Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Modeling and mapping total freight traffic in China using NPP-VIIRS nighttime light composite data

Modeling and mapping total freight traffic in China using NPP-VIIRS nighttime light composite data GIScience & Remote Sensing, 2015 Vol. 52, No. 3, 274–289, http://dx.doi.org/10.1080/15481603.2015.1022420 Modeling and mapping total freight traffic in China using NPP-VIIRS nighttime light composite data a a b c d a Kaifang Shi , Bailang Yu *, Yingjie Hu , Chang Huang , Yun Chen , Yixiu Huang , a a Zuoqi Chen and Jianping Wu Key Laboratory of Geographic Information Science, Ministry of Education, East China Normal University, Shanghai 200241, China; Department of Geography, University of California Santa Barbara, Santa Barbara, CA 93106, USA; College of Urban and Environmental Science, Northwest University, Xi’an 710127, China; CSIRO Land and Water, Canberra, ACT 2601, Australia (Received 10 October 2014; accepted 20 February 2015) In early 2013, the first global Suomi National Polar-orbiting Partnership (NPP) visible infrared imaging radiometer suite (VIIRS) nighttime light composite data were released. Up to present, few studies have been conducted to evaluate the ability of NPP-VIIRS data to estimate the amount of freight traffic. This paper provides an exploratory evaluation on the NPP-VIIRS data for estimating the total freight traffic (TFT) in China, in comparison with the results derived from the Defense Meteorological Satellite Program-Operational Linescan System (DMSP-OLS) night- time stable light composite data. We first corrected the original NPP-VIIRS data by employing a simple method to remove the outliers. The total nighttime light (TNL) which is measured by the sum value of all pixels from the nighttime light composite data was then regressed on TFT at the provincial level of China. Finally, the spatial distribution patterns of TFT were produced from the corrected NPP-VIIRS and DMSP- OLS data, respectively, and validated by the TFT statistics of 244 prefectures. The results have demonstrated that the corrected NPP-VIIRS data are more suitable for modeling TFT in China than the DMSP-OLS data. Keywords: NPP-VIIRS; DMSP-OLS; total freight traffic; nighttime light; China 1. Introduction Regional total freight traffic (TFT) is the total amount of goods which has arrived and been unloaded in a certain period (Wu et al. 2011). TFT is important to understanding the development of transport infrastructures and regional economic status. However, TFT data are sometimes difficult to acquire, especially at multiple spatial scales. Therefore, survey- ing TFT at different scales is an important and challenging task for policy making and academic research. The major data sources used in previous TFT studies are the statistical data released by government agencies for administrative units (Guler and Vitosoglu 2012;Lindholm 2010). While statistical data provide useful numeric records for TFT, the spatial distribution of those records is missing. Compared to traditional socioeconomic census, nighttime light data provide a new data source for estimating TFT. Studies have demonstrated that night- time light data can not only count TFT at administrative units but also reveal the spatial patterns of TFT (Tian et al. 2014). The Defense Meteorological Satellite Program’s *Corresponding author. Email: blyu@geo.ecnu.edu.cn © 2015 Taylor & Francis GIScience & Remote Sensing 275 Operational Linescan System (DMSP-OLS) nighttime light composite data provided by the National Geophysical Data Center (NGDC) at National Oceanic and Atmospheric Administration (NOAA) States were commonly used as the sole effective nighttime light data resource for modeling and estimating socioeconomic factors (Tian et al. 2014;He etal. 2012, 2013;Yangetal. 2013; Elvidge et al. 1997, 1999; Witmer and O’Loughlin 2011; Propastin and Kappas 2012; Butt 2012). However, the DMSP-OLS data have a number of shortcomings, such as saturation on bright lights, six bit quantization, coarse spatial resolu- tion, and lack of in-flight calibration (He et al. 2012; Min et al. 2013;Shi,Huang, et al. 2014; Li, Ge, and Chen 2014; Li, Chen, and Chen 2013;Forbes 2013), which often bring in uncertainties to socioeconomic data modeling (Elvidge, Baugh, et al. 2013;Letuetal. 2010; Yu et al. 2014; Wei et al. 2014; Zhang, He, and Liu 2014; Liu et al. 2012; Long and Singh 2013). In early 2013, the NOAA/NGDC released a new generation of nighttime light composite data from the Visible Infrared Imaging Radiometer Suite (VIIRS) Day-Night Band (DNB) carried by the Suomi National Polar-orbiting Partnership (NPP) Satellite (Hillger et al. 2013). The NPP-VIIRS data were regarded as having better qualities compared with the DMSP-OLS data (Shi, Huang, et al. 2014; Shi, Yu, et al. 2014). First, the NPP-VIIRS data do not have the problem of oversaturation, due to a wider radiometric detection range. Second, the VIIRS-DNB data are a huge leap forward in capability from the OLS, with global data at 742 m spatial resolution (Ma et al. 2014; Elvidge, Zhizhin, et al. 2013). In order to composite nighttime lights, different image types, including the VIIRS-DNB data, the VIIRS flag data and the M15 flag data, were resampled to a resolution of 15 arc-second. For detailed description about the compositing algorithm, readers are recommended to refer to Baugh et al. (2013). Consequently, the NPP-VIIRS data have a higher spatial resolution (15 arc-second, about 500 m) than the DMSP-OLS (30 arc-second, about 1000 m). Thirdly, the NPP-VIIRS data hold onboard calibration (not available for the DMSP-OLS data) by using solar diffuser, which increases the data quality. A detailed information of this composite data have been reported by Elvidge, Baugh, et al. (2013) and Baugh et al. (2013). Some studies have estimated gross domestic product (GDP) and electric power consumption (EPC) using the NPP-VIIRS data at the multiple scales in China (Shi, Yu, et al. 2014; Li et al. 2013), but to the best of our knowledge, no related research has focused on their potential of the TFT estimation. The main objective of this paper is to test whether the NPP-VIIRS data can be used as a potential data source in modeling TFT at multiple scales in China through a comparative analysis with the DMSP-OLS data. To fulfill this objective, a method was used to correct the original NPP-VIIRS data. The total nighttime light (TNL) which is measured by the sum value of all pixels from the nighttime light composite data were regressed on the TFT at the provincial level of China. The absolute and relative errors between the two nighttime light composite data were calculated to evaluate their capabilities for modeling TFT. Finally, maps of the spatial distribution patterns of TFT were produced for different nighttime light composite data and validated by the TFT records of 244 prefectures in China. 2. Study area and data This paper takes mainland China as the case study area. Since the start of its economic reforms in 1978, China has experienced a tremendous growth of TFT, with an increase from 2489 million tons in 1978 to 40,994 million tons in 2012 (an increase of 15.47 times). However, the development of delivery vehicles and transportation infrastructure 276 K. Shi et al. lags behind the rapid economic growth in China. Scarcity of transport capacity and traffic jams have hindered the normal flow of goods. Therefore, accurately modeling and mapping freight traffic demands can facilitate the distribution of transportation resources and help reduce the freight shortages. Data-sets used in this paper consist of five parts. The 2012 NPP-VIIRS nighttime light composite data (two months composite) were obtained from the NOAA/NGDC website (http://ngdc.noaa.gov/eog/viirs/download_viirs_ntl.html). VIIRS DNB data collected on nights with low moonlight during 18–26 April 2012 and 11–23 October 2012 were composited into a single data-set. Version 4 DMSP-OLS nighttime stable light composite data for 2012 were used in this study. This data-set was also retrieved from the NOAA/ NGDC website (http://ngdc.noaa.gov/eog/dmsp/downloadV4composites.html). The statisti- cal TFT data for provincial and prefectural units (31 provinces and 244 prefectures) in China were obtained from 2013 China Statistical Yearbook and 2011 China City Statistical Yearbook, respectively. It is noted that the 2013 China Statistical Yearbook records the provincial level TFT in 2012, and the 2011 China City Statistical Yearbook records the prefectural level TFT in 2010. Since TFT data of 2012 for prefectural level were not available yet, the 2010 TFT data were adopted as alternatives in this study. Four high- quality Landsat 8 Operational Land Imager and the Thermal Infrared Sensor (OLI-TIRS) images were downloaded from Geospatial Data Cloud operated by the Chinese Academy of Sciences (http://www.gscloud.cn/, accessed 6 August 2013). The administrative boundaries for provinces and prefectures in China of 2008 were obtained from the National Geomatics Center of China (NGCC). A brief description of data-sets is listed in Table 1. The DMSP-OLS and NPP-VIIRS data of China were extracted from the global data-sets by using a mask polygon of the national boundary of China with a 50 km buffer and resampled into the same spatial resolution of 500 m (Figure 1). All data-sets were projected to the Lambert Azimuthal Equal Area Projection with reference to WGS84 Datum. 3. Methods 3.1. Correction of the NPP-VIIRS nighttime light composite data As noted by some studies (Schroeder et al. 2014; Elvidge, Baugh, et al. 2013), the original NPP-VIIRS data are a preliminary product. Bright noise surfaces, such as gas flares, fires, oilfields, volcanoes and background noise, have not been filtered out. Therefore, they may not be able to reflect the true TFT. A simple method to remove the outliers has thus been employed to reduce those interference factors from the original NPP-VIIRS data (Li et al. 2013; Shi, Yu, et al. 2014). The detailed procedures are as follows. First, a mask with all pixels with DN value of 0 from the 2012 DMSP-OLS data was generated. Next, the mask Table 1. Description of the data-set used in this study. Data source Product description Year NPP-VIIRS Yearly nighttime light composite 2012 DMSP-OLS Yearly nighttime stable light composite 2012 TFT The statistical data derived from China Statistical Yearbook 2012 and and China City Statistical Yearbook 2010 Landsat 8 Four images covering four cities, bands 4, 5, and 6 2013 OLI-TIRS Administrative The administrative boundaries for provinces and prefectures 2008 boundaries GIScience & Remote Sensing 277 60°0′0″E 70°0′0″E 80°0′0″E 90°0′0″E 100°0′0″E 130°0′0″E 140°0′0″E 120°0′0″E 150°0′0″E Radiance –2 –1 (nWcm sr ) 2815.18 (a) –0.41 South China Sea 100°0′0″E 110°0′0″E 80°0′0″E 90°0′0″E 120°0′0″E 130°0′0″E 60°0′0″E 70°0′0″E 80°0′0″E 90°0′0″E 100°0′0″E 130°0′0″E 140°0′0″E 120°0′0″E 150°0′0″E DN (b) South China Sea 100°0′0″E 80°0′0″E 90°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E Figure 1. The nighttime light data of mainland China in 2012. (a) The original NPP-VIIRS nighttime light composite data; (b) the DMSP-OLS nighttime stable light data. multiplies the 2012 NPP-VIIRS data to generate the primary corrected NPP-VIIRS data. Since the lighted areas