Obtaining accurate and up-to-date information on the spatial dimensions of gross domestic product (GDP) and electric power consumption (EPC) is important to understanding a country’s social and economic status. Previous studies on GDP and EPC have mainly used statistical data for administrative units [1
]. For instance, using GDP statistical data, Mehrotra et al.
] evaluated the dynamics of GDP growth for China against alternative indicators. Michieka et al.
] investigated the relationship among GDP, electricity production and coal consumption in mainland China from 1971 to 2009. However, statistical data only provide numeric records for the socioeconomic situation of a specific region (e.g., census or administrative region), and the spatial distribution of those records is not explicitly represented. Therefore, appropriate approaches should be used as complements to the statistics data for estimating and mapping the spatial patterns of those socioeconomic indicators, such as GDP and EPC, in a statistical area [7
Compared to the traditional socioeconomic census, remote sensing techniques provide a suitable method for describing the spatial distribution of GDP and EPC. Nighttime light data are typical remote sensing data used to map GDP and EPC [8
]. Such data can help investigate the socioeconomic situation at a large spatial scale with a comparatively low cost [12
]. Consequently, it has become the preferred choice for modeling spatial distribution of GDP and EPC in national or continental scales. Traditionally, the nighttime light data acquired by the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP-OLS) archived by the National Oceanic and Atmospheric Administration’s National Geophysical Data Center (NOAA/NGDC) of the United States comprise the only useful dataset for estimating and mapping socioeconomic indicators, such as GDP and EPC [16
]. For example, De Souza et al.
] evaluated Brazil’s 2001 energy crisis by using DMSP-OLS data. Propastin and Kappas [22
] revealed that the DMSP-OLS data became an effective tool for the monitoring of both spatial and temporal variability of the examined socioeconomics. Min et al.
] detected rural electrification in Africa using DMSP-OLS data. Furthermore, Zhao et al.
] produced a GDP change map of China based on a regression between DMSP-OLS data and the population.
However, pixel saturation in the DMSP-OLS data could reduce the correlation between the socioeconomic activity and the nighttime light data [9
].This issue results from the OLS sensor’s low radiometric resolution. As the OLS sensor’s radiance ranges from 10−1
° to 10−8
under normal operation, pixels with radiance greater than 10−8
(which often existed in the centers of large cities) would not be distinguished [26
]. DMSP-OLS data have pixel values ranging from 0 to 63. Due to the aforementioned reasons, such saturated pixels are all given the value of 63. This effect introduced inaccuracies to GDP and EPC modeling in some areas, especially in the centers of large cities with strong artificial lighting [10
A new generation of nighttime light data sensed by the Visible Infrared Imaging Radiometer Suite (VIIRS) carried by the Suomi National Polar-Orbiting Partnership (Suomi NPP) satellite was released by NOAA/NGDC in early 2013 [17
]. VIIRS is one of the five instruments (others are the Advanced Technology Microwave Sounder [32
], the Cross-Track Infrared Sounder [33
], the Ozone Mapping Profiler Suite [34
], and the Clouds and the Earth’s Radiant Energy System [36
]) on the satellite, and is configured to collect visible and infrared imagery and radiometric measurements of the land, atmosphere, cryosphere, and oceans [38
]. The first global NPP-VIIRS nighttime light data were generated by the Earth Observation Group, NOAA National Geophysical Data Center. VIIRS day/night band data collected on nights with zero moonlight during 18–26 April 2012 and 11–23 October 2012 was composited into a single dataset [31
Compared with the DMSP-OLS data, the better qualities of NPP-VIIRS data are threefold. First, as a new generation of nighttime light data, the NPP-VIIRS data feature a higher spatial resolution (15 arc-second, about 500 m) than the DMSP-OLS data (30 arc-second, about 1,000 m). Second, the NPP-VIIRS data do not have the issue of over-saturation existing in the DMSP-OLS data, since a wider radiometric detection range has been used. The VIIRS day/night band on Suomi NPP has a specified dynamic range of approximately seven orders of magnitude from 3 × 10−9
to 0.02 W·cm−2
]. Third, The NPP-VIIRS data employ onboard calibration (not available for the DMSP-OLS data) which increases the data quality. More details are available in [45
]. Li et al.
