A Saturated Light Correction Method for DMSP ... - Semantic Scholar
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
389
A Saturated Light Correction Method for DMSP/OLS Nighttime Satellite Imagery Husi Letu, Masanao Hara, Gegen Tana, and Fumihiko Nishio
Abstract—Several studies have clarified that electric power consumption can be estimated from the Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) stable light imagery. As digital numbers (DNs) of stable light images are often saturated in the center of city areas, we developed a saturated light correction method for the DMSP/OLS stable light image using the nighttime radiance calibration image of the DMSP/OLS. The comparison between the nonsaturated part of the stable light image for 1999 and the radiance calibration image for 1996–1997 in major areas of Japan showed a strong linear correlation (R2 = 92.73) between the DNs of both images. Saturated DNs of the stable light image could therefore be corrected based on the correlation equation between the two images. To evaluate the new saturated light correction method, a regression analysis is performed between statistic data of electric power consumption from lighting and the cumulative DNs of the stable light image before and after correcting for the saturation effects by the new method, in comparison to the conventional method, which is, the cubic regression equation method. The results show a stronger improvement in the determination coefficient with the new saturated light correction method (R2 = 0.91, P = 1.7 · 10−6 < 0.05) than with the conventional method (R2 = 0.81, P = 2.6 · 10−6 < 0.05) from the initial correlation with the uncorrected data (R2 = 0.70, P = 4.5 · 10−6 < 0.05). The new method proves therefore to be very efficient for saturated light correction. Index Terms—Defense Meteorological Satellite (DMSP), Operational Linescan System (OLS), calibration image, saturated light, stable light image.
Program radiance
I. I NTRODUCTION
S
ATELLITES can uniformly observe a wide area in a short time and can monitor spatial land surface information and atmospheric environment at increasingly finer resolutions [1]–[8]. Among these satellites, the Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) has a unique capability for monitoring nighttime light present at the earth’s surface which includes human settlements (stable light), clouds affected by moonlight, forest fires, fishing boats and gas flares [9]. Stable light is one of the most important comManuscript received November 2, 2010; revised February 6, 2011, April 18, 2011, July 1, 2011, and August 29, 2011; accepted October 23, 2011. Date of publication December 27, 2011; date of current version January 20, 2012. H. Letu is with the Research and Information Center, Tokai University, Tokyo 151-0063, Japan (e-mail: [email protected]). M. Hara is with the VTI Research Institute, VisionTech Inc., Tsukuba City 305-0045, Japan (e-mail: [email protected]). G. Tana is with the Graduate School of Science, Chiba University, Chiba 263-8522, Japan (e-mail: [email protected]). F. Nishio is with the Centre for Environmental Remote Sensing (CEReS), Chiba University, Chiba 263-8522, Japan (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TGRS.2011.2178031
ponents for monitoring electric power consumption and economic activity, factors that rely mainly on statistical data. However, some countries and regions lack statistical data or have insufficiently accurate data on electric power consumption. Previous studies [10]–[15] have reported a good correlation between electric power consumption and digital numbers (DNs) of the DMSP/OLS stable light image. DNs of the stable light image are saturated in the center of city areas where nighttime light is very strong [16]; however, this saturation of DNs is not found in the nighttime radiance calibration DMSP/OLS images. Elvidge et al. [17] estimated electric power consumption from the radiance calibration image extracted from DMSP/OLS low-light nighttime imagery for 1996–1997 and pointed out that radiance calibration images can accurately estimate electric power consumption. Ziskin et al. [18] developed the new radiance calibration image for 2006. Although longterm radiance calibration images are not currently available, time-series stable light images can be obtained since 1992 from the National Oceanic and Atmosphere Administration (NOAA)/National Geophysical Data Center (NGDC) (http:// www.ngdc.noaa.gov/dmsp/download.html). Baugh et al. [19] developed a recent stable light product for 2009. For the monitoring of long-term changes of electric power consumption from nighttime images, the correction of the saturated DNs of the stable light images is needed. Using stable light images for 1993–2002, Chand et al. [20] analyzed the correlation between increased nighttime light and electric power consumption to evaluate electric power consumption patterns over India. However, they did not correct for the saturated DNs of the stable light images. Letu et al. [16] therefore developed a saturated light correction method using the cubic regression equation (CRE) for the correction of the DMSP/OLS-VIS stable light image of 1999. The CRE is obtained from the DNs of nonsaturated pixels by a four-dimensional simultaneous equation based on the least-square method. This method can correct the profile of DNs of saturated light but cannot estimate the saturated DNs at the image level. In this study, we developed a saturated light correction method of the stable light image using the radiance calibration nighttime image to estimate the electric power consumption. We calculated a regression function from the stable light image for 1999 and the radiance calibration images for 1996–1997. We assumed that nighttime light intensity did not change in the saturated DN areas of the stable light images during 1996–1999 and corrected the saturated DNs of the stable light image from the radiance calibration image. The cumulative DNs of the stable light image before and after the saturation light were estimated in the Kanto, Kansai, Chubu areas of Japan. To evaluate
0196-2892/$26.00 © 2011 IEEE
390
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
TABLE I S TATISTIC DATA OF THE E LECTRIC P OWER C ONSUMPTION F ROM L IGHTING FOR 1999 (×106 kwh)
this new method, regression analyses were performed between cumulative DNs of the stable light image and electric power consumption from lighting. II. S ATELLITE AND DATA C OLLECTION The DMSP/OLS is operated by the United States Department of Defense and acquires broadband visible near-infrared (VNIR) image data. With the OLS-VNIR data, it is possible to detect clouds illuminated by moonlight, lights from cities, towns, and industrial sites, gas flares, and ephemeral events such as fires and lighting-illuminated clouds [21]. The high contrast between lighted and unlighted areas and the sensor’s spatial resolution makes it a useful tool to identify areas of intense human activity [22], [23]. Each DMSP satellite has a 101-min period, which is sun-synchronous near the circular polar orbit with an altitude of 830 km above the surface of the earth. The OLS sensor collects images across a 3000-km swath, providing global coverage twice per day. The combination of day/night and dawn/dusk satellites allows monitoring of global information such as clouds and city lights every 6 h (NOAA/NGDC homepage: http://www.ngdc.noaa.gov/dmsp/). DMSP/OLSVIS 1999 data provided by the Satellite Image Data Base (SIDaB) of the Ministry of Agriculture, Forestry and Fisheries Agriculture Information Resources System (AGROPEDIA) of Japan (http://rms1.agsearch.agropedia.affrc.go.jp/sidab/indexja.html) were used to map the stable light image. The SIDaB converts 6-bit (0–63) OLS-VIS data distributed by the National Oceanic and Atmospheric Administration/National Geophysical Data Center (NOAA/NGDC) into 8-bit (0–255) data using the following function: DN8 bit = DN6 bit × 255/63. The primary mission of the OLS sensor is to observe the nighttime moon-lit cloud cover for global meteorological forecasting for the Air Force. It uses a photomultiplier tube at least four orders of magnitude more sensitive than those used in any other satellite system. The DMSP/OLS sensor monitors almost all the artificial nighttime light from urban areas [24]–[35]. Therefore, DNs of the OLS-VIS nighttime image in the center of city areas are easily saturated depending on the sensitivity of the sensor. Elvidge et al. [17] created a radiance calibration nighttime light image using DMSP/OLS low-light imaging data. This image was made using 28 nights of data for the period 1996–1997, downloaded from the NOAA/NGDC. Statistic data of the electric power consumption from lighting for 1999 was obtained from the Japan Business History Institute (http://www.jbhi.or.jp/toukei.html) and used for evaluating the saturated light correction method of the stable light images. Table I shows the electric power consumption from lighting
Fig. 1.