detected by the DMSP-OLS data are consistently larger than those of the NPP-VIIRS data, it is believed that all of the lighted areas related to human activity are still reserved in the primary corrected NPP-VIIRS data (Small, Pozzi, and Elvidge 2005). Then, the final corrected NPP-VIIRS data were generated by setting the maximum values which depend upon economic development of three megacities (Beijing, Shanghai, and Guangzhou) in China. As the three cities are the biggest and most developed cities in China, the pixel values of the other areas should not exceed the maximum values of these three megacities. However, due to the regional difference in socioeconomic status, 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 278 K. Shi et al. atmospheric condition, as well as landscape, a unique threshold value cannot completely remove the remaining outliers for all the regions in China. In this paper, we subdivide China into 10 regions (Wang et al. 2013), including Northeast China (NEC), Inner Mongolia (IM), Southwest China (SWC), Tibet (TB), South China (SC), Xinjiang (XJ), Northwest China (NWC), North China (NC), East China (EC), and Central China (CC) (Figure 2). The optimal threshold value for each region was determined based on the maximum radiance value of the largest representative cities in the specific region. These values were derived from the original NPP-VIIRS data and listed in Table 2. Due to 70°0′0″E 110°0′0″E 130°0′0″E 140°0′0″E 60°0′0″E 80°0′0″E 90°0′0″E 150°0′0″E NEC XJ IM NC NWC EC TB CC SWC Regional boundary SC National boundary South China Sea 90°0′0″E 80°0′0″E 100°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E Figure 2. Ten regions of mainland China. Note: NEC represents Northeast China; IM represents Inner Mongolia; SWC represents Southwest China; TB represents Tibet. SC is South China; XJ is Xinjiang; NWC is Northwest China; NC is North China; EC is East China; CC is Central China. Table 2. Optimal threshold values for removing outliers in different regions of China. Threshold Threshold −2 −1 −2 −1 District Representative cities (nW cm sr ) District Representative cities (nW cm sr ) NEC Shenyang, 177.19 XJ Urumqi 180.61 Changchun, Harbin IM Hohhot 107.23 NWC Xi’an, Lanzhou, 182.59 Yinchuan SWC Chengdu, Kunming, 184.81 NC Beijing, Taiyuan, 235.13 Guiyang Shijiazhuang TB Lhasa 165.81 EC Shanghai, Nanjing, 148.78 Jinan SC Guangzhou, 105.19 CC Wuhan, Zhengzhou, 116.88 Shenzhen, Nanning Changsha 20°0′0″N 30°0′0″N 40°0′0″N 10°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N GIScience & Remote Sensing 279 limited space, the original NPP-VIIRS data of selected cities within only five regions (North China, Northeast China, Northwest China, East China, and South China) are illustrated in Figure 3. 116°0′0″E 117°0′0″E 112°0′0″E 113°0′0″E 113°40′0″E 115°0′0″E (a) (b) (c) Radiance Radiance Radiance 235.13 96.01 114.64 Beijing 0 0 Taiyuan 0 Shijiazhuang 0 116°0′0″E 117°0′0″E 112°0′0″E 113°0′0″E 113°20′0″E 114°40′0″E 126°0′0″E 128°0′0″E 130°0′0″E 125°0′0″E 126°0′0″E 127°0′0″E 122°0′0″E 123°0′0″E 124°0′0″E 125°0′0″E (d) (e) (f) Radiance Radiance Radiance 124.39 158.03 177.19 Shenyang Harbin 0 Changchun 0 0 125°0′0″E 127°0′0″E 129°0′0″E 124°0′0″E 125°0′0″E 127°0′0″E 122°20′0″E 123°20′0″E 124°20′0″E 126°0′0″E 108°0′0″E 109°0′0″E 103°0′0″E 104°0′0″E 105°40′0″E 106°0′0″E 106°20′0″E 106°40′0″E (g) (h) (i) Radiance 128.23 Radiance Radiance 135.34 182.59 Xi’an Lanzhou Yinchuan 0 0 108°0′0″E 109°0′0″E 103°0′0″E 104°0′0″E 105°40′0″E 106°0′0″E 106°20′0″E 106°40′0″E 117°0′0″E 121°0′0″E 122°0′0″E 119°0′0″E 120°0′0″E 116°20′0″E 117°40′0″E 118°20′0″E (j) (k) (l) Radiance Radiance Radiance 148.78 115.57 47.01 Jinan Shanghai Nanjing 0 0 0 122°0′0″E 118°0′0″E 116°40′0″E 117°20′0″E 118°0′0″E 121°0′0″E 119°0′0″E 116°0′0″E 113°0′0″E 114°0′0″E 114°0′0″E 108°20′0″E 109°20′0″E 110°20′0″E (n) (m) (o) Radiance Radiance Radiance 103.76 105.19 70.58 Nanning Guangzhou 0 Shenzhen 0 0 113°0′0″E 114°0′0″E 114°0′0″E 107°20′0″E 108°20′0″E 109°20′0″E 110°20′0″E Figure 3. The original NPP-VIIRS data of the selected cities for five different regions in 2012. (a), (b), and (c) are the representative cities in North China; (d), (e), and (f) are the representative cities in Northeast China; (g), (h), and (i) are the representative cities in Northwest China; (j), (k), and (l) are the representative cities in East China; (m), (n), and (o) are the representative cities in South China; The solid lines in all figures are the administrative boundaries. 41°0′0″N 23°0′0″N 24°0′0″N 31°0′0″N 34°0′0″N 35°0′0″N 47°0′0″N 40°0′0″N 45°0′0″N 46°0′0″N 23°0′0″N 31°0′0″N 40°0′0″N 41°0′0″N 34°0′0″N 35°0′0″N 44°0′0″N 45°0′0″N 46°0′0″N 23°0′0″N 32°0′0″N 38°0′0″N 36°0′0″N 37°0′0″N 44°0′0″N 45°0′0″N 23°0′0″N 44°0′0″N 45°0′0″N 38°0′0″N 32°0′0″N 36°0′0″N 37°0′0″N 22°20′0″N 23°20′0″N 24°20′0″N 36°40′0″N 37°20′0″N 41°40′0″N 42°20′0″N 43°0′0″N 38°0′0″N 38°40′0″N 38°20′0″N 38°40′0″N 22°0′0″N 23°0′0″N 24°0′0″N 36°20′0″N 37°0′0″N 38°20′0″N 41°20′0″N 42°0′0″N 42°40′0″N 37°40′0″N 38°20′0″N 38°40′0″N 280 K. Shi et al. After that, each pixel whose radiance value was larger than that of the representative cities in the each region was modified to be a new value determined by its surrounding pixel values. For detailed approaches to correct the original NPP-VIIRS data and depict the corrected results, readers are recommended to refer to Shi, Yu, et al. (2014). After this process, all pixel values in the corrected NPP-VIIRS data of China were less than 235.13. 3.2. Establishment of a regression model There are many linear regression models commonly used by related studies to estimate socioeconomic factors using the nighttime light data (Cheng et al. 2014; Shi, Yu, et al. 2014). After investigating various existing models, a linear relationship between the log of TFT and the log of TNT was regarded as better correlating between TFT and TNT to accurately estimate TFT using nighttime light composite data at the provincial level: lgT ¼ wlgE þ c (1) where T is the statistical TFT of a provincial unit, E represents the TNL, which is calculated by the sum of all pixels in a provincial unit, and w and c are parameters determined by the regression analysis of sample data. Next, relative errors were examined to evaluate the capacity of TNL in estimating TFT as P  T RE ¼  100% (2) where RE is the relative error, P represents the estimated TTF, and T denotes the statistical TFT. 3.3. Spatialization of TFT To map the spatial distribution of TFT, TFT from the province level needs to be disaggregated to the pixel level based on the assumption that a more developed area generally has brighter lights and larger freight traffic demands. With reference to the method of Tian et al. (2014), each provincial TFT is disaggregated to individual pixel based on a pixel’s radiance value (or a DN value): ST HP ¼  R (3) SL where HP represents the amount of distributed TFT of a pixel, ST is a provincial p i statistical TFT, SL represents TNL of a province, and R is a radiance value (or a DN value) of a pixel. 4. Results 4.1. Regression results The regression results between TNL and TFT are shown in Figure 4. The R of the corrected NPP-VIIRS data and TFT was 0.817 (Figure 4a), whereas that of the TNL from GIScience & Remote Sensing 281 4 4 y = 0.3006x + 1.2397 y = 1.3108x – 4.2581 y = 1.1719x – 4.5163 2 2 2 R = 0.028 R = 0.817 R = 0.746 3 3 3 2 2 2 1 1 (a) (c) (b) 0 0 45 6 7 45 6 7 lg TNL of corrected NPP-VIIRS lg TNL of DMSP-OLS lg TNL of original NPP-VIIRS Figure 4. Relationships between the total freight traffic (TFT) and the total nighttime light (TNL) for 31 province units in China. (a) The TFT and the TNL of corrected NPP-VIIRS; (b) the TFT and the TNL of DMSP-OLS; (c) the TFT and the TNL of original NPP-VIIRS. DMSP-OLS data and TFT is 0.746 (Figure 4b). Moreover, the R of the TNL from the original NPP-VIIRS data and TFT is as low as 0.028 (Figure 4c). The results indicate that the corrected NPP-VIIRS can better reflect the TFT than the DMSP-OLS data and the original NPP-VIIRS data at the provincial level. Table 3 shows the absolute and relative errors calculated based on the regression analysis. On the whole, the absolute errors of the corrected NPP-VIIRS data are much lower than those of the DMSP-OLS data. There are nine provincial units with low absolute errors less than 100 million tons when TFT is predicted from the corrected NPP-VIIRS data. Contrarily, the DMSP-OLS data exhibit a low capacity in predicting TFT, with five provincial units with low absolute errors less than 100 million tons. Meanwhile, it can also be seen that the relative errors of the estimated total freight traffic (ETFT) in 2012 derived from the DMSP-OLS data in Tibet and Shaanxi are 312.63% and −0.70%, respectively, which is very different. By contrast, the counterparts of ETFT in 2012 derived from the corrected NPP-VIIRS data are 221.60% and −0.92%, respectively. To evaluate the capability of the two types of nighttime light data more comprehen- sively, the absolute relative errors (AREs) were divided into four classes: 0–20% as low, 20–40% as moderate, 40–60% as high, >60% as very high. The percentage of each class was calculated for the different estimation approaches. For instance, when calculating TFT from the corrected NPP-VIIRS data, there are 10 provincial units with low among the 31 provinces, and the ARE of low is 32.26%. Based on the ARE, the predictability for different types of nighttime light data has been quantified in Table 4. The corrected NPP- VIIRS data showed a fine capacity in estimating TFT, with low and moderate percentage of 32.26% and 29.04%, respectively, which are relatively higher than those of the DMSP- OLS data (29.03% and 22.58%, respectively). Moreover, the percentages of high and very high for TFT using the corrected NPP-VIIRS (19.35% and 19.35%, respectively) are lower than those using the DMSP-OLS data (25.81% and 22.58%, respectively). From the above analysis, we can see that the corrected NPP-VIIRS data have a better ability in estimating TFT than the DMSP-OLS data at the provincial level. The better estimated results mainly benefit from the higher radiometric detection range and higher resolution of the NPP-VIIRS data. However, a series of obvious errors still exist. As listed in Table 3, Beijing, Jiangsu, Guangdong, Zhejiang, and Tianjin are all comparatively overestimated (Table 3). The five provinces are the economically more developed pro- vinces in China. They consume and produce the largest quantities of goods. However, these provinces mainly produce high-tech products and light industrial goods that lead to high outputs (GDP) but relatively small total weights of the finished products as well as lg TFT (million tons) lg TFT (million tons) lg TFT (million tons) 282 K. Shi et al. Table 3. Accuracy of the estimated total freight traffic at the provincial level of China. Corrected NPP-VIIRS DMSP-OLS TFT ETFT ETFT (million (million AE RE (million AE RE (million tons) (%) (million tons) (%) Province tons) tons) tons) Beijing 261.62 559.28 297.66 113.77 463.50 201.88 77.17 Tianjin 460.15 554.63 94.48 20.53 418.17 −41.98 −9.12 Hebei 2191.30 2047.36 −143.94 −6.57 2653.22 461.92 21.08 Shanxi 1446.08 1477.74 31.66 2.19 1607.77 161.69 11.18 Inner 1899.42 1128.62 −770.8 −40.58 1717.66 −181.76 −9.57 Mongolia Liaoning 2067.89 1472.44 −595.45 −28.79 1590.74 −477.15 −23.07 Jilin 548.08 555.26 7.18 1.31 805.52 257.44 46.97 Heilongjiang 652.31 1016.37 364.06 55.81 1888.70 1236.39 189.54 Shanghai 940.38 1022.53 82.15 8.74 369.23 −571.15 −60.74 Jiangsu 2200.08 4560.24 2360.16 107.28 3512.43 1312.35 59.65 Zhejiang 1918.17 2909.02 990.85 51.66 2080.81 162.64 8.48 Anhui 3124.37 1392.42 −1731.95 −55.43 1336.44 −1787.93 −57.23 Fujian 843.45 1537.49 694.04 82.29 1189.11 345.66 40.98 Jiangxi 1271.96 621.12 −650.84 −51.17 601.41 −670.55 −52.72 Shandong 3336.03 3028.49 −307.54 −9.22 3789.21 453.18 13.58 Henan 2721.15 1914.87 −806.28 −29.63 2405.64 −315.51 −11.59 Hubei 1229.45 1024.48 −204.97 −16.67 892.30 −337.15 −27.42 Hunan 1910.52 1006.64 −903.88 −47.31 713.32 −1197.2 −62.66 Guangdong 2560.77 4814.57 2253.8 88.01 3310.67 749.9 29.28 Guangxi 1613.56 1092.42 −521.14 −32.30 917.49 −696.07 −43.14 Hainan 268.80 216.46 −52.34 −19.47 291.45 22.65 8.43 Chongqing 864.74 556.22 −308.52 −35.68 377.69 −487.05 −56.32 Sichuan 1743.49 1436.96 −306.53 −17.58 1089.28 −654.21 −37.52 Guizhou 526.55 355.25 −171.3 −32.53 402.11 −124.44 −23.63 Yunnan 687.35 931.85 244.5 35.57 1137.46 450.11 65.49 Tibet 11.27 36.23 24.96 221.60 46.49 35.22 312.63 Shaanxi 1367.27 1354.65 −12.62 −0.92 1357.65 −9.62 −0.70 Gansu 458.32 481.17 22.85 4.99 655.06 196.74 42.93 Qinghai 134.84 102.44 −32.4 −24.03 140.80 5.96 4.42 Ningxia 411.13 266.61 −144.52 −35.15 277.20 −133.93 −32.58 Xinjiang 587.94 1155.21 567.27 96.49 1403.22 815.28 138.67 Note: TFT is the statistical total freight traffic, ETFT is the estimated total freight traffic, AE is the absolute error, and RE is the relative error in percentage. Table 4. Different classes of estimated accuracies for total freight traffic (TFT) at the provincial level. NPP-VIIRS and TFT DMSP-OLS and TFT Class Quantity ARE (%) Quantity ARE (%) Very high 6 19.35 7 22.58 High 6 19.35 8 25.81 Moderate 9 29.04 7 22.58 Low 10 32.26 9 29.03 Note: ARE is the absolute relative error in percentage. GIScience & Remote Sensing 283 the materials required for their production (Tian et al. 2014). Weights of freight traffic denoted by one unit of TNL in these provinces were smaller than that of nationwide average level. Thus, relatively overestimation appears when TNL was used to estimate TFT in those provinces. Meanwhile, estimated results of Tibet, Heilongjiang, and Xinjiang provinces are much higher than the recorded statistics. There were many deserts, snowy mountains, and oilfields in those three provinces, the overestimation might be ascribed to the negative effects caused by the background noises of NPP-VIIRS data (Shi, Yu, et al. 2014). Although our proposed method could remove a lot of outliers, the residual noises still affect the accuracy of estimation. On the contrary, some provinces’ TFT, including Inner Mongolia, Liaoning, Anhui and Jiangxi, were underestimated. Those provinces are moderately developed provinces and provide raw materials in China. Every year goods with large masses and volumes, such as coal, iron ore and timber, are shipped via railways to these provinces, which lead to relatively large TFT but correspond to very limited light from train stations. Thus, although these provinces have relatively higher TFT, their TFT are still underestimated due to relatively low TNL. Actually, as percen- tages of railway freight traffic (RFT) to TFT are very small (i.e., 9.67% in 2012), TNL still have strong relationship with TFT at the provincial level in China. 4.2. Spatialization results Based on the spatialization model equation in Section 3.3, we then produced the spatial maps of 2012 TFT by using the corrected NPP-VIIRS data and DMSP-OLS data, respectively (Figure 5). There is a significant difference between the two maps. In the corrected NPP-VIIRS spatial TFT map (Figure 5a), the high TFTs are concentrated in some inland metropolitan areas and industrial cities, whereas the TFTs are relatively dispersed in the DMSP-OLS spatial TFT map (Figure 5b). This could be easily discovered from a visual comparison among the corrected NPP-VIIRS data, DMSP-OLS data and a fine resolution reference, Landsat 8 OLI-TIRS images (Figure 6). The four sampling cities with different levels of urbanization and socioeconomic development include Shanghai (located in Eastern China, economic center), Nanjing (located in Eastern China, the capital of Jiangsu province), Maanshan (located in central China, a city of Anhui province in middle size) and Hechuan (located in western China, a small city of Chongqing city). The relative high TFT of the four cities spread out to the urban areas including urban forests and rural areas in the DMSP-OLS data (Figure 6e–h), regardless of the developed levels and locations of the cities. On the contrary, the high TFT in the corrected NPP-VIIRS data (Figure 6i–l) are concentrated in urban areas which compared to the Landsat 8 OLI-TIRS images (Figure 6a–d). As we known, the urban areas usually have larger population and produce and consume more goods which lead to a large freight transport demand. Thus, we can conclude that TFT is mainly concentrated in urban areas, while urban forests and rural areas only have small TFT. Consequently, it could be easily certified that the corrected NPP-VIIRS data can reflect human activities and support a more accurate spatialization of TFT. Figures 5 and 6 showed the pixel-based TFT estimated from provincial statistical data. We then used prefectural-level statistical TFT to validate the spatial TFT results in Figure 5. We collected 244 prefectural TFT and used the prefecture boundary to calculate the TFT for each prefecture from spatial TFT maps. Two indicators, root-mean-square ), were used to evaluate the accuracies of TFT error (RMSE) and regression coefficient (R maps. Figure 7 shows the validation results of estimated TFT for 244 prefectures. Results derived from the corrected NPP-VIIRS data in Figure 7a have higher R (0.674) and lower 284 K. Shi et al. 60°0′0″E 70°0′0″E 80°0′0″E 90°0′0″E 110°0′0″E 130°0′0″E 140°0′0″E 150°0′0″E TFT (million tons) 0 – 1 1 – 2 2 – 3 3 – 4 (a) South China Sea >4 80°0′0″E 90°0′0″E 100°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E 60°0′0″E 80°0′0″E 90°0′0″E 110°0′0″E 130°0′0″E 140°0′0″E 150°0′0″E 70°0′0″E TFT (million tons) 0 – 1 1 – 2 2 – 3 3 – 4 (b) >4 South China Sea 80°0′0″E 90°0′0″E 100°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E Figure 5. The TFT maps estimated from nighttime light composite data. (a) is the corrected NPP-VIIRS spatial TFT map; (b) is the DMSP-OLS spatial TFT map. RMSE (93.87) than those from DMSP-OLS data in Figure 7b (R of 0.207, RMSE of 216.55). They have proved that the corrected NPP-VIIRS data have more accurate results than the DMSP-OLS data at the pixel level. 5. Discussion The study is based on the assumption that there is a positive relationship between TFT and nighttime lights. Because the developed regions usually have a large population, advanced industry, and convenient transportation, they would produce and consume more goods which lead a large freight transport demand (McKinnon and Woodburn 1996; Tian et al. 2014). Meanwhile, it was well-known that nighttime lights could be used to model socioeconomic indicators, such as GDP and population (Tian et al. 2014; Shi, Yu, et al. 2014). Therefore, we proposed that the nighttime light data could be as a proxy for TFT estimation. In this study, regression analysis was firstly employed to quantify the relation- ship between TFT and TNL. The results have proved that regression analysis is simple but 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N GIScience & Remote Sensing 285 118°0′0″E 119°0′0″E 121°0′0″E 122°0′0″E 118°20′0″E 118°40′0″E 119°0′0″E 106°0′0″E 106°20′0″E 106°40′0″E (a) (b) (c) (d) 121°0′0″E 122°0′0″E 118°0′0″E 119°0′0″E 118°30′0″E 118°50′0″E 106°0′0″E 106°20′0″E 106°40′0″E 121°0′0″E 122°0′0″E 118°0′0″E 119°0′0″E 118°20′0″E 118°40′0″E 119°0′0″E 106°0′0″E 106°20′0″E 106°40′0″E (e) (f) (g) (h) 121°0′0″E 122°0′0″E 118°0′0″E 119°0′0″E 118°30′0″E 118°50′0″E 106°0′0″E 106°20′0″E 106°40′0″E 119°0′0″E 121°0′0″E 122°0′0″E 118°0′0″E 118°20′0″E 118°40′0″E 119°0′0″E 106°0′0″E 106°20′0″E 106°40′0″E (i) (j) (k) (l) 122°0′0″E 121°0′0″E 118°0′0″E 119°0′0″E 118°30′0″E 118°50′0″E 106°0′0″E 106°20′0″E 106°40′0″E TFT (million tons) 0 0 – 1 3 – 4 >4 1 – 2 2 – 3 Figure 6. Comparisons among the Landsat 8 OLI-TIRS images, the DMSP-OLS data and the corrected NPP-VIIRS data of four typical cities (Shanghai, Nanjing, Maanshan, and Hechuan). (a), (b), (c), and (d) are Landsat 8 OLI-TIRS images; (e), (f), (g), and (h) are DMSP-OLS spatial TFT maps; (i), (j), (k), and (l) are corrected NPP-VIIRS TFT maps. The solid lines in all figures are the administrative boundaries. y = 0.6726x + 5179 RMSE = 93.87 y = 0.2954x + 8857 RMSE = 216.55 2 2 R = 0.674 (million tons) 800 (million tons) 800 R = 0.207 600 600 400 400 200 200 (a) (b) 0 0 0 2 00 4 00 6 00 800 0 200 400 600 800 1000 1200 Estimated TFT by DMSP-OLS (milliontons) Estimated TFT by corrected NPP-VIIRS (million tons) Figure 7. Relationships between estimated and statistical TFT for 244 prefectures of China. (a) is the relationship between TFT estimated from corrected NPP-VIIRS data and statistical TFT; (b) is the relationship between TFT estimated from DMSP-OLS data and statistical TFT. reliable in estimating TFT at the provincial level in China. In addition, since each provincial TFT was disaggregated to individual pixel based on a pixel value, the disag- gregated TFT could confine the estimated errors within a provincial unit. Spatial 31°0′0″N 32°0′0″N 31°0′0″N 32°0′0″N 31°0′0″N 32°0′0″N Statistical TFT (million tons) 31°0′0″N 31°0′0″N 31°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 31°20′0″N 31°40′0″N 31°40′0″N 31°20′0″N 31°40′0″N 31°20′0″N Statistical TFT (million tons) 31°30′0″N 31°30′0″N 31°30′0″N 30°0′0″N 30°20′0″N 30°0′0″N 30°20′0″N 30°0′0″N 30°20′0″N 30°0′0″N 30°0′0″N 30°20′0″N 30°20′0″N 30°0′0″N 30°20′0″N 286 K. Shi et al. variations of TFT could be clearly exhibited by TFT maps. The amount of TFT gradually decreases from urban areas to suburban areas. Thus, the NPP-VIIRS data were also relatively creditable in modeling TFT at the pixel level. At the same time, although our study has proved that our method has a good performance in modeling TFT at the multiple scales, it still contains a few limitations. For example, we were not confident enough in identifying the source of errors caused by confusion of brightness values between a central business district (CBD) and a dense forested area. The major contributions of this study are twofold. First, previous studies have proved that the DMSP-OLS data have a potential to be used as a proxy for TFT estimation (Tian et al. 2014), but no related work reported the ability of NPP-VIIRS data to estimate TFT. This study has demonstrated that the NPP-VIIRS data have a stronger capacity in modeling TFT than those of the DMSP-OLS data for the first time. This finding will benefit the further utilization of NPP-VIIRS data and provide a powerful tool for the socioeconomic indicator estimation using remotely sensed data. Second, for the released NPP-VIIRS data have not been filtered to remove light detections associated with fires, gas flares, volcanoes, and background noise, these noises could limit the accuracy and reliability in TFT estimation. In this study, a new correction method