] employed NPP-VIIRS data to estimate gross regional products (GRP) in China and demonstrated that the data have a strong capacity in modeling regional economic indicators at the national scale. However, to the best of our knowledge, few studies have been conducted to investigate the potential of NPP-VIIRS nighttime light data for estimating GDP and EPC at multiple scales. In addition, there is still a lack of comparison between the GDP and EPC estimation using NPP-VIIRS data and the estimation using DMSP-OLS data. Therefore, a comprehensive assessment of the advantages of the new dataset to estimate such socioeconomic indicators would provide a better understanding of the data quality, as well as support further analysis in future research.
This study aims to investigate the potential of NPP-VIIRS data for estimating GDP and EPC in provincial and prefecture units of mainland China. The structure of the paper is organized as follows. A detailed description of study area and data will be presented in Section 2. The data processing and linear regression methods used in this study will be described in Section 3. We will then present the estimation results and discuss the advantages and limits of NPP-VIIRS data in modeling GDP and EPC at multiple scales. Finally, we summarize results and draw conclusions in the last section.
2. Case Study Area and Data
2.1. Case Study Area: Mainland China
This study takes mainland China as the case study area. Hong Kong, Macao, and Taiwan are not included in this research due to the lack of relevant statistical data. Since the economic reform and the adoption of open-door policy in 1979, China has been undergoing rapid economic development. The GDP of China keeps on increasing in the last several decades at an unprecedented high speed. Meanwhile, the EPC of China continues increasing as well, providing support to the economic activities.
The administrative divisions of mainland China contain five levels; there is a level for the province, prefecture, county, township, and village. The provincial level is the highest level of administrative division. There are 31 provincial-level divisions, including 22 provinces, four municipalities (Beijing, Shanghai, Tianjin, and Chongqing), and five autonomous regions. Generally, a province consists of several prefectural-level units, and a prefecture includes several counties. This study employs provincial and prefectural levels to perform a multi-scale analysis. At the provincial level, all the provincial units are included in our analysis. At the prefectural level, 268 out of 333 prefectures in mainland China are selected for this study. The 268 prefectures belong to 27 provincial units. The areas of the prefectural divisions in municipalities are much smaller than those in provinces and autonomous regions, and cannot be accurately identified from nighttime light data. Consequently, all the prefectural divisions of four municipalities are neglected in the prefectural-level analysis. Other neglected prefectural units are located in western China due to a lack of their GDP and EPC statistical data.
2.2. Data Collections
The only available composite NPP-VIIRS nighttime light data of the year 2012 were obtained from website of NOAA/NGDC ( http://ngdc.noaa.gov/eog/viirs/download_viirs_ntl.html
]. Unlike DMSP-OLS data, the NPP-VIIRS data have not been filtered to remove light detections associated with fires, gas flares, volcanoes or aurora. Also, the background noise has not been subtracted. For example, the weak light reflected by snowcapped mountains and dry lake beds is recorded in the data, as well. It should be noted that there are some pixels with negative DN values in original NPP-VIIRS data. As there is no description about those pixels in the metadata of original NPP-VIIRS data, we assume the negative DN values of those pixels are caused by background noise and outliers from data processing.
The DMSP-OLS data can be divided into three types: the stable light data, the cloud-free coverage, and the data with no further filtering. Among the three types of data, the stable light data contain lights from cities, towns and other sites with persistent lighting, and have removed ephemeral events (e.g., fires, gas flares, volcanoes and background noise) [46
]. In September 2013, NOAA/NGDC updated Version 4 DMSP-OLS Nighttime Lights Time Series dataset, and provided the stable light datasets from 1992 to 2012. Since the NPP-VIIRS data were only available for the year 2012, we use the DMSP-OLS stable light data of the year 2012 in this study (available at http://ngdc.noaa.gov/eog/dmsp/downloadV4composites.html
The boundary data of administrative divisions of mainland China, including national, provincial, and prefectural boundaries, were obtained from the National Geomatics Center of China. The nighttime light images of China were extracted from the global datasets of NPP-VIIRS data (Figure 1
) and DMSP-OLS data (Figure 2
) by using a mask polygon of the national boundary of China with a 50 km buffer. All the data were projected into the Lambert Azimuthal Equal Area Projection and resampled to the spatial resolution of 500 m × 500 m (cell size).
The GDP and EPC statistical data for provincial and prefectural units in China were obtained from the China Statistical Yearbook and the China City Statistical Yearbook. Unfortunately, GDP and EPC data of the year 2012 are not available yet. The GDP data of the year 2011 for provincial and prefectural units, the EPC data of the year 2011 for provincial unit, and the EPC data of the year 2010 for prefectural units were adopted as alternatives.