Electric power consumption area in Japan.
in the study areas. In this study, the spatial distribution of the nighttime stable light in Japan was divided according to the main electric power supply areas (Fig. 1). III. S ATURATED L IGHT C ORRECTION M ETHOD A. Generation of Stable Light Image and Radiance Calibration Image Fig. 2 shows the DNs of the stable light image and radiance calibration image for the Kanto area, Japan. DNs of both images were sorted and plotted in a gradually increasing order. Saturated DNs can be seen with the data from the stable light image, but not with the data from the radiance calibration image. To correct for the saturated DNs of the stable light image using the radiance calibration image, we compared the generation methods between the stable light image for 1999 and the radiance calibration image for 1996–1997 (Fig. 3). The procedure to generate the stable light image is as follows [14]. 1) A 36-piece composite image is created using the 10-day maximum value composition (MVC) method from DMSP/OLS time-series imagery for 1999. The 10-day MVC method extracts the maximum value of the DNs from the 10-day multitemporal layer satellite image. 2) Light from clouds illuminated by moonlight, gas flares, and ephemeral events such as fires and solar glare is
LETU et al.: SATURATED LIGHT CORRECTION METHOD
391
removed from the 36-piece composite image using a noise reduction filter (NRF). 3) The stable light image is extracted from the 36-piece time-series composite image using NRF. NRF finds periodic components from time-series satellite data, performs the correction of the deficit data, and removes abnormal values. In this study, NRF was used to extract the artificial nighttime light from DMSP data after removing clouds and other disturbance noise. Equation (1) expresses the function of the conventional NRF method. The NRF method calculates the estimated DNs (ft ) and the periodic components from DNs of the time-series 10-day MVC imagery. Detailed algorithms and processes for the image composite have been previously described by Hara et al. [13] N 2πkl 2πkl t +C2l+1 cos t C2l sin ft = C0 +C1 t+ M M l=1 (1)
Fig. 2. (a) DNs of stable light image for the year 1999 and (b) radiance calibration image for the period 1996–1997 in Kanto, Japan.
Fig. 3. areas.
Stable light image (a) and radiance calibration image (b) of the study
where C0 is the direct current component and is obtained from the least-square method; C1 , C2l , and C2l+1 are the coefficients; kl is the frequency k of the periodic component l (i.e., 1, 2, 3, 4, 6, or 12 months); N is the amount of data; M is the period; and t expresses the days (1–365 days). Conventionally, the direct current component (C0 ) means the stable light image. The amplitude is the 1-year cycle component noise in the direct current component except for the stable light image [6]. Thus, (2), a modification of (1), can be used for calculating the amplitude of the periodic components. The stable light image used in this study was generated by subtracting the amplitude of the 1-year cycle component from the direct current component, i.e., N 2πkl t + θl αl sin ft = C0 + C1 t + M l=2 C2l+1 2 + C2 θl = tan−1 (2) αl = C2l 2l+1 C2l where αl is the amplitude of the periodic component and θl is the phase of the periodic component. The basic procedure to generate the radiance calibration image is as follows [16]. 1) After eliminating clouds using thermal infrared (TIR) band data, the geolocation of the light is identified. For this step, an automatic light detection algorithm was developed to calculate light occurring from the time series cloud-free images. 2) Cloud-free compositing is performed within two overlapping gain ranges: high and low. The high and low gain composite images are filtered to eliminate many ephemeral events such as fires or lightning, which occur only once during light detection. 3) High and low gain cloud-free composites are combined and averaged. The radiance calibrated average DNs from each image are weighted by the total number of light detections. NRF was used to remove noise for generation of the stable light image. The methods for visually selecting a TIR data
392
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
Fig. 5. Relationship between DNs of the stable light image and the radiance calibration image.
Fig. 4. Electric power consumption from lighting during 1996–1999.
threshold are applied, and composition of the high- and lowgain images is filtered to eliminate the noise for processing the radiance calibration image. Therefore, noise components such as clouds affected by moonlight, forest fire, fishing boats, and gas flares can finally be eliminated from both images, and the MVC method can be used on both nighttime images. B. Correlation Between Stable Light Image and Radiance Calibration Image Stable light image for 1999 and radiance calibration image for 1996–1997 were generated by the same MVC method. To correct the saturated DN areas of the stable light image for 1999 using radiance calibration image for 1996–1997, we need to confirm that there was very little change in light intensity during 1996–1999. Fig. 4 shows the electric power consumption in each area of Japan during 1996–1999; these data were obtained from the Japan Business History Institute. The source of the nighttime light by human settlements was the electric power consumption from lighting. We can see that this light changes very little (< 9.5%) during 1996–1999. Ichinose et al. [36] previously investigated the intensity of the nighttime city light using the population density data of East Asia during 1992–2002. They reported that nighttime light intensity changed very little in the center of the city area. Thus, it is assumed that the DNs of the stable light image did not change in the saturated DN areas of Japan during 1996–1999. A regression analysis of DNs between the stable light image and radiance calibration image was therefore performed after plotting the DNs of both images in a gradually increasing order. Fig. 5 shows the relationship between DNs of the stable light image and radiance calibration image in the Kanto, Kansai, and Chubu areas. A linear relationship was found between DNs of the two images in each area for DNs of the radiance calibration image lower than 20. DNs of the stable light image were almost saturated when DNs of the radiance calibration image were larger than 20. Equation (3) is a regression function of the DNs of the nonsaturated part of stable and radiance calibrated imagery (Fig. 5). It was applied to adjust the DNs of the radiance calibration image to the stable light image for saturation light correction Ysta = 17Xrad − 102
(3)
Fig. 6.
Flowchart of the new saturated light correction method.
where Y is DNs of the stable light image, and X is DNs of the radiance calibration image. The stable light image for 1999 is 8-bit data downloaded from SIDaB. In the case of extracting the stable light image using 6-bit data from NOAA/NGDC, regression function (3) changes to Ysta = 63/255(17Xrad − 102). C. Saturated Light Correction Method Fig. 6 shows the flowchart of the saturated light correction method. Radiance calibration image for 1996–1997 was used for correcting the saturated DNs of the DMSP/OLS stable light image for 1999. We analyzed the radiance calibrated DNs in the saturated zone of the stable lights and applied regression equation (3) to the radiance calibration image for calculation of the corrected values. Details of data processing by this method comprise the following steps. 1) The regression equation (3) is derived from the nonsaturated part of the relationship between stable light image and radiance calibration image.
LETU et al.: SATURATED LIGHT CORRECTION METHOD
Fig. 7.
393
Comparison of the stable light image and the radiance calibration image in the Kanto, Kansai, and Chubu areas.
TABLE II C UMULATIVE DNs OF THE E LECTRIC P OWER S UPPLY A REA OF JAPAN
2) The DNs of the radiance calibration image is adjusted to the stable light image using the calculated regression equation (3). 3) The saturated DN area is extracted from the stable light image to generate the saturation area mask. 4) The DNs in the mask areas are extracted from the corrected radiance calibration image generated in step 2. 5) The saturated DNs of the stable light image are corrected with that of the extracted radiance calibration image.