integrating DMSP- OLS data and regional optimal thresholds are proposed. This method takes the regional difference in landscape, socioeconomic status, and atmospheric condition into account, and could not only remove obvious outliers, but also reserve most of pixels with correct radiance values. Since the NPP-VIIRS data are a newly released data source, only three scenes of images are available, including the composites of April and October in 2012, January in 2013, April in 2014, which are insufficient to conduct a comprehensive evaluation. As NOAA/NGDC is working to produce more NPP-VIIRS data with higher quality, further research can be focused on multitemporal image analysis for disclosing the spatiotemporal pattern of TFT. 6. Conclusions TFT is an important socioeconomic indicator for understanding the development of transport infrastructures and regional economic status. Therefore, accurate information about TFT in a timely manner has been regarded as an important and challenging task for the government department as well as academic community. This study highlighted the ability of using the NPP-VIIRS data for modeling and mapping TFT in China. Based on a simple method to remove the outliers of original data, the corrected NPP-VIIRS data were generated. The sum value of all pixels from the nighttime light composite data in a provincial unit were regressed on the TFT at the provincial level of China. The estimation results suggest the R of TFT between the corrected NPP-VIIRS data and statistical data is as high as 0.817, which is higher than that of the value between the DMSP-OLS data and statistical data (R = 0.746). The spatial TFT maps were produced from the corrected NPP-VIIRS data and DMSP-OLS data, respectively. The spatialization results of TFT also prove that the corrected NPP-VIIRS data are able to produce more accurate spatial TFT map in comparison with the DMSP-OLS data. The validation results show that the corrected NPP-VIIRS data have a higher R (0.674) and a lower RMSE (93.87), whereas the DMSP-OLS data have the worse results with R of 0.207 and RMSE of 216.55. In conclusion, this study reveals that the corrected NPP-VIIRS data are a more reliable data source for modeling TFT in China. GIScience & Remote Sensing 287 Acknowledgments The authors thank three anonymous reviewers and the editor for their constructive comments and suggestions. Disclosure statement No potential conflict of interest was reported by the authors. Funding This study is supported by the National Natural Science Foundation of China [grant number 41471449]; the Natural Science Foundation of Shanghai [grant number 14ZR1412200]; the Innovation Program of Shanghai Municipal Education Commission [grant number 15ZZ026]; the Fundamental Research Funds for the Central Universities of China; the State Scholarship Fund sponsored by the China Scholarship Council [grant number 201406140007]. ORCID Bailang Yu http://orcid.org/0000-0001-5628-0003 Chang Huang http://orcid.org/0000-0002-8621-4581 References Baugh, K., F.-C. Hsu, C. D. Elvidge, and M. Zhizhin. 2013. “Nighttime Lights Compositing Using the VIIRS Day-night Band: Preliminary Results.” Proceedings of the Asia-Pacific Advanced Network 35: 70–86. doi:10.7125/APAN.35.8. Butt, M. J. 2012. “Estimation of Light Pollution Using Satellite Remote Sensing and Geographic Information System Techniques.” GIScience & Remote Sensing 49 (4): 609–621. doi:10.2747/ 1548-1603.49.4.609. Cheng, Y., Z. Wang, X. Ye, and Y. Wei. 2014. “Spatiotemporal Dynamics of Carbon Intensity from Energy Consumption in China.” Journal of Geographical Sciences 24 (4): 631–650. doi:10.1007/s11442-014-1110-6. Elvidge, C. D., K. E. Baugh, J. B. Dietz, T. Bland, P. C. Sutton, and H. W. Kroehl. 1999. “Radiance Calibration of DMSP-OLS Low-light Imaging Data of Human Settlements.” Remote Sensing of Environment 68 (1): 77–88. doi:10.1016/S0034-4257(98)00098-4. Elvidge, C. D., K. E. Baugh, E. A. Kihn, H. W. Kroehl, and E. R. Davis. 1997. “Mapping City Lights with Nighttime Data from the DMSP Operational Linescan System.” Photogrammetric Engineering and Remote Sensing 63 (6): 727–734. Elvidge, C. D., K. E. Baugh, M. Zhizhin, and F. Hsu. 2013. “Why VIIRS Data Are Superior to DMSP for Mapping Nighttime Lights.” Proceedings of the Asia-Pacific Advanced Network 35: 62–69. doi:10.7125/APAN.35.7. Elvidge, C. D., M. Zhizhin, F. Hsu, and K. Baugh. 2013. “What Is So Great About Nighttime VIIRS Data for the Detection and Characterization of Combustion Sources?” Proceedings of the Asia- Pacific Advanced Network 35: 33–48. doi:10.7125/APAN.35.5. Forbes, D. J. 2013. “Multi-scale Analysis of the Relationship Between Economic Statistics and DMSP-OLS Night Light Images.” GIScience & Remote Sensing 50 (5): 483–499. Guler, H., and Y. Vitosoglu. 2012. “Estimation of Freight Transportation.” Proceedings of the ICE- Transport 166 (3): 174–185. doi:10.1680/tran.10.00065. He, C., Q. Ma, T. Li, Y. Yang, and Z. F. Liu. 2012. “Spatiotemporal Dynamics of Electric Power Consumption in Chinese Mainland from 1995 to 2008 Modeled Using DMSP/OLS Stable Nighttime Lights Data.” Journal of Geographical Sciences 22 (1): 125–136. doi:10.1007/ s11442-012-0916-3. He, C., Q. Ma, Z. Liu, and Q. Zhang. 2013. “Modeling the Spatiotemporal Dynamics of Electric Power Consumption in Mainland China Using Saturation Corrected DMSP/OLS Nighttime Stable Light Data.” International Journal of Digital Earth 6 (1): 1–22. 288 K. Shi et al. Hillger, D., T. Kopp, T. Lee, D. Lindsey, C. Seaman, S. Miller, J. Solbrig, et al. 2013. “First-Light Imagery from Suomi NPP VIIRS.” Bulletin of the American Meteorological Society 94 (7): 1019–1029. doi:10.1175/bams-d-12-00097.1. Letu, H., M. Hara, H. Yagi, K. Naoki, G. Tana, F. Nishio, and O. Shuhei. 2010. “Estimating Energy Consumption from Night-time DMPS/OLS Imagery after Correcting for Saturation Effects.” International Journal of Remote Sensing 31 (16): 4443–4458. doi:10.1080/ Li, X., F. Chen, and X. Chen. 2013. “Satellite-Observed Nighttime Light Variation as Evidence for Global Armed Conflicts.” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 6 (5): 2302–2315. doi:10.1109/jstars.2013.2241021. Li, X., L. Ge, and X. Chen. 2014. “Quantifying Contribution of Land Use Types to Nighttime Light Using an Unmixing Model.” IEEE Geoscience and Remote Sensing Letters 11 (10): 1667–1671. doi:10.1109/lgrs.2014.2304496. Li, X., H. Xu, X. Chen, and C. Li. 2013. “Potential of NPP-VIIRS Nighttime Light Imagery for Modeling the Regional Economy of China.” Remote Sensing 5 (6): 3057–3081. doi:10.3390/ rs5063057. Lindholm, M. 2010. “A Sustainable Perspective on Urban Freight Transport: Factors Affecting Local Authorities in the Planning Procedures.” Procedia-Social and Behavioral Sciences 2 (3): 6205–6216. doi:10.1016/j.sbspro.2010.04.031. Liu, Z., C. He, Q. Zhang, Q. Huang, and Y. Yang. 2012. “Extracting the Dynamics of Urban Expansion in China Using DMSP-OLS Nighttime Light Data from 1992 to 2008.” Landscape and Urban Planning 106 (1): 62–72. doi:10.1016/j.landurbplan.2012.02.013. Long, D., and V. P. Singh. 2013. “An Entropy-Based Multispectral Image Classification Algorithm.” IEEE Transactions on Geoscience and Remote Sensing 51 (12): 5225–5238. doi:10.1109/ tgrs.2013.2272560. Ma, T., C. Zhou, T. Pei, S. Haynie, and J. Fan. 2014. “Responses of Suomi-NPP VIIRS-Derived Nighttime Lights to Socioeconomic Activity in China’s Cities.” Remote Sensing Letters 5 (2): 165–174. doi:10.1080/2150704x.2014.890758. McKinnon, A. C., and A. Woodburn. 1996. “Logistical Restructuring and Road Freight Traffic Growth.” Transportation 23 (2): 141–161. doi:10.1007/BF00170033. Min, B., K. M. Gaba, O. F. Sarr, and A. Agalassou. 2013. “Detection of Rural Electrification in Africa Using DMSP-OLS Night Lights Imagery.” International Journal of Remote Sensing 34 (22): 8118–8141. doi:10.1080/01431161.2013.833358. Propastin, P., and M. Kappas. 2012. “Assessing Satellite-Observed Nighttime Lights for Monitoring Socioeconomic Parameters in the Republic of Kazakhstan.” GIScience & Remote Sensing 49 (4): 538–557. doi:10.2747/1548-1603.49.4.538. Schroeder, W., P. Oliva, L. Giglio, and I. A. Csiszar. 2014. “The New VIIRS 375m Active Fire Detection Data Product: Algorithm Description and Initial Assessment.” Remote Sensing of Environment 143: 85–96. doi:10.1016/j.rse.2013.12.008. Shi, K., C. Huang, B. Yu, B. Yin, Y. Huang, and J. Wu. 2014. “Evaluation of NPP-VIIRS Night- time Light Composite Data for Extracting Built-up Urban Areas.” Remote Sensing Letters 5 (4): 358–366. doi:10.1080/2150704X.2014.905728. Shi, K., B. Yu, Y. Huang, Y. Hu, B. Yin, Z. Chen, L. Chen, and J. Wu. 2014. “Evaluating the Ability of NPP-VIIRS Nighttime Light Data to Estimate the Gross Domestic Product and the Electric Power Consumption of China at Multiple Scales: A Comparison with DMSP-OLS Data.” Remote Sensing 6 (2): 1705–1724. doi:10.3390/rs6021705. Small, C., F. Pozzi, and C. D. Elvidge. 2005. “Spatial Analysis of Global Urban Extent from DMSP- OLS Night Lights.” Remote Sensing of Environment 96 (3–4): 277–291. doi:10.1016/j. rse.2005.02.002. Tian, J., N. Zhao, E. L. Samson, and S. Wang. 2014. “Brightness of Nighttime Lights as a Proxy for Freight Traffic: A Case Study of China.” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 7 (1): 206–212. doi:10.1109/jstars.2013.2258892. Wang, X., P. Xiao, X. Fen, and H. Li. 2013. “Extraction of Large Scale Urban Area Information in China Using DMSP-OLS Nighttime Light Data.” Remote Sensing for Land & Resources 25 (3): 159–164. Wei, Y., H. Liu, W. Song, B. Yu, and C. Xiu. 2014. “Normalization of Time Series DMSP-OLS Nighttime Light Images for Urban Growth Analysis with Pseudo Invariant Features.” Landscape and Urban Planning 128: 1–13. doi:10.1016/j.landurbplan.2014.04.015. GIScience & Remote Sensing 289 Witmer, F. D. W., and J. O’Loughlin. 2011. “Detecting the Effects of Wars in the Caucasus Regions of Russia and Georgia Using Radiometrically Normalized DMSP-OLS Nighttime Lights Imagery.” GIScience & Remote Sensing 48 (4): 478–500. doi:10.2747/1548-1603.48.4.478. Wu, W., Y. Cao, S. Liang, and C. Xiao. 2011. “Study on Dynamics of Freight Volume in the Yangtze River Delta since the Mid-1980s.” Resource and Environment in the Yangtze Basin 20 (7): 843–847. Yang, Y., C. He, Q. Zhang, L. Han, and S. Du. 2013. “Timely and Accurate National-scale Mapping of Urban Land in China Using Defense Meteorological Satellite Program’s Operational Linescan System Nighttime Stable Light Data.” Journal of Applied Remote Sensing 7 (1): 073535. doi:10.1117/1.JRS.7.073535. Yu, B., S. Shu, H. Liu, W. Song, J. Wu, L. Wang, and Z. Chen. 2014. “Object-based Spatial Cluster Analysis of Urban Landscape Pattern Using Nighttime Light Satellite Images: A Case Study of China.” International Journal of Geographical Information Science 28 (11): 2328–2355. doi:10.1080/13658816.2014.922186. Zhang, Q., C. He, and Z. Liu. 2014. “Studying Urban Development and Change in the Contiguous United States Using Two Scaled Measures Derived from Nighttime Lights Data and Population Census.” GIScience & Remote Sensing 51 (1): 63–82. doi:10.1080/15481603.2014.883212. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png GIScience & Remote Sensing Taylor & Francis

Modeling and mapping total freight traffic in China using NPP-VIIRS nighttime light composite data

Loading next page...
