectance. Water bodies have a relatively higher reflectance in the blue light band, which accounts for the blue seawater as we see it, but a lower reflectance in the near-infrared and mid-infrared bands.
Our study used the Modified Normalized Difference Water Index (MNDWI) proposed by Xu Hanqiu8 to extract and mask the water bodies.
(2) Cloud cover masking
Cloud cover inevitably impacts on the accuracy of remote sensing images. In areas under cloud cover, the reflectance values collected by remote sensing satellites are composed of two parts: first, the reflectance value measured by solar radiation emitted through the clouds; second, the reflectance value measured by solar radiation reflected by ground objects emitted back through the clouds. Cloud cover, in other words, affects the reflectance of ground features in each band and therefore interferes with impervious layer extraction. Cloud cover has to be masked on each scene of the remote sensing image. For this end, we used ENVI (Environment for Visualizing Images), a software application equipped with mature tools for automatic cloud-cover detection, to generate the cloud masking files.
2.3.3   Modified Normalized Difference Water Index
(1) Normalized Difference Impervious Surface Index
As different ground features have different reflection spectra in the remote sensing image, the surface types can be identified by their spectral characteristics. An impervious layer typically has a high reflectance in the thermal infrared band but a low reflectivity in the near-infrared band. In both bands, the impervious layer has similar spectral characteristics with soil, water and bare land. Detailed band information is often needed for clear distinction. Research found that water bodies have a higher reflectance in the green light band of visible wavelengths.9 In this study, we used the Modified Normalized Difference Water Index (MNDWI) proposed by Xu Hanqiu8 to identify water bodies. The MNDWI is expressed in formula (1):
where Green denotes the green light band, and MIR is the mid-infrared band.
Also, considering the higher reflectance of vegetation in the near-infrared band, we amplified the differentiation between strong and weak radiation for more effective impervious layer extraction. Specifically, as per the Normalized Difference Impervious Surface Index (NDISI), a ratio operation was performed by using the stronger reflection band of the impervious layer as the numerator and the weak reflection band as the denominator to amplify the distinction between strong and weak reflection bands, thereby enhancing the effect of the impervious layer while suppressing that of other ground features. The impervious layer can be expressed by the NDISI index10 composed of compound bands as shown in formula (2):
where NIR, MIR and TIR are the near-infrared, mid-infrared and thermal infrared bands of the image, respectively; MNDWI is the modified normalized difference water body index.
(2) Modified Normalized Difference Impervious Surface Index
Through the ratio calculation in formula (2), NDISI helped enhance the effect of the thermal infrared band while weakening that of water bodies and near- and mid-infrared bands, enabling an effective distinction between the impervious layer and bare soil, water body, bare land, etc. However, NDISI did not consider the impact of vegetation on impervious layer extraction. Located in a tropical region, Hainan Island is covered by a vast vegetation area throughout the year. As vegetation might be mistaken as the impervious layer by NDISI, the vegetation index was introduced into band combination for more accurate extraction. After a comprehensive comparison, we selected the Normalized Difference Vegetation Index (NDVI) to characterize the vegetation.
Meanwhile, as water bodies had been masked earlier by the MNDWI, only essential NDVI bands were introduced into the new index and non-essential MNDWI bands were removed.
The M_NDISI is expressed in formula (3):
The NDVI is expressed in formula (4):
NDVI was obtained through a band operation by the M_NDISI method. The reflectance values in the thermal infrared and NDVI bands were stretched to 0 – 255 for obtaining the M_NDISI. Finally, a threshold value was established for extracting the impervious layers.