IV. R ESULTS AND D ISCUSSION A. Comparison of the Stable Light Image After Correction for Saturation Effect The combined saturated area of the 1999 stable light image for the Kanto, Kansai, and Chubu areas is 87.03% of the total saturated area of Japan. This implies that the Kanto, Kansai, and Chubu areas are the main source of the nighttime light for all of Japan. Fig. 7 shows the radiance calibration image, original stable light image, and stable light image after correction in
394
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
the Kanto, Kansai, and Chubu areas. DNs of the radiance calibration image in the center of city areas were higher than those in the surrounding areas. In the stable light image, DNs were almost saturated in the center of city areas, and these saturated DN areas corresponded with the high DN area in the radiance calibration image. In the corrected stable light image, light intensity was clearly distributed in the center of city areas. Table II shows the cumulative DNs in each area of Japan. The cumulative DNs increased after the saturation effects were removed. The ratio of the cumulative DNs, the difference (stable light after and before correction) over the cumulative DNs of stable light, was the highest in the Kanto area (63.75%), followed by the Kansai (58.51%) and Chubu areas (55.55%). B. Verification of the Saturated Light Correction Method Almost all DMSP/OLS nighttime stable light is emitted from the illumination of human settlements, suggesting a linear correlation between DNs of stable light and electric power consumption from lighting. To evaluate the new saturated light correction method described in this study, a regression analysis was performed between cumulative DNs of the stable light image and statistic data of electric power consumption from lighting in electric power supply areas of Japan. The conventional CRE correction method and our new method were then compared. CRE is a method to correct the saturated DNs of the nighttime stable light image using CREs obtained from the nonsaturated DNs [16]. Fig. 8 shows the relationship between cumulative DNs of the stable light image and electric power consumption from lighting. The highest electric power consumption from lighting was in the Kanto area, and the lowest was in the Okinawa area. Before correction, the results for the Kansai, Kanto, Tohoku, and Hokkaido areas varied greatly from the regression line [Fig. 8(a)]. After correction by the conventional CRE method [Fig. 8(b)], they fit well around the regression line and even better by our new method [Fig. 8(c)]; the coefficient of determination, R2 , increased from 81.2% (CRE method) to 91.3% (new method, P < 0.05). Based on these results of images in Japan, this new saturated light correction method proves to be useful for estimating the electric power consumption from lighting and should be applied to wider areas in future studies. V. S UMMARY AND C ONCLUSION A new saturation light estimation method for DMSP/OLS stable light imagery was developed and evaluated by comparing the corrected stable light image for 1999 with the radiance calibration image for 1996–1997 and by using statistical data of the electric power consumption from lighting in major areas of Japan. To correct the saturation light of the stable light image for 1999 using the radiance calibration image for 1996–1997, we first compared the generation method of the stable light image and radiance calibration image; the same MVC method was used in the process of both nightlight images. We thus investigated the relationship between the DNs of the stable
Fig. 8. Relationship between cumulative DNs of the stable light image 1 Okinawa, 2 Sikoku, 3 and electric power consumption from lighting ( 4 Chugoku, 5 Kyusyu, 6 Chubu). (a) Before saturated light corHokuriku, 2 −6 rection (R = 0.70, P = 4.5 × 10 < 0.05, dotted line means 95% confidence intervals). (b) After saturated light correction by the CRE method (R2 = 0.81, P = 2.6 × 10−6 < 0.05). (c) After saturated light correction by the new method described in Fig. 4(R2 = 0.91, P = 1.7 × 10−6 < 0.05).
light image and radiance calibration image in major areas of Japan. The comparison between the nonsaturated part of the stable light image and the radiance calibration image showed a strong linear correlation (R2 = 92.73%) between the DNs of both images. DNs of the radiance calibration image were adjusted to the stable light image using the regression function, and then the adjusted radiance calibration image was used to estimate the saturated DNs areas of the stable light image. To evaluate this new method, we first compared the stable light image before and after correcting for saturation effects in the Kanto, Kansai, and Chubu areas. A clear intensity distribution of the stable light image in the saturated areas was obtained after the correction. Regression analysis was then performed between statistical data of the electric power consumption from
LETU et al.: SATURATED LIGHT CORRECTION METHOD
lighting and cumulative DNs of the stable light image corrected by conventional method and our new method. As stable light by human activities almost always comprises light emitted from urban areas and streets, we estimated electric power consumption by lighting from the stable light image by correcting for saturated light using our new method. A high correlation was found between cumulative DNs and electric power consumption from lighting; this correlation was stronger after applying the new correction method. Thus, this new method proves to be very efficient for correction of DMSP/OLS stable light images. In future research, the new saturated light correction method should be applied to stable light images from the NOAANGDC server. To do so, the following two conditions must be met. • Nighttime light intensity of the stable light image in the saturated area does not change during the observation time of the stable light image and radiance calibration image. • There is a strong relationship between the radiance calibration image and the DNs of the stable light image in nonsaturated areas. This topic was not addressed in this study, but will be investigated in the near future. ACKNOWLEDGMENT The authors thank the DMSP program office of SIDaB of Japan for providing DMSP/OLS data and NOAA/NGDC for providing radiance calibration images. R EFERENCES [1] P. Mayaux, H. Eva, J. Gallego, A. H. Strahler, M. Herold, S. Agrawal, S. Naumov, E. E. De Miranda, C. M. Bella, C. Ordoyne, Y. Kopin, and P. S. Roy, “Validation of the global land cover 2000 map,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 7, pp. 1728–1739, Jul. 2006. [2] M. Herold, C. Woodcock, A. Gregprio, P. Mayaux, A. Elward, L. Latham, and J. Schmullius, “A joint initiative for harmonization and validation of land cover datasets,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 7, pp. 1719–1727, Jul. 2006. [3] M. Herold, C. E. Woodcock, T. R. Loveland, J. Townshend, M. Brady, and C. Schmullius, “Land cover observations as part of a global earth observation system of systems (GEOSS): Progress, activities, and prospects,” IEEE Syst., vol. 2, no. 3, pp. 414–423, Sep. 2008. [4] L. M. See and S. Fritz, “A method to compare and improve land cover datasets: Application to the GLC-2000 and MODIS land cover products,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 7, pp. 1740–1746, Jul. 2006. [5] P. W. Rosenkranz, “Retrieval of temperature and moisture profiles from AMSU-A and AMSU-B measurements,” IEEE Trans. Geosci. Remote Sens., vol. 39, no. 11, pp. 2429–2435, Nov. 2001. [6] P. Yang, L. Zhang, G. Hong, S. L. Nasiri, B. A. Baum, H.-L. Huang, M. D. King, and S. Platnick, “Differences between Collection 004 and 005 MODIS ice cloud optical/microphysical products and their impact on radiative forcing simulations,” IEEE Trans. Geosci. Remote Sens., vol. 45, no. 9, pp. 2886–2899, Sep. 2007. [7] F. Karbou, C. Prigent, L. Eymard, and J. R. Pardo, “Microwave land emissivity calculations using AMSU measurements,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 5, pp. 948–959, May 2005. [8] C. Surussavadee and D. H. Staelin, “Comparison of AMSU millimeterwave satellite observations, MM5/TBSCAT predicted radiances, and electromagnetic models for hydrometeors,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 10, pp. 2667–2678, Oct. 2006. [9] C. D. Elvidge, K. E. Baugh, E. A. Kihn, H. W. Kroehl, E. R. Davis, and C. Davis, “Relation between satellite observed visible-near infrared emission, population, and energy consumption,” Int. J. Remote Sens., vol. 18, no. 6, pp. 1373–1379, 1997.