 
/lp/taylor-francis/modeling-and-mapping-total-freight-traffic-in-china-using-npp-viirs-xQzMyzZUr0

References (36)

Publisher
Taylor & Francis
Copyright
© 2015 Taylor & Francis
ISSN
1943-7226
eISSN
1548-1603
DOI
10.1080/15481603.2015.1022420
Publisher site
See Article on Publisher Site

Abstract

GIScience & Remote Sensing, 2015 Vol. 52, No. 3, 274–289, http://dx.doi.org/10.1080/15481603.2015.1022420 Modeling and mapping total freight traffic in China using NPP-VIIRS nighttime light composite data a a b c d a Kaifang Shi , Bailang Yu *, Yingjie Hu , Chang Huang , Yun Chen , Yixiu Huang , a a Zuoqi Chen and Jianping Wu Key Laboratory of Geographic Information Science, Ministry of Education, East China Normal University, Shanghai 200241, China; Department of Geography, University of California Santa Barbara, Santa Barbara, CA 93106, USA; College of Urban and Environmental Science, Northwest University, Xi’an 710127, China; CSIRO Land and Water, Canberra, ACT 2601, Australia (Received 10 October 2014; accepted 20 February 2015) In early 2013, the first global Suomi National Polar-orbiting Partnership (NPP) visible infrared imaging radiometer suite (VIIRS) nighttime light composite data were released. Up to present, few studies have been conducted to evaluate the ability of NPP-VIIRS data to estimate the amount of freight traffic. This paper provides an exploratory evaluation on the NPP-VIIRS data for estimating the total freight traffic (TFT) in China, in comparison with the results derived from the Defense Meteorological Satellite Program-Operational Linescan System (DMSP-OLS) night- time stable light composite data. We first corrected the original NPP-VIIRS data by employing a simple method to remove the outliers. The total nighttime light (TNL) which is measured by the sum value of all pixels from the nighttime light composite data was then regressed on TFT at the provincial level of China. Finally, the spatial distribution patterns of TFT were produced from the corrected NPP-VIIRS and DMSP- OLS data, respectively, and validated by the TFT statistics of 244 prefectures. The results have demonstrated that the corrected NPP-VIIRS data are more suitable for modeling TFT in China than the DMSP-OLS data. Keywords: NPP-VIIRS; DMSP-OLS; total freight traffic; nighttime light; China 1. Introduction Regional total freight traffic (TFT) is the total amount of goods which has arrived and been unloaded in a certain period (Wu et al. 2011). TFT is important to understanding the development of transport infrastructures and regional economic status. However, TFT data are sometimes difficult to acquire, especially at multiple spatial scales. Therefore, survey- ing TFT at different scales is an important and challenging task for policy making and academic research. The major data sources used in previous TFT studies are the statistical data released by government agencies for administrative units (Guler and Vitosoglu 2012;Lindholm 2010). While statistical data provide useful numeric records for TFT, the spatial distribution of those records is missing. Compared to traditional socioeconomic census, nighttime light data provide a new data source for estimating TFT. Studies have demonstrated that night- time light data can not only count TFT at administrative units but also reveal the spatial patterns of TFT (Tian et al. 2014). The Defense Meteorological Satellite Program’s *Corresponding author. Email: blyu@geo.ecnu.edu.cn © 2015 Taylor & Francis GIScience & Remote Sensing 275 Operational Linescan System (DMSP-OLS) nighttime light composite data provided by the National Geophysical Data Center (NGDC) at National Oceanic and Atmospheric Administration (NOAA) States were commonly used as the sole effective nighttime light data resource for modeling and estimating socioeconomic factors (Tian et al. 2014;He etal. 2012, 2013;Yangetal. 2013; Elvidge et al. 1997, 1999; Witmer and O’Loughlin 2011; Propastin and Kappas 2012; Butt 2012). However, the DMSP-OLS data have a number of shortcomings, such as saturation on bright lights, six bit quantization, coarse spatial resolu- tion, and lack of in-flight calibration (He et al. 2012; Min et al. 2013;Shi,Huang, et al. 2014; Li, Ge, and Chen 2014; Li, Chen, and Chen 2013;Forbes 2013), which often bring in uncertainties to socioeconomic data modeling (Elvidge, Baugh, et al. 2013;Letuetal. 2010; Yu et al. 2014; Wei et al. 2014; Zhang, He, and Liu 2014; Liu et al. 2012; Long and Singh 2013). In early 2013, the NOAA/NGDC released a new generation of nighttime light composite data from the Visible Infrared Imaging Radiometer Suite (VIIRS) Day-Night Band (DNB) carried by the Suomi National Polar-orbiting Partnership (NPP) Satellite (Hillger et al. 2013). The NPP-VIIRS data were regarded as having better qualities compared with the DMSP-OLS data (Shi, Huang, et al. 2014; Shi, Yu, et al. 2014). First, the NPP-VIIRS data do not have the problem of oversaturation, due to a wider radiometric detection range. Second, the VIIRS-DNB data are a huge leap forward in capability from the OLS, with global data at 742 m spatial resolution (Ma et al. 2014; Elvidge, Zhizhin, et al. 2013). In order to composite nighttime lights, different image types, including the VIIRS-DNB data, the VIIRS flag data and the M15 flag data, were resampled to a resolution of 15 arc-second. For detailed description about the compositing algorithm, readers are recommended to refer to Baugh et al. (2013). Consequently, the NPP-VIIRS data have a higher spatial resolution (15 arc-second, about 500 m) than the DMSP-OLS (30 arc-second, about 1000 m). Thirdly, the NPP-VIIRS data hold onboard calibration (not available for the DMSP-OLS data) by using solar diffuser, which increases the data quality. A detailed information of this composite data have been reported by Elvidge, Baugh, et al. (2013) and Baugh et al. (2013). Some studies have estimated gross domestic product (GDP) and electric power consumption (EPC) using the NPP-VIIRS data at the multiple scales in China (Shi, Yu, et al. 2014; Li et al. 2013), but to the best of our knowledge, no related research has focused on their potential of the TFT estimation. The main objective of this paper is to test whether the NPP-VIIRS data can be used as a potential data source in modeling TFT at multiple scales in China through a comparative analysis with the DMSP-OLS data. To fulfill this objective, a method was used to correct the original NPP-VIIRS data. The total nighttime light (TNL) which is measured by the sum value of all pixels from the nighttime light composite data were regressed on the TFT at the provincial level of China. The absolute and relative errors between the two nighttime light composite data were calculated to evaluate their capabilities for modeling TFT. Finally, maps of the spatial distribution patterns of TFT were produced for different nighttime light composite data and validated by the TFT records of 244 prefectures in China. 2. Study area and data This paper takes mainland China as the case study area. Since the start of its economic reforms in 1978, China has experienced a tremendous growth of TFT, with an increase from 2489 million tons in 1978 to 40,994 million tons in 2012 (an increase of 15.47 times). However, the development of delivery vehicles and transportation infrastructure 276 K. Shi et al. lags behind the rapid economic growth in China. Scarcity of transport capacity and traffic jams have hindered the normal flow of goods. Therefore, accurately modeling and mapping freight traffic demands can facilitate the distribution of transportation resources and help reduce the freight shortages. Data-sets used in this paper consist of five parts. The 2012 NPP-VIIRS nighttime light composite data (two months composite) were obtained from the NOAA/NGDC website (http://ngdc.noaa.gov/eog/viirs/download_viirs_ntl.html). VIIRS DNB data collected on nights with low moonlight during 18–26 April 2012 and 11–23 October 2012 were composited into a single data-set. Version 4 DMSP-OLS nighttime stable light composite data for 2012 were used in this study. This data-set was also retrieved from the NOAA/ NGDC website (http://ngdc.noaa.gov/eog/dmsp/downloadV4composites.html). The statisti- cal TFT data for provincial and prefectural units (31 provinces and 244 prefectures) in China were obtained from 2013 China Statistical Yearbook and 2011 China City Statistical Yearbook, respectively. It is noted that the 2013 China Statistical Yearbook records the provincial level TFT in 2012, and the 2011 China City Statistical Yearbook records the prefectural level TFT in 2010. Since TFT data of 2012 for prefectural level were not available yet, the 2010 TFT data were adopted as alternatives in this study. Four high- quality Landsat 8 Operational Land Imager and the Thermal Infrared Sensor (OLI-TIRS) images were downloaded from Geospatial Data Cloud operated by the Chinese Academy of Sciences (http://www.gscloud.cn/, accessed 6 August 2013). The administrative boundaries for provinces and prefectures in China of 2008 were obtained from the National Geomatics Center of China (NGCC). A brief description of data-sets is listed in Table 1. The DMSP-OLS and NPP-VIIRS data of China were extracted from the global data-sets by using a mask polygon of the national boundary of China with a 50 km buffer and resampled into the same spatial resolution of 500 m (Figure 1). All data-sets were projected to the Lambert Azimuthal Equal Area Projection with reference to WGS84 Datum. 3. Methods 3.1. Correction of the NPP-VIIRS nighttime light composite data As noted by some studies (Schroeder et al. 2014; Elvidge, Baugh, et al. 2013), the original NPP-VIIRS data are a preliminary product. Bright noise surfaces, such as gas flares, fires, oilfields, volcanoes and background noise, have not been filtered out. Therefore, they may not be able to reflect the true TFT. A simple method to remove the outliers has thus been employed to reduce those interference factors from the original NPP-VIIRS data (Li et al. 2013; Shi, Yu, et al. 2014). The detailed procedures are as follows. First, a mask with all pixels with DN value of 0 from the 2012 DMSP-OLS data was generated. Next, the mask Table 1. Description of the data-set used in this study. Data source Product description Year NPP-VIIRS Yearly nighttime light composite 2012 DMSP-OLS Yearly nighttime stable light composite 2012 TFT The statistical data derived from China Statistical Yearbook 2012 and and China City Statistical Yearbook 2010 Landsat 8 Four images covering four cities, bands 4, 5, and 6 2013 OLI-TIRS Administrative The administrative boundaries for provinces and prefectures 2008 boundaries GIScience & Remote Sensing 277 60°0′0″E 70°0′0″E 80°0′0″E 90°0′0″E 100°0′0″E 130°0′0″E 140°0′0″E 120°0′0″E 150°0′0″E Radiance –2 –1 (nWcm sr ) 2815.18 (a) –0.41 South China Sea 100°0′0″E 110°0′0″E 80°0′0″E 90°0′0″E 120°0′0″E 130°0′0″E 60°0′0″E 70°0′0″E 80°0′0″E 90°0′0″E 100°0′0″E 130°0′0″E 140°0′0″E 120°0′0″E 150°0′0″E DN (b) South China Sea 100°0′0″E 80°0′0″E 90°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E Figure 1. The nighttime light data of mainland China in 2012. (a) The original NPP-VIIRS nighttime light composite data; (b) the DMSP-OLS nighttime stable light data. multiplies the 2012 NPP-VIIRS data to generate the primary corrected NPP-VIIRS