2.3.4   Perpendicular Impervious Index
(1) Index construction
In contrast to soil and vegetation, the impervious layer has similar spectral values in the blue and near-infrared bands. Based on this, Tian Yugang et al.11 proposed a method for impervious layer extraction where the blue and near-infrared bands were selected to construct the PII index, as expressed in formula (5):
where Pblue and PNIR are the spectral values in the blue light and near-infrared bands, respectively; m and n are the coefficients of the blue light and near-infrared bands, respectively; and C is a constant. m, n and C can be obtained through the reference line equation. Let the reference line equation of PII be y=ax+b, the relative vertical distance from the point of the pixel on the two-dimensional plane to the reference line is expressed by equation (6):
where a and b denote the slope and intercept of the reference line, respectively; and x and y are the spectral values of the pixel in the blue and near-infrared bands, respectively. When the coefficient of PII takes equation (7), PII=D, the slope of its contour is the same as that of the reference line. As shown in Figure 1, when the distance from the sampling point to the reference line is zero (PII=0), the area between the reference line and the x-axis is the area when D>0, and the area between the reference line and the y-axis is the area when D<0.

Figure 1   Schematic diagram illustrating the distance from the sampling point to the reference line
(2) Parameter estimation
In the study area composed of the five Landsat 8 remote sensing image scenes, we selected 100 bare soil samples and 100 impervious layer samples for obtaining their spectral values in the blue and near-infrared bands. Least square fitting was performed on the two.
As the samples showed varied degrees of dispersion due to complex surface features, a standard deviation was calculated to adjust the fitted soil line and the impervious layer line. The reference line of PII was determined as the angular bisectors of the translated soil line and the impervious layer line.
Let the standard deviation of the bare soil sample, counted as its vertical distance to the soil line, be . Likewise, let the standard deviation of the impervious layer sample, counted as its vertical distance to the impervious layer line, be . Translate the lines according to their standard deviations, and the angular bisectors of the two are taken as the reference line of PII.
Suppose the original soil line equation be y = asx+bs, the impervious layer line equation be y = aix + b, the adjusted soil line equation be y = asx + bsa, and the adjusted impervious layer line equation be y = aix + bia. Let and , and formula (8) can be expressed as follows:
Substitute a and b into equation (7) to obtain the PII coefficient.
As per the PII method, we first selected a small amount of pure soil samples and impervious layer samples of an even quality in the study area, to obtain the spectral values of each sample in blue and near-infrared bands, the soil and impervious layer fit lines, and also the impervious layer reference line. Then an impervious layer index was constructed mathematically. After that, a band operation was performed to obtain the calculation result of each pixel point on the Landsat image. Finally, a threshold was determined for extracting the impervious layers.
2.3.5   Impervious surface mapping
The impervious layers extracted by M_NDISI and PII methods were spliced to obtain two data maps for the impervious surface of Hainan Island, respectively. However, each of the methods had its deficiencies: the large shadow area projected by the vast mountainous terrain of the Island might be mistaken as impervious surfaces by the M_NDISI method, due to their similar spectral features; likewise, the PII method might mistake bare soil as impervious surface, due to the similar spectral features of the two.
Given this, we performed logical derivation and calculation on the results obtained by both methods to offset the deficiencies of each other, which effectively minimized errors in the impervious surface extraction.
3.   Sample description
After the five scenes of Landsat 8 remote sensing images were processed, we used M_NDISI and PII methods to map the impervious surface. As shown in Figure 2, red line denotes the boundary of the impervious surface.

Figure 2   The impervious surface map of Hainan Island (2018)
4.   Quality control and assessment
For accuracy analysis, the data were validated against the 2017 GF-2 fused remote sensing images with a resolution of 1 m, and through visual interpretation, to obtain the values of the impervious layer. For this end, we randomly selected 1,000 pixel points on the Landsat 8 images, and calculated the extraction results of each pixel point through visual interpretation to obtain the data accuracy (Table 2).
Table 2   Accuracy analysis
Visual interpretation
Our results
Impervious layerPermeable layerTotalUser’s accuracy
Impervious layer4296349287.2%
Permeable layer5645250889.0%
Producer’s accuracy88.5%87.8%
Overall accuracy88.1%
The validation results show our dataset has an overall accuracy of 88.1%, and a Kappa coefficient of 0.762, which are satisfactory outcomes. Our method is also proved to have a higher accuracy compared with M_NDISI or PII methods, as shown in Table 3.