395
[10] C. D. Elvidge, M. L. Imhoff, K. E. Baugh, V. R. Hobson, I. Nelson, and J. B. Dietz, “Nighttime light of the world: 1994–1995,” ISPRS J. Photogramm. Remote Sens., vol. 56, no. 2, pp. 81–99, 2001. [11] C. N. H. Doll, J. P. Muller, and C. D. Elvidge, “Nighttime imagery as a tool for globe mapping of socioeconomic parameters and greenhouse gas emissions,” Ambio, vol. 29, no. 3, pp. 157–162, May 2000. [12] S. Amaral, C. Camara, A. M. V. Monteiro, J. A. Quintanilha, and C. D. Elvidge, “Estimating population and energy consumption in Brazilian Amazonia using DMSP nighttime satellite data,” Comput. Environ. Urban Syst., vol. 29, no. 2, pp. 179–195, 2005. [13] M. Hara, S. Okada, H. Yagi, T. Moriyama, K. Shigehar, and Y. Sugimori, “Developing and evaluation of the noise reduction filter for the time-series satellite images,” (in Japanese), J. Photogramm. Remote Sens., vol. 42, no. 5, pp. 48–59, 2003. [14] H. Letu, M. Hara, S. Okada, H. Yagi, K. Kotake, K. Naoki, and F. Nisio, “Extraction of stable light from DMSP/OLS night-time image,” (in Japanese), J. Adv. Mar. Sci. Technol., vol. 14, pp. 21–28, 2008. [15] M. R. Raupach, P. J. Rayner, and M. Paget, “Regional variations in spatial structure of nightlights, population density and fossil-fuel CO2 emissions,” Energy Policy, vol. 38, no. 9, pp. 4756–4764, Sep. 2010. [16] H. Letu, M. Hara, H. Yagi, K. Naoki, G. Tana, F. Nisio, and S. Okada, “Estimating energy consumption from nighttime DMSP/OLS imagery after correcting for saturation effects,” Int. J. Remote Sens., vol. 31, no. 16, pp. 4443–4458, 2010. [17] C. D. Elvidge, K. E. Baugh, V. R. Hobson, J. B. Dietz, T. Bland, P. C. Sutton, and H. W. Kroehl, “Radiance calibration of DMSP-OLS lowlight imaging data of human settlements,” Remote Sens. Environ., vol. 68, no. 1, pp. 77–88, 1999. [18] K. Baugh, C. Elvidge, T. Ghosh, and D. Ziskin, “Development of a 2009 stable lights product using DMSP-OLS data,” in Proc. 30th Asia-Pacific Adv. Netw. Meeting, 2010, pp. 114–130. [19] D. Ziskin, K. Baugh, F. Chi Hsu, T. Ghosh, and C. Elvidge, “Methods used for the 2006 radiance lights,” in Proc. 30th Asia-Pacific Adv. Netw. Meeting, 2010, pp. 131–142. [20] T. R. K. Chand, K. V. S. Badarinath, C. D. Elvidge, and B. T. Tuttle, “Spatial characterization of electrical power consumption patterns over India using temporal DMSP-OLS night-time satellite data,” Int. J. Remote Sens., vol. 30, no. 3, pp. 647–661, 2009. [21] C. D. Elvidge, K. E. Baugh, E. A. Kihn, H. W. Kroehl, and E. R. Davis, “Mapping city lights with nighttime data from the DMSP operational linescan system,” Photogramm. Eng. Remote Sens., vol. 63, no. 6, pp. 727–734, Jun. 1997. [22] T. A. Croft, “Burning waste gas in oil fields,” Nature, vol. 245, pp. 375– 376, Oct. 1973. [23] T. A. Croft, “Nighttime images of the earth from space,” Sci. Amer., vol. 239, pp. 86–98, 1978. [24] M. L. Imhoff, W. T. Lawrence, C. D. Elvidge, T. Paul, E. Levine, M. Prevalsky, and V. Brown, “Using nighttime DMSP/OLS images of city lights to estimate the impact of urban land use on soil resources in the U.S.,” Remote Sens. Environ., vol. 59, no. 1, pp. 105–117, Jan. 1997. [25] M. L. Imhoff, W. T. Lawrence, D. C. Stutzer, and C. D. Elvidge, “Technique for using composite DMSP/OLS “city lights” satellite data to accurately map urban areas,” Remote Sens. Environ., vol. 61, no. 3, pp. 361–370, Sep. 1997. [26] M. Henderson, E. T. P. YehGong, C. D. Elvidge, and K. Baugh, “Validation of urban boundaries derived from global nighttime satellite imagery,” Int. J. Remote Sens., vol. 24, no. 3, pp. 595–609, 2003. [27] P. C. Sutton, “A scale-adjusted measure of “urban sprawl” using nighttime satellite imagery,” Remote Sens. Environ., vol. 86, no. 3, pp. 353–369, Aug. 2003. [28] P. C. Sutton, M. J. Taylor, and C. D. Elvidge, “Using DMSP OLS imagery to characterize urban populations in developed and developing countries,” in Remote Sensing of Urban and Suburban Areas, T. Rashed and C. Jürgens, Eds. Berlin, Germany: Springer-Verlag, 2010, pp. 329–348. [29] S. V. Nghiem, D. Balk, E. Rodriguez, G. Neumanna, A. Sorichetta, C. Small, and C. D. Elvidge, “Observations of urban and suburban environments with global satellite scatterometer data,” ISPRS. J. Photogramm. Remote Sens., vol. 64, no. 4, pp. 367–380, Jul. 2009. [30] C. Small, F. Pozzi, and C. D. Elvidge, “Spatial analysis of global urban extent from DMSP-OLS nighttime lights,” Remote Sens. Environ., vol. 96, no. 3/4, pp. 277–291, Jun. 2005. [31] K. P. Gallo, C. D. Elvidge, L. Yang, and B. C. Reed, “Trends in nighttime city lights and vegetation indices associated with urbanization within the conterminous USA,” Int. J. Remote Sens., vol. 25, no. 10, pp. 2003–2007, 2003.
396
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
[32] C. Milesi, C. D. Elvidge, R. R. Nemani, and S. W. Runnings, “Assessing the impact of urban land development on net primary productivity in the southeastern United States,” Remote Sens. Environ., vol. 86, no. 3, pp. 273–432, Aug. 2003. [33] R. Welch, “Monitoring urban population and energy utilization patterns from satellite data,” Remote Sens. Environ., vol. 9, no. 1, pp. 1–9, 1980. [34] R. Welch and S. Zupko, “Urbanized area energy patterns from DMSP data,” Photogramm. Eng. Remote Sens., vol. 46, no. 2, pp. 201–207, 1980. [35] T. W. Owen, K. P. Gallo, C. D. Elvidge, and K. E. Baugh, “Using DMSPOLS light frequency data to categorize urban environments associated with US climate observing stations,” Int. J. Remote Sens., vol. 19, no. 17, pp. 3451–3456, 1998. [36] T. Ichinose, “Monitoring the Asian economic activities from nighttime satellite data,” (in Japanese), in Proc. Nat. Inst. Environ. Studies Exhib. Symp., Tokyo, Japan, Jul. 19, 2001.
Husi Letu received the B.S. and M.S. degrees in geography from Inner Mongolia Normal University, Hohhot, China, in 1999 and 2002, and the Ph.D. degree in geosciences and remote sensing from the Center for Environmental Remote Sensing (CEReS), Chiba University, Chiba, Japan, in 2010. He is currently a Post Doctor in the Research and Information Center, Tokai University, Tokyo, Japan. His research interests include remote sensing applications, image processing, and atmospheric remote sensing.
Masanao Hara received the B.S. degree from the Shibaura Institute of Technology, Tokyo, Japan, in 1970, and Ph.D. degree from Chiba University, Chiba, Japan, in 2004. He is currently the President of VisionTech Inc., Tsukuba City, Japan, also a Co-researcher of the Center for Environmental Remote Sensing (CEReS), Chiba University, and a Visiting Professor in Udayana University, Bali, Indonesia. His research interests include remote sensing image processing and analysis to extract spatial information from time series and multitemporal imagery observed by an Earth Observation Satellite. He is a member of the board of directors of AMSTEC, SICE, RSSJ, and JSPRS.
Gegen Tana received the Bachelor’s degree in electronic information engineering from the Central University for Nationalities, Beijing, China, in 2004, and the Master’s degree from the Graduate School of Science and Technology, Chiba University, Chiba, Japan, in 2008, where she is currently working toward the Ph.D. degree in the Department of Earth Sciences. Her research interests are global/regional land cover map and wetlands mapping.
Fumihiko Nishio received the B.S. degree in geophysics from Tohoku University, Sendai, Japan, in 1970, and the M.S. and Ph.D. degrees in geophysics from Hokkaido University, Hokkaido, Japan, in 1972 and 1982, respectively. He joined the Japanese Antarctic Research Expedition (JARE) and the National Institute of Polar Research, in 1974. He has conducted research in various aspects of ice sheet dynamics, meteorite concentration mechanism, and radio echo sounding in Antarctica. He was a Visiting Scientist with the Scott Polar Research Institute, Cambridge, U.K., in 1984. He was leading the Japanese satellite program with the ERS-1 and JERS-1 SAR research for ice sheet and sea ice in Antarctica, in 1987. He was a Professor of Hokkaido University of Education in 1991 and started a satellite program of sea ice in Okhotsk Sea with the National Space Development Agency of Japan (NASDA, currently JAXA). Since 2000, he has been engaged in remote sensing of the cryosphere in the polar regions as a Professor with the Center for Environmental Remote Sensing, Chiba University, Chiba, Japan. He also led the JARE-43 expedition in 2001–2003 and the remote sensing of cryosphere in polar regions. His current research interests include the detection of global climate change from satellite data of the cryosphere in the polar regions and air-sea-ice interactions. Dr. Nishio is a member of the International Glaciological Society, the Remote Sensing Society of Japan, the Meteorological Society of Japan, and the Japanese Society of Snow and Ice.