data. Since the lighted areas detected by the DMSP-OLS data are consistently larger than those of the NPP-VIIRS data, it is believed that all of the lighted areas related to human activity are still reserved in the primary corrected NPP-VIIRS data (Small, Pozzi, and Elvidge 2005). Then, the final corrected NPP-VIIRS data were generated by setting the maximum values which depend upon economic development of three megacities (Beijing, Shanghai, and Guangzhou) in China. As the three cities are the biggest and most developed cities in China, the pixel values of the other areas should not exceed the maximum values of these three megacities. However, due to the regional difference in socioeconomic status, 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 278 K. Shi et al. atmospheric condition, as well as landscape, a unique threshold value cannot completely remove the remaining outliers for all the regions in China. In this paper, we subdivide China into 10 regions (Wang et al. 2013), including Northeast China (NEC), Inner Mongolia (IM), Southwest China (SWC), Tibet (TB), South China (SC), Xinjiang (XJ), Northwest China (NWC), North China (NC), East China (EC), and Central China (CC) (Figure 2). The optimal threshold value for each region was determined based on the maximum radiance value of the largest representative cities in the specific region. These values were derived from the original NPP-VIIRS data and listed in Table 2. Due to 70°0′0″E 110°0′0″E 130°0′0″E 140°0′0″E 60°0′0″E 80°0′0″E 90°0′0″E 150°0′0″E NEC XJ IM NC NWC EC TB CC SWC Regional boundary SC National boundary South China Sea 90°0′0″E 80°0′0″E 100°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E Figure 2. Ten regions of mainland China. Note: NEC represents Northeast China; IM represents Inner Mongolia; SWC represents Southwest China; TB represents Tibet. SC is South China; XJ is Xinjiang; NWC is Northwest China; NC is North China; EC is East China; CC is Central China. Table 2. Optimal threshold values for removing outliers in different regions of China. Threshold Threshold −2 −1 −2 −1 District Representative cities (nW cm sr ) District Representative cities (nW cm sr ) NEC Shenyang, 177.19 XJ Urumqi 180.61 Changchun, Harbin IM Hohhot 107.23 NWC Xi’an, Lanzhou, 182.59 Yinchuan SWC Chengdu, Kunming, 184.81 NC Beijing, Taiyuan, 235.13 Guiyang Shijiazhuang TB Lhasa 165.81 EC Shanghai, Nanjing, 148.78 Jinan SC Guangzhou, 105.19 CC Wuhan, Zhengzhou, 116.88 Shenzhen, Nanning Changsha 20°0′0″N 30°0′0″N 40°0′0″N 10°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N GIScience & Remote Sensing 279 limited space, the original NPP-VIIRS data of selected cities within only five regions (North China, Northeast China, Northwest China, East China, and South China) are illustrated in Figure 3. 116°0′0″E 117°0′0″E 112°0′0″E 113°0′0″E 113°40′0″E 115°0′0″E (a) (b) (c) Radiance Radiance Radiance 235.13 96.01 114.64 Beijing 0 0 Taiyuan 0 Shijiazhuang 0 116°0′0″E 117°0′0″E 112°0′0″E 113°0′0″E 113°20′0″E 114°40′0″E 126°0′0″E 128°0′0″E 130°0′0″E 125°0′0″E 126°0′0″E 127°0′0″E 122°0′0″E 123°0′0″E 124°0′0″E 125°0′0″E (d) (e) (f) Radiance Radiance Radiance 124.39 158.03 177.19 Shenyang Harbin 0 Changchun 0 0 125°0′0″E 127°0′0″E 129°0′0″E 124°0′0″E 125°0′0″E 127°0′0″E 122°20′0″E 123°20′0″E 124°20′0″E 126°0′0″E 108°0′0″E 109°0′0″E 103°0′0″E 104°0′0″E 105°40′0″E 106°0′0″E 106°20′0″E 106°40′0″E (g) (h) (i) Radiance 128.23 Radiance Radiance 135.34 182.59 Xi’an Lanzhou Yinchuan 0 0 108°0′0″E 109°0′0″E 103°0′0″E 104°0′0″E 105°40′0″E 106°0′0″E 106°20′0″E 106°40′0″E 117°0′0″E 121°0′0″E 122°0′0″E 119°0′0″E 120°0′0″E 116°20′0″E 117°40′0″E 118°20′0″E (j) (k) (l) Radiance Radiance Radiance 148.78 115.57 47.01 Jinan Shanghai Nanjing 0 0 0 122°0′0″E 118°0′0″E 116°40′0″E 117°20′0″E 118°0′0″E 121°0′0″E 119°0′0″E 116°0′0″E 113°0′0″E 114°0′0″E 114°0′0″E 108°20′0″E 109°20′0″E 110°20′0″E (n) (m) (o) Radiance Radiance Radiance 103.76 105.19 70.58 Nanning Guangzhou 0 Shenzhen 0 0 113°0′0″E 114°0′0″E 114°0′0″E 107°20′0″E 108°20′0″E 109°20′0″E 110°20′0″E Figure 3. The original NPP-VIIRS data of the selected cities for five different regions in 2012. (a), (b), and (c) are the representative cities in North China; (d), (e), and (f) are the representative cities in Northeast China; (g), (h), and (i) are the representative cities in Northwest China; (j), (k), and (l) are the representative cities in East China; (m), (n), and (o) are the representative cities in South China; The solid lines in all figures are the administrative boundaries. 41°0′0″N 23°0′0″N 24°0′0″N 31°0′0″N 34°0′0″N 35°0′0″N 47°0′0″N 40°0′0″N 45°0′0″N 46°0′0″N 23°0′0″N 31°0′0″N 40°0′0″N 41°0′0″N 34°0′0″N 35°0′0″N 44°0′0″N 45°0′0″N 46°0′0″N 23°0′0″N 32°0′0″N 38°0′0″N 36°0′0″N 37°0′0″N 44°0′0″N 45°0′0″N 23°0′0″N 44°0′0″N 45°0′0″N 38°0′0″N 32°0′0″N 36°0′0″N 37°0′0″N 22°20′0″N 23°20′0″N 24°20′0″N 36°40′0″N 37°20′0″N 41°40′0″N 42°20′0″N 43°0′0″N 38°0′0″N 38°40′0″N 38°20′0″N 38°40′0″N 22°0′0″N 23°0′0″N 24°0′0″N 36°20′0″N 37°0′0″N 38°20′0″N 41°20′0″N 42°0′0″N 42°40′0″N 37°40′0″N 38°20′0″N 38°40′0″N 280 K. Shi et al. After that, each pixel whose radiance value was larger than that of the representative cities in the each region was modified to be a new value determined by its surrounding pixel values. For detailed approaches to correct the original NPP-VIIRS data and depict the corrected results, readers are recommended to refer to Shi, Yu, et al. (2014). After this process, all pixel values in the corrected NPP-VIIRS data of China were less than 235.13. 3.2. Establishment of a regression model There are many linear regression models commonly used by related studies to estimate socioeconomic factors using the nighttime light data (Cheng et al. 2014; Shi, Yu, et al. 2014). After investigating various existing models, a linear relationship between the log of TFT and the log of TNT was regarded as better correlating between TFT and TNT to accurately estimate TFT using nighttime light composite data at the provincial level: lgT ¼ wlgE þ c (1) where T is the statistical TFT of a provincial unit, E represents the TNL, which is calculated by the sum of all pixels in a provincial unit, and w and c are parameters determined by the regression analysis of sample data. Next, relative errors were examined to evaluate the capacity of TNL in estimating TFT as P  T RE ¼  100% (2) where RE is the relative error, P represents the estimated TTF, and T denotes the statistical TFT. 3.3. Spatialization of TFT To map the spatial distribution of TFT, TFT from the province level needs to be disaggregated to the pixel level based on the assumption that a more developed area generally has brighter lights and larger freight traffic demands. With reference to the method of Tian et al. (2014), each provincial TFT is disaggregated to individual pixel based on a pixel’s radiance value (or a DN value): ST HP ¼  R (3) SL where HP represents the amount of distributed TFT of a pixel, ST is a provincial p i statistical TFT, SL represents TNL of a province, and R is a radiance value (or a DN value) of a pixel. 4. Results 4.1. Regression results The regression results between TNL and TFT are shown in Figure 4. The R of the corrected NPP-VIIRS data and TFT was 0.817 (Figure 4a), whereas that of the TNL from GIScience & Remote Sensing 281 4 4 y = 0.3006x + 1.2397 y = 1.3108x – 4.2581 y = 1.1719x – 4.5163 2 2 2 R = 0.028 R = 0.817 R = 0.746 3 3 3 2 2 2 1 1 (a) (c) (b) 0 0 45 6 7 45 6 7 lg TNL of corrected NPP-VIIRS lg TNL of DMSP-OLS lg TNL of original NPP-VIIRS Figure 4. Relationships between the total freight traffic (TFT) and the total nighttime light (TNL) for 31 province units in China. (a) The TFT and the TNL of corrected NPP-VIIRS; (b) the TFT and the TNL of DMSP-OLS; (c) the TFT and the TNL of original NPP-VIIRS. DMSP-OLS data and TFT is 0.746 (Figure 4b). Moreover, the R of the TNL from the original NPP-VIIRS data and TFT is as low as 0.028 (Figure 4c). The results indicate that the corrected NPP-VIIRS can better reflect the TFT than the DMSP-OLS data and the original NPP-VIIRS data at the provincial level. Table 3 shows the absolute and relative errors calculated based on the regression analysis. On the whole, the absolute errors of the corrected NPP-VIIRS data are much lower than those of the DMSP-OLS data. There are nine provincial units with low absolute errors less than 100 million tons when TFT is predicted from the corrected NPP-VIIRS data. Contrarily, the DMSP-OLS data exhibit a low capacity in predicting TFT, with five provincial units with low absolute errors less than 100 million tons. Meanwhile, it can also be seen that the relative errors of the estimated total freight traffic (ETFT) in 2012 derived from the DMSP-OLS data in Tibet and Shaanxi are 312.63% and −0.70%, respectively, which is very different. By contrast, the counterparts of ETFT in 2012 derived from the corrected NPP-VIIRS data are 221.60% and −0.92%, respectively. To evaluate the capability of the two types of nighttime light data more comprehen- sively, the absolute relative errors (AREs) were divided into four classes: 0–20% as low, 20–40% as moderate, 40–60% as high, >60% as very high. The percentage of each class was calculated for the different estimation approaches. For instance, when calculating TFT from the corrected NPP-VIIRS data, there are 10 provincial units with low among the 31 provinces, and the ARE of low is 32.26%. Based on the ARE, the predictability for different types of nighttime light data has been quantified in Table 4. The corrected NPP- VIIRS data showed a fine capacity in estimating TFT, with low and moderate percentage of 32.26% and 29.04%, respectively, which are relatively higher than those of the DMSP- OLS data (29.03% and 22.58%, respectively). Moreover, the percentages of high and very high for TFT using the corrected NPP-VIIRS (19.35% and 19.35%, respectively) are lower than those using the DMSP-OLS data (25.81% and 22.58%, respectively). From the above analysis, we can see that the corrected NPP-VIIRS data have a better ability in estimating TFT than the DMSP-OLS data at the provincial level. The better estimated results mainly benefit from the higher radiometric detection range and higher resolution of the NPP-VIIRS data. However, a series of obvious errors still exist. As listed in Table 3, Beijing, Jiangsu, Guangdong, Zhejiang, and Tianjin are all comparatively overestimated (Table 3). The five provinces are the economically more developed pro- vinces in China. They consume and produce the largest quantities of goods. However, these provinces mainly produce high-tech products and light industrial goods that lead to high outputs (GDP) but relatively small total weights of the finished products as well as lg TFT (million tons) lg TFT (million tons) lg TFT (million tons) 282 K. Shi et al. Table 3. Accuracy of the estimated total freight traffic at the provincial level of China. Corrected NPP-VIIRS DMSP-OLS TFT ETFT ETFT (million (million AE RE (million AE RE (million tons) (%) (million tons) (%) Province tons) tons) tons) Beijing 261.62 559.28 297.66 113.77 463.50 201.88 77.17 Tianjin 460.15 554.63 94.48 20.53 418.17 −41.98 −9.12 Hebei 2191.30 2047.36 −143.94 −6.57 2653.22 461.92 21.08 Shanxi 1446.08 1477.74 31.66 2.19 1607.77 161.69 11.18 Inner 1899.42 1128.62 −770.8 −40.58 1717.66 −181.76 −9.57 Mongolia Liaoning 2067.89 1472.44 −595.45 −28.79 1590.74 −477.15 −23.07 Jilin 548.08 555.26 7.18 1.31 805.52 257.44 46.97 Heilongjiang 652.31 1016.37 364.06 55.81 1888.70 1236.39 189.54 Shanghai 940.38 1022.53 82.15 8.74 369.23 −571.15 −60.74 Jiangsu 2200.08 4560.24 2360.16 107.28 3512.43 1312.35 59.65 Zhejiang 1918.17 2909.02 990.85 51.66 2080.81 162.64 8.48 Anhui 3124.37 1392.42 −1731.95 −55.43 1336.44 −1787.93 −57.23 Fujian 843.45 1537.49 694.04 82.29 1189.11 345.66 40.98 Jiangxi 1271.96 621.12 −650.84 −51.17 601.41 −670.55 −52.72 Shandong 3336.03 3028.49 −307.54 −9.22 3789.21 453.18 13.58 Henan 2721.15 1914.87 −806.28 −29.63 2405.64 −315.51 −11.59 Hubei 1229.45 1024.48 −204.97 −16.67 892.30 −337.15 −27.42 Hunan 1910.52 1006.64 −903.88 −47.31 713.32 −1197.2 −62.66 Guangdong 2560.77 4814.57 2253.8 88.01 3310.67 749.9 29.28 Guangxi 1613.56 1092.42 −521.14 −32.30 917.49 −696.07 −43.14 Hainan 268.80 216.46 −52.34 −19.47 291.45 22.65 8.43 Chongqing 864.74 556.22 −308.52 −35.68 377.69 −487.05 −56.32 Sichuan 1743.49 1436.96 −306.53 −17.58 1089.28 −654.21 −37.52 Guizhou 526.55 355.25 −171.3 −32.53 402.11 −124.44 −23.63 Yunnan 687.35 931.85 244.5 35.57 1137.46 450.11 65.49 Tibet 11.27 36.23 24.96 221.60 46.49 35.22 312.63 Shaanxi 1367.27 1354.65 −12.62 −0.92 1357.65 −9.62 −0.70 Gansu 458.32 481.17 22.85 4.99 655.06 196.74 42.93 Qinghai 134.84 102.44 −32.4 −24.03 140.80 5.96 4.42 Ningxia 411.13 266.61 −144.52 −35.15 277.20 −133.93 −32.58 Xinjiang 587.94 1155.21 567.27 96.49 1403.22 815.28 138.67 Note: TFT is the statistical total freight traffic, ETFT is the estimated total freight traffic, AE is the absolute error, and RE is the relative error in percentage. Table 4. Different classes of estimated accuracies for total freight traffic (TFT) at the provincial level. NPP-VIIRS and TFT DMSP-OLS and TFT Class Quantity ARE (%) Quantity ARE (%) Very high 6 19.35 7 22.58 High 6 19.35 8 25.81 Moderate 9 29.04 7 22.58 Low 10 32.26 9 29.03 Note: ARE is the absolute relative error in percentage. GIScience & Remote Sensing 283 the materials required for their production (Tian et al. 2014). Weights of freight traffic denoted by one unit of TNL in these provinces were smaller than that of nationwide average level. Thus, relatively overestimation appears when TNL was used to estimate TFT in those provinces. Meanwhile, estimated results of Tibet, Heilongjiang, and Xinjiang provinces are much higher than the recorded statistics. There were many deserts, snowy mountains, and oilfields in those three provinces, the overestimation might be ascribed to the negative effects caused by the background noises of NPP-VIIRS data (Shi, Yu, et al. 2014). Although our proposed method could remove a lot of outliers, the residual noises still affect the accuracy of estimation. On the contrary, some provinces’ TFT, including Inner Mongolia, Liaoning, Anhui and Jiangxi, were underestimated. Those provinces are moderately developed provinces and provide raw materials in China. Every year goods with large masses and volumes, such as coal, iron ore and timber, are shipped via railways to these provinces, which lead to relatively large TFT but correspond to very limited light from train stations. Thus, although these provinces have relatively higher TFT, their TFT are still underestimated due to relatively low TNL. Actually, as percen- tages of railway freight traffic (RFT) to TFT are very small (i.e., 9.67% in 2012), TNL still have strong relationship with TFT at the provincial level in China. 4.2. Spatialization results Based on the spatialization model equation in Section 3.3, we then produced the spatial maps of 2012 TFT by using the corrected NPP-VIIRS data and DMSP-OLS data, respectively (Figure 5). There is a significant difference between the two maps. In the corrected NPP-VIIRS spatial TFT map (Figure 5a), the high TFTs are concentrated in some inland metropolitan areas and industrial cities, whereas the TFTs are relatively dispersed in the DMSP-OLS spatial TFT map (Figure 5b). This could be easily discovered from a visual comparison among the corrected NPP-VIIRS data, DMSP-OLS data and a fine resolution reference, Landsat 8 OLI-TIRS images (Figure 6). The four sampling cities with different levels of urbanization and socioeconomic development include Shanghai (located in Eastern China, economic center), Nanjing (located in Eastern China, the capital of Jiangsu province), Maanshan (located in central China, a city of Anhui province in middle size) and Hechuan (located in western China, a small city of Chongqing city). The relative high TFT of the four cities spread out to the urban areas including urban forests and rural areas in the DMSP-OLS data (Figure 6e–h), regardless of the developed levels and locations of the cities. On the contrary, the high TFT in the corrected NPP-VIIRS data (Figure 6i–l) are concentrated in urban areas which compared to the Landsat 8 OLI-TIRS images (Figure 6a–d). As we known, the urban areas usually have larger population and produce and consume more goods which lead to a large freight transport demand. Thus, we can conclude that TFT is mainly concentrated in urban areas, while urban forests and rural areas only have small TFT. Consequently, it could be easily certified that the corrected NPP-VIIRS data can reflect human activities and support a more accurate spatialization of TFT. Figures 5 and 6 showed the pixel-based TFT estimated from provincial statistical data. We then used prefectural-level statistical TFT to validate the spatial TFT results in Figure 5. We collected 244 prefectural TFT and used the prefecture boundary to calculate the TFT for each prefecture from spatial TFT maps. Two indicators, root-mean-square ), were used to evaluate the accuracies of TFT error (RMSE) and regression coefficient (R maps. Figure 7 shows the validation results of estimated TFT for 244 prefectures. Results derived from the corrected NPP-VIIRS data in Figure 7a have higher R (0.674) and lower 284 K. Shi et al. 60°0′0″E 70°0′0″E 80°0′0″E 90°0′0″E 110°0′0″E 130°0′0″E 140°0′0″E 150°0′0″E TFT (million tons) 0 – 1 1 – 2 2 – 3 3 – 4 (a) South China Sea >4 80°0′0″E 90°0′0″E 100°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E 60°0′0″E 80°0′0″E 90°0′0″E 110°0′0″E 130°0′0″E 140°0′0″E 150°0′0″E 70°0′0″E TFT (million tons) 0 – 1 1 – 2 2 – 3 3 – 4 (b) >4 South China Sea 80°0′0″E 90°0′0″E 100°0′0″E 110°0′0″E 120°0′0″E 130°0′0″E Figure 5. The TFT maps estimated from nighttime light composite data. (a) is the corrected NPP-VIIRS spatial TFT map; (b) is the DMSP-OLS spatial TFT map. RMSE (93.87) than those from DMSP-OLS data in Figure 7b (R of 0.207, RMSE of 216.55). They have proved that the corrected NPP-VIIRS data have more accurate results than the DMSP-OLS data at the pixel level. 5. Discussion The study is based on the assumption that there is a positive relationship between TFT and nighttime lights. Because the developed regions usually have a large population, advanced industry, and convenient transportation, they would produce and consume more goods which lead a large freight transport demand (McKinnon and Woodburn 1996; Tian et al. 2014). Meanwhile, it was well-known that nighttime lights could be used to model socioeconomic indicators, such as GDP and population (Tian et al. 2014; Shi, Yu, et al. 2014). Therefore, we proposed that the nighttime light data could be as a proxy for TFT estimation. In this study, regression analysis was firstly employed to quantify the relation- ship between TFT and TNL. The results have proved that regression analysis is simple but 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N 20°0′0″N 30°0′0″N 40°0′0″N GIScience & Remote Sensing 285 118°0′0″E 119°0′0″E 121°0′0″E 122°0′0″E 118°20′0″E 118°40′0″E 119°0′0″E 106°0′0″E 106°20′0″E 106°40′0″E (a) (b) (c) (d) 121°0′0″E 122°0′0″E 118°0′0″E 119°0′0″E 118°30′0″E 118°50′0″E 106°0′0″E 106°20′0″E 106°40′0″E 121°0′0″E 122°0′0″E 118°0′0″E 119°0′0″E 118°20′0″E 118°40′0″E 119°0′0″E 106°0′0″E 106°20′0″E 106°40′0″E (e) (f) (g) (h) 121°0′0″E 122°0′0″E 118°0′0″E 119°0′0″E 118°30′0″E 118°50′0″E 106°0′0″E 106°20′0″E 106°40′0″E 119°0′0″E 121°0′0″E 122°0′0″E 118°0′0″E 118°20′0″E 118°40′0″E 119°0′0″E 106°0′0″E 106°20′0″E 106°40′0″E (i) (j) (k) (l) 122°0′0″E 121°0′0″E 118°0′0″E 119°0′0″E 118°30′0″E 118°50′0″E 106°0′0″E 106°20′0″E 106°40′0″E TFT (million tons) 0 0 – 1 3 – 4 >4 1 – 2 2 – 3 Figure 6. Comparisons among the Landsat 8 OLI-TIRS images, the DMSP-OLS data and the corrected NPP-VIIRS data of four typical cities (Shanghai, Nanjing, Maanshan, and Hechuan). (a), (b), (c), and (d) are Landsat 8 OLI-TIRS images; (e), (f), (g), and (h) are DMSP-OLS spatial TFT maps; (i), (j), (k), and (l) are corrected NPP-VIIRS TFT maps. The solid lines in all figures are the administrative boundaries. y = 0.6726x + 5179 RMSE = 93.87 y = 0.2954x + 8857 RMSE = 216.55 2 2 R = 0.674 (million tons) 800 (million tons) 800 R = 0.207 600 600 400 400 200 200 (a) (b) 0 0 0 2 00 4 00 6 00 800 0 200 400 600 800 1000 1200 Estimated TFT by DMSP-OLS (milliontons) Estimated TFT by corrected NPP-VIIRS (million tons) Figure 7. Relationships between estimated and statistical TFT for 244 prefectures of China. (a) is the relationship between TFT estimated from corrected NPP-VIIRS data and statistical TFT; (b) is the relationship between TFT estimated from DMSP-OLS data and statistical TFT. reliable in estimating TFT at the provincial level in China. In addition, since each provincial TFT was disaggregated to individual pixel based on a pixel value, the disag- gregated TFT could confine the estimated errors within a provincial unit. Spatial 31°0′0″N 32°0′0″N 31°0′0″N 32°0′0″N 31°0′0″N 32°0′0″N Statistical TFT (million tons) 31°0′0″N 31°0′0″N 31°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 32°0′0″N 31°20′0″N 31°40′0″N 31°40′0″N 31°20′0″N 31°40′0″N 31°20′0″N Statistical TFT (million tons) 31°30′0″N 31°30′0″N 31°30′0″N 30°0′0″N 30°20′0″N 30°0′0″N 30°20′0″N 30°0′0″N 30°20′0″N 30°0′0″N 30°0′0″N 30°20′0″N 30°20′0″N 30°0′0″N 30°20′0″N 286 K. Shi et al. variations of TFT could be clearly exhibited by TFT maps. The amount of TFT gradually decreases from urban areas to suburban areas. Thus, the NPP-VIIRS data were also relatively creditable in modeling TFT at the pixel level. At the same time, although our study has proved that our method has a good performance in modeling TFT at the multiple scales, it still contains a few limitations. For example, we were not confident enough in identifying the source of errors caused by confusion of brightness values between a central business district (CBD) and a dense forested area. The major contributions of this study are twofold. First, previous studies have proved that the DMSP-OLS data have a potential to be used as a proxy for TFT estimation (Tian et al. 2014), but no related work reported the ability of NPP-VIIRS data to estimate TFT. This study has demonstrated that the NPP-VIIRS data have a stronger capacity in modeling TFT than those of the DMSP-OLS data for the first time. This finding will benefit the further utilization of NPP-VIIRS data and provide a powerful tool for the socioeconomic indicator estimation using remotely sensed data. Second, for the released NPP-VIIRS data have not been filtered to remove light detections associated with fires, gas flares, volcanoes, and background noise, these noises could limit the accuracy and reliability in TFT estimation. In