Three scenes of Landsat 8 OLI-TIRS images (free of cloud) are used to provide a visual evaluation for the quality of nighttime light data in this study. Two images were acquired on 14 April 2013, and another one was acquired on 21 April 2013. These images were downloaded from the Geospatial Data Cloud ( http://www.gscloud.cn/
). They cover three cities at the prefectural level in China, including Chengdu (the capital of Sichuan Province in western China), Wuhan (the capital of Hubei Province in central China), and Nanjing (the capital of Jiangsu Province in eastern China).
The regression results show the corrected NPP-VIIRS nighttime light data have a better performance in estimating GDP and EPC than the DMSP-OLS data at both the provincial and prefectural levels. The better estimated results mainly benefit from the higher resolution and radiometric detection range of NPP-VIIRS nighttime light data. This could be easily discovered from a visual comparison among DMSP-OLS data, NPP-VIIRS data, and a fine-resolution reference, Landsat 8 OLI-TIRS images, for three different cities in China (Figure 7
). It should be noted that since Landsat 7 Enhanced Thematic Mapper Plus (ETM+) images of 2012 have banded noise, the newly released Landsat 8 OLI-TIRS images of 2013 are a better choice to be the reference data. Moreover, considering that the temporal difference between Landsat 8 OLI-TIRS images of 2013 and nighttime light data of 2012 is slight, we believe that an observation among them can be used to compare the spatial consistency of urban areas in two types of nighttime light data. The three sampling cities include Chengdu (located in western China), Wuhan (located in central China), and Nanjing (located in eastern China). The lit areas of the three cities spread out the urban built-up areas (see Landsat 8 images) considerably, regardless of the developed levels and locations of the cities. In addition, the regions of water bodies and urban forests in Wuhan and Nanjing are also illuminated in DMSP-OLS data. This “blooming” phenomenon was also reported by other researchers [51
] and would obstruct a better estimation of socioeconomic indicators from DMSP-OLS data. On the contrary, the NPP-VIIRS data can display the patterns of urban built-up areas which compares to the Landsat 8 OLI-TIRS images. As one can see from Figure 7
, the approximate boundaries of rivers and lakes can be identified in the NPP-VIIRS data. Unlike the DMSP-OLS data, the DN values of the pixels in NPP-VIIRS data located at the central areas of those cities are no longer identically 63. The varied DN values in NPP-VIIRS data can reflect human activities and support a more accurate GDP and EPC estimation.
Although the high resolution and wide radiometric detection range benefit the GDP and EPC estimation, the noise in the first global NPP-VIIRS nighttime light data affects the accuracy of the estimation at multiple scales. As shown in Figure 5
and Figure 6
, the R2
value of the TNL from the original NPP-VIIRS data with either GDP or EPC is much lower than corresponding estimated values by using corrected NPP-VIIRS data. Our method for NPP-VIIRS data preprocessing can significantly reduce the negative effects caused by the brightness from oilfield, dry riverbed, and desert, which is still recorded in the raw NPP-VIIRS data. Meanwhile, some researchers (e.g., [17
]) have proposed some methods for detecting wildfires or removing the noise from NPP-VIIRS data. Those efforts would help to improve the quality of the new-generation nighttime light data in future.
Due to absence of the statistical GDP and EPC in 2012, we also have to use the 2012 corrected NPP-VIIRS data to model GDP and EPC in 2011 at the provincial level. In addition, because of the same reason, we also have to use the 2012 corrected NPP-VIIRS data to model GDP in 2011 and EPC in 2010 at the prefectural level, respectively. Although there is a one-year gap between the corrected NPP-VIIRS data and GDP and EPC at the provincial level, the corrected NPP-VIIRS data demonstrated a good performance in modeling GDP and EPC. At the prefectural level, the corrected NPP-VIIRS data could also have a correlation in modeling GDP. However, as there is the two-year gap between the corrected NPP-VIIRS data and EPC at the prefectural level, the corrected NPP-VIIRS data are not perfectly correlated with the EPC. In fact, other studies also used nighttime light images and statistical data in different years for the regression analysis, also due to unavailable data [19
]. We can thus infer that using the corrected NPP-VIIRS data to model GDP and EPC in the same year could produce more accurate results.