389
A Saturated Light Correction Method for DMSP/OLS Nighttime Satellite Imagery Husi Letu, Masanao Hara, Gegen Tana, and Fumihiko Nishio
Abstract—Several studies have clarified that electric power consumption can be estimated from the Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) stable light imagery. As digital numbers (DNs) of stable light images are often saturated in the center of city areas, we developed a saturated light correction method for the DMSP/OLS stable light image using the nighttime radiance calibration image of the DMSP/OLS. The comparison between the nonsaturated part of the stable light image for 1999 and the radiance calibration image for 1996–1997 in major areas of Japan showed a strong linear correlation (R2 = 92.73) between the DNs of both images. Saturated DNs of the stable light image could therefore be corrected based on the correlation equation between the two images. To evaluate the new saturated light correction method, a regression analysis is performed between statistic data of electric power consumption from lighting and the cumulative DNs of the stable light image before and after correcting for the saturation effects by the new method, in comparison to the conventional method, which is, the cubic regression equation method. The results show a stronger improvement in the determination coefficient with the new saturated light correction method (R2 = 0.91, P = 1.7 · 10−6 < 0.05) than with the conventional method (R2 = 0.81, P = 2.6 · 10−6 < 0.05) from the initial correlation with the uncorrected data (R2 = 0.70, P = 4.5 · 10−6 < 0.05). The new method proves therefore to be very efficient for saturated light correction. Index Terms—Defense Meteorological Satellite (DMSP), Operational Linescan System (OLS), calibration image, saturated light, stable light image.
Program radiance
I. I NTRODUCTION
S
ATELLITES can uniformly observe a wide area in a short time and can monitor spatial land surface information and atmospheric environment at increasingly finer resolutions [1]–[8]. Among these satellites, the Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) has a unique capability for monitoring nighttime light present at the earth’s surface which includes human settlements (stable light), clouds affected by moonlight, forest fires, fishing boats and gas flares [9]. Stable light is one of the most important comManuscript received November 2, 2010; revised February 6, 2011, April 18, 2011, July 1, 2011, and August 29, 2011; accepted October 23, 2011. Date of publication December 27, 2011; date of current version January 20, 2012. H. Letu is with the Research and Information Center, Tokai University, Tokyo 151-0063, Japan (e-mail: [email protected]). M. Hara is with the VTI Research Institute, VisionTech Inc., Tsukuba City 305-0045, Japan (e-mail: [email protected]). G. Tana is with the Graduate School of Science, Chiba University, Chiba 263-8522, Japan (e-mail: [email protected]). F. Nishio is with the Centre for Environmental Remote Sensing (CEReS), Chiba University, Chiba 263-8522, Japan (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TGRS.2011.2178031
ponents for monitoring electric power consumption and economic activity, factors that rely mainly on statistical data. However, some countries and regions lack statistical data or have insufficiently accurate data on electric power consumption. Previous studies [10]–[15] have reported a good correlation between electric power consumption and digital numbers (DNs) of the DMSP/OLS stable light image. DNs of the stable light image are saturated in the center of city areas where nighttime light is very strong [16]; however, this saturation of DNs is not found in the nighttime radiance calibration DMSP/OLS images. Elvidge et al. [17] estimated electric power consumption from the radiance calibration image extracted from DMSP/OLS low-light nighttime imagery for 1996–1997 and pointed out that radiance calibration images can accurately estimate electric power consumption. Ziskin et al. [18] developed the new radiance calibration image for 2006. Although longterm radiance calibration images are not currently available, time-series stable light images can be obtained since 1992 from the National Oceanic and Atmosphere Administration (NOAA)/National Geophysical Data Center (NGDC) (http:// www.ngdc.noaa.gov/dmsp/download.html). Baugh et al. [19] developed a recent stable light product for 2009. For the monitoring of long-term changes of electric power consumption from nighttime images, the correction of the saturated DNs of the stable light images is needed. Using stable light images for 1993–2002, Chand et al. [20] analyzed the correlation between increased nighttime light and electric power consumption to evaluate electric power consumption patterns over India. However, they did not correct for the saturated DNs of the stable light images. Letu et al. [16] therefore developed a saturated light correction method using the cubic regression equation (CRE) for the correction of the DMSP/OLS-VIS stable light image of 1999. The CRE is obtained from the DNs of nonsaturated pixels by a four-dimensional simultaneous equation based on the least-square method. This method can correct the profile of DNs of saturated light but cannot estimate the saturated DNs at the image level. In this study, we developed a saturated light correction method of the stable light image using the radiance calibration nighttime image to estimate the electric power consumption. We calculated a regression function from the stable light image for 1999 and the radiance calibration images for 1996–1997. We assumed that nighttime light intensity did not change in the saturated DN areas of the stable light images during 1996–1999 and corrected the saturated DNs of the stable light image from the radiance calibration image. The cumulative DNs of the stable light image before and after the saturation light were estimated in the Kanto, Kansai, Chubu areas of Japan. To evaluate
0196-2892/$26.00 © 2011 IEEE
390
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
TABLE I S TATISTIC DATA OF THE E LECTRIC P OWER C ONSUMPTION F ROM L IGHTING FOR 1999 (×106 kwh)
this new method, regression analyses were performed between cumulative DNs of the stable light image and electric power consumption from lighting. II. S ATELLITE AND DATA C OLLECTION The DMSP/OLS is operated by the United States Department of Defense and acquires broadband visible near-infrared (VNIR) image data. With the OLS-VNIR data, it is possible to detect clouds illuminated by moonlight, lights from cities, towns, and industrial sites, gas flares, and ephemeral events such as fires and lighting-illuminated clouds [21]. The high contrast between lighted and unlighted areas and the sensor’s spatial resolution makes it a useful tool to identify areas of intense human activity [22], [23]. Each DMSP satellite has a 101-min period, which is sun-synchronous near the circular polar orbit with an altitude of 830 km above the surface of the earth. The OLS sensor collects images across a 3000-km swath, providing global coverage twice per day. The combination of day/night and dawn/dusk satellites allows monitoring of global information such as clouds and city lights every 6 h (NOAA/NGDC homepage: http://www.ngdc.noaa.gov/dmsp/). DMSP/OLSVIS 1999 data provided by the Satellite Image Data Base (SIDaB) of the Ministry of Agriculture, Forestry and Fisheries Agriculture Information Resources System (AGROPEDIA) of Japan (http://rms1.agsearch.agropedia.affrc.go.jp/sidab/indexja.html) were used to map the stable light image. The SIDaB converts 6-bit (0–63) OLS-VIS data distributed by the National Oceanic and Atmospheric Administration/National Geophysical Data Center (NOAA/NGDC) into 8-bit (0–255) data using the following function: DN8 bit = DN6 bit × 255/63. The primary mission of the OLS sensor is to observe the nighttime moon-lit cloud cover for global meteorological forecasting for the Air Force. It uses a photomultiplier tube at least four orders of magnitude more sensitive than those used in any other satellite system. The DMSP/OLS sensor monitors almost all the artificial nighttime light from urban areas [24]–[35]. Therefore, DNs of the OLS-VIS nighttime image in the center of city areas are easily saturated depending on the sensitivity of the sensor. Elvidge et al. [17] created a radiance calibration nighttime light image using DMSP/OLS low-light imaging data. This image was made using 28 nights of data for the period 1996–1997, downloaded from the NOAA/NGDC. Statistic data of the electric power consumption from lighting for 1999 was obtained from the Japan Business History Institute (http://www.jbhi.or.jp/toukei.html) and used for evaluating the saturated light correction method of the stable light images. Table I shows the electric power consumption from lighting
Fig. 1.