this study, a new correction method integrating DMSP- OLS data and regional optimal thresholds are proposed. This method takes the regional difference in landscape, socioeconomic status, and atmospheric condition into account, and could not only remove obvious outliers, but also reserve most of pixels with correct radiance values. Since the NPP-VIIRS data are a newly released data source, only three scenes of images are available, including the composites of April and October in 2012, January in 2013, April in 2014, which are insufficient to conduct a comprehensive evaluation. As NOAA/NGDC is working to produce more NPP-VIIRS data with higher quality, further research can be focused on multitemporal image analysis for disclosing the spatiotemporal pattern of TFT. 6. Conclusions TFT is an important socioeconomic indicator for understanding the development of transport infrastructures and regional economic status. Therefore, accurate information about TFT in a timely manner has been regarded as an important and challenging task for the government department as well as academic community. This study highlighted the ability of using the NPP-VIIRS data for modeling and mapping TFT in China. Based on a simple method to remove the outliers of original data, the corrected NPP-VIIRS data were generated. The sum value of all pixels from the nighttime light composite data in a provincial unit were regressed on the TFT at the provincial level of China. The estimation results suggest the R of TFT between the corrected NPP-VIIRS data and statistical data is as high as 0.817, which is higher than that of the value between the DMSP-OLS data and statistical data (R = 0.746). The spatial TFT maps were produced from the corrected NPP-VIIRS data and DMSP-OLS data, respectively. The spatialization results of TFT also prove that the corrected NPP-VIIRS data are able to produce more accurate spatial TFT map in comparison with the DMSP-OLS data. The validation results show that the corrected NPP-VIIRS data have a higher R (0.674) and a lower RMSE (93.87), whereas the DMSP-OLS data have the worse results with R of 0.207 and RMSE of 216.55. In conclusion, this study reveals that the corrected NPP-VIIRS data are a more reliable data source for modeling TFT in China. GIScience & Remote Sensing 287 Acknowledgments The authors thank three anonymous reviewers and the editor for their constructive comments and suggestions. Disclosure statement No potential conflict of interest was reported by the authors. Funding This study is supported by the National Natural Science Foundation of China [grant number 41471449]; the Natural Science Foundation of Shanghai [grant number 14ZR1412200]; the Innovation Program of Shanghai Municipal Education Commission [grant number 15ZZ026]; the Fundamental Research Funds for the Central Universities of China; the State Scholarship Fund sponsored by the China Scholarship Council [grant number 201406140007]. ORCID Bailang Yu http://orcid.org/0000-0001-5628-0003 Chang Huang http://orcid.org/0000-0002-8621-4581 References Baugh, K., F.-C. Hsu, C. D. Elvidge, and M. Zhizhin. 2013. “Nighttime Lights Compositing Using the VIIRS Day-night Band: Preliminary Results.” Proceedings of the Asia-Pacific Advanced Network 35: 70–86. doi:10.7125/APAN.35.8. Butt, M. J. 2012. “Estimation of Light Pollution Using Satellite Remote Sensing and Geographic Information System Techniques.” GIScience & Remote Sensing 49 (4): 609–621. doi:10.2747/ 1548-1603.49.4.609. Cheng, Y., Z. Wang, X. Ye, and Y. Wei. 2014. “Spatiotemporal Dynamics of Carbon Intensity from Energy Consumption in China.” Journal of Geographical Sciences 24 (4): 631–650. doi:10.1007/s11442-014-1110-6. Elvidge, C. D., K. E. Baugh, J. B. Dietz, T. Bland, P. C. Sutton, and H. W. Kroehl. 1999. “Radiance Calibration of DMSP-OLS Low-light Imaging Data of Human Settlements.” Remote Sensing of Environment 68 (1): 77–88. doi:10.1016/S0034-4257(98)00098-4. Elvidge, C. D., K. E. Baugh, E. A. Kihn, H. W. Kroehl, and E. R. Davis. 1997. “Mapping City Lights with Nighttime Data from the DMSP Operational Linescan System.” Photogrammetric Engineering and Remote Sensing 63 (6): 727–734. Elvidge, C. D., K. E. Baugh, M. Zhizhin, and F. Hsu. 2013. “Why VIIRS Data Are Superior to DMSP for Mapping Nighttime Lights.” Proceedings of the Asia-Pacific Advanced Network 35: 62–69. doi:10.7125/APAN.35.7. Elvidge, C. D., M. Zhizhin, F. Hsu, and K. Baugh. 2013. “What Is So Great About Nighttime VIIRS Data for the Detection and Characterization of Combustion Sources?” Proceedings of the Asia- Pacific Advanced Network 35: 33–48. doi:10.7125/APAN.35.5. Forbes, D. J. 2013. “Multi-scale Analysis of the Relationship Between Economic Statistics and DMSP-OLS Night Light Images.” GIScience & Remote Sensing 50 (5): 483–499. Guler, H., and Y. Vitosoglu. 2012. “Estimation of Freight Transportation.” Proceedings of the ICE- Transport 166 (3): 174–185. doi:10.1680/tran.10.00065. He, C., Q. Ma, T. Li, Y. Yang, and Z. F. Liu. 2012. “Spatiotemporal Dynamics of Electric Power Consumption in Chinese Mainland from 1995 to 2008 Modeled Using DMSP/OLS Stable Nighttime Lights Data.” Journal of Geographical Sciences 22 (1): 125–136. doi:10.1007/ s11442-012-0916-3. He, C., Q. Ma, Z. Liu, and Q. Zhang. 2013. “Modeling the Spatiotemporal Dynamics of Electric Power Consumption in Mainland China Using Saturation Corrected DMSP/OLS Nighttime Stable Light Data.” International Journal of Digital Earth 6 (1): 1–22. 288 K. Shi et al. Hillger, D., T. Kopp, T. Lee, D. Lindsey, C. Seaman, S. Miller, J. Solbrig, et al. 2013. “First-Light Imagery from Suomi NPP VIIRS.” Bulletin of the American Meteorological Society 94 (7): 1019–1029. doi:10.1175/bams-d-12-00097.1. Letu, H., M. Hara, H. Yagi, K. Naoki, G. Tana, F. Nishio, and O. Shuhei. 2010. “Estimating Energy Consumption from Night-time DMPS/OLS Imagery after Correcting for Saturation Effects.” International Journal of Remote Sensing 31 (16): 4443–4458. doi:10.1080/ Li, X., F. Chen, and X. Chen. 2013. “Satellite-Observed Nighttime Light Variation as Evidence for Global Armed Conflicts.” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 6 (5): 2302–2315. doi:10.1109/jstars.2013.2241021. Li, X., L. Ge, and X. Chen. 2014. “Quantifying Contribution of Land Use Types to Nighttime Light Using an Unmixing Model.” IEEE Geoscience and Remote Sensing Letters 11 (10): 1667–1671. doi:10.1109/lgrs.2014.2304496. Li, X., H. Xu, X. Chen, and C. Li. 2013. “Potential of NPP-VIIRS Nighttime Light Imagery for Modeling the Regional Economy of China.” Remote Sensing 5 (6): 3057–3081. doi:10.3390/ rs5063057. Lindholm, M. 2010. “A Sustainable Perspective on Urban Freight Transport: Factors Affecting Local Authorities in the Planning Procedures.” Procedia-Social and Behavioral Sciences 2 (3): 6205–6216. doi:10.1016/j.sbspro.2010.04.031. Liu, Z., C. He, Q. Zhang, Q. Huang, and Y. Yang. 2012. “Extracting the Dynamics of Urban Expansion in China Using DMSP-OLS Nighttime Light Data from 1992 to 2008.” Landscape and Urban Planning 106 (1): 62–72. doi:10.1016/j.landurbplan.2012.02.013. Long, D., and V. P. Singh. 2013. “An Entropy-Based Multispectral Image Classification Algorithm.” IEEE Transactions on Geoscience and Remote Sensing 51 (12): 5225–5238. doi:10.1109/ tgrs.2013.2272560. Ma, T., C. Zhou, T. Pei, S. Haynie, and J. Fan. 2014. “Responses of Suomi-NPP VIIRS-Derived Nighttime Lights to Socioeconomic Activity in China’s Cities.” Remote Sensing Letters 5 (2): 165–174. doi:10.1080/2150704x.2014.890758. McKinnon, A. C., and A. Woodburn. 1996. “Logistical Restructuring and Road Freight Traffic Growth.” Transportation 23 (2): 141–161. doi:10.1007/BF00170033. Min, B., K. M. Gaba, O. F. Sarr, and A. Agalassou. 2013. “Detection of Rural Electrification in Africa Using DMSP-OLS Night Lights Imagery.” International Journal of Remote Sensing 34 (22): 8118–8141. doi:10.1080/01431161.2013.833358. Propastin, P., and M. Kappas. 2012. “Assessing Satellite-Observed Nighttime Lights for Monitoring Socioeconomic Parameters in the Republic of Kazakhstan.” GIScience & Remote Sensing 49 (4): 538–557. doi:10.2747/1548-1603.49.4.538. Schroeder, W., P. Oliva, L. Giglio, and I. A. Csiszar. 2014. “The New VIIRS 375m Active Fire Detection Data Product: Algorithm Description and Initial Assessment.” Remote Sensing of Environment 143: 85–96. doi:10.1016/j.rse.2013.12.008. Shi, K., C. Huang, B. Yu, B. Yin, Y. Huang, and J. Wu. 2014. “Evaluation of NPP-VIIRS Night- time Light Composite Data for Extracting Built-up Urban Areas.” Remote Sensing Letters 5 (4): 358–366. doi:10.1080/2150704X.2014.905728. Shi, K., B. Yu, Y. Huang, Y. Hu, B. Yin, Z. Chen, L. Chen, and J. Wu. 2014. “Evaluating the Ability of NPP-VIIRS Nighttime Light Data to Estimate the Gross Domestic Product and the Electric Power Consumption of China at Multiple Scales: A Comparison with DMSP-OLS Data.” Remote Sensing 6 (2): 1705–1724. doi:10.3390/rs6021705. Small, C., F. Pozzi, and C. D. Elvidge. 2005. “Spatial Analysis of Global Urban Extent from DMSP- OLS Night Lights.” Remote Sensing of Environment 96 (3–4): 277–291. doi:10.1016/j. rse.2005.02.002. Tian, J., N. Zhao, E. L. Samson, and S. Wang. 2014. “Brightness of Nighttime Lights as a Proxy for Freight Traffic: A Case Study of China.” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 7 (1): 206–212. doi:10.1109/jstars.2013.2258892. Wang, X., P. Xiao, X. Fen, and H. Li. 2013. “Extraction of Large Scale Urban Area Information in China Using DMSP-OLS Nighttime Light Data.” Remote Sensing for Land & Resources 25 (3): 159–164. Wei, Y., H. Liu, W. Song, B. Yu, and C. Xiu. 2014. “Normalization of Time Series DMSP-OLS Nighttime Light Images for Urban Growth Analysis with Pseudo Invariant Features.” Landscape and Urban Planning 128: 1–13. doi:10.1016/j.landurbplan.2014.04.015. GIScience & Remote Sensing 289 Witmer, F. D. W., and J. O’Loughlin. 2011. “Detecting the Effects of Wars in the Caucasus Regions of Russia and Georgia Using Radiometrically Normalized DMSP-OLS Nighttime Lights Imagery.” GIScience & Remote Sensing 48 (4): 478–500. doi:10.2747/1548-1603.48.4.478. Wu, W., Y. Cao, S. Liang, and C. Xiao. 2011. “Study on Dynamics of Freight Volume in the Yangtze River Delta since the Mid-1980s.” Resource and Environment in the Yangtze Basin 20 (7): 843–847. Yang, Y., C. He, Q. Zhang, L. Han, and S. Du. 2013. “Timely and Accurate National-scale Mapping of Urban Land in China Using Defense Meteorological Satellite Program’s Operational Linescan System Nighttime Stable Light Data.” Journal of Applied Remote Sensing 7 (1): 073535. doi:10.1117/1.JRS.7.073535. Yu, B., S. Shu, H. Liu, W. Song, J. Wu, L. Wang, and Z. Chen. 2014. “Object-based Spatial Cluster Analysis of Urban Landscape Pattern Using Nighttime Light Satellite Images: A Case Study of China.” International Journal of Geographical Information Science 28 (11): 2328–2355. doi:10.1080/13658816.2014.922186. Zhang, Q., C. He, and Z. Liu. 2014. “Studying Urban Development and Change in the Contiguous United States Using Two Scaled Measures Derived from Nighttime Lights Data and Population Census.” GIScience & Remote Sensing 51 (1): 63–82. doi:10.1080/15481603.2014.883212.

Journal

GIScience & Remote SensingTaylor & Francis

Published: May 4, 2015

Keywords: NPP-VIIRS; DMSP-OLS; total freight traffic; nighttime light; China

There are no references for this article.