In addition, the modeled GDP and EPC are lower than the actual socioeconomic data (GDP and EPC) in some provincial-level units, such as Chongqing, Guangxi, Hebei, Henan, Hubei, Hunan, Jiangxi and Shandong (Table 1
). These provincial-level units are medium-developed regions in mainland China. The utilization of a sole regression model for estimating the GDP and EPC of all the provinces is liable to underestimating the socioeconomic indicators in those regions. In that circumstance, an appropriate subdivision of provincial-level units into several regions could help improve the estimate accuracy [19
]. Meanwhile, estimated results of several provinces, such as Xinjiang, Tibet, Yunnan, Shaanxi, and Heilongjiang, are much higher than the statistics data (Table 1
). Because there are many dry riverbeds, deserts, and snowy mountains in those regions, this overestimation may be ascribed to the negative effects caused by the background noises of NPP-VIIRS data. Although our corrected method has removed 11,419.75 km2
lit areas and decreased 6,257,364 TNL value in mainland China (see Table 4
), the residual noise still affects the estimated accuracy. The total nighttime light (TNL) of some provinces in western China (such as Xinjiang, Yunnan, and Tibet) are still high, as shown in Table 4
Although the corrected NPP-VIIRS nighttime light data have improved the accuracy of GDP and EPC estimation, the proposed models still contain some uncertainty, caused by the following factors. First, even though many studies have proven the nighttime light images can be used to estimate GDP, EPC, and other socioeconomic variables [8
], the relationship between the TNL and those variables is an empirical relationship which cannot be viewed as an absolute law. Second, the reliability of statistical data is a key factor affecting the modeling accuracy since such data are the basis for the linear regressions. Third, the NPP-VIIRS nighttime light data released by NOAA/NGDC are raw data in which noise still exists. Thus, there is still room for improving the data quality, and more methods could be applied to the data correction process (such as [17
]). In addition, other techniques, such as application of the entropy-based approach [53
] could also be applied to make a prediction of GDP and EPC using NPP-VIIRS nighttime light data.
The nighttime light data records artificial light on the Earth’s surface and can be used to represent human activities and estimate socioeconomic indicators. Traditionally, the DMSP-OLS stable nightlight data are the only effective nighttime light data to estimate GDP and EPC. In early 2013, the first global NPP-VIIRS nighttime light data were released by the Earth Observation Group. As a new-generation nighttime light data, NPP-VIIRS data have a higher spatial resolution and a wider radiometric detection range compared with the DMSP-OLS data. At the same time, the noise in the original NPP-VIIRS nighttime light data still affects the accuracy of the estimation at multiple scales considerably. In this study, a sequence of preprocessing procedures is used to reduce the negative impacts from the background noise of the original data. The results prove that our proposed method is effective for reducing the negative effects caused by the lights from oilfield, dry riverbed, and desert in the original NPP-VIIRS data and can provide a more reliable dataset for the GDP and EPC estimation.
Through a case study of mainland China, this research investigates the ability of NPP-VIIRS data to estimate GDP and EPC at two levels: provincial and prefectural. The linear regression analysis results reveal that the corrected NPP-VIIRS data have better performance in estimating GDP and EPC than the DMSP-OLS data. At the provincial level, the R2 values of the corrected NPP-VIIRS data with GDP and EPC are 0.8702 and 0.8961, respectively, which are higher than those values of the DMSP-OLS data (0.7315 and 0.8208). In the provincial-level divisions, over 60% and 70% administrative units have high-accuracy estimated results of GDP and EPC, respectively. At the prefectural level, the R2 of the corrected NPP-VIIRS data with GDP is 0.8088, while that of the DMSP-OLS data is 0.6644. With respect to EPC, the R2 of NPP-VIIRS data is as low as 0.5144, which is still higher than the R2 value (0.3019) from the DMSP-OLS data. The estimation of different nighttime light data for GDP and EPC of prefectural-level units in each province also shows that the corrected NPP-VIIRS nighttime light data have a better performance in estimating GDP than EPC than the DMSP-OLS data at the prefecture level. The average R2 values of the corrected NPP-VIIRS data with GDP and EPC are higher than those values with DMSP-OLS data. In conclusion, the results prove that the corrected NPP-VIIRS data are more reliable for estimating GDP and EPC than the DMSP-OLS data at multiple scales.
Since the NPP-VIIRS data comprise an emerging data source, all the released data are taken from one sequence of images acquired in 2012, which hinders a more comprehensive evaluation. As NOAA/NGDC may produce more and better-quality NPP-VIIRS nighttime light data in the future, we could also apply a multi-temporal analysis of the data in wider fields.