Electric power consumption area in Japan.
in the study areas. In this study, the spatial distribution of the nighttime stable light in Japan was divided according to the main electric power supply areas (Fig. 1). III. S ATURATED L IGHT C ORRECTION M ETHOD A. Generation of Stable Light Image and Radiance Calibration Image Fig. 2 shows the DNs of the stable light image and radiance calibration image for the Kanto area, Japan. DNs of both images were sorted and plotted in a gradually increasing order. Saturated DNs can be seen with the data from the stable light image, but not with the data from the radiance calibration image. To correct for the saturated DNs of the stable light image using the radiance calibration image, we compared the generation methods between the stable light image for 1999 and the radiance calibration image for 1996–1997 (Fig. 3). The procedure to generate the stable light image is as follows [14]. 1) A 36-piece composite image is created using the 10-day maximum value composition (MVC) method from DMSP/OLS time-series imagery for 1999. The 10-day MVC method extracts the maximum value of the DNs from the 10-day multitemporal layer satellite image. 2) Light from clouds illuminated by moonlight, gas flares, and ephemeral events such as fires and solar glare is
LETU et al.: SATURATED LIGHT CORRECTION METHOD
391
removed from the 36-piece composite image using a noise reduction filter (NRF). 3) The stable light image is extracted from the 36-piece time-series composite image using NRF. NRF finds periodic components from time-series satellite data, performs the correction of the deficit data, and removes abnormal values. In this study, NRF was used to extract the artificial nighttime light from DMSP data after removing clouds and other disturbance noise. Equation (1) expresses the function of the conventional NRF method. The NRF method calculates the estimated DNs (ft ) and the periodic components from DNs of the time-series 10-day MVC imagery. Detailed algorithms and processes for the image composite have been previously described by Hara et al. [13] N 2πkl 2πkl t +C2l+1 cos t C2l sin ft = C0 +C1 t+ M M l=1 (1)
Fig. 2. (a) DNs of stable light image for the year 1999 and (b) radiance calibration image for the period 1996–1997 in Kanto, Japan.
Fig. 3. areas.
Stable light image (a) and radiance calibration image (b) of the study
where C0 is the direct current component and is obtained from the least-square method; C1 , C2l , and C2l+1 are the coefficients; kl is the frequency k of the periodic component l (i.e., 1, 2, 3, 4, 6, or 12 months); N is the amount of data; M is the period; and t expresses the days (1–365 days). Conventionally, the direct current component (C0 ) means the stable light image. The amplitude is the 1-year cycle component noise in the direct current component except for the stable light image [6]. Thus, (2), a modification of (1), can be used for calculating the amplitude of the periodic components. The stable light image used in this study was generated by subtracting the amplitude of the 1-year cycle component from the direct current component, i.e., N 2πkl t + θl αl sin ft = C0 + C1 t + M l=2 C2l+1 2 + C2 θl = tan−1 (2) αl = C2l 2l+1 C2l where αl is the amplitude of the periodic component and θl is the phase of the periodic component. The basic procedure to generate the radiance calibration image is as follows [16]. 1) After eliminating clouds using thermal infrared (TIR) band data, the geolocation of the light is identified. For this step, an automatic light detection algorithm was developed to calculate light occurring from the time series cloud-free images. 2) Cloud-free compositing is performed within two overlapping gain ranges: high and low. The high and low gain composite images are filtered to eliminate many ephemeral events such as fires or lightning, which occur only once during light detection. 3) High and low gain cloud-free composites are combined and averaged. The radiance calibrated average DNs from each image are weighted by the total number of light detections. NRF was used to remove noise for generation of the stable light image. The methods for visually selecting a TIR data
392
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
Fig. 5. Relationship between DNs of the stable light image and the radiance calibration image.
Fig. 4. Electric power consumption from lighting during 1996–1999.
threshold are applied, and composition of the high- and lowgain images is filtered to eliminate the noise for processing the radiance calibration image. Therefore, noise components such as clouds affected by moonlight, forest fire, fishing boats, and gas flares can finally be eliminated from both images, and the MVC method can be used on both nighttime images. B. Correlation Between Stable Light Image and Radiance Calibration Image Stable light image for 1999 and radiance calibration image for 1996–1997 were generated by the same MVC method. To correct the saturated DN areas of the stable light image for 1999 using radiance calibration image for 1996–1997, we need to confirm that there was very little change in light intensity during 1996–1999. Fig. 4 shows the electric power consumption in each area of Japan during 1996–1999; these data were obtained from the Japan Business History Institute. The source of the nighttime light by human settlements was the electric power consumption from lighting. We can see that this light changes very little (< 9.5%) during 1996–1999. Ichinose et al. [36] previously investigated the intensity of the nighttime city light using the population density data of East Asia during 1992–2002. They reported that nighttime light intensity changed very little in the center of the city area. Thus, it is assumed that the DNs of the stable light image did not change in the saturated DN areas of Japan during 1996–1999. A regression analysis of DNs between the stable light image and radiance calibration image was therefore performed after plotting the DNs of both images in a gradually increasing order. Fig. 5 shows the relationship between DNs of the stable light image and radiance calibration image in the Kanto, Kansai, and Chubu areas. A linear relationship was found between DNs of the two images in each area for DNs of the radiance calibration image lower than 20. DNs of the stable light image were almost saturated when DNs of the radiance calibration image were larger than 20. Equation (3) is a regression function of the DNs of the nonsaturated part of stable and radiance calibrated imagery (Fig. 5). It was applied to adjust the DNs of the radiance calibration image to the stable light image for saturation light correction Ysta = 17Xrad − 102
(3)
Fig. 6.
Flowchart of the new saturated light correction method.
where Y is DNs of the stable light image, and X is DNs of the radiance calibration image. The stable light image for 1999 is 8-bit data downloaded from SIDaB. In the case of extracting the stable light image using 6-bit data from NOAA/NGDC, regression function (3) changes to Ysta = 63/255(17Xrad − 102). C. Saturated Light Correction Method Fig. 6 shows the flowchart of the saturated light correction method. Radiance calibration image for 1996–1997 was used for correcting the saturated DNs of the DMSP/OLS stable light image for 1999. We analyzed the radiance calibrated DNs in the saturated zone of the stable lights and applied regression equation (3) to the radiance calibration image for calculation of the corrected values. Details of data processing by this method comprise the following steps. 1) The regression equation (3) is derived from the nonsaturated part of the relationship between stable light image and radiance calibration image.
LETU et al.: SATURATED LIGHT CORRECTION METHOD
Fig. 7.
393
Comparison of the stable light image and the radiance calibration image in the Kanto, Kansai, and Chubu areas.
TABLE II C UMULATIVE DNs OF THE E LECTRIC P OWER S UPPLY A REA OF JAPAN
2) The DNs of the radiance calibration image is adjusted to the stable light image using the calculated regression equation (3). 3) The saturated DN area is extracted from the stable light image to generate the saturation area mask. 4) The DNs in the mask areas are extracted from the corrected radiance calibration image generated in step 2. 5) The saturated DNs of the stable light image are corrected with that of the extracted radiance calibration image.