Table 3   Accuracy comparison
Method \ Accuracy evaluationOverall accuracyKappa
M_NDISI method86.3%0.726
PII method86.1%0.722
By comparison, it can be seen that the extraction results have a significantly higher accuracy than the original data obtained by the M_NDISI and PII methods.
5.   Value and significance
This study obtained the impervious surface map of Hainan Island based on the 2018 Landsat 8 remote sensing data. The map can be used to help monitor the ecological environment changes and urban structures of Hainan Island. It also provides data support for research on the Island’s ecological environment and other social fields, and is of great value for the province’s policy formulation, development planning and resource utilization.
Li DR, Luo H & Shao ZF. Review of impervious surface mapping using remote sensing technology and its application. Geomatics and Information Science of Wuhan University 41(2016): 569–577, 703.
Jr Arnold CL& Gibbons CJ. Impervious surface coverage: the emergence of a key environmental indicator. Journal of the American Planning association 62(1996): 243–258.
Deng C & Wu C. BCI: a biophysical composition index for remote sensing of urban environments. Remote Sensing of Environment 127(2012): 247–259.
Cui QY, Pan Y &Yang X. Beijing plain area of remote sensing images based on Landsat 8 impermeable layer coverage estimates. Journal of Capital Normal University (Natural Sciences Edition) 36(2015): 89–92.
Weng Q. Remote sensing of impervious surfaces in the urban areas: requirements, methods, and trends. Remote Sensing of Environment 117(2012): 34–49.
Jin J, Wang B & Zhang LM. Estimating urban impervious surface distribution by linear spectral mixture analysis. Journal of Fudan University (Natural Science) 49(2010): 197–208, 273–274.
Wang XK, Meng QY, Liu Y et al. A dataset of impervious surface in Sanya City from 2004 to 2015. Science Data Bank 4(2018). DOI: 10.11922/sciencedb.721.
Xu HQ. A study on information extraction of water body with the modified normalized difference water index (MNDWI). Journal of Remote Sensing 9(2005): 589–595.
Du YY & Zhou CH. Automatically extracting remote sensing information for water bodies. Journal of Remote Sensing 2(1998): 264–269.
Xu HQ. A new remote sensing index for fastly extracting impervious surface information. Geomatics and Information Science of Wuhan University 11(2008): 1150–1153, 1211.
Tian YG, Xu Y & Yang XN. Perpendicular impervious index for remote sensing of multiple impervious surface extraction in cities. Acta Geodaetica et Cartographica Sinica 46(2017): 468–477.
Data citation
1. Li Q, Yan D, Ling J et al. An impervious surface map of Hainan Island (2018). Science Data Bank, 2019. (2019-12-27). DOI: 10.11922/sciencedb.835.
Article and author information
How to cite this article
Li Q, Yan D, Ling J et al. An impervious surface map of Hainan Island (2018). China Scientific Data 4(2019). DOI: 10.11922/csdata.2019.0038.zh.
Li Qingwen
automatic extraction of impervious layers from Landsat TM images.
MSc; research area: remote sensing image processing.
Yan Dongmei
overall technical design for dataset production and data quality control.
PhD, Professor; research area: remote sensing image processing and application, spatial big data mining and analysis.
Ling Jianmei
Landsat-8 image data pre-processing, field verification and inspection, accuracy analysis, etc.
MSc; research area: forestry information engineering.
Huang Qingqing
Landsat-8 image acquisition and sorting, field verification and inspection.
PhD, Assistant Professor; research area: remote sensing image processing and application.
Key Research and Development Plan of Hainan Province (Grant No. ZDYF2018001)
Publication records
Published: Dec. 31, 2019 ( VersionsEN1
Released: Sept. 10, 2019 ( VersionsZH2
Published: Dec. 31, 2019 ( VersionsZH3