IV. R ESULTS AND D ISCUSSION A. Comparison of the Stable Light Image After Correction for Saturation Effect The combined saturated area of the 1999 stable light image for the Kanto, Kansai, and Chubu areas is 87.03% of the total saturated area of Japan. This implies that the Kanto, Kansai, and Chubu areas are the main source of the nighttime light for all of Japan. Fig. 7 shows the radiance calibration image, original stable light image, and stable light image after correction in
394
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
the Kanto, Kansai, and Chubu areas. DNs of the radiance calibration image in the center of city areas were higher than those in the surrounding areas. In the stable light image, DNs were almost saturated in the center of city areas, and these saturated DN areas corresponded with the high DN area in the radiance calibration image. In the corrected stable light image, light intensity was clearly distributed in the center of city areas. Table II shows the cumulative DNs in each area of Japan. The cumulative DNs increased after the saturation effects were removed. The ratio of the cumulative DNs, the difference (stable light after and before correction) over the cumulative DNs of stable light, was the highest in the Kanto area (63.75%), followed by the Kansai (58.51%) and Chubu areas (55.55%). B. Verification of the Saturated Light Correction Method Almost all DMSP/OLS nighttime stable light is emitted from the illumination of human settlements, suggesting a linear correlation between DNs of stable light and electric power consumption from lighting. To evaluate the new saturated light correction method described in this study, a regression analysis was performed between cumulative DNs of the stable light image and statistic data of electric power consumption from lighting in electric power supply areas of Japan. The conventional CRE correction method and our new method were then compared. CRE is a method to correct the saturated DNs of the nighttime stable light image using CREs obtained from the nonsaturated DNs [16]. Fig. 8 shows the relationship between cumulative DNs of the stable light image and electric power consumption from lighting. The highest electric power consumption from lighting was in the Kanto area, and the lowest was in the Okinawa area. Before correction, the results for the Kansai, Kanto, Tohoku, and Hokkaido areas varied greatly from the regression line [Fig. 8(a)]. After correction by the conventional CRE method [Fig. 8(b)], they fit well around the regression line and even better by our new method [Fig. 8(c)]; the coefficient of determination, R2 , increased from 81.2% (CRE method) to 91.3% (new method, P < 0.05). Based on these results of images in Japan, this new saturated light correction method proves to be useful for estimating the electric power consumption from lighting and should be applied to wider areas in future studies. V. S UMMARY AND C ONCLUSION A new saturation light estimation method for DMSP/OLS stable light imagery was developed and evaluated by comparing the corrected stable light image for 1999 with the radiance calibration image for 1996–1997 and by using statistical data of the electric power consumption from lighting in major areas of Japan. To correct the saturation light of the stable light image for 1999 using the radiance calibration image for 1996–1997, we first compared the generation method of the stable light image and radiance calibration image; the same MVC method was used in the process of both nightlight images. We thus investigated the relationship between the DNs of the stable
Fig. 8. Relationship between cumulative DNs of the stable light image 1 Okinawa, 2 Sikoku, 3 and electric power consumption from lighting ( 4 Chugoku, 5 Kyusyu, 6 Chubu). (a) Before saturated light corHokuriku, 2 −6 rection (R = 0.70, P = 4.5 × 10 < 0.05, dotted line means 95% confidence intervals). (b) After saturated light correction by the CRE method (R2 = 0.81, P = 2.6 × 10−6 < 0.05). (c) After saturated light correction by the new method described in Fig. 4(R2 = 0.91, P = 1.7 × 10−6 < 0.05).
light image and radiance calibration image in major areas of Japan. The comparison between the nonsaturated part of the stable light image and the radiance calibration image showed a strong linear correlation (R2 = 92.73%) between the DNs of both images. DNs of the radiance calibration image were adjusted to the stable light image using the regression function, and then the adjusted radiance calibration image was used to estimate the saturated DNs areas of the stable light image. To evaluate this new method, we first compared the stable light image before and after correcting for saturation effects in the Kanto, Kansai, and Chubu areas. A clear intensity distribution of the stable light image in the saturated areas was obtained after the correction. Regression analysis was then performed between statistical data of the electric power consumption from
LETU et al.: SATURATED LIGHT CORRECTION METHOD
lighting and cumulative DNs of the stable light image corrected by conventional method and our new method. As stable light by human activities almost always comprises light emitted from urban areas and streets, we estimated electric power consumption by lighting from the stable light image by correcting for saturated light using our new method. A high correlation was found between cumulative DNs and electric power consumption from lighting; this correlation was stronger after applying the new correction method. Thus, this new method proves to be very efficient for correction of DMSP/OLS stable light images. In future research, the new saturated light correction method should be applied to stable light images from the NOAANGDC server. To do so, the following two conditions must be met. • Nighttime light intensity of the stable light image in the saturated area does not change during the observation time of the stable light image and radiance calibration image. • There is a strong relationship between the radiance calibration image and the DNs of the stable light image in nonsaturated areas. This topic was not addressed in this study, but will be investigated in the near future. ACKNOWLEDGMENT The authors thank the DMSP program office of SIDaB of Japan for providing DMSP/OLS data and NOAA/NGDC for providing radiance calibration images. R EFERENCES [1] P. Mayaux, H. Eva, J. Gallego, A. H. Strahler, M. Herold, S. Agrawal, S. Naumov, E. E. De Miranda, C. M. Bella, C. Ordoyne, Y. Kopin, and P. S. Roy, “Validation of the global land cover 2000 map,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 7, pp. 1728–1739, Jul. 2006. [2] M. Herold, C. Woodcock, A. Gregprio, P. Mayaux, A. Elward, L. Latham, and J. Schmullius, “A joint initiative for harmonization and validation of land cover datasets,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 7, pp. 1719–1727, Jul. 2006. [3] M. Herold, C. E. Woodcock, T. R. Loveland, J. Townshend, M. Brady, and C. Schmullius, “Land cover observations as part of a global earth observation system of systems (GEOSS): Progress, activities, and prospects,” IEEE Syst., vol. 2, no. 3, pp. 414–423, Sep. 2008. [4] L. M. See and S. Fritz, “A method to compare and improve land cover datasets: Application to the GLC-2000 and MODIS land cover products,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 7, pp. 1740–1746, Jul. 2006. [5] P. W. Rosenkranz, “Retrieval of temperature and moisture profiles from AMSU-A and AMSU-B measurements,” IEEE Trans. Geosci. Remote Sens., vol. 39, no. 11, pp. 2429–2435, Nov. 2001. [6] P. Yang, L. Zhang, G. Hong, S. L. Nasiri, B. A. Baum, H.-L. Huang, M. D. King, and S. Platnick, “Differences between Collection 004 and 005 MODIS ice cloud optical/microphysical products and their impact on radiative forcing simulations,” IEEE Trans. Geosci. Remote Sens., vol. 45, no. 9, pp. 2886–2899, Sep. 2007. [7] F. Karbou, C. Prigent, L. Eymard, and J. R. Pardo, “Microwave land emissivity calculations using AMSU measurements,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 5, pp. 948–959, May 2005. [8] C. Surussavadee and D. H. Staelin, “Comparison of AMSU millimeterwave satellite observations, MM5/TBSCAT predicted radiances, and electromagnetic models for hydrometeors,” IEEE Trans. Geosci. Remote Sens., vol. 44, no. 10, pp. 2667–2678, Oct. 2006. [9] C. D. Elvidge, K. E. Baugh, E. A. Kihn, H. W. Kroehl, E. R. Davis, and C. Davis, “Relation between satellite observed visible-near infrared emission, population, and energy consumption,” Int. J. Remote Sens., vol. 18, no. 6, pp. 1373–1379, 1997.
395
[10] C. D. Elvidge, M. L. Imhoff, K. E. Baugh, V. R. Hobson, I. Nelson, and J. B. Dietz, “Nighttime light of the world: 1994–1995,” ISPRS J. Photogramm. Remote Sens., vol. 56, no. 2, pp. 81–99, 2001. [11] C. N. H. Doll, J. P. Muller, and C. D. Elvidge, “Nighttime imagery as a tool for globe mapping of socioeconomic parameters and greenhouse gas emissions,” Ambio, vol. 29, no. 3, pp. 157–162, May 2000. [12] S. Amaral, C. Camara, A. M. V. Monteiro, J. A. Quintanilha, and C. D. Elvidge, “Estimating population and energy consumption in Brazilian Amazonia using DMSP nighttime satellite data,” Comput. Environ. Urban Syst., vol. 29, no. 2, pp. 179–195, 2005. [13] M. Hara, S. Okada, H. Yagi, T. Moriyama, K. Shigehar, and Y. Sugimori, “Developing and evaluation of the noise reduction filter for the time-series satellite images,” (in Japanese), J. Photogramm. Remote Sens., vol. 42, no. 5, pp. 48–59, 2003. [14] H. Letu, M. Hara, S. Okada, H. Yagi, K. Kotake, K. Naoki, and F. Nisio, “Extraction of stable light from DMSP/OLS night-time image,” (in Japanese), J. Adv. Mar. Sci. Technol., vol. 14, pp. 21–28, 2008. [15] M. R. Raupach, P. J. Rayner, and M. Paget, “Regional variations in spatial structure of nightlights, population density and fossil-fuel CO2 emissions,” Energy Policy, vol. 38, no. 9, pp. 4756–4764, Sep. 2010. [16] H. Letu, M. Hara, H. Yagi, K. Naoki, G. Tana, F. Nisio, and S. Okada, “Estimating energy consumption from nighttime DMSP/OLS imagery after correcting for saturation effects,” Int. J. Remote Sens., vol. 31, no. 16, pp. 4443–4458, 2010. [17] C. D. Elvidge, K. E. Baugh, V. R. Hobson, J. B. Dietz, T. Bland, P. C. Sutton, and H. W. Kroehl, “Radiance calibration of DMSP-OLS lowlight imaging data of human settlements,” Remote Sens. Environ., vol. 68, no. 1, pp. 77–88, 1999. [18] K. Baugh, C. Elvidge, T. Ghosh, and D. Ziskin, “Development of a 2009 stable lights product using DMSP-OLS data,” in Proc. 30th Asia-Pacific Adv. Netw. Meeting, 2010, pp. 114–130. [19] D. Ziskin, K. Baugh, F. Chi Hsu, T. Ghosh, and C. Elvidge, “Methods used for the 2006 radiance lights,” in Proc. 30th Asia-Pacific Adv. Netw. Meeting, 2010, pp. 131–142. [20] T. R. K. Chand, K. V. S. Badarinath, C. D. Elvidge, and B. T. Tuttle, “Spatial characterization of electrical power consumption patterns over India using temporal DMSP-OLS night-time satellite data,” Int. J. Remote Sens., vol. 30, no. 3, pp. 647–661, 2009. [21] C. D. Elvidge, K. E. Baugh, E. A. Kihn, H. W. Kroehl, and E. R. Davis, “Mapping city lights with nighttime data from the DMSP operational linescan system,” Photogramm. Eng. Remote Sens., vol. 63, no. 6, pp. 727–734, Jun. 1997. [22] T. A. Croft, “Burning waste gas in oil fields,” Nature, vol. 245, pp. 375– 376, Oct. 1973. [23] T. A. Croft, “Nighttime images of the earth from space,” Sci. Amer., vol. 239, pp. 86–98, 1978. [24] M. L. Imhoff, W. T. Lawrence, C. D. Elvidge, T. Paul, E. Levine, M. Prevalsky, and V. Brown, “Using nighttime DMSP/OLS images of city lights to estimate the impact of urban land use on soil resources in the U.S.,” Remote Sens. Environ., vol. 59, no. 1, pp. 105–117, Jan. 1997. [25] M. L. Imhoff, W. T. Lawrence, D. C. Stutzer, and C. D. Elvidge, “Technique for using composite DMSP/OLS “city lights” satellite data to accurately map urban areas,” Remote Sens. Environ., vol. 61, no. 3, pp. 361–370, Sep. 1997. [26] M. Henderson, E. T. P. YehGong, C. D. Elvidge, and K. Baugh, “Validation of urban boundaries derived from global nighttime satellite imagery,” Int. J. Remote Sens., vol. 24, no. 3, pp. 595–609, 2003. [27] P. C. Sutton, “A scale-adjusted measure of “urban sprawl” using nighttime satellite imagery,” Remote Sens. Environ., vol. 86, no. 3, pp. 353–369, Aug. 2003. [28] P. C. Sutton, M. J. Taylor, and C. D. Elvidge, “Using DMSP OLS imagery to characterize urban populations in developed and developing countries,” in Remote Sensing of Urban and Suburban Areas, T. Rashed and C. Jürgens, Eds. Berlin, Germany: Springer-Verlag, 2010, pp. 329–348. [29] S. V. Nghiem, D. Balk, E. Rodriguez, G. Neumanna, A. Sorichetta, C. Small, and C. D. Elvidge, “Observations of urban and suburban environments with global satellite scatterometer data,” ISPRS. J. Photogramm. Remote Sens., vol. 64, no. 4, pp. 367–380, Jul. 2009. [30] C. Small, F. Pozzi, and C. D. Elvidge, “Spatial analysis of global urban extent from DMSP-OLS nighttime lights,” Remote Sens. Environ., vol. 96, no. 3/4, pp. 277–291, Jun. 2005. [31] K. P. Gallo, C. D. Elvidge, L. Yang, and B. C. Reed, “Trends in nighttime city lights and vegetation indices associated with urbanization within the conterminous USA,” Int. J. Remote Sens., vol. 25, no. 10, pp. 2003–2007, 2003.
396
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 50, NO. 2, FEBRUARY 2012
[32] C. Milesi, C. D. Elvidge, R. R. Nemani, and S. W. Runnings, “Assessing the impact of urban land development on net primary productivity in the southeastern United States,” Remote Sens. Environ., vol. 86, no. 3, pp. 273–432, Aug. 2003. [33] R. Welch, “Monitoring urban population and energy utilization patterns from satellite data,” Remote Sens. Environ., vol. 9, no. 1, pp. 1–9, 1980. [34] R. Welch and S. Zupko, “Urbanized area energy patterns from DMSP data,” Photogramm. Eng. Remote Sens., vol. 46, no. 2, pp. 201–207, 1980. [35] T. W. Owen, K. P. Gallo, C. D. Elvidge, and K. E. Baugh, “Using DMSPOLS light frequency data to categorize urban environments associated with US climate observing stations,” Int. J. Remote Sens., vol. 19, no. 17, pp. 3451–3456, 1998. [36] T. Ichinose, “Monitoring the Asian economic activities from nighttime satellite data,” (in Japanese), in Proc. Nat. Inst. Environ. Studies Exhib. Symp., Tokyo, Japan, Jul. 19, 2001.
Husi Letu received the B.S. and M.S. degrees in geography from Inner Mongolia Normal University, Hohhot, China, in 1999 and 2002, and the Ph.D. degree in geosciences and remote sensing from the Center for Environmental Remote Sensing (CEReS), Chiba University, Chiba, Japan, in 2010. He is currently a Post Doctor in the Research and Information Center, Tokai University, Tokyo, Japan. His research interests include remote sensing applications, image processing, and atmospheric remote sensing.
Masanao Hara received the B.S. degree from the Shibaura Institute of Technology, Tokyo, Japan, in 1970, and Ph.D. degree from Chiba University, Chiba, Japan, in 2004. He is currently the President of VisionTech Inc., Tsukuba City, Japan, also a Co-researcher of the Center for Environmental Remote Sensing (CEReS), Chiba University, and a Visiting Professor in Udayana University, Bali, Indonesia. His research interests include remote sensing image processing and analysis to extract spatial information from time series and multitemporal imagery observed by an Earth Observation Satellite. He is a member of the board of directors of AMSTEC, SICE, RSSJ, and JSPRS.
Gegen Tana received the Bachelor’s degree in electronic information engineering from the Central University for Nationalities, Beijing, China, in 2004, and the Master’s degree from the Graduate School of Science and Technology, Chiba University, Chiba, Japan, in 2008, where she is currently working toward the Ph.D. degree in the Department of Earth Sciences. Her research interests are global/regional land cover map and wetlands mapping.
Fumihiko Nishio received the B.S. degree in geophysics from Tohoku University, Sendai, Japan, in 1970, and the M.S. and Ph.D. degrees in geophysics from Hokkaido University, Hokkaido, Japan, in 1972 and 1982, respectively. He joined the Japanese Antarctic Research Expedition (JARE) and the National Institute of Polar Research, in 1974. He has conducted research in various aspects of ice sheet dynamics, meteorite concentration mechanism, and radio echo sounding in Antarctica. He was a Visiting Scientist with the Scott Polar Research Institute, Cambridge, U.K., in 1984. He was leading the Japanese satellite program with the ERS-1 and JERS-1 SAR research for ice sheet and sea ice in Antarctica, in 1987. He was a Professor of Hokkaido University of Education in 1991 and started a satellite program of sea ice in Okhotsk Sea with the National Space Development Agency of Japan (NASDA, currently JAXA). Since 2000, he has been engaged in remote sensing of the cryosphere in the polar regions as a Professor with the Center for Environmental Remote Sensing, Chiba University, Chiba, Japan. He also led the JARE-43 expedition in 2001–2003 and the remote sensing of cryosphere in polar regions. His current research interests include the detection of global climate change from satellite data of the cryosphere in the polar regions and air-sea-ice interactions. Dr. Nishio is a member of the International Glaciological Society, the Remote Sensing Society of Japan, the Meteorological Society of Japan, and the Japanese Society of Snow